Delft University of Technology

Spin Qubits in Silicon and Germanium

Lawrie, W.I.L.

DOI
10.4233/uuid:97c4ea24-9672-4e0b-b7a5-e3a48258c871
Publication date
2022

Citation (APA)
Lawrie, W. I. L. (2022). Spin Qubits in Silicon and Germanium. [Dissertation (TU Delft), Delft University of
Technology]. https://doi.org/10.4233/uuid:97c4ea24-9672-4e0b-b7a5-e3a48258c871

Important note
To cite this publication, please use the final published version (if applicable).
Please check the document version above.

Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent
of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Takedown policy
Please contact us and provide details if you believe this document breaches copyrights.
We will remove access to the work immediately and investigate your claim.

This work is downloaded from Delft University of Technology.

For technical reasons the number of authors shown on this cover page is limited to a maximum of 10.
S PIN Q UBITS IN S ILICON AND G ERMANIUM
S PIN Q UBITS IN S ILICON AND G ERMANIUM

Dissertation

for the purpose of obtaining the degree of doctor

at Delft University of Technology,
by the authority of the Rector Magnificus, prof. dr. ir. T.H.J.J. van der Hagen,
chair of the Board for Doctorates,
to be defended publicly on
30 June 2022 at time 12:30 o’clock

by

William Iain Leonard LAWRIE

Master of Science in Physics,

The University of Melbourne, Australia
born in Perth, Australia.
This dissertation has been approved by the promotor.

Composition of the doctoral committee:

Rector Magnificus, chairperson
Dr. ir. M. Veldhorst, Delft University of Technology, promotor
Dr. G. Scappucci, Delft University of Technology, copromotor

Independent members:
Prof. dr. ir. L. DiCarlo, Delft University of Technology
Prof. dr. E. P. A. M. Bakkers, Eindhoven University of Technology
Prof. dr. G. Katsaros, Institute of Science and Technology Austria
Dr. E. Greplova, Delft University of Technology
Dr. M. Wimmer, Technical University of Delft, reserve member

Other members:
Dr. J. Helsen, Centrum Wiskunde & Informatica

Keywords: quantum computing, quantum dots, spin qubits, semiconductors,

germanium, silicon, electrons, holes, qubit arrays
Printed by: Gildeprint

Copyright © 2022 by W. I. L. Lawrie

Casimir PhD Series, Delft-Leiden 2022-18
ISBN 978-90-8593-529-2
Cover design by William Lawrie
An electronic version of this dissertation is available at
http://repository.tudelft.nl/.
 C ONTENTS

Summary xi
Samenvatting xv
1 Introduction 1
1.1 Primer on Quantum Information . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Quantum Bits, Superposition and Entanglement . . . . . . . . . . 2
1.1.2 Quantum gates. . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 What makes a good quantum bit? . . . . . . . . . . . . . . . . . . . . . . 4
1.3 How does this work fit in? . . . . . . . . . . . . . . . . . . . . . . . . . . 6
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Electron and Hole Spin Qubits in Semiconductor Quantum Dots 11
2.1 Semiconductors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Semiconductor Quantum Dots . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Coloumb Oscillations . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Charge Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 Virtual Gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Spin Qubits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Zeeman Splitting. . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Lamor Precission . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.3 Qubit State Manipulation . . . . . . . . . . . . . . . . . . . . . . 20
2.3.4 Exchange Interaction . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3.5 Two-Qubit Exchange Gates . . . . . . . . . . . . . . . . . . . . . 23
2.3.6 Spin Readout. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Electron spin qubits in Silicon quantum dots . . . . . . . . . . . . . . . . 25
2.4.1 Valley Splitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.2 Electron Spin Resonance. . . . . . . . . . . . . . . . . . . . . . . 28
2.5 Hole spin qubits in Germanium quantum dots . . . . . . . . . . . . . . . 28
2.5.1 The Valence Band . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.2 Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5.3 Lifting Spin Degeneracy . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.4 Hyperfine Interaction . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5.5 Electric Dipole Spin Resonance . . . . . . . . . . . . . . . . . . . 34
2.6 Qualifying a Quantum System . . . . . . . . . . . . . . . . . . . . . . . 34
2.6.1 Spin Relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6.2 Dephasing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6.3 Gate Fidelity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

v
vi C ONTENTS

3 Quantum Dot Arrays in Silicon and Germanium 45

3.1 Mobility and Stack by Platform . . . . . . . . . . . . . . . . . . . . . . . 47
3.2 Integration Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Singly Occupied Quantum Dot Arrays. . . . . . . . . . . . . . . . . . . . 52
3.4 Cross Capacitance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5 Future Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.7 SiMOS Fabrication Improvements . . . . . . . . . . . . . . . . . . . . . 55
3.7.1 Gate Anneal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.7.2 ALD Window Etch . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.8 Extended Fabrication Recipe . . . . . . . . . . . . . . . . . . . . . . . . 56
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Silicon 65
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4 Tuneable Coupling and Isolation of Single Electrons in Silicon Quantum Dots 67
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2.1 Charge Sensing and Readout. . . . . . . . . . . . . . . . . . . . . 68
4.2.2 Tunnel Couple Control. . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5 Universal quantum logic in hot silicon qubits 75
5.1 Hot Qubits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.2 Large-scale quantum computing in SiMOS . . . . . . . . . . . . . . . . . 76
5.3 Single Qubit Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.4 Two-Qubit Quantum Logic at 1.1 K . . . . . . . . . . . . . . . . . . . . . 79
5.5 Temperature Effects on Quantum Coherence . . . . . . . . . . . . . . . . 81
5.6 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.6.1 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.6.2 Single-qubit randomized benchmarking . . . . . . . . . . . . . . . 82
5.6.3 Two-qubit randomized benchmarking . . . . . . . . . . . . . . . . 83
5.7 Extended Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.8 Fit of the full exchange spectrum . . . . . . . . . . . . . . . . . . . . . . 86
5.9 Noise model and noise fitting . . . . . . . . . . . . . . . . . . . . . . . . 86
5.10 Temperature dependence of the dephasing time . . . . . . . . . . . . . . 87
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6 Design and integration of single-qubit rotations and two-qubit gates in sili-
con above one Kelvin 91
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2.1 Silicon quantum dot device . . . . . . . . . . . . . . . . . . . . . 92
6.2.2 Qubit characterization . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.3 Hot, fast, and high-fidelity CPHASE gates . . . . . . . . . . . . . . 94
C ONTENTS vii

6.2.4 Hot, fast and high-fidelity SWAP operation. . . . . . . . . . . . . . 95

6.2.5 Hot two-qubit gate performance . . . . . . . . . . . . . . . . . . . 98
6.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.4 Reconstruction of the spin state probabilities . . . . . . . . . . . . . . . . 98
6.5 Temperature effects on readout and initialization. . . . . . . . . . . . . . 99
6.6 Exchange spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.7 Two-qubit gate simulations . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.7.1 Physical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.7.2 Adiabatic and diabatic CPHASE gates . . . . . . . . . . . . . . . . 101
6.7.3 SWAP gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Germanium 107
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7 A Single-Hole Spin Qubit 109
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.2.1 Single hole quantum dot and Pauli spin blockade . . . . . . . . . . 110
7.2.2 Qubit operation . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.1 Fabrication process . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.2 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.3.3 Measurement details. . . . . . . . . . . . . . . . . . . . . . . . . 116
7.3.4 Virtual gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.4 Extended Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
8 Spin Relaxation Benchmarks and Individual Qubit Addressability for Holes
in Quantum Dots 123
8.1 Device and Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.2 Spin Blockade for Hole Fillings . . . . . . . . . . . . . . . . . . . . . . . 127
8.3 Spin Relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.4 Qubit Addressability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
8.6 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.6.1 Tunnel Rate Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.6.2 Relative Lever Arm . . . . . . . . . . . . . . . . . . . . . . . . . . 132
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
9 A Four-Qubit Germanium Quantum Processor 137
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
9.2 A Two-by-Two Qubit Quantum Dot Array . . . . . . . . . . . . . . . . . . 138
9.3 Single Qubit Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
9.4 One, Two and Threefold Conditional Rotations . . . . . . . . . . . . . . . 139
9.5 Conditional Phase Gates Between Neighbouring Qubit Pairs . . . . . . . . 142
9.6 Demonstration of a Four Qubit Greenberger-Horne-Zeilinger State. . . . . 145
viii C ONTENTS

9.7 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

9.7.1 Device fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.7.2 Device screening . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.7.3 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.7.4 Virtual gate matrices . . . . . . . . . . . . . . . . . . . . . . . . . 147
9.7.5 Latched readout mechanism. . . . . . . . . . . . . . . . . . . . . 149
9.7.6 Conditional rotation gates . . . . . . . . . . . . . . . . . . . . . . 150
9.7.7 Shift in Rabi frequency due to exchange . . . . . . . . . . . . . . . 151
9.7.8 Fitting of the two-qubit decay data. . . . . . . . . . . . . . . . . . 152
9.7.9 Specification of the CROT and CZ gate times. . . . . . . . . . . . . 152
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

10 Simultaneous driving of semiconductor spin qubits at the fault-tolerant thresh-

old 161
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.2.1 Single Qubit Benchmarking . . . . . . . . . . . . . . . . . . . . . 163
10.2.2 N -copy Randomized Benchmarking . . . . . . . . . . . . . . . . . 165
10.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
10.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
10.4.1 Device Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . 167
10.4.2 Generation of the Single-Qubit Clifford Set . . . . . . . . . . . . . 168
10.4.3 Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
10.4.4 Simultaneous Elementary Gate Tuning . . . . . . . . . . . . . . . 168
10.5 Initialization and Readout . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.6 Clifford Gate Set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
10.7 Two-Copy Randomized Benchmarking . . . . . . . . . . . . . . . . . . . 171
10.7.1 Structure of the matrix decays . . . . . . . . . . . . . . . . . . . . 172
10.7.2 Different types of two-copy RB . . . . . . . . . . . . . . . . . . . . 172
10.7.3 Even same-system (ES) two-copy RB. . . . . . . . . . . . . . . . . 173
10.7.4 Odd same-system (OS) two-copy RB . . . . . . . . . . . . . . . . . 174
10.7.5 Even different-system (DS) two-copy RB . . . . . . . . . . . . . . . 174
10.8 Two-copy RB in practice . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.8.1 Lower Bound on Average Fidelity . . . . . . . . . . . . . . . . . . 175
10.8.2 Symmetric and Anti-symmetric input states for two-copy RB . . . . 176
10.9 Direct comparison of two-copy benchmarking and two-qubit simultane-
ous randomized benchmarking . . . . . . . . . . . . . . . . . . . . . . . 177
10.9.1 Simulation of the Clifford gate superoperator . . . . . . . . . . . . 178
10.9.2 Simulation of the clifford experiments . . . . . . . . . . . . . . . . 179
10.9.3 Comparison between the results . . . . . . . . . . . . . . . . . . . 179
10.10 Summary of N -copy benchmarking data. . . . . . . . . . . . . . . . . . 182
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
C ONTENTS ix

11 Quantum Coherence in Germanium Hole Spin Qubits 185

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
11.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
11.2.1 Quantum Coherences . . . . . . . . . . . . . . . . . . . . . . . . 187
11.2.2 Probing the 73 Ge Larmor Procession . . . . . . . . . . . . . . . . . 189
11.2.3 Spin Dephasing due to Hyperfine Noise . . . . . . . . . . . . . . . 190
11.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
11.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
11.4.1 Device Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . 193
11.4.2 Fitting Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
11.4.3 Readout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
11.5 Extended Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
11.5.1 Hole Spin Qubit g-Factors . . . . . . . . . . . . . . . . . . . . . . 194
11.5.2 Observation of 73 Ge Lamor Precession at High Field . . . . . . . . . 194
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
12 Conclusion & Outlook 197
12.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
12.2 What next? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
12.2.1 In Germanium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
12.2.2 In Silicon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
12.2.3 In General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Acknowledgements 203
Curriculum Vitæ 207
List of Publications 209
 S UMMARY
Quantum computers based on semiconductor quantum dots are proving promising contenders
for large scale quantum information processing. In particular, group IV based semiconductor
hosts containing an abundance of nuclear spin-zero isotopes have made considerable headway
into fulfilling the requirements of a universal quantum computer. Silicon (Si) and germanium
(Ge) are two elements that have played important roles in the history of classical computing,
and are now poised to do the same in the future of quantum computation. In this thesis, we
advance efforts in developing a quantum computer in both Si, using electron spins in Si-metal-
oxide-semiconductor (SiMOS) hosted quantum dots, and hole spins in planar germanium quan-
tum wells (Ge/SiGe).
In chapter 3, we develop fabrication strategies for quantum dot arrays in SiMOS, Si/SiGe and
Ge/SiGe platforms. In each case, a fabrication recipe is developed with considerations accord-
ing to each of the different platforms challenges and opportunities. Using these recipes, we create
quantum dot arrays in silicon and germanium, depletable down to single charge occupancy in all
cases. Additionally, we study the cross-capacitance of the electrostatic gates on adjacent quantum
dots, finding that SiMOS requires the smallest virtual corrections to gate voltages.
Using these fabrication recipes, we then focus on the SiMOS platform, creating devices for quan-
tum information processing. In chapter 4, we show that tunnel couple control is possible between
quantum dots in silicon MOS, using dedicated barrier gates. We demonstrate tuneable tunnel
rates and tunnnel couplings via a barrier gate, showing tunnel rates can be suppressed to below
1 Hz, and tunnel couplings are tuneable between 3 GHz and 13 GHz, overcoming a longstanding
challenge for electrons in silicon-MOS quantum dots.
Operation of spin qubits at higher temperatures will permit strategies such as on chip integration
of electronics, which may be key for scaling up spin qubits. In chapter 5, we demonstrate a uni-
versal set of quantum gates in a pair of SiMOS electron spin qubits, operated at 1.1 K. We find high
single qubit fidelities for both qubits of around 99%, and find that dephasing times are not limited
by temperature in the measured range, between 300 mK and 1.25 K. Furthermore in chapter 6,
we demonstrate a portfolio of two qubit gates in the same device, and construct composite gates
that should overcome limitations due to the finite Zeeman energy differences of the qubits. These
results mark a milestone in the scale-up of spin qubits.
The results using electron spin qubits in SiMOS show that spin qubits can be operated at higher
temperatures, where the cooling power of dilution refrigerators is much higher. SiMOS as a mate-
rial host however still faces several issues in terms of scale-up. First, the interface disorder results
in the unwanted formation of quantum dots, as well as the strong localization of their wavefunc-
tions, making reliable qubit interactions non-trivial. This is reflected by the relatively lower mobil-
ity, and higher percolation densities when compared to other platforms as measured in chapter 3.
We therefore switch gears to hole spins in Ge/SiGe, advancing the platform through a number of
experiments.
Building on the success of the platform, we utilize the fabrication strategies developed in chap-
ter 3 to fabricate quantum dot arrays in Ge/SiGe, which we deplete to single hole occupancy, and
demonstrate the first single hole spin qubit in chapter 7. Exploiting the native spin orbit interac-
tion, we achieve coherent, fast, all-electrical driving via electric dipole spin resonance (EDSR) of a
single hole spin qubit, and characterize it’s spin dephasing and relaxation times. We also find that

xi
xii S UMMARY

the g -factor is highly dependent on the electric field environment.

By tuning the coupling between the hole spin qubit and its charge reservoirs, we find that spin re-
laxation times can be significantly improved, and we measure single hole spin qubit spin lifetimes
of 32 ms. We also find that g -factors are influenced significantly by neighbouring gate potentials,
providing both an opportunity for ease of qubit addressability, as well as raising the challenge of
enhanced charge noise coupling in the platform.
In an effort to reduce charge noise, we fabricated wafers on quantum wells grown deeper below
the heterostructure surface. This innovation led to reduced charge noise levels in hole spin qubits
and facilitated the work in chapter 9. Here we present a two-by-two array of hole spin qubits,
demonstrating universal quantum logic via single qubit manipulations, as well two-qubit CZ and
CROT gates between all neighbouring qubit pairs. We further show the expected eight-fold split-
ting of a qubit resonance frequency as a result of four exchange coupled spins, allowing for both a
native Toffoli-like three qubit gate, and even a four qubit controlled rotation gate. To showcase the
control of the qubit array, we generate a four-qubit Greenberger-Horne-Zeilinger state in the first
ever four qubit algorithm performed on a quantum processor hosted by semiconductor quantum
dots.
Quantum computers will require very high operational fidelities, and will likely require the simul-
taneous operation of many qubits in close proximity, leading to qubit cross-talk. In chapter 10,
we examine single qubit gate fidelities using Randomized Benchmarking protocols. We find that
single hole spin qubits, when operated individually can exhibit native gate fidelities of 99.990(1)%,
marking the state of the art for single qubit operational accuracy in spin qubit platforms. Further-
more, we investigate how cross-talk when operating multiple qubits simultaneously affects single
qubit gate fidelities. To do this, we develop a new randomized benchmarking protocol called "N -
Copy Benchmarking", where the same Random Clifford sequences are applied simultaneously to
each of the N qubit in the experiment. This permits extraction of a single qubit gate fidelity that ap-
proximates gate fidelity in the absence of qubit-qubit interactions, for example, residual exchange
coupling. We find two-copy native gate fidelities above 99.905(8)%, and four-copy fidelities of
99.34(4)%, indicating that classical crosstalk is a major consideration when considering quantum
computing scale up.
Having explicitly investigated hole spin qubit relaxation times and gate fidelities, we also inves-
tigated their dephasing times to better understand the processes that limit coherence times. We
found that as a function of magnetic field, a T 2∗ ∝ 1/B proportionality described the behaviour
of spin dephasing between 0.1 T and 1 T magnetic field strengths. This trend is consistent with
the expectation that charge noise coupled to the spin eigenstates via the spin-orbit interaction is
the limiting noise process for hole spin qubits in Ge/SiGe. We found also that by implementing
refocussing pulses, coherence could be extended as high as 504 µs. To do this, we implemented
Carl-Purcell-Meiboom-Gill pulse sequences, of varying numbers of refocussing π-pulses. These
pulse sequences exhibit a characteristic filter fucntion based on this number, that effectively re-
duces low frequency noise effects on qubit coherence, but also enhances certain frequencies noise.
We exploit this feature to investigate the sensitivity of hole spins to the 73 Ge nuclei present in the
quantum well with an abundance of approximately 7.3%. By modelling the traces resulting from
CPMG sequences, we extract a hyperfine coupling strength for the hole spins, and estimate a de-
phasing time limited by hyperfine interaction. We find this time to be within the same order of
magnitude as the charge noise limited dephasing time, at an in-plane magnetic field of 0.25 T.
From this we conclude that, in light of the rapidity of advancements in planar germanium mate-
rial quality, isotopic purification will likely become necessary in the near future in order to extend
quantum coherence times of hole spins in Ge/SiGe.
S UMMARY xiii

William I. L. Lawrie
 S AMENVATTING
Kwantum computers gebaseerd op halfgeleider kwantumdots blijken veelbelovende kandidaten
voor kwantum-informatie-verwerking op grote schaal. In het bijzonder hebben op groep IV half-
geleider gebaseerde materialen met een overvloed van nucleare spin-nul isotopen een aanzien-
lijke vooruitgang doorgemaakt richting het vervullen van de eisen van een universele kwantum
computer. Silicium (Si) en germanium (Ge) zijn twee elementen die belangrijke rollen hebben
gespeeld in de geschiedenis van klassieke computers, en zijn nu klaar om hetzelfde te doen in
de toekomst van kwantum computers. In deze thesis maken we vordering met het ontwikkelen
van een kwantum computer in zowel Si, gebruikmakende van elektron spins in Si-metaal-oxide-
halfgeleider (SiMOS) genestelde kwantum dots, en gat spins in planaire germanium kwantum put-
ten (Ge/SiGe

In hoofdstuk 3 ontwikkelen we fabricage strategieën voor kwantum dot rasters in de SiMOS,

Si/SiGe en Ge/SiGe platforms. In elke casus is een fabricage recept ontwikkeld waarbij we de uit-
dagingen en kansen van de verschillende platformen overwegen. Gebruikmakende van deze re-
cepten maken we kwantum dot rasters in silicium en germanium, in ieder geval uitputbaar tot
enkele ladings bezetting. Aanvullend daarop bestuderen we de kruis-capaciteit van de elektro-
statische electrodes op aangrenzende kwantum dots, waarbij we bevinden dat SiMOS de kleinste
virtuele correcties op de spanningen nodig heeft.
Gebruikmakende van deze fabricage recepten leggen we dan de focus op het SiMOS platform,
waarin we chips voor kwantum informatie verwerking maken. In hoofdstuk 4, laten we zien dat
controle over de tunnel koppeling mogelijk is tussen kwantumdots in SiMOS, gebruikmakende
van afzonderlijke barriëre elektrodes. We laten afstembare tunnel snelheden en tunnel koppe-
lingen zien via een barriëre elektrode, waarbij we laten zien dat tunnel snelheden tot onder 1 Hz
gedempt kunnen worden, en dat tunnel koppelingen afstembaar tussen 3 GHz en 13 GHz zijn,
waarbij een lang-staande uitdaging voor electronen in SiMOS kwantum dots overkomen wordt.

Het functioneren van spin kwantumbits op hogere temperaturen zal strategieën zoals op-
chip-integratie van electronica mogelijk maken, hetgeen de sleutel kan zijn voor het opschalen
van spin kwantumbits. In hoofdstuk 5 demonstreren we een universele set van kwantum poorten
in een paar SiMOS elektron spin kwantumbits, werkend op 1.1 K. We vinden hoge enkel kwan-
tumbit betrouwbaarheden van rond de 99% en vinden dat defaseringstijden in het gemeten be-
reik (tussen 300 mK en 1.25 K) niet gelimiteerd worden door temperatuur. Bovendien laten we in
hoofdstuk 6 een portfolio van twee kwantumbit poorten op dezelfde chip zien, waarbij we samen-
gestelde poorten opbouwen die de limieten kunnen overkomen die gesteld worden als gevolg van
het eindige Zeeman energie verschil tussen de kwantumbits. Deze resultaten tonen een belang-
rijke mijlpaal aan in het opschalen van spin kwantumbits.

De resultaten van elektron spin kwantumbits in SiMOS laten zien dat spin kwantumbits op
hogere temperaturen kunnen functioneren, waar het koelvermogen van verdunnings-koelkasten
veel groter is. Als materiaal wordt SiMOS echter nog steeds geconfronteerd met meerder pro-
blemen wat betreft het opschalen. Allereerst heeft de wanorde op het grensvlak als resultaat dat
er ongewenste kwantumdots gevormd worden, zowel als sterke localisatie van hun golffuncties,

xv
xvi S AMENVATTING

waardoor betrouwbare kwantumbit interacties bemoeilijkt worden. Dit blijkt uit de relatief la-
gere mobiliteit en hogere percolatiedichtheden vergeleken met andere platforms zoals gemeten
in hoofdstuk 3. Daarom veranderen we koers naar gat spins in Ge/SiGe, waarbij we het platform
door een serie experimenten vooruithelpen.
Bouwend op het ucces van het platform benutten we de fabricage strategiën die in hoofdstuk 3
ontwikkeld zijn om kwantumdot rasters in Ge/SiGe te fabriceren, welke we uitputten tot bezetting
met een enkel gat, en demonstreren de eerste enkelgats-spinkwantumbit in hoofdstuk 7. Door het
uitbuiten van de intrinsieke spin-baan interactie behalen we coherente, snelle, puur elektrische
aandrijving via elektrische dipool spin resonantie (EDSR) van een enkelgats-spinkwantumbit, en
karakteriseren we diens spin defasering en relaxatie tijden. We bevinden ook dat de g -factor sterk
afhankelijk is van de elektrische omgeving.
Door het afstemmen van de koppeling tussen de gat-spinkwantumbits en diens ladingsreservoirs
komen we erachter dat spin-relaxatietijden sterk verbeterd kunnen worden, en we meten enkelgats-
spinkwantumbit spin levensduren van 32 ms. We bevinden ook dat g -factors significant beïnvloed
worden door aangrenzende elektrode potentialen, hetgeen zowel voor een kans voor gemakkelijke
addresseerbaarheid voor kwantumbits zorgt als voor een uitdaging met betrekking tot versterkte
ladingsruis-koppeling in het platform.
In een poging tot het verminderen van ladingsruis fabriceren we wafers met kwantum putten die
dieper onder de heterostructuur gegroeid zijn. Deze inovatie leidde tot verminderde ladingsruis-
niveaus in gat-spinkwantumbits en facilliteerde het werk in hoofdstuk 9. Hier presenteren we een
twee-bij-twee raster van gat-spinkwantumbits, hetgeen universele kwantum logica laat zien via
enkele kwantumbits manipulatie, samen met twee-kwantumbit CZ en CROT poorten tussen alle
aangrenzende kwantumbit paren. Verder laten we ook de verwachte achtvoudige splitsing van
een kwantumbit resonantiefrequentie zien als resultaat van vier wisselwerkings gekoppelde spins,
hetgeen zowel een intrinsieke Toffoli-achtige drie kwnatumbit poort laat zien als zelfs een vier
kwantumbit gecontroleerde rotatie poort. Om de controle over het kwantumbitraster uit te lichten
genereren we een vier-kwantumbit Greenberger-Horne-Zeilinger staat in het eerste vier kwnatum-
bits algoritme ooit uitgevoerd op een kwantum processor genesteld in halfgeleider kwantumdots.

Kwantum computers vereisen erg hoge operationele betrouwbaarheid, en zullen waarschijn-

lijk de gelijktijdige operatie van vele kwantumbits in directie nabijheid, hetgeen leidt tot kwan-
tumbit cross-talk. In hoofdstuk 10 onderzoeken we enkel-kwantumbit poort betrouwbaarheden
met Randomized Benchmarking protocollen. We bevinden dat enkel-gat-spinkwantumbits in-
trinsieke poort betrouwbaarheden van 99.990(1)% kunnen laten zien bij individuële operatie, het-
geen de state of the art voor enkel kwantumbits operationele nauwkeurigheid in spin kwantumbit
platformen markeert. Daarbovenop zoeken we uit hoe cross-talk resulterende uit het gelijktijdig
opereren van meerder kwantumbits de enkel kwantumbit poort betrouwbaarheid beïnvloed. Om
dit te doen ontwikkelen we een nieuw randomized benchmarking protocol genaamd "N -Copy
Benchmarking", waarbij dezelfde willekeurige Random Clifford reeks gelijktijdig op elk van de N
kwantumbits in het experiment wordt toegepast. Dit stelt ons in staat om de enkel kwantumbit
poort betrouwbaarheid te bepalen die benaderd wat de betrouwebaarheid zou zijn bij het ont-
breken van kwantumbit-kwantumbit interacties zoals residuële wisselwerkingskoppeling. We be-
palen two-copy intrinsieke poort betrouwbaarheden van boven de 99.905(8)%, en four-copy be-
trouwbaarheden van 99.34(4)%, aangevend dat klassieke cross-talk een belangrijke afweging is bij
het opschalen van kwantum computers.
Nu we expliciet gat kwantumbit spin relaxatie tijden en poort betrouwbaarheden onderzocht heb-
ben onderzoeken we ook hun defaseer tijden om beter de processen die coherentie tijden beper-
ken te begrijpen. We vonden dat als functie van magnetisch veld een a T 2∗ ∝ 1/B evenredigheid
het gedrag van de spin defasering beschrijft tussen 0.1 T en 1 T magnetisch veldsterkte. Deze
S AMENVATTING xvii

trend is consistent met de verwachting dat ladingsruis gekoppeld aan spin eigenstaten via spin-
baan interactie het limiterende ruis proces voor gat-spin-kwantumbits in Ge/SiGe is. We vonden
ook dat door het implementeren van herfocus pulsen coherentie verhoogd kon worden tot wel
504 µs. Om dit te kunnen doen implementeren we Carl-Purcell-Meiboom-Gill pulse reeksen van
verschillende antallen herfocus π-pulsen. Deze pulsreeksen laten een karakteristieke filter functie
gebaseerd op die aantal zien, welke effectief lage frequentie ruis effecten op kwantumbit coheren-
tie vermindert, maar ook de ruis op bepaalde frequenties versterkt. We buiten deze eigenschap uit
om de gevoeligheid van gat spinkwantumbits voor 73 Ge nuclei te onderzoeken, welke aanwezig
zijn in de kwantumput voor ongeveer 7.3%. Door het modelleren van de curves als resultaat van
CPMG reeksen leiden we een hyperfine koppelings kracht af voor de gat spins, en we benaderen
de defaserings tijd gelimiteerd door hyperfine interactie. We vinden dat deze tijd in dezelfde or-
degrootte ligt als de ladingsruis gelimiteerde defaserings tijd, bij een in-vlak magnetisch veld van
0.25 T. Hieruit concluderen we dat, in het licht van de snelheid van vooruitgang in planair germa-
nium materiaal kwaliteit, isotopisch purificeren waarschijnlijk in de nabije toekomst nodig wordt
om kwantum coherentie tijden van gat spins in Ge/SiGe te verlengen.
 Strewth
/stru:θ/

Exclamation
Used to express surprise or dismay.1

1 Oxford English Dictionary

 1
I NTRODUCTION
Classical mechanics is the physics of music. Quantum mechanics is the music of physics
Klaus Mølmer

Quantum computing promises to revolutionize the way we process information, and the complexity
of the computational tasks we can achieve. The first proposals to utilize quantum systems as a com-
putational resource came about in the early 1980’s [1]. It was soon shown that in order to simulate
a quantum system efficiently, one would need a computer that evolved according to quantum me-
chanical laws [2]. Among other important early contributions, these two statements birthed a field
of quantum information processing, spurred on by the idea that quantum computers may be able to
perform certain tasks much faster than classical ones. Throughout the 1990s, algorithms exploiting
the quantum mechanical nature of reality were developed that confirmed this belief, resulting in an
immense global effort to build a useful quantum computer.

In this section we review basic quantum information and quantum mechanical principles required
for the work in this thesis. We examine the criteria for making a good qubit, and where this work fits
into the greater field of quantum computing.

1
 2 1. I NTRODUCTION

1 1.1. P RIMER ON Q UANTUM I NFORMATION

The theory encapsulating the behaviour of quantum systems at the physical level is called quan-
tum mechanics. The theory describing how quantum information can be manipulated is called
quantum information theory. Quantum computing involves the processing of information con-
tained within a quantum system. In this chapter, we review the basics of quantum computing
utilized in this thesis.

1.1.1. Q UANTUM B ITS , S UPERPOSITION AND E NTANGLEMENT

The theories of both classical and quantum information rely on the manipulation of data. At it’s
most fundamental unit, this data takes the form of an abstract two level system called a bit if
classical, and a qubit if quantum. The reason for this distinction is because the physical properties
of a qubit give rise to additional complexities in the complete description of its state. Figure 1.1

(a) (b) |0〉

0
|Ψ〉
^
x
^

1 y
|1〉

Figure 1.1: Comparison of the properties of classical and quantum bits. The classical bit (left) can be com-
pletely described by one piece of information, the value of the variable a. The qubit’s state can be represented
by an arbitrary vector of unit magnitude |Ψ〉, on the surface of a sphere. Two complex valued variables a and b
describe the state of a single isolated qubit.

compares a description of a classical state to a quantum state. In the former case, a complete
description of the state is possible with a single variable, a, which can only take one of two possible
discrete values a ∈ {0, 1}. The quantum state, while still binary, is much richer. A qubit’s state can
be depicted as a unit vector using Dirac notation as
µ ¶
a
|Ψ〉 = a |0〉 + b |1〉 = (1.1)
b

where a, b ∈ C, and |a|2 +|b|2 = 1. When this vector points towards the bottom or top of the sphere,
we recover the classical bit states of 0 or 1 respectively. However, an infinite number of states
exist in between, reflecting the ability for a quantum state to exist in both states at once. This
phenomenon is called superposition, and constitutes one of the intrinsically quantum properties
that lead to the predicted advantages of quantum computing [3–6].
A second important and advantageous property for quantum computation is the phenomenon
of entanglement. To illustrate entanglement, let’s first consider the state space of a system of two
qubits |Ψ〉1 = α1 |0〉 + β1 |1〉 and |Ψ〉2 = α2 |0〉 + β2 |1〉. The state space of the full quantum system
is then
α1 α2
 
 α1 β2 
|Ψ〉12 = |Ψ〉1 ⊗ |Ψ〉2 = α1 α2 |00〉 + α1 β2 |01〉 + β1 α2 |10〉 + β1 β2 |11〉 =  
 β1 α2  (1.2)
β1 β2

where ’⊗’ is the tensor product operation, and we have taken the convention |ψ1 〉1 |ψ2 〉2 = |ψ1 ψ2 〉.
Notice that we have to specify four separate complex variables to describe a two-qubit quantum
1.1. P RIMER ON QUANTUM I NFORMATION 3

system, whereas a two-bit system could be described with only two real variables. In fact, as a
function of the system size N , the amount of information required to represent a quantum system 1
scales as 2N , while classical systems scale as N . One can quickly see that classical computers will
have a hard time simulating quantum systems of even modest sizes, with a system of about 250
qubits requiring more bits than there are estimated atoms in the universe [7].
To illustrate the concept of entanglement, it is useful to classify the state of a system of qubits into
two types: separable states, and inseparable states. Separable states are quantum systems that can
be expressed as the product of the individual qubit states. For example, considering the present
two-qubit system, |Ψ〉12 is already by definition a separable state, since it was defined to be the
tensor product of two individual qubit states |Ψ〉12 = |Ψ〉1 ⊗ |Ψ〉2 . Separable states are also called
product states for this reason. Inseparable states on the other hand, cannot be expressed as the
tensor product of two separate quantum systems. Consider the state |Ψ〉12 = p1 (|00〉 + |11〉). This
2
state cannot be written as the product of two individual qubit states, and is therefore inseparable.
We call a state that is inseparable, an entangled state. The physical consequence of a state being
entangled, is that the full state information of an individual qubit is not stored locally. The ability
to create quantum states with non-local properties lies at the heart of the fields of quantum infor-
mation and communication.

Another useful representation of a qubit’s state, is via its density matrix. This is particularly useful
when describing systems of noisy, or entangled quantum systems. The density matrix of a qubit
state composed of some probabilistic mixture of states |ψ j 〉 is defined as

ρ=
X
p j |ψ j 〉 〈ψ j | (1.3)
j

where the probabilities p j add up to one. If a quantum state is known, ie. we can always choose
some projective measurement that will return the state with complete certainty, then it is said to be
a pure state. If no such basis exists, such as for qubits who’s state’s have decohered, or are entangled
with other qubits, they are said to be mixed states. The density matrix of a quantum state allows
us to measure the outcome probability of some system observable  simply as p = Tr { Âρ} where
Tr {·} is the trace of a matrix.

1.1.2. Q UANTUM GATES

Operations that change the state of a systems of quantum bits are called gates. For a system of
N qubits, a quantum gate can be represented mathematically as 2N × 2N matrix. Quantum gates
must satisfy the property of unitarity, that is, for a gate U , it’s inverse must be equal to its conjugate
transpose, ie. U †U = I .
To begin, let us consider operations on a single qubit state. Single qubit operations are equiv-
alent to rotations on the Bloch sphere. It is therefore useful to define quantum gates in terms of
rotational operators around the mutually orthogonal axes of the Bloch sphere (Fig. 1.1b). Let’s
define X θ , Yθ and Zθ as rotations of angle θ around the x, y and z axes, where [8]

cos θ2 −i sin θ2
à !
Xθ = (1.4)
−i sin θ2 cos θ2

cos θ2 − sin θ2
à !
Yθ = (1.5)
sin θ2 cos θ2
à θ !
e −i 2 0
Zθ = θ . (1.6)
0 ei 2
 4 1. I NTRODUCTION

These gates unlock the full state space of a single qubit. When we set θ = π, we recover a special
1 set of quantum gates, known as the Pauli operators
µ ¶
0 1
X π = σx = (1.7)
1 0
µ ¶
0 i
Yπ = σ y = (1.8)
−i 0
µ ¶
1 0
Zπ = σz = . (1.9)
0 −1

The same principle can be applied for rotations of θ = π/2, who’s matrices are shown in Figure 1.2
These types of rotations allow the generation of superposition states.

In order to achieve universal access to a two-qubit state space, one needs a two-qubit entangling
gate [9]. Here it becomes useful to introduce different types of qubits in quantum computation.
We usually refer to a qubit whose state is changed as a target qubit, while a qubit who’s state deter-
mines whether or not action occurs on the target qubit is called a control qubit [10].

In this thesis, we work with three main two-qubit entangling gates: The controlled-NOT (CNOT),
The controlled-Z (CZ) and the SWAP. The CNOT gate is a two-qubit gate which performs a σx gate
on a target qubit, if the control qubit is in the |1〉 state, and does nothing otherwise. The controlled-
Z gate is the same, but by applying a σz gate on a target qubit. The SWAP gate exchanges the states
of the qubits.pThe SWAP gate does not actually permit universal access to the two qubit state space,
however the SW AP does. These gates are all shown in Figure 1.2.

Since universal quantum logic is possible using a single two-qubit entangling gate in addition to
single qubit rotations, it follows that each of the two-qubit gates presented above can be compiled
in terms of a combination of single and two-qubit gates. Furthermore, any arbitrary unitary opera-
tion on any number of qubits can be decomposed into only single- and two-qubit quantum gates.
This is indeed the case, however sometimes it can be more efficient to utilize a variety of quantum
gates when operating a quantum computer. This is particularly prudent since real implementa-
tions of qubits are very sensitive to their environments, and can lose their quantum information
rapidly. We now turn to the criteria that determine whether a quantum system is suitable as a
qubit candidate.

1.2. W HAT MAKES A GOOD QUANTUM BIT ?

There are many eligible quantum systems with which to build quantum computers at the hard-
ware level. This might seem overwhelming, but luckily there are some well established guide-
lines that help us assess whether a particular qubit is suitable or not. These guidelines were pub-
lished by theoretical physicist David DiVincenzo in 2000 [11], and are have served as the qualitative
benchmarks against which to compare quantum computing platforms. There are seven criteria,
though these are frequently referred to as ’5+2’ since the first five pertain to quantum computa-
tion, while the final two relate to quantum communication. However, some proposals for quan-
tum computing architectures rely on medium or long range communication elements for their
qubits [12], usually on the same chip.

• A scalable physical system with well characterized qubits

A well characterized qubit is simply a two level quantum system that is split off from other
1.2. W HAT MAKES A GOOD QUANTUM BIT ? 5

Gate Action Representation Matrix Form 1

Pauli-X/2 Xπ/2 1 1 -i
√2 -i 1

Pauli-Y/2 Yπ/2 1 1 -1
√2 1 1

Pauli-Z/2 Zπ/2 1 1 0
√2 0 i

H 1 1 1
Hadamard
√2 1 -1

1 0 0 0
Controlled-NOT 0 1 0 0
0 0 0 1
0 0 1 0

1 0 0 0
Controlled-Z 0 1 0 0
0 0 1 0
0 0 0 -1

1 0 0 0
SWAP 0 0 1 0
0 1 0 0
0 0 0 1

1 0 ... 0 0
0 1 ... 0 0
Toffoli ... ... ... ... ...
0 0 ... 0 1
0 0 ... 1 0
8x8

Figure 1.2: Summary of single- two- and three-qubit gates. Rows represent single- two- and three- qubit gates.
The action column shows examples of their action on a single qubit in the Bloch sphere. A arrow indicates an
example starting state, while a red arrow indicates the final state. Red arcs indicate the path through state space
traced by the qubit over the duration of the gate. The remaining columns show how they are represented in a
quantum circuit, and their mathematical matrix structure.

energy states by a reasonable energy. Examples of such a a system are the spin eigenstates
of a hole or electron in a magnetic field [13], the vibrational modes of ions confined to po-
tential wells [14], and the oscillations of cooper pairs across tunnel junctions between su-
perconducting islands [15]. The requirement for a scalable physical system is an extremely
complicated condition, and one that continues to baffle the quantum computing commu-
nity as the number of qubits on state-of-the-art quantum computers becomes larger. At it’s
most idealistic, scalability states that for a given platform, there exists a unit-cell such that
going from one unit cell to a large number (say 109 ) should introduce no emergent prob-
lems. In the case of classical computing, reduction in transistor size, power requirements,
 6 1. I NTRODUCTION

electronic signal control and mulitplexing contributed to the scalable nature of the transis-
1 tor, and are why billions of transistors can be controlled with around a thousand pins on
modern CPUs. Generally speaking for quantum computers, a single qubit needs several DC
electronic connections, several AC signals, and isolation of the quantum state, leading to an
unscalable number of interconnects per qubit [16]. This metric, known as "Rent’s Rule", is
a good way to disambiguate scalability claims [17].
• The ability to reliably initialize qubits into a known state
Quantum computers perform algorithms on qubits. This requires precise knowledge of
what the state of the qubit was to begin with. It is not surprising that this is a requirement,
however it can be difficult to achieve accurate state initialization. One tactic is to simply
wait for a system to relax into it’s ground state. However this time frame is often impracti-
cably long, leading to investigation into other high fidelity state initialization methods.
• Long relevant coherence times
Quantum states are delicate - any interaction with unaccounted for physical systems will
result in information loss, either due to quantum state relaxation or decoherence. As a re-
sult, the timeframes over which information can be reliably stored in a quantum computer
must be long enough to endure a given algorithm.
• A Universal set of quantum gates
Computing with quantum systems requires the ability to access and exploit it’s full state
space. A set of gates that are able to achieve this is called a universal set. Technically, due
to the infinite nature of the state space of a qubit, no finite set of quantum gates exists that
is universal. The gate set {X θ , Yθ , Zθ , CNOT} constitutes a universal gate set of infinite size,
however arbitrary qubit rotations can in practice be easily performed (see chapters 5 and 9
for example).
• Measurement capability
In order to extract information from a quantum algorithm we need to be able to measure
the state of a quantum system. Like initialization, quantum measurement fidelities are sus-
ceptible to error and must also be reliable. Measurement can be non-trivial in practice,
for example, direct measurement of the magnetic moment of an electron spin is difficult.
Measurement techniques for spin qubit systems are elaborated upon in Chapter 2.

1.3. H OW DOES THIS WORK FIT IN ?

Having seen what makes a good qubit, we are now in a position to place the work contained in
this thesis into context. We have seen that there are a multitude of challenges that quantum com-
puting faces, as the field attempts to build processors with increasing numbers of qubits. We have
also seen that there are a multitude of approaches to quantum computing, each offering different
solutions to the many challenges we face moving forward. As a result, it is a non-trivial task to
narrow down which quantum computing implementation is most suitable.

In the sub-field of spin qubits, several diverse qubit candidates exist. However, a certain class of
qubits, called semiconductor quantum dot spin qubits share a unique advantage, in that they bear
a high degree of similarity to the transistor: the physical representation of the bit, upon which
modern computers are based. The potential of bootstrapping off the the wealth of knowledge and
experience contained within the semiconductor industry is a highly attractive prospect that posi-
tions semiconductor spin qubit platforms as promising candidates for quantum computing scale
up.
1.3. H OW DOES THIS WORK FIT IN ? 7

The work in this thesis pertains to experiments on spin qubits defined in silicon and germanium
semiconductor hosts. Namely, electron spins defined in silicon quantum dots, and hole spins de- 1
fined in germanium quantum dots. A particular emphasis is placed on assessing both platforms’
suitability for quantum computing efforts, and attempting to draw conclusions about which as-
pects of each candidate may limit future progress. The structure of the thesis is as follows:
• Chapter 2 contains a theoretical background required to understand the experiments and
conclusions of the chapters that follow it, introducing semiconductor quantum dots, elec-
tron spin qubits in silicon, and hole spin qubits in planar germanium. Experimental tech-
niques and setups are also introduced here.
• Chapter 3 Introduces a fabrication strategy for producing quantum dot arrays in group IV
based platforms, and examines the advantages of electron spin qubits in silicon metal-oxide
semiconductor, as well as silicon quantum wells. This chapter also motivates the operation
of qubits at higher temperatures to facilitate cryogenic on-chip integration of electronics,
easing scale up requirements.
The next two chapters examine quantum computing using electron spins in SiMOS, with an em-
phasis on demonstration of qubit operation at high temperature.
• Chapter 4 shows the first demonstration of a tuneable tunnel-coupling between quantum
dots in silicon-MOS, adding a highly versatile new tool to operating qubits in this platform.
• Chapter 5 covers the first demonstration of universal two-qubit logic at elevated tempera-
tures, showing that high temperature operation of spin qubits is possible. We explore the
behaviour of spin dephasing times as a function of temperature, finding little dependence.
We also find that single qubit gate fidelities obtained via randomized benchmarking, can
exceed 99%, even at 1.1 K.
• Chapter 6 presents a two-qubit gate portfolio, operated at high temperatures. SWAP, CPHASE
and CROT gates are all demonstrated in the same device by utilizing adiabaticity and dia-
baticity to overcome the finite Zeeman energy difference present between the qubits.
The following chapters are devoted to experiments in planar germanium. Scaling the number
of qubits and characterization of their quality are the primary focus points.
• Chapter 7 Demonstrates the first single hole spin qubit, serving as a starting point for all fu-
ture measurements. We quantify the coherence the qubit, and demonstrate coherent con-
trol of it’s spin state.
• Chapter 8 demonstrates long spin relaxation times for single- and few-hole quantum dots,
as well as studies the effects of gate voltages on qubit g-factor. These results show that spin
relaxation will not be a limiting factor for hole spins defined in planar germanium.
• Chapter 9 builds on the work in previous chapters, by demonstrating a four-qubit germa-
nium quantum processor, defined in a 2x2 array of hole spin qubits. We show full control
of single- and two-qubit operations for nearest neighbour operations in the array, as well
as excellent control of qubit pair exchange, allowing for the execution of native 3-qubit Tof-
foli like and 4-qubit gates CCC-NOT gates. We also demonstrate fast CZ gates between all
qubit nearest neighbour pairs, and showcase our qubit control and quality by creating a
fully entangled four-qubit GHZ state, marking the state-of-the-art for semiconductor spin
qubit quantum computing efforts.
• Chapter 10 demonstrates the ability to operate up to four qubits simultaneously, while still
maintaining good control fidelity. This alleviates alleviates some of the requirements placed
on spin coherence, since simultaneous qubit operation will reduce qubit idle time. We also
report state-of-the-art single qubit control fidelties across all semiconductor quantum dot
platforms.
 8 R EFERENCES

• Chapter 11 studies the primary cause of spin decoherence in germanium hole spin qubits.
1 We find this to be primarily charge noise, but also find that noise due to the non-zero nu-
clear spin isotopes of natural germanium causes a significant decohering effect on hole spin
qubits, presenting an argument for the isotopic purification of the platform.

The final chapter concludes by discussing future research directions for silicon and germa-
nium spin qubits, clarifying the immanent hurdles of both platforms, and avenues that may open
the way to practical quantum devices.

R EFERENCES
[1] Benioff, P. The computer as a physical system: A microscopic quantum mechanical Hamil-
tonian model of computers as represented by Turing machines. Journal of Statistical Physics
22, 563–591 (1980).

[2] Feynman, R. P. Simulating physics with computers. International Journal of Theoretical

Physics 21, 467–488 (1982).

[3] Shor, P. W. Algorithms for quantum computation: Discrete logarithms and factoring. In
Proceedings - Annual IEEE Symposium on Foundations of Computer Science, FOCS, 124–134
(IEEE Computer Society, 1994).

[4] Grover, L. K. A fast quantum mechanical algorithm for database search. In Proceedings of the
Annual ACM Symposium on Theory of Computing, vol. Part F1294, 212–219 (Association for
Computing Machinery, 1996).

[5] Deutsch, D. & Jozsa, R. Rapid solution of problems by quantum computer - Deutsch and
Jozsa.pdf. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences
553–558 (1992).

[6] Zalka, C. Simulating quantum systems on a quantum computer. Proceedings of the Royal
Society A: Mathematical, Physical and Engineering Sciences 454, 313–322 (1998).

[7] Gott III, J. R. et al. A Map of the Universe. Astrophys. J. 624, 463–484 (2005).

[8] Quantum inspire knowledge base: Rotation operators. URL https://www.

quantum-inspire.com/kbase/rotation-operators/.

[9] Williams, C. P. Explorations in Quantum Computing. Texts in Computer Science

(Springer London, London, 2011), 2 edn. URL http://link.springer.com/10.1007/
978-1-84628-887-6.

[10] Nielsen, M. A. & Chuang, I. L. Quantum Computation and Quantum Information (Cambridge
University Press, 2010).

[11] DiVincenzo, D. P. The physical implementation of quantum computation. arXiv 0002077 48,
771–783 (2000).

[12] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Inf. 3, 1–10 (2017).

[13] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Physical Review A -
Atomic, Molecular, and Optical Physics 57, 120–126 (1998).
R EFERENCES 9

[14] Cirac, J. I. & Zoller, P. Quantum computations with cold trapped ions. Physical Review Letters
74, 4091–4094 (1995). 1
[15] Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Na-
ture 574, 505–510 (2019).

[16] Franke, D. P., Clarke, J. S., Vandersypen, L. M. & Veldhorst, M. Rent’s rule and extensibility in
quantum computing. Microprocessors and Microsystems 67, 1–7 (2019).

[17] Landman, B. S. & Russo, R. L. On a Pin Versus Block Relationship For Partitions of Logic
Graphs. IEEE Transactions on Computers C-20, 1469–1479 (1971).
 2
E LECTRON AND H OLE S PIN Q UBITS
IN S EMICONDUCTOR Q UANTUM
D OTS
In this section, we introduce the concepts pertaining to the implementations of quantum bits used
throughout the thesis: Spin qubits defined in semiconductor quantum dots. We begin by intuitively
introducing quantum dots in planar semiconductor platforms, and the techniques common to the
operation of semiconductor quantum dot devices, including charge sensing and virtual gates. We
continue by introducing the concept of a spin qubit, and its behaviour in stationary and oscillating
magnetic fields, the exchange interaction, how to implement single- and two-qubit gates, and dis-
cuss how to perform spin readout. Next we introduce the spin qubit systems of electrons silicon and
holes in germanium, defined in quantum dots, reviewing their respective physical considerations.
Finally we review noise sources acting on spin qubits in silicon and germanium, and experimental
methods used to quantify the performance of a quantum system.

11
 12 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

2.1. S EMICONDUCTORS
Crystalline materials in solid state physics can be classified as conductors, semiconductors, or in-
sulators. The main parameters that determine which of these categories a given material falls in to
are the bandgap E g , and the Fermi level E F . For a crystalline material without defects, no energy

2 states exist within it’s bandgap. It is defined as the difference in energy between the highest va-
lence band state, and the lowest conduction band state, representing how much energy is required
to excite a charge carrier from a valence bond, to a free carrier. For insulators, this gap is very large,
and transport through them is very difficult. Semiconductors have smaller band gaps, allowing for
excitations of electrons from valence to conduction bands. To put this in perspective, the band
gaps of the semiconductors silicon and germanium are about 1.1 eV and 0.72 eV respectively [1],
while for insulator SiO2 , depending on growth quality, it can be as high as 9.6 eV. For conductors,
the conduction and valence bands overlap, requiring no excitations to generate free carriers. This
results in favorable conduction properties. The Fermi level of a material is defined as the energy
at which the occupation probability of the next available energy state is one half. For the exam-
ples of conductors and semiconductors just described, the Fermi level lies at the intrinsic level,
which is in the exact energy center of the bandgap. However, the Fermi level can also be raised
or lowered, by introducing a greater density of allowed states via techniques such as doping [2].
Adding dopants to a semiconductor enhances their conduction properties and in extreme limits,
can cause transitions between semiconductors and metals [3].

Often, by combining semiconductors, metals, and insulators together, we can create devices that
function in practical ways. In this thesis, we focus on three semiconductor heterostructures. Sili-
con is used as a host material in both metal-oxide-semiconductor (SiMOS) as shown in figure 2.1a,
and in Si/SiGe heterostructures as shown in figure 2.1b. In the former case, electrons accumulate
at the Si-SiO2 interface when a positive voltage is applied to a metallic gate deposited above the in-
sulating oxide layer. For Si/SiGe, the band energies of the Si and SiGe layers are different, creating
a quantum well, where electrons become trapped in the silicon layer. Figure 2.1c shows a Ge/SiGe
heterostructure, where a quantum well is also formed that permits trapping of either conduction
band electrons or valence band holes. This thesis focuses on hole spins in Ge quantum wells. Here,
a negative potential is required to accumulate holes in the quantum well.

2.2. S EMICONDUCTOR Q UANTUM D OTS

In semiconductors, quantum dots are clusters of charge carriers that are spatially confined in such
a way that their energies become measurably quantized [4]. The confinement depends on the ge-
ometry of the host medium, however the work in this thesis pertains to the case where quantum
dots are defined in a planar geometry (Fig 2.2). The cross-sectional view shows charges (white cir-
cles) accumulating at a semiconductor-insulator interface, below metallic gates (dark grey). These
gates are electrostatically biased to attract charges, overcoming the repulsive energy between the
individual charges. This forms a small island of charge in the semiconductor bulk, constituting
a quantum dot. It’s important to note that the depictions of quantum dots in this chapter are
intended to cater to a classical intuition. In reality, the way that charges arrange themselves in
quantum dots is complex and for a more thorough and accurate description, consult ref. [4]. Fig-
ure 2.2c shows a simulation of the arrangement of charges within a circular quantum dot of radius
25 nm. Each geometry is two-fold degenerate due to spin, which we will come to in 2.3.

In order to confine quantum dots in practice, it is usually necessary to have several metallic
gates in order to create a well controlled potential profile. Figure 2.2 shows a typical strategy em-
2.2. S EMICONDUCTOR QUANTUM D OTS 13

(a)
VP=0 VP>0

SiO2

28
Si
2
EVB ECB EVB ECB
U U

(b)
VP=0 VP>0
Si0.7Ge0.3

28
Si

Si0.7Ge0.3
EVB ECB EVB ECB
U U

(c)
VP=0 VP<0

Si0.2Ge0.8

Ge

Si0.2Ge0.8
EVB ECB EVB ECB
U U

Figure 2.1: Summary of the semiconductor heterostructures studied in this thesis. (a) Silicon-MOS. Elec-
trons are confined to the Si-Si02 interface. In equillibrium, conduction and valence bands bend slightly due
to strain induced by the semiconductor-oxide lattice mismatch. When a positive voltage Vp is applied to a
metallic gate, electrons accumulate. (b) Si/SiGe heterostructures. Electrons accumulate in the Si quantum
well, when a negative voltage is applied. (c) Ge/SiGe heterostructures. The nature of the materials parameters
permit electrons or holes to become trapped inside a Ge quantum well. However we focus on holes in this
case. When a negative potential is applied, holes become confined in the valence band within the quantum
well.

ployed in the design of nanoscale devices, whereby charges are attracted beneath a plunger gate
(blue), and repulsed by flanking barrier gates (red).

2.2.1. C OLOUMB O SCILLATIONS

In order to load a quantum dot with charges, a nearby reservoir must be coupled to it. Reservoirs
are regions in the semiconductor bulk with a practically infinite number of charges. Figure 2.4
shows how reservoirs can be used to controllably load charges into a quantum dot. Two reservoirs
(green) contact the semiconductor bulk through windows in the insulator. They are controllably
coupled to the quantum dot via the barrier gate voltages. We can imagine an experiment where
one measures the current between the two reservoir leads, which we will call source (the left lead)
and drain (the right lead). By setting the chemical potential of the source higher than that of the
 14 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

(a) (b) (c)

Top-Down Cross-Section n = 1,2 n = 3,4 n = 5,6

Quantum Dot
n = 7,8 n = 9,10 n = 11,12

Figure 2.2: Planar semiconductor quantum dots. (a) A top down view shows darker circles representing metal-
lic gates, in an array on the insulating surface of a semiconductor wafer. These gates control the potentials that
allow for the accumulation and depletion of charges that form quantum dots. (b) The Cross-sectional view
shows charges accumulating in the semiconductor bulk underneath the metallic gates, constituting quantum
dots. (c) More realistic depiction of the first 12 wavefunctions of spin-1/2 charges distributed in a quantum dot
defined by a circular potential of radius 25 nm. Simulated using KWANT [5]

(a) (b)
VBL = 0 VP > Vgap VBR = 0 VBL < 0 VP > Vgap VBR < 0

Figure 2.3: Confinement of quantum dots. (a) A voltage is applied to the plunger gate (blue) resulting in
charges becoming confined to the interface of the semiconductor bulk and insulator. The charges are still
mostly unconfined in the lateral dimension. (b) Repulsive voltages are applied to the barrier gates (red). This
causes lateral confinement of the quantum dot, underneath the plunger gate.

drain µS > µD , we create a source-drain bias VSD = α(µS − µD ) where α is the lever arm, such that
it is favorable for current to flow [6]. The quantum dot also has a chemical potential µdot dictated
by the voltage on the plunger gate VP . Unlike the reservoir, the quantum dot has discrete allowed
energy levels, in this case dictated by the number of charges it contains. The chemical potential
of the N’th charge occupation of a quantum dot is defined as µN = µN −1 + E add . If µS ≥ µN , then
a charge will tunnel from the source onto the quantum dot such that there are N charges. Simi-
larly, if µD ≤ µN , a charge from the quantum dot will tunnel into the drain, leaving N − 1 charges.
In order for current to flow between the source-drain leads, an available quantum dot state must
exist within the bias window such that µS < µN < µD . The left panels show the situation where no
allowed charge occupation state exists within the bias window. Resultantly, no current flows for
this plunger gate voltage (red star). This situation is called "Coloumb blockade". The right panels
show the case where the plunger gate voltage has tuned the chemical potential of the quantum dot
such that µD < µN < µS , allowing current to flow, and resulting in a sharp current peak (blue star).
The sharpness of the current peaks in the lower panel due to the lifting of Coloumb blockade is a
very useful tool when sensing the motion of nearby charges, as we will now see.
2.2. S EMICONDUCTOR QUANTUM D OTS 15

2.2.2. C HARGE S ENSING

The abrupt nature of the sharp transitions between current flow and Coloumb blockade in a quan-
tum dot provides a useful tool in quantum devices, known as charge sensing. Charge sensing is
2
the ability to detect when a quantum dot changes occupancy. That is, when a charge is added or
removed from the quantum dot. It is important to know the charge occupation of a quantum dot,
since it is common practice to define a qubit as the spin of a single charge particle (more of this in
section 2.3. It is also useful for a technique called spin-to-charge conversion, used in the reading
out of a qubit’s spin state, which will be discussed later. For now, we will consider the general con-
cept of charge sensing.
Consider a system, where a quantum dot coupled to two reservoirs (like the case above) is de-
fined, close to another quantum dot coupled to a different reservoir. This is shown in figure 2.5.
We will call the former system a charge sensor. Here, we assume that current cannot flow between
the charge sensor and the quantum dot, rather they are capacitively coupled. We also assume for
ease of explanation that there is no capacitive coupling of the gates to the quantum dots, except
via their respective plunger gates. The breaking of this assumption will be explained later in the
context of ’virtual gates’. Let us assume that the charge sensor is currently tuned in such a way that
there is an allowed energy level of the sensor quantum dot inside the source-drain bias window.
This means that current is flowing between the leads. Let us also assume that the other quantum
dot has an occupation N . If we now sweep the plunger gate controlling the chemical potential of
this quantum dot, we will reach a regime where it is energetically favourable for a new charge to
tunnel into the dot, bringing the occupation to N + 1. If this happens, the capacitance of this dot
will change and therefore due to it’s capcative coupling to the sensor dot, so too will the chem-
ical potential of the charge sensor, leading to a change in current. Since the Coloumb peaks of
the quantum sensor are sharp, this allows us a sufficiently large differential signal between charge
states to deterministically fill a quantum dot to an arbitrary occupancy.

2.2.3. V IRTUAL G ATES

This principle can be extended to sensing multiple quantum dots. The bottom panels of figure 2.5
show a double-dot charge sensing configuration. Here, a second quantum dot is also capacitively
coupled to the charge sensor. Figure 2.5f is called a charge stability diagram (CSD), and shows the
charge occupation of the two quantum dots as a function of VP 1 and VP 2 . Note that the charge
transitions of one dot are not dependent on the value of the other dot. This is a result of the as-
sumption that there is no capacitive coupling between the quantum dots. In reality, there will be
some dependence of the chemical potential of a quantum dot on the gate voltages in the device.
This motivates the use of virtual gates, a practice where multiple gate voltages are changed at once
to alter a given quantum dots chemical potential, without affecting others in the system. Consider
the device in Figure 2.6a, where a double quantum dot system is defined with two plunger gates
P1 and P2 , separated by a barrier gate B12 . The charge state of the system is detected by a charge
sensor. We want to define virtual gates ṼP 1 , ṼP 2 , ṼB 12 , as a function of the real gate voltages VP 1 ,
VP 2 , VB 12 such that a change in the virtual gate voltage does not affect the position of the charge
transition lines of the dots unintentionally. Therefore we define changes in the virtual gates as [7]:

ṼP 1 α12 α13 VP 1

  
1
 
 ṼB 12  =  0 1 0   VB 12  (2.1)
ṼP 2 α31 α32 1 VP 2
 16 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

2
(a) (c)

VP VP

µN+1
(b) (d)
µN+1
µN
µS µS
µD µN µD
µN-1
µN-1

(e)

N-1 N N+1 N+2

I

VP

Figure 2.4: Quantum dot energy levels and transport. Green indicates a reservoir that is tunnel coupled to the
quantum dot defined under the plunger gate (blue). Two reservoirs exist on the left and right of the quantum
dot, called source and drain respectively. The source and drain leads are biased such that µS > µD promoting
current flow between the leads if Coloumb blockade is lifted. Left panels indicate a state of Coloumb blockade,
where current cannot flow between the source-drain leads since no states exist within the source-drain bias
window. When the plunger gate is tuned slightly, the energy level µN can be brought into the bias window
permitting current to flow. This happens at regularly spaced peaks in energy based on the addition energy of
the quantum dot and the lever arm of the plunger gate.
2.2. S EMICONDUCTOR QUANTUM D OTS 17

2
(a) (b) (c)

αEadd

Pdot
Isensor

Isensor
Quantum Dot

Charge Sensor

Psensor VSensor VDot

Isensor
(d) (e) (f)

N,N+2 N+1,N+2 N+2,N+2

P1 P2
VP2
Isensor

N,N+1 N+1,N+1 N+2,N+1

N,N N+1,N N+2,N

Psensor VSensor VP1

Figure 2.5: Charge sensing in single- and double- quantum dots. (a,b,c) correspond to a charge sensor sensing
the charge occupation of a single quantum dot. The chemical potential of sensor µsensor and dot µdot are
controlled by Vsensor and Vdot respectively. The charge sensor is capacitively coupled to quantum dot, such
that µsensor depends on the charge occupation of the quantum dot. We assume system is tuned such that
there are N charges in the quantum dot, and that the charge sensor current is maximised on top of a Coloumb
peak (yellow star). Changes in the voltage on an increase in Vdot of more than αE add where α is the plunger-
dot lever arm and E add is the addition energy of then quantum dot, will result in a change in the quantum
dot occupancy. Via capacitive coupling, this will affect µsensor and a different current will flow (orange star).
Therefore, sensor current corresponds to a direct measure of charge occupancy. (d,e,f)) show the case where
a sensor senses two dots. Here, one plunger for each dot is used to control the chemical potentials. A two-
dimensional sweep of the plunger gates results in a pattern of many different currents passing through the
dot, corresponding to different charge occupancy’s.
 18 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

(a) (b) (c) Isensor

P1 B12 P2

N,N+1 N+1,N+1

VP2

VP1
2

VP2
VP1 VP2

~
N,N N+1,N

VB12

VB12
VP1 VP2 ~
VP1

Figure 2.6: Virtual gate matrices. (a) Double dot system defined by voltages on real gates VP 1 , VP 2 , VB 12 .
Charge state of the double dot system is determined by the charge sensor below. (b) Tracking the shift in
energy of a charge transition under P1 (left) or P2 (right) as a function of other gates in the system. (c) Charge
stability diagram sweeping the virtual plungers ṼP 1 and ṼP 2 .

where αi j = δV j /δVi is the change in voltage of the charge transition under real gate i due to a
change of V j on real gate j . The diagonals of this matrix are of course unity, and the entries for the
barrier gates are set to zero, since there is (hopefully) not a quantum dot under these gates. Con-
structing a charge stability diagram by sweeping virtual gates ṼP 1 and ṼP 2 instead, would yield the
charge stability diagram in Fig 2.6c, where the virtual gates have influence on the chemical poten-
tial only of their respective quantum dots. It is common practice to define charge transitions with
respect to a different virtual gate space, called detuning-energy (² −U ) space. For a given double-
dot system containing a set number of charges n, one can define a detuning axis perpendicular
to the charge anticrossings such that varying the detuning (effectively creating a difference in real
plunger gate voltages up to a lever arm) results in different distributions of the n charges amongst
the quantum dots. U can then be thought of as the chemical potential of the quantum dot. Figure
2.7a shows charge stability diagrams in the two different virtual gate spaces.

Quantum dots can couple, such that they mutually effect their energies. Notice that a change in
charge occupation results in a slight shift in the charge transition (See Fig. 2.6c). This is due to the
additional mutual capacitance that exists between the dots themselves, hence the dots are capaci-
tively coupled together. Another form of coupling between dots is called tunnel coupling, which is
related to the overlap of wavefunction of adjacent quantum dots, leading to tunneling of charges
between dots. The quantum dot device in figure 2.6a has a barrier gate B12 , which is intended
to control the tunnel coupling between quantum dots. With energy, detuning, charge occupation
and tunnel coupling defined, qualitatively, we can model and simulate simplistic multi-quantum
dot systems. Consider a double quantum dot system such as the one in figure 2.6a. If we as-
sume charge occupation is the only important quantum number, then the eigenstates of such a
system will be i , j |n i , n j 〉 where n i , j ∈ N represents the number of charges on a quantum dot.
P

The Hamiltonian describing the system can be written in terms of an on-site energy HE , an inter-
dot coupling Hamiltonian H t , an internal Coloumb repulsion term HU and an inter-dot Coloumb
repulsion term HV [8].

X Ui
²i nˆi − t i j (ĉ i† ĉ j + h.c.) +
X X X
H = HE + H t + HU + HV = nˆi (nˆi − 1) + Vi j nˆi nˆj (2.2)
i i 6= j i 2 i 6= j
2.3. S PIN QUBITS 19

(a) (b)
tc = 0 tc>0 tc>>0

VU
VP2

VU
~

N,N N,N N,N

~
VP1 Vε Vε Vε Vε
2
P1 B12 P2 B12 B12 B12

tc
U

tc~0 tc>0 tc>>0

ε

Figure 2.7: Detuning-Energy Virtual Gates and Tunnel Coupling. (a) Conversion from virtual plunger gate
space to energy-detuning space. In the latter, the detuning axis V² represents the skew in chemical potentials
of the two dots, which is related to the difference in plunger gate voltages. The energy axis VU represents the
eigenenergies of the charge occupation levels of the system. (b) Tunnel coupling and it’s effect on the charge
stability diagrams. In the limit of zero tunnel coupling, the barrier between quantum dots is high, and so
there is no mixing of energy levels near the charge anticrossing resulting in rigid charge transition lines. For
moderate tunnel coupling, the barrier is lowered and state mixing occurs, resulting in bending of the charge
transition lines near the anticrossing. In the limit of high tunnel coupling, an effective single quantum dot is
formed.

where n̂ i is the number operator on the i ’th quantum dot (n̂ i |n i , n j 〉 = n i |n i , n j 〉), ĉ and ĉ † are
the annihilation and creation operators, Ui sets the strength of the internal Coloumb repulsion of
the i ’th quantum dot, and Vi j sets the strength of the inter-dot Coloumb repulsion between the
i ’th and j ’th quantum dots.

We have now derived the Hamiltonian for a system of quantum dots, filled with charges. This
is sufficient to build a first-order intuition about the dynamics of quantum dots, but does not pro-
vide us with a useful qubit in which to perform quantum computation. The two-level quantum
system that we will actually define the quantum information in, will be encoded in the spin quan-
tum number of the charges contained within the quantum dot.

2.3. S PIN Q UBITS

According to classical electrodynamics, a spinning charged body will generate a magnetic field
similar to a bar magnet, depending on it’s angular momentum. This analogy is illustrative for
quantum mechanics, where an electron in a quantum dot can be described by it’s spin quantum
number. In this section, we examine two level spin-1/2 systems, which constitute the basic build-
ing blocks for the qubit implementations in this thesis. A spin-1/2 system, has two eigenstates,
commonly referred to as spin-down |↓〉 and spin-up |↑〉. We can therefore represent the state of a
spin qubit as
α
µ ¶
|Ψ〉 = α |↓〉 + β |↑〉 = (2.3)
β

with α, β ∈ C and |α2 + β2 | = 1.

 20 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

2.3.1. Z EEMAN S PLITTING

q
When a magnetic dipole M = 2m L is placed in an external magnetic field B, it forms a system with
energy E = −M · B. Spin-1/2 systems are magnetic dipoles, with angular momentum S z = ħ2 σz ,
and thus their energy states can be written analogously, as

2 E=
g µB
S·B (2.4)
ħ
where g is the gyromagnetic ratio of the spin (here assumed to be spatially isotropic), µB is the
Bohr magneton and ħ is the reduced Plank’s constant. In general, the gyromagnetic ratio is not
←→
isotopic in all directions, and is better represented by a g -tensor g = diag(g x , g y , g z ), such that
equation 2.4 can be rewritten more generally as
µB ← →
E= S· g ·B (2.5)
ħ
If we now (without loss of generality) assume a magnetic field is applied in the ẑ direction, the
dot product between magnetic field and spin simplifies, resulting in two energy eigenstates from
the two allowed spin angular momenta. The difference in energy between these eigenstates is
called the Zeeman energy. We can write down a Hamiltonian for the spin-1/2 system in an external
magnetic field, as
g z µB B z
HZ = σz . (2.6)
2

2.3.2. L AMOR P RECISSION

If we consider the case where the spin vector is not perfectly p aligned to the external magnetic
field, for example a system prepared in the state |ψ0 〉 = 1/ 2(|↑〉 + |↓〉), ie. the cardinal x̂-axis
on equator of the Bloch sphere, then the time evolution of the system will be given by |ψ(t )〉 =
exp[i H Z t /ħ] |ψ0 〉. If, for example, we examine the behaviour of the projection of the S x spin com-
ponent, evaluating < S x >= 〈ψ(t )| S x |ψ(t )〉, we will observe oscillatory behaviour, according to

g z µB B Z
µ ¶
ħ
〈S x 〉 = cos t (2.7)
2 ħ

indicating that the spin vector will precess with a frequency

g z µB B Z
ωz = (2.8)
ħ
knwon as the Lamor precession frequency [9]. Note that in the case that the system was prepared
in an eigenstate, the spin would still precess at this frequency, however there would be no change
to the spin observables as time evolves.

2.3.3. Q UBIT S TATE M ANIPULATION

Rotations between the |↓〉 and |↑〉 eigenstates of the qubit can be performed by applying an os-
cillating magnetic field at the resonance frequency of the qubits, perpendicular to the stationary
Zeeman field. The driving Hamiltonian H AC can be written in terms of a AC magnetic field of
strength B AC applied in the ~x axis of the Bloch sphere

g x B AC ((A IQ )e −i ωa t + (A ∗ )e i ωa t
 ³ ´ 
µB  0 IQ
H AC =  (2.9)
g x B AC (A IQ )e −i ωa t + (A ∗ )e i ωa t
³ ´
2 0
IQ
2.3. S PIN QUBITS 21

where ωa is the applied microwave frequency, and A IQ = I + iQ represents the composition

of in-phase (I ) and quadrature (Q) components of the microwave signal. The total Hamiltonian of
the driven system H = H Z + H AC , for magnetic field B = (B AC , 0, B z ) becomes

g x B AC (A IQ )e −i ωa t + (A ∗ )e i ωa t
 ³ ´ 
µB  g z Bz IQ
H=
2 g x B AC (A IQ )e −i ωa t + (A ∗
³
IQ
)e i ωa t
´
g z Bz
 (2.10)
2
In it’s current form, this Hamiltonian describes the qubit’s state in the lab frame, meaning it’s dy-
namics are seen as a spin vector is precessing around the externally applied magnetic field while
rotating between eigenstates as a result of an AC magnetic field. This is inconvenient to visu-
alise and treat mathematically. Therefore, to analyze qubit dynamics, it is prudent to transform
this Hamiltonian to a rotating frame describing the interaction between the qubit and microwave
field, but not the spin precession. This is done through a transformation HRF = U HU † − i ħ ∂U
∂t
U†
giving a rotating frame Hamiltonian

ωz − ωa ωx (A ∗ ) + ωx (A IQ )e −2i ωa t
à !
ħ IQ
HRF = . (2.11)
2 ωx (A IQ ) + ωx (A ∗
IQ
)e 2i ωa t −ωz + ωa

Here, U = e i H Z t /ħ is the time evolution unitary of the stationary Hamiltonian, and

ωx(y) = g x(y) µB B x(y) /ħ. Since the terms e ±2i ωa t are fast oscillations, they will time-average out
to zero over meaningful operation timescales. Discarding these terms is called the Rotating Wave
Approximation (RWA) [10] and gives a final Hamiltonian
à !
ħ ωz − ωa ωx (A ∗
IQ
)
H= . (2.12)
2 ωx (A IQ ) ωa − ωz
It is not difficult to see how this Hamiltonian then decomposes into it’s respective Pauli operators.
Clearly by choosing I = 1 and Q = 0 for the microwave, we will rotate the qubit state around the
x̂-axis of the Bloch sphere, and by choosing I = 0 and Q = 1, we will perform ŷ-axis rotations, pro-
viding us with a complete set of single qubit operations.

The time evolution of this Hamiltonian (assume I = 1, Q = 0) acting on an initial state |ψ0 〉 then
gives the system dynamics of a spin-1/2 particle in a magnetic field, being driven by an oscillating
magnetic field
iHt
½ ¾
|ψ(t )〉 = U † exp − U |ψ0 〉 (2.13)
ħ
For the example of an initial state |ψ0 〉 = |↓〉, we can find the spin-up probability p ↑ = 〈↑ |ψ(t )〉 to
be q 
ω2x ω2x + (ωa − ωz )2
p↑ = sin2  t. (2.14)
ω2x + (ωa − ωz )2 2ħ

The quantity ωa − ωz is called the detuning, and respresents how far from resonance the driving
microwave field is from the Lamor precesssion of the qubit. The oscillatory evolution of a qubit
state due to a driving field is called a Rabi oscillation. We can see the effect of the detuning of a
driving pulse on the quantum state evolution of a qubit in Figure 2.8.

2.3.4. E XCHANGE I NTERACTION

We have now seen how a single spin qubit can be manipulated, using an AC-magnetic field. In
order to perform two-qubit gates, spin qubits need a form of coupling such that their states can
be come entangled. This occurs natively as a result of the exchange interaction, which manifests
 22 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

(a) (b) (c)

p|↑〉
0Ι 1
|↓〉 |↓〉

tPulse
Ι ΙΙ

Ι ΙΙ
|↑〉 - |↑〉
a 0

Figure 2.8: Rabi Rotations of a Spin Qubit. (a) and (c) Correspond to specific cases where the detuning fre-
quency is zero, and non-zero respectively. In the former case, perfect oscillations between the north and south
poles of the Bloch sphere occur at the Rabi speed. In the latter case, the detuned microwave frequency causes
faster rotations between the initial state and the final state, which is no longer the |↑〉 state. (b) Rabi Chevron
patterns, showing the behavior of the qubit state for a driving pulse time t , and a detuning ωa − ωz .

due to the indistinguishability of electrons. An effective exchange interaction arises in double

quantum dot systems due to the coupling between the singlet states. To illustrate this, consider
the the Fermi-Hubbard model (See equation 2.2, but add spin to the summation of on-site energy)
for a double quantum dot occupied by two spins. Without considering orbital states, the relevant
basis states are given by {|↓↓〉 , |↓↑〉 , |↑↓〉 , |↑↑〉 , |S(2, 0)〉 , |S(0, 2)〉}. For an on-site interaction energy U ,
energy detuning ² with respect to the (1,1) and (0,2) charge anticrossings (see section 2.2.3), and
tunnel coupling t 0 , we can write the 6x6 Hamiltonian as [11]
 E +E 
1 2
2 0 0 0 0 0
E 1 −E 2
0 0 0 t0 t0
 
2
 
E 2 −E 1
 
 0 0 2 0 −t 0 −t 0 
H = E 1 +E 2
. (2.15)
0 0 0 − 0 0
 
2
 
U −²
 
 0 t0 −t 0 0 0 
0 t0 −t 0 0 0 U +²

where E 1(2) is the Zeeman energy of the qubit in the left(right) quantum dot. This Hamiltonian can
be rebroadcast into one that only considers effective interactions between the lowest four energy
eigenstates, assuming that U ±² > t 0 > 0. This is done via a Schreiffer-Wolf transformation [12, 13],
and yields an effective Hamiltonian in the two qubit subspace {|↓↓〉 , |↓↑〉 , |↑↓〉 , |↑↑〉} [11, 14]
 E 1 +E 2 
2 0 0 0
E 1 −E 2
0 − 2J J
0
 
H =
 2 2 
(2.16)
J E 2 −E 1
− 2J

 0 2 2 0 
E 1 +E 2
0 0 0 − 2

where J is the effective exchange interaction, defined as

t 02 t 02
J= + . (2.17)
U − ² + (E 1 − E 2 )/2 U + ² + (E 2 − E 1 )/2

Controlling J in this Hamiltonian is integral to designing two-qubit gate operations.

2.3. S PIN QUBITS 23

2.3.5. T WO -Q UBIT E XCHANGE G ATES

There are three main two-qubit gates based on the exchange interaction, that occur in different
regimes of exchange interaction strength J compared to Zeeman energy difference ∆E z = |E 1 −E 2 |.
In the limit of high exchange such that J /∆E z >> 1, one can approximate the Hamiltonian using
the anti-parallel spin states such that
µ
−J /2 J /2
¶ 2
H= (2.18)
J /2 −J /2

Therefore, by turning on the exchange interaction, the qubits will exchange p states at a rate exactly
dictated by the size of J . If one pulses the exchange for a time t = 1/(2J ), a SW AP gate will occur,
which is a universal two-qubit gate. Depending on the achievable exchange interaction, SWAP
gates can be executed in very fast times. One problem with SWAP style gates however, is that any
finite Zeeman energy difference between qubits will result in an imperfect state exchange, which
limits the achievable accuracy. One can overcome this in two ways. The first is to perform a SWAP
gate by pulsing the exchange between the two qubits at their Zeeman energy difference frequency,
cancelling unwanted rotations due to a finite Zeeman energy difference. This is called a resonant
SWAP gate [15]. Alternatively, composite pulses can be constructed with timing considerations
that overcome these issues [16].

A second type of two-qubit exchange gate, is the controlled-phase gate (C-Phase), which occurs in
the regime where J /∆E z << 1. In this regime, the anti-parallel spin states will acquire phase at a
different rate, relative to the parallel spin states. The unitary describing a C-Phase gate is

1 0 0 0
 
 0 1 0 0 
UCPHASE = 
 0
 (2.19)
0 1 0 
0 0 0 ei φ
By timing the phase accumulation such that φ = π, the gate will perform a controlled-Z rotation,
another universal two-qubit gate. CZ-gates can also be performed in very fast times, and have
been shown with fidelities above 99% [17]. Errors from residual exchange can result in unwanted
SWAP oscillations, however these can either be mitigated by ensuring constant adiabaticity when
pulsing J , or by timing the gate with the periodicity of the SWAP oscillation.

The final two-qubit exchange gate is known as a controlled rotation (CROT) gate. It occurs since
increasing the exchange interaction between two qubits, lowers the anti-parallel state energies
without lowering the parallel state energies, resulting in a splitting of each qubit’s resonance fre-
quency, dependent on the state of the other.
q
f |↓↓〉→|↑↓〉 = Ē z − J 2 + ∆E z2 /2 − J /2 (2.20)
q
f |↓↑〉→|↑↑〉 = Ē z − J 2 + ∆E z2 /2 + J /2 (2.21)
q
f |↓↓〉→|↓↑〉 = Ē z + J 2 + ∆E z2 /2 − J /2 (2.22)
q
f |↑↓〉→|↑↑〉 = Ē z + J 2 + ∆E z2 /2 + J /2 (2.23)

where Ē z = (E 1 + E 2 )/2. Due to the frequency differences, controlled rotations can be performed
on one qubit based on the states of the others. Controlled rotations have two main challenges.
The first is the achieveable driving speed, which is limited by the single qubit gate rotations speed.
This speed is not nessecarily low, however gates who’s speed is based on the size of the exchange
 24 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

interaction like SWAP and CZ gates are typically much faster. The second challenge is that since
CROT’s are typically performed in the low-exchange regime, the gates must be carefully shaped
such that cross-talk effects are taken into account to prevent unwanted transitions between other
states. Fidelities as high as 98% have been reported for controlled rotation gates in silicon MOS
[18], and above 99% in Si/SiGe [19].
2
2.3.6. S PIN R EADOUT
With the concepts of two-level spin systems and of quantum dots under our belts, we can now
address an important requirement in building a quantum computer: readout. Readout is the final
operation performed on a quantum computer, whereby a qubit state is measured, collapsing the
wavefunction of the quantum state after computation. In order to read out the spin state of a qubit
in a quantum dot, one must somehow map the spin degree of freedom onto a charge state, such
that the readout signal is large enough. We do this since the magnetic moment of a spin is typi-
cally much too small to measure directly. Instead, the process of spin-to-charge conversion allows
readout of a charge state via a charge sensor or reflected phase of a microwave signal [20].

There are two main types of spin-to-charge conversion for semiconductor spin qubits. The first is
known as Elzermann readout, and involves spin selective tunnelling of a charge between a quan-
tum dot and a reservoir [21]. Here, the Zeeman energy splitting of a qubits spin states are tuned
such that they straddle the Fermi-level of a nearby reservoir. In this way, during readout, a spin
up state will tunnel from the dot into the reservoir, resulting in a jump in the sensor current. After
some time, a spin down state will tunnel back into the dot, causing the current to change back.
This process is summarised in figure 2.9a. The deterministic filling of a spin-down state also pro-
vides a useful initialization mechanism for the qubit. Elzerman readout has been used extensively
in early spin qubit experiments[22–24], however it has some major pitfalls. Firstly, it relies on
large Zeeman splittings. If this is not possible for reasonable magnetic field strengths, then there
will be readout errors. Additionally, the unloading and loading occurs at a random time, placing
constraints on readout bandwidth. Finally, the reservoir must be close to zero-temperature, or
the thermal broadening of the Fermi-energy will also lead to readout errors. This makes it an in-
compatible strategy for high temperature qubit operation [25], and therefore we will examine an
alternative readout method.

The Pauli exclusion principle prevents two fermions from existing with the exact same quantum
numbers. In a quantum dot, two spins cannot occupy the same quantum dot ground state if they
share the same spin. This facilitates a spin-to-charge conversion technique called Pauli spin block-
ade (PSB). In PSB, a spin state in one quantum dot is read out using the spin state of another quan-
tum dot. The term measurement qubit refers to the qubit who’s state we want to read out, and the
term helper qubit refers to the qubit who’s state is used for readout. Typically, we know what the
helper qubit spin state is prior to measurement, allowing for direct readout of the measurement
qubit. To illustrate PSB, consider a quantum double dot system, where a single spin qubit defined
in a quantum dot has completed some arbitrary quantum operation, and is now ready for mea-
surement. The helper qubit is known to be in the spin down state. By tilting the detuning between
the two quantum dots such that it becomes energetically favorable for two-electrons to exist in the
helper quantum dot, the measurement qubit will tunnel in only if it was in the spin up state, since
two spin down states are not permitted to co-exist in the same quantum dot. The current detected
by the charge sensor will therefore either detect a (2,0) configuration (IUnblocked ) or a (1,1) config-
uration (IBlocked ). Figure 2.9b summarises the PSB protocol.

The above examples have been presented in a (mostly) platform agnostic way. In other words,
 2.4. E LECTRON SPIN QUBITS IN S ILICON QUANTUM DOTS 25

(a) (b)
Blocked 2,1

1,1 2,0

VU
Not Blocked

1,0
2
Vε
I II III

Isensor
Isensor

(b) IBlocked

IUnblocked

t t

Figure 2.9: Spin-to-charge conversion of a spin qubit. (a) Elzerman readout. Spin state energy levels are tuned
such that they straddle the Fermi level of a reservoir. Only a spin up state can tunnel into the reservoir (red
trace), resulting in a sensor signal. A spin down qubit can then tunnel back into the quantum dot, resulting in
a sensor change, and a spin down initialization of the qubit. Spin down qubits cannot tunnel into the reservoir
as there are no free states below it’s chemical potential (black trace), therefore no current change occurs in
the sensor. (b) Pauli Spin Blockade readout. An helper qubit (left dot) is used to read out the spin state of a
measurement qubit (right dot). Preparing the helper qubit in a spin down state allows spin selective tunnelling
from the measurement dot to the helper dot, based on the measurement qubit’s state. Choosing a readout
point in the spin blockade window (triangle in the CSD), one can observe a difference in current response when
the measurement qubit is spin up or spin down. Note that sensor currents in (a) and (b) are only sketches, and
are not real or simulated data.

we are yet to apply our concept of spin qubits to the case of electrons in silicon, or holes in germa-
nium. Therefore, the current discussion will shift gears to talk about the In the following sections,
we examine the specific physics of both electrons in silicon, and holes in germanium.

2.4. E LECTRON SPIN QUBITS IN S ILICON QUANTUM DOTS

The electron is a fundamental, negatively charged particle. It is also a spin-1/2 particle, making it a
logical qubit candidate. It is also therefore crucial to quantum computation, that an electron’s spin
degree of freedom exists in an environment where unwanted interactions are minimized. Silicon,
in contrast to group III-V materials like GaAs, contains a low abundance of spin non-zero nuclei,
making it an attractive candidate for quantum computation. Additionally, silicon can be isotopi-
cally purified [26], effectively eliminating unwanted interactions between nuclear and electron
spins. The utilization of isotopically purified silicon as a host for electron spin qubits substantially
improved the time for which quantum information can be stored [22, 27], and has therefore dras-
tically increased the platforms viability. Improvements in the quality of the silicon-oxide [28], and
material growth have also led to a very quiet environment in which to host qubits, earning the Si-
MOS platform the prestigious title of ‘the semiconductor vacuum’ [29].

2.4.1. VALLEY S PLITTING

In order to understand the behaviour of electron spin qubits in silicon quantum dots, we can begin
by examining the band structure of silicon. The conduction band of silicon contains a minimum
about 85 % along the energy dispersion line between the Γ-point and X -point [30]. In the case of
an electron in bulk silicon, this minimum, called a valley, will occur at the same energy for each
cardinal axis in momentum space, leading to a six-fold degenerate energy spectrum (Figure 2.10b
 26 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

(a) (b) I: FZ = 0 II: FZ > 0 III: FZ > 0, B > 0

U
kz |x±〉,|y±〉

|z+〉
ky |x±〉,|y±〉,|z±〉 ~ ~〉
|↑,z
|z+〉
2 kx
~
|z±〉 Ez ~〉
|↓,z
+

∆VS |z~− 〉
+
~ 〉
|↑,z
Ez −
~ 〉
|↓,z
|z- 〉 −

Figure 2.10: Energy structure of Silicon quantum dots due to valleys. (a) Conduction band electrons in the
silicon bulk experience an energy minima in each direction in k-space between the Γ-point and X -point [30].
(b) I: A 6-folk valley degeneracy exists for conduction band electrons in bulk silicon. II: the application of an
electric field in the out-of-plane ẑ direction results in the splitting of this degeneracy into two-lower lying ẑ
valleys and four degenerate states in the in-plane directions at much higher energies [31]. These two lower
lying valley states are mixed by the strong interface confinement, resulting in new eigenstates |z̃ ± 〉 split by an
energy ∆V S called the valley splitting [32]. III: The application of a magnetic field results in the lifting of spin
degeneracy, resulting in four lowest energy eigenstates, the lowest two of which, |↓, z̃ − 〉 and |↑, z̃ − 〉, we use qubit
encoding.

I) [30]. When we confine an electron at an interface to form a quantum dot with an electric field
F z in the ẑ direction, the electron will experience a much higher effective mass in the confinement
direction than within the plane of the quantum well, splitting the degeneracy into a four-fold de-
generacy containing the four (±k x , ±k y ) valleys with a high energy, leaving the two lowest energy
valleys in the k̂ z direction [31] (Figure 2.10b II). The presence of an electric field also results in the
mixture of these valleys, such that their eigenstates are combinations of the bulk |z ± 〉 valleys [33].
These lower two valley eigenstates |z̃ ± 〉 are split by an energy gap ∆V S , called the valley splittting.
In SiMOS and Si/SiGe platforms the singular nature of the interfaces at which electrons are con-
fined has a major impact on the size of ∆V S [34]. The size of ∆V S is typically between 0.03-0.3 meV
in Si/SiGe[35, 36], and between about 0.1-1 meV in SiMOS [23, 37] with evidence of tuneability via
electrostatic gates [23, 38]. These values are comparable to typical Zeeman energies of electron
spins in silicon and as a result, strategies to deal with the valley states in silicon are integral to
achieving good electron spin qubits.

Consider a single electron spin defined in a silicon quantum dot. The four lowest energy eigen-
states will now be given by it’s spin and valley degrees of freedom, such that |1〉 = |↓, z̃ − 〉, |2〉 =
|↑, z̃ − 〉, |3〉 = |↓, z̃ + 〉 and |4〉 = |↑, z̃ + 〉. Opposite valley states can couple via the spin orbit interac-
tion, such that the states [39]

1 − a 1/2 1 + a 1/2
¶ µ µ ¶
|2̃〉 = |2〉 − |3〉 , (2.24)
2 2
1 + a 1/2 1 − a 1/2
µ ¶ µ ¶
|3̃〉 = |2〉 + |3〉 (2.25)
2 2
p
describe the energy eigenstates, where a = −(∆v s −ωz )/ (∆v s − ωz )2 − ∆2 for coupling ∆ and Zee-
man energy ω Z . Figure 2.11a shows how these four lowest energy spin-valley eigenstates behave
as a function of magnetic field. We can see that above a critical magnetic field, the spin eigen-
states no longer define the lowest energy eigenstates of the system. At magnetic fields close to
this critical value, spin-valley mixing causes significantly enhanced spin relaxation times [39]. The
2.4. E LECTRON SPIN QUBITS IN S ILICON QUANTUM DOTS 27

(a) (b) Isensor

+ +
T- S T0 T+
+ + (c)

|4〉 -
(↑,↓)
δEz (2,1)
S (↓,↑)
|2〉
|3〉 (↑,↑)
Energy

N,N
∆VS ∆ (↑,↓) (a) (1,1) L (2,0)

VU
U

(↓,↑)
|2〉
|3〉 (↓,↓)
∆VS
|1〉 (1,0)
I
B -U Vε
(d) + + +
- +
(↓,↓) (↓,↑) (↑,↓) (↑,↑) S T- S T0 T+
~ + +
|z+〉
Valley

~ - - -
|z− 〉
L R
Dot

Figure 2.11: Impact of valley splitting for electron spin qubits in silicon quantum dots. (a) Four lowest energy
levels of a single electron spin in a quantum dot as a function of applied magentic field. |1〉, |2〉, |3〉, and |4〉,
represent |↓, z̃ − 〉, |↑, z̃ − 〉, |↓, z̃ + 〉, |↑, z̃ + 〉 respectively. Spin-valley mixing can occur due to a finite spin-orbit
interaction, leading to a coupling of size ∆ at a critical magnetic field [38]. Above this magnetic field, the spin
states no longer define the lowest energy qubit states. Additionally, close to this field, spin relaxation rates are
rapidly enhanced [39]. (b) Valley splitting affects ability to read out a spin state. First nine energy levels of two
electron spins in a quantum double dot system. Blue or red shading indicates a preferred (1,1) or (2,0) charge
configuration respectively. Spin blockade readout is affected by the presence of Valley states as readout is only
possible within the detuning window ²V S before the first excited valley state T+ − crosses the (↓, ↓) state, resulting
in ambiguity of spin state readout above this detuning value. [40] (c) Effect of valley splitting on spin readout at
spin blockade enabled anticrossings. During a spin-blockade readout sequence, a random spin state is loaded
into the left dot by moving from I to L. The right dot is initialized in the spin down state. The readout point
is then swept across the parameter space shown in the charge stability diagram. Valley splitting truncates the
readout window, limiting available readout parameter space, and affecting readout fidelity [39]. (d) Summary
of state labelling for (b).
 28 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

presence of valley states also affects Pauli Spin blockade readout. To illustrate this, we can simu-
late a double quantum dot system in silicon, with a total of two electrons. Here we simulate the
first nine spin-valley eigenstates considering the (1,1)-(2,0) anticrossing. Figure 2.11b shows the
resulting energy eigenstates as a function of detuning ² between the dots. The readout window for
a spin blockade style measurement is limited to the detuning space ²V S , between the lowest valley
2 S20 and first excited valley T− states. Figure 2.11c shows how a spin blockade readout sequence
behaves as a result of valleys. The pulse sequence to produce such a measurement involves ini-
tialization of a spin down electron in the right dot, followed by the loading of a random spin in the
left dot. The readout window is then swept and the resulting sensor current is recorded. The spin
blockade window is truncated due to the valley splitting. This has the consequence of both reduc-
ing the parameter space in which one can read out spin states, and also reduces the fidelity of the
readout itself. It is therefore crucial for electron spin qubits in silicon to achieve valley splittings
significantly larger than the Zeeman energies of the qubits. Strategies to reliably increase valley
splittings, particularly in the Si/SiGe platform have been investigated, including, as mentioned,
electric tuning [23, 38], and by introducing small amounts of Ge into the quantum well, which has
been shown to help increase valley splittings [41].

2.4.2. E LECTRON S PIN R ESONANCE

Performing quantum gates on electron spin qubits relies on delivering oscillating magnetic fields
perpendicular to the quantization axis. This can be achieved in different ways, however for now,
we will examine the technique of electron spin resonance (ESR) since it is the spin driving tech-
nique used for all experiments in this thesis on electron spin qubits. ESR involves the delivery of an
oscillating magnetic field to a qubit, by sending an AC-current through a nearby antenna. Achiev-
able single qubit gate times using this method are typically on the order of 1 MHz, and very high
control fidelities have been obtained [22]. From a device point of view, this antenna must dissipate
minimal heat, therefore aluminium (Al), niobium titanium nitride (NbTiN), and other supercon-
ducting materials are usually chosen for its fabrication. Additionally, in order to minimize loss,
it must be designed with impedance matching in mind, often resulting in large, bulky structures
[42]. This means it is difficult to implement more than a single antennae in a device, which raises
the concern of individual qubit addressability. Typically in SiMOS, variations in the Silicon-Oxide
interface result in highly varying g-factors, and therefore individually addressable qubits. How-
ever in Si/SiGe, disorder is typically much lower, and qubit g-factors are closer. This problem can
be mitigated by designing micro-magnets such that the Zeeman differences between qubits are
enlarged, however the introduction of a spatially varying magnetic field enhances the coupling of
charge noise to the qubit eigenstates, resulting in lower relaxation and coherence times. The use
of micromagnets also permits the design of artificial spin-orbit coupling, that allows for electrical
driving of the qubit using electric dipole spin resonance (EDSR). This technique will be expanded
on when considering qubit manipulation using holes in germanium.

2.5. H OLE SPIN QUBITS IN G ERMANIUM QUANTUM DOTS

When an electron in the valence band receives enough energy to excite to the conduction band
of a semiconductor, it leaves behind a hole. Holes are pseudoparticles, that can be thought of as
positively charged particles that can move throughout a semiconductor, and crucially, are suitable
qubit candidates. Valence band holes exhibit p-like Bloch wave vectors, which is a consequence
of their l = 1 azimuthal quantum number. This means that their wavefunctions have no overlap
with nuclear sites in their host, eliminating the contact hyperfine contribution to dephasing (see
section 2.5.4), but more importantly, facilitates a spin orbit coupling δSO L · S between the spin ħS
and orbital ħL angular momentum of the hole [43]. The following sections derive the properties of
2.5. H OLE SPIN QUBITS IN G ERMANIUM QUANTUM DOTS 29

hole spins in planar germanium quantum dots.

2.5.1. T HE VALENCE B AND

The topmost valence band for bulk germanium would be six-fold degenerate at the Γ-point, in the
absence of a spin-orbit interaction [44]. The presence of a spin-orbit interaction lifts these degen-
eracies, which is important for creating a well define two-level spin system to use as a qubit. To 2
derive the spin orbit effects, we therefore want to understand the effect of the spin orbit operator
on the first valence band δSO L · S |Ψ〉. Spin orbit coupling results in a Hamiltonian that no longer
commutes with S or L, which are therefore no longer good quantum numbers. The total angular
momentum however, J = S + L is a good quantum number. We can take the square of J, and rear-
range to find that L·S = 1/2 J2 −L2 −S2 . We know that L2 |Ψ〉 = ħ2 l (l +1) |Ψ〉, S2 |Ψ〉 = ħ2 s(s +1) |Ψ〉,
and J2 |Ψ〉 = ħ2 j ( j + 1) |Ψ〉, leaving us with the expression

ħ2 δSO
δSO L · S |Ψ〉 = ( j ( j + 1) − l (l + 1) − s(s + 1)) |Ψ〉 . (2.26)
2
since m s and m l are not good quantum numbers, we cannot say anything about them, except that
they can either be parallel or anti-parallel leading to j = 1/2 or j = 3/2. We can immediately assign
l = 1 and s = 1/2 since these are properties of the band we are considering, allowing us to calculate
the eigenenergies of the above equation as δSO /2 for j = 3/2 and −δSO for j = 1/2. The projections
of j therefore give rise to a four fold degenerate band at energy δSO /2 and a two fold degenerate
band at energy −δSO . The difference in energy between these sets of bands is called the spin or-
bit gap ∆O = 3δSO /2, and is defined at the Γ-point, where k x = k y = k z = 0 [45]. In germanium,
∆0 ≈ 0.3 eV, which separates it sufficiently from the j = 3/2 bands.

The energy dispersion of these bands around the Γ-point can be modelled using a Luttinger-Kohn
(LK) Hamiltonian[46]. The LK Hamiltonian in general can be applied to a multitude of semicon-
ductors, however for semiconductors with high inversion symmetry, it can be written in a sim-
plified way by making the spherical approximation according to the eigenstates |m J 〉 ∈ {|−3/2〉 ,
|−1/2〉 , |1/2〉 , |3/2〉} [47]

P +Q L M 0
 
 L∗ P −Q 0 M 
HLK = 
 M∗
 (2.27)
0 P −Q −L 
0 M∗ −L ∗ P +Q
where
ħ2
P =− γ1 k 2 (2.28)
2m 0
ħ2
Q= γs (2k z2 − k x2 − k 2y ) (2.29)
2m 0
ħ2 p
L= 2 3γs k − k z (2.30)
2m 0
ħ2 p 2
M= 3γs k − (2.31)
2m 0

with k 2 = k · k and k the wave vector, m 0 the free electron mass, k ± = k x ± i k y , and γ1 and γs =
(2γ2 + 3γ3 )/5 the Luttinger parameters [48]. This hamiltonian yields two sets of two-fold degener-
ate eigenenergies, E H H = −ħ2 k 2 /2m H H and E LH = −ħ2 k 2 /2m LH , belonging to the |m J 〉 = |±3/2〉
and |m J 〉 = |±1/2〉 eigenstates respectively, where m H H = m 0 /(γ1 −2γs ) and m LH = m 0 /(γ1 +2γs ).
 30 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

The subscripts ’LH’ and ’HH’ stand for light hole and heavy hole respectively, and are labelled as
such since the substitution of the relevant Luttinger parameters for Ge yield m H H = 0.33m 0 and
m LH = 0.04m 0 . We can immediately see that away from the Γ-point, the energy dispersion of
these LH and HH bands will be different, though they remain degenerate at |k| = 0.

2 The system we have defined in the above scenario corresponds to hole states in the valence band,
in germanium bulk. However the system we want knowledge about is how hole states behave
in a quantum dot, specifically in a confinement potential and in the quantum well of a Ge/SiGe
heterostructure. To begin, let’s examine a hole confined in a hard wall potential, as might be ex-
pected at the interfaces of a quantum well. The strong confinement in the ẑ-direction leads to
a quantization of momentum such that k z = πn/L z where n corresponds to the excitation har-
monic number, and L z is the width of the quantum dot. We will introduce another assumption
here, whereby the off-diagonal terms in equation 2.27 are negligible when considering the macro-
scopic energy dispersion of the heavy hole light hole bands near the Γ-point, which turns out to be
a good approximation for quantum wells grown on the high symmetry [001] crystallographic axis
[44, 45] as is the case for the quantum wells used throughout this thesis [49, 50]. As a result, the
eigenenergies E H H and E LH are shifted to
à !
ħ2 ¢ n H H π2 ¡
k ||2 γ1 − 2γs + γ
¡ ¢
EH H = − 1 − 2γ s (2.32)
2m 0 L 2z
à !
ħ2 ¢ n LH π2 ¡
k ||2 γ1 + 2γs + γ
¡ ¢
E LH = − 1 + 2γ s (2.33)
2m 0 L 2z

where k ||2 = k|| · k|| = k x2 + k 2y , and we have approximated that the sub-band excitations in the con-
finement potential of the LH and HH states are not coupled such that they can be described by
their own quantum numbers n LH and n H H respectively. The above equations have an important
consequence for energy dispersion when considering the planar wavevectors k|| in that even at
k|| = (0, 0), there is an energy gap separating the LH and HH states of

2γs ħ2 π2
∆LH ,H H = (2.34)
m 0 L 2z

which therefore lifts the LH-HH degeneracy. This energy splitting is important for ensuring that
the two-level quantum system is well defined and isolated from other states. A second interest-
ing consequence is that re-evaluating the effective light hole and heavy hole masses yields in a
somewhat confusing result in that m H H < m LH .

2.5.2. S TRAIN
The sandwiching of a Germanium quantum well between two SiGe layers imparts a strain on the
quantum well due to the slight lattice constant mismatch of the two different semiconductors.
The degree of the lattice mismatch and therefore strain, is controlled by the relative composition
x of Si and Ge in the Six Ge1−x layers. The effect of strain on energy dispersion relations requires
additional treatment of the Hamiltonian that the previous LK Hamiltonian does not take into ac-
count. The strain tensor in a material in general a 3x3 matrix with elements ²i j where i , j ∈ {x, y, z}
represent the cartesian coordinates. For the case of a strained, planar quantum well, the strain
is uni-axial [49], allowing for a simplification of the tensor such that ²xx = ² y y , and off-diagonal
elements vanish such that ²i j = 0 if i 6= j . Additionally, the strain quite small, and approximately
linear, allowing one to write ²zz = −2(C 12 /C 11 )²xx where C 11 = 129.2 GPa and C 12 = 47.9 GPa are
stiffness coefficients in Ge [51]. The value of ²xx is a function of the lattice mismatch, and can be
2.5. H OLE SPIN QUBITS IN G ERMANIUM QUANTUM DOTS 31

approximated ²xx = (a(x)−a 0 )/a 0 where a 0 is the relaxed lattice constant of the Ge quantum well,
and a(x) is that of the Six Ge1−x layers [52].

In this form, the LK Hamiltonian can easily adapt for strain by adjusting augmenting the matri-
ces [53]
2
P → P + P² (2.35)
Q → Q +Q ² (2.36)

where

P ² = a v ²xx + ² y y + ²zz
¡ ¢
(2.37)
Q ² = b v /2 ²xx + ² y y − 2²zz
¡ ¢
(2.38)

with a v = 2.0 eV and b v = −2.16 eV are deformation potentials [54]. The value of ²xx = −0.0063 is
used as it comes from calculations pertaining to the exact quantum wells used in this thesis[49].
The negative value of this strain indicates that it is compressive in the plane of the quantum well.
It is immediately aparent that strain of this type will lead to a shift in EH H upwards, and ELH
downwards, both of magnitude Q ² . This increases the splitting of the two bands by ∆²LH ,H H =
2Q ² ≈ 49 meV.

2.5.3. L IFTING S PIN D EGENERACY

At this point, we have a material platform that hosts a well defined, and well separated spin qubit,
in the form of the heavy-hole bands. However, these bands are currently degenerate in spin and
our model does not yet explain their behaviour when this degerneracy is lifted, let alone how to do
this. A good starting point is to examine the spin orbit interaction and it’s effect on the spin states,
particularly those corresponding to the heavy holes.

Spin orbit interaction can manifest in a quantum system that exhibits structural inversion asym-
mety (SIA) or bulk inversion asymmetry (BIA) [45]. The latter effect describes lattices that are not
the same under point reflection, for example zinc-blende structures like GaAs. Germanium how-
ever is symmetric under bulk inversion, and therefore does not exhibit a spin-orbit interaction
due to BIA (called a Dresselhaus-type spin orbit interaction) [55]. SIA, however, can occur due to
the confinement potential of the quantum well and the metallic gates used to define the quantum
dot. This type of spin orbit interaction is called a Rashba spin orbit interaction [56], and acts on
the heavy-hole spin manifold according to the Hamiltonian [57]
3 3
HR,SO = i α1 (k + σ+ − k − σ− ) + i α2 (k + σ− k − σ+ ) + i α3 (k + σ+ − k − σ− )k 2 . (2.39)

Here, σ± = σx ± i σ y , and αi are the i ’th order Rashba parameters that dictate the strength of the
interaction due to various symmetries in HLK . For crystals grown along the high symmetry axis,
α1 and α3 are typically small compared to α2 , when making the assumption that HLK is effectively
spherical in symmetry [47]. Thus the α2 term is primarily responsible for lifting spin degeneracy
within the heavy hole and light hole manifolds when considering their energy dispersion, in the
absence of a magnetic field.

We now come to the behaviour of the spin eigenstates under an externally applied magnetic field.
For holes in the upmost valence band, this is given by a Zeeman Hamiltonian of the form [48]

H Z = −2κµB B · J − 2qµB B · J3 (2.40)

 32 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

where κ and q are Luttinger parameters. If we restrict ourselves to the case of heavy holes, then
the above simplifies to

µ ¶
27 3
q µB B z σz − qµB B x σx − B y σ y .
¡ ¢
H Z = − 3κ + (2.41)
2 4 2

We can see immediately that the effect of an applied magnetic field will differ greatly depending
on its orientation. For the in-plane directions, the effective g-factor is isotopic (g x = g y = g || = 3q).
However out of plane, we have g z = g ⊥ = 6κ + 13.5q. Taking typical values of these Luttinger
paramters, we find g || ≈ 0.2 and g ⊥ ≈ 21.4 [58]. The experiments conducted on hole spin qubits in
germanium in this thesis are performed using a (mostly) in plane magnetic field. In this case, the
Zeeman splitting of the heavy hole-like states is perturbed by a spin-orbit term given by [59–61]

3γs κµB ³ 2 2
´
HSO,Z = B − k− σ+ + B + k + σ− (2.42)
m 0 ∆H H ,LH

where B ± = B x ± i B y . Interestingly, the dependence of this term on ∆H H ,LH , which depends on

the confinement potential, results in an electrically-tuneable g-factor, a valuable tool when con-
sidering qubit addressability. However, the dependence of the eigenenergies of hole spin qubits
on electric fields makes susceptible to charge noise, which turns out to be a major source of deco-
herence for the Ge/SiGe platform.

2.5.4. H YPERFINE I NTERACTION

The spin and orbital angular momentum degrees of freedom in a hole qubit can interract directly
with other spin and orbital degrees of freedom in their environment. Naturally, this process is in-
tegral to the operation of a quantum computer to generate entanglement or shuttle information
etc. However, this process can also occur between spin systems in an undesireable way. In semi-
conductors, a major source of unwanted information loss occurs due to the presence of non-zero
spin nuclear isotopes of the host material. This interaction between nuclear spins and qubits is
called the Hyperfine interaction.

Semiconductor hosts such as GaAs have a 100% nuclear non-zero spin population, which makes
storing quantum information a non-trivial matter. Group IV based platforms such as silicon and
germanium however, have relatively low compositions of spin non-zero isotopes (in natural sili-
con, about 4.7 % and in natural germanium about 7.8 %). However, these populations still play a
role in limiting coherence times of qubits defined in these platforms. The hyperfine interaction
between a spin qubit (hole or electron) and a nuclear spin bath can be written down generally as

X
HH f = A k S · Ik (2.43)
j

where A k is the hyperfine interaction strength between the hole and the k’th nuclei in the bath, S is
the qubit spin and Ik is the spin of the k’th nucleus. The strength of this interaction A k is material
dependent, and also depends on the degrees of freedom available to the qubit. To illustrate this,
we can write down the three separate hyperfine interaction mechanism Hamiltonians available to
a electron or hole when coupling to the k’th nuclear spin (assuming a single non-zero nuclear spin
2.5. H OLE SPIN QUBITS IN G ERMANIUM QUANTUM DOTS 33

species) [62]

µ0 8π
h 1k = γs γk δ(rk )S · Ik (2.44)
4π 3
µ0 3(nk · S)(nk · Ik ) − S · Ik
h 2k = γs γk (2.45)
4π r 3 (1 + d /r k )
k 2
µ0 L ·I
h 3k = γs γk 3 k k (2.46)
4π r k (1 + d /r k )

where γs = 2µB , γk = g k µN , µB is the Bohr magneton, g k is the nuclear g-factor, µN is the nu-
clear magneton, rk = r − Rk is the electron-spin position operator relative to the nucleus, d ≈
Z × 1.5 × 10−15 m is a length describing nuclear dimensions, Z is the nuclear charge, nk = rk /r k ,
and S and Lk = rk × p are the spin and orbital angular momentum operators of the qubit respec-
tively. These correspond respectively to a direct contact hyperifne interaction, a dipole-dipole
type coupling, and a coupling of the orbital angular momentum of the qubit to the nuclear spin.
In the case of an electron in the lowest conduction band, the spherical symmetry of the Bloch
wavefunctions eliminates the possibility of a dipole-dipole like interaction, and since there is no
orbital angular momentum, we have h 2k = h 3k = 0, such the the hyperfine interaction of electrons
is purely due to the overlap of the qubit and nuclear wavefunctions. In the case of a hole, p-like
symmetry results in nodes at nuclear spin sites such that h 1k = 0, while the other two mechanisms
are possible. The elimination of h 1k type couplings have led to the prediction that the hyperfine in-
teractions for hole spins will have negligible contributions. However the resulting terms as a direct
consequence of Bloch wavefunctions with p-like symmetry still contribute non-negligibly. We can
see that (following ref. [62]) the ratio of the hyperfine coupling strength for holes and electrons in
a given material can be approximated by

A h 1 Zeff (κ, 4p) 3

µ ¶
= (2.47)
Ae 5 Zeff (κ, 4s)

where Zeff (κ, 4p(4s)) describes the effective screening potential experienced by holes (electrons)
in 4p(4s) atomic orbitals of nuclear species κ. Taking these values for the 73 Ge nuclear spin from
ref. [63], we can see that A h /A e ≈ 0.12. This indicates that while less, hyperfine coupling in hole
spin qubits is still a relevant contribution to the noise processes inhibiting their operation.

Unlike electrons, the hyperfine interaction for holes can be highly anisotropic. The interaction
of holes in the quasi-two dimensional limit (such as holes confined in quantum dots within a
quantum well) takes on an Ising-like form
X h
HH f = A k s z I kz (2.48)
k

Such that only the z-components of the hole and nuclear spins couple strongly. This anisotropy
has some implications regarding optimal magnetic field alignment, since A || << A ⊥ , where per-
pendicular in this case indicates perpendicular to the plane of the quantum well. By aligning a
magnetic field in plane, the strong out-of-plane hyperfine coupling term is suppressed due to the
Ising like nature of H H f . This is investigated in length in Chapter 11.

The non-zero nuclear spin isotope of germanium, 73 Ge, has a spin I = 9/2 quantum number.
Nuclear spins with quantum numbers greater than spin-1/2 can exhibit a nuclear quadrupole
interaction, that shifts the nuclear energy levels of the satellite transitions [64]. Therefore when
 34 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

describing a 73 Ge nuclear spin bath, one must consider an additional term in the nuclear Hamil-
tonian
ωQ X z 2
γI B z I kz +
X
Hnuc = − (I ) (2.49)
k 2 k k

2
where γI is the Lamor precession frequency of 73 Ge and ωQ is the quadrupolar interaction strength.
In bulk, magnetic resonance experiments have measured ωQ ≈ 100 Hz [65], however this term is
sensitive to local electric fields, therefore the strain in the quantum well can drastically enlarge the
magnitude [66].

2.5.5. E LECTRIC D IPOLE S PIN R ESONANCE

The spin orbit interaction experienced by holes in Ge/SiGe facilitates a coupling of electric field
oscillations to the spin of the hole qubit. Driving a hole spin in this way is called electric dipole
spin resonance (EDSR), and allows for very fast manipulation speeds of spin qubits compared to
conventional ESR. EDSR is typically achieved by modulating the electric potential of the metallic
gates close to the spin qubits in a semiconductor device, causing the qubit oscillate within it’s con-
finement potential. The spin orbit interaction results in the hole seeing an effective AC-magnetic
field, resulting in spin rotations.

Exciting spin transitions between the heavy-hole eigenstates |0, 0, ±3/2〉 (where the indices cor-
respond to excited state orbital n, azimuthal quantum number l and spin m J ) cannot be achieved
by a magnetic dipole interaction, since ∆m J = 3. Transitions become possible, however, in the
picture where the spin flip occurs via a virtual transition to the first excited state orbital of the
quantum dot (n = 1) in the confinement potential [60]. The AC-electric field provides this change,
and since it also moves the qubit within the confinement potential itself, the spin orbit interaction
facilitates spin flips. This requires a term in the spin orbit Hamiltonian that permits a change in
the orbital excitation of ∆n = 1, which is only found in the cubic-symmetric α3 term in equation
2.39 [52].

Driving via EDSR in the presence of a spin-orbit interaction usually allows for much faster qubit
manipulations than achievable with ESR [67]. ESR strip-lines are also limited by superconduct-
ing critical currents dictated by their design and material. On the other hand, oscillating electric
fields can be readily achieved by modulating the electrostatic potentials on metallic gates already
present on a device (such as plunger or barrier gates), circumventing the need for additional bulky
structures such as microwave antennas. However, the dependence of a systems eigenstates on
electric field fluctuations enhances the coupling of charge noise to the system, resulting in a trade-
off between fast driving and quantum coherence.

2.6. Q UALIFYING A Q UANTUM S YSTEM

Quantum systems are constantly interacting and exchanging energy (and therefore information)
with other quantum systems around them. The trick is to isolate qubits such that they only interact
with other qubits when desired, and as little as possible with the outside world. Unfortunately this
is never completely possible, and therefore it is important to know the timescales over which one
can reliably store information in, or operate on qubits in a quantum computer. Information loss
can occur in a number of ways. In the case of a single qubit, information loss can be separated into
two processes, called spin-relaxation, and dephasing.
2.6. QUALIFYING A QUANTUM S YSTEM 35

2.6.1. S PIN R ELAXATION

Spin relaxation is quantified by a timescale called T 1 . It represents the time that a qubit can re-
main in it’s higher energy state, before relaxing to it’s ground state. Relaxation of this type occurs
due to noise that couples perpendicular to the quantization axis of the qubit. For electrons in sili-
con qubits, this typically occurs due to charge noise in the surrounding 2DEG’s and metallic gates
[68, 69]. This charge noise couples to the spin degree of freedom weakly via the spin-orbit cou- 2
pling, resulting in spin-relaxation [39]. Due to this relatively weak effect, spin relaxation times of
more than a second have been measured in silicon quantum dots [38, 70]. The large spin orbit
coupling of holes in Ge/SiGe results in a stronger effect with respect to the case of silicon, and
therefore hole spin relaxation times have been limited to a few hundred microseconds [71, 72]. We
demonstrate a drastic improvement to this timescale in chapter 8.

In order to measure spin relaxation of a single qubit, one must initialize a qubit in it’s higher energy
spin state. One then waits for a time, and then reads out the final state of the spin. By repeating
this process over many measurements and averaging the traces, a characteristic decay will occur
related to the probability of finding the spin still in its |↑〉 state (See Figure 2.12a). Performing this
measurement as a function of wait time yields an exponential decay with characteristic decay time
T 1 . Long spin relaxation times are necessary to the operation of a quantum computer, however
they are typically not the limiting timescale for spin qubits in general.

2.6.2. D EPHASING
Depashing is the loss of phase coherence of a qubit. When a qubit is bought to a superposition
state, noise that couples to the quantization axis of the qubit will perturb the eigenstates of the
system, which will in turn cause the qubit to acquire phase at a different rate. This can happen
in the presence of a fluctuating nuclear spin bath via the hyperfine interaction, or due to charge
noise coupling to the spin eigenstates via the spin-orbit interaction. These processes destroy the
phase information of the qubit, and are currently the limiting mechanism for qubit scale up.

To measure dephasing, we define a timescale T 2 . This timescale is upper bounded by T 1 since

a relaxation process while in a superposition also does not preserve phase coherence, however
in spin qubits, it is always experimentally measured to be much lower. We define the the ’free-
induction-decay’ timescale T 2∗ to be the rate at which phase coherence p is lost. This is measured
by preparing the qubit in the |↓〉 state, then bringing it to the state 1/ 2 (|↓〉 + i |↑〉 by applying an
X π/2 pulse, waiting for a time, then applying a X −π/2 pulse to return it to the |↓〉 state, and reading
out the final spin state (See Figure 2.12b). Averaging these measurements for different wait times
results in an exponential decay with characteristic time T 2∗ . It is typical to apply the pulses with
a small but precise detuning, as it aids in the accuracy of the resulting fit. The resulting decay
becomes an exponential envelope modulated by a sinusoid of the form
n o
P |↓〉 = A sin (2π f τ + φ) exp −(τ)/T 2∗ (2.50)

where A represents the difference in readout value between the |↓〉 and fully decohered states, f is
the oscillation frequency of the sinusoid due to the detuning, φ allows for a phase offset, and τ is
the wait time. This pulse sequence is called a Ramsey sequence.

Noise acting on a qubit comes from different sources. One such noise source, which in itself is
very general and contains many comprising noise sources, is charge noise. Charge noise typically
follows a 1/ωα spectral density [73]. If the noise acting on a qubit is constant over some time, then
it is possible to recover phase coherence using refocussing pulses [74]. One pulse sequence that
 36 2. E LECTRON AND H OLE S PIN QUBITS IN S EMICONDUCTOR QUANTUM D OTS

(a)

|↑〉
t

(b)

|↓〉
t
Xπ/2 X−π/2

Figure 2.12: Spin relaxation and spin dephasing measurements. (a) A spin relaxation measurement occurs
by preparing the qubit in the spin-up state, then waiting a time t , before measuring it. Over time, noise in the
sytstem will cause the spin to relax to it’s ground state, resulting in an exponential decay as a function of t . The
decay constant of this exponential is called the spin relaxation time, or T 1 . (b) Spin dephasing rates can be
measured using Ramsey pulse sequence. The spin state is prepared in the |↓〉 state. Then, it is brought to the
equator of the Bloch sphere by an X π/2 pulse, where it is left to freely precess for a time t , before being projected
back onto the ẑ-axis of the Bloch sphere, and read out in the σz Pauli basis. Dephasing of the quantum state
due to various noise sources will result in imperfect state recovery, resulting in an exponential decay with
characteristic decay constant T 2∗ .
2.6. QUALIFYING A QUANTUM S YSTEM 37

(a)

N 2
|↓〉 t/4N t/2N t/2N t/4N

Xπ/2 Yπ X−π/2

(b) 100
N=1
N=2
N=4
2

N=8
|F| /(N )

N = 16
N = 32
2

N = 64

0 0.5 1 1.5 2 2.5 3 3.5 4

Figure 2.13: CPMG pulse sequence and filter functions. A CPMG pulse begins like a Ramsey pulse sequnce,
by bringing the spin state to the equator. However, the time t over which the spin would then freely precess, is
divided up into smaller time partitions, with equally spaced Yπ pulses separating them. These pulses invert the
parity of the noise acting on the qubit, such that for quasi-static noise, the effect is reversed and the qubit phase
is recovered. By dividing a time t up into smaller and smaller intervals with refocussing pulses, the ceiling of
noise frequency considered to be ’quasi-static’ is raised, resulting in an extended coherence time, quantified by
T 2,CPMG , the characteristic decay rate of the resulting exponential. (b) The CPMG pulse sequence is effectively
a filter function, that transmits certain frequencies of noise. The filter function transmission function is plotted
for different numbers of refocussing pulses N .

extends phase coherence is called a Hahn spin echo pulse sequence, and is performed by applying
an extra X π pulse half way through the wait time of a Ramsey sequence. The resulting decay will
yield a different time scale T 2H , longer than T 2∗ since the qubit is less sensitive to charge noise.
One can think of the pulse sequence as a filter function that reduces sensitivity of the qubit to low
frequency noise. For a Hahn echo sequence, this filter function is given by [75]

F H (ωτ) = 8 sin4 (πωτ/2) (2.51)

for a pulse sequence of length τ, and noise frequency ω.

This technique can be extended by adding additional refocussing pulses in the same time interval
at regular partitions. This is called a Carl-Purcell-Meiboom-Gill (CPMG) sequence (Fig. 2.13a),
and has the effect of further increasing the band of noise frequency that a qubit is insensitive to. It
has a filter function given by [75]

8 4 sin2 (Nπ ωτ/2)

F CPMG (Nπ , ωτ) = sin (ωτ/4) (2.52)
πω2 cos2 (ωτ/2)

where Nπ represents the number of refocussing pulses in the sequence. Several filter functions
are plotted in Fig. 2.12b for various Nπ .
 38 R EFERENCES

Using refocussing techniques, one can effectively extend quantum coherence far beyond it’s free
induction decay value. In general, For a spectral density of noise given by S(ω), the coherence of a
qubit undergoing a CPMG pulse sequence after a time τ will behave as [76, 77]
µ Z ∞
S(ω)
¶
C (Nπ , τ) = exp − d ω 2 F (Nπ , ωτ) . (2.53)
ω
2 −∞

As a final note, CPMG filter functions can also be used to narrow sensitivity to a particular fre-
quency of noise, making them useful tools to explore the various noise sources acting upon them.
We will see this in chapter 11, where we use CPMG pulse sequences to analyse and quantify the
effect of the 73 Ge nuclear spin on the coherence of hole spin qubits defined in Ge/SiGe.

2.6.3. G ATE F IDELITY

Noise sources that occur in semiconductor devices will limit the timescales over which a quantum
computer can be reliably operated, and therefore reduce their overall effectiveness. Additionally,
quantum computers can also be affected by systematic mistakes due to errors in the calibration
of the resonance frequency of the qubits, errors in the pulse times, cross-talk between qubits etc.
Both environmental noise sources and systematic errors reduce the effectiveness of a quantum
computers ability to execute quantum gates. It is useful to know the average error rates of these
quantum gates, since they effectively determine the achievable complexity of the quantum algo-
rithms that can be executed on a quantum processor.
One widely used technique to extract the error rates of a quantum processor is Randomized Bench-
marking [78]. Randomized benchmarking utilizes the Clifford group, which is a set of unitary op-
erations that map elements of the Pauli group to other elements of the Pauli group. In randomized
benchmarking, the general procedure goes according to:
• (1) prepare the quantum system in a known state |ψ0 〉,
• (2) perform a sequence of N − 1 operations selected at random from the Clifford group,
• (3) perform a recovery operation (also a Clifford) that returns the system to the original state
|ψ0 〉 and finally,
• (4) calculate the return probability of the final state 〈Ψ f |ΨI 〉 to the original state.
By repeating this sequence for increasing N , error rates will cause a reduction in the return prob-
ability, resulting in a decay from which the circuit level error rate can be calculated. Randomized
benchmarking produces an average error rate, however error rates may be different as different er-
ror channels exist for different quantum operations. Therefore it is important to benchmark with
a large enough random subset of Clifford gates such that the errors are effectively averaged out.
There are 24 single qubit Cliffords (C 1 ), 11540 two-qubit Cliffords (C 2 ), 92897280 three-qubit Clif-
fords (C 3 ). This can make it non-trivial to perform randomized benchmarking for larger numbers
of qubits, however these considerations don’t apply in this thesis, as only single qubit randomized
benchmarking is performed.

R EFERENCES
[1] Thornton, S. T. & Rex, A. F. Modern physics for scientists and engineers (Saunder’s College Pub.,
Fort Worth, 1993).

[2] Yu, P. Y. & Cardona, M. Fundamentals of semiconductors physics and materials properties
(Springer, 2010).

[3] Alexander, M. N. & Holcomb, D. F. The semiconductor-to-metal transition in n-type silicon.

Solid State Commun. 6, 355–358 (1968).
R EFERENCES 39

[4] Hanson, R., Kouwenhoven, L. P., Petta, J. R., Tarucha, S. & Vandersypen, L. M. Spins in few-
electron quantum dots. Reviews of Modern Physics 79, 1217–1265 (2007).

[5] Groth, C. W., Wimmer, M., Akhmerov, A. R. & Waintal, X. Kwant: A software package for
quantum transport. New J. Phys. 16, 063065 (2014).

[6] Van der Wiel, W. G. et al. Electron transport through double quantum dots. Reviews of Modern
2
Physics 75, 1–22 (2003).

[7] Volk, C. et al. Loading a quantum-dot based “Qubyte” register. npj Quantum Information 5,
1–8 (2019).

[8] Hensgens, T. et al. Quantum simulation of a Fermi-Hubbard model using a semiconductor

quantum dot array. Nature 548, 70–73 (2017).

[9] Levitt, M. H. Spin dynamics : basics of nuclear magnetic resonance (John Wiley & Sons, 2008).

[10] Wu, Y. & Yang, X. Strong-coupling theory of periodically driven two-level systems. Physical
Review Letters 98, 013601 (2007).

[11] Burkard, G., Loss, D. & DiVincenzo, D. P. Coupled quantum dots as quantum gates. Physical
Review B 59, 2070 (1999).

[12] Golovach, V. N., Borhani, M. & Loss, D. Electric-dipole-induced spin resonance in quantum
dots. Phys. Rev. B 74, 165319 (2006).

[13] Schrieffer, J. R. & Wolff, P. A. Relation between the Anderson and Kondo Hamiltonians. Phys.
Rev. 149, 491–492 (1966).

[14] Meunier, T., Calado, V. E. & Vandersypen, L. M. K. Efficient controlled-phase gate for single-
spin qubits in quantum dots. Phys. Rev. B 83, 121403 (2011).

[15] Sigillito, A. J., Gullans, M. J., Edge, L. F., Borselli, M. & Petta, J. R. Coherent transfer of quan-
tum information in a silicon double quantum dot using resonant SWAP gates. npj Quantum
Information 5, 1–7 (2019).

[16] Petit, L. et al. High-fidelity two-qubit gates in silicon above one Kelvin. arXiv 2007.09034
(2020).

[17] Xue, X. et al. Quantum logic with spin qubits crossing the surface code threshold. Nature
601, 343–347 (2022).

[18] Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 569, 532–536
(2019).

[19] Noiri, A. et al. Fast universal quantum control above the fault-tolerance threshold in silicon.
Nature 2022 601:7893 601, 338–342 (2021).

[20] Zheng, G. et al. Rapid gate-based spin read-out in silicon using an on-chip resonator. Nature
Nanotechnology 14, 742–746 (2019).

[21] Eizerman, J. M. et al. Single-shot read-out of an individual electron spin in a quantum dot.
Nature 430, 431–435 (2004).
 40 R EFERENCES

[22] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nature Nanotechnology 9, 981–985 (2014).

[23] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[24] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon. Nature 555,
2 633–637 (2018).

[25] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 1–10 (2017).

[26] Itoh, K. M. & Watanabe, H. Isotope engineering of silicon and diamond for quantum com-
puting and sensing applications. MRS Communications 4, 143–157 (2014).

[27] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nature Nanotechnology 13, 102–106 (2018).

[28] Sabbagh, D. et al. Quantum Transport Properties of Industrial Si 28 / Si O2 28. Physical Review
Applied 12, 014013 (2019).

[29] Steger, M. et al. Quantum information storage for over 180 s using donor spins in a 28Si
"semiconductor vacuum". Science 336, 1280–1283 (2012).

[30] Phillips, J. C. Band structure of silicon, germanium, and related semiconductors. Phys. Rev.
125, 1931–1936 (1962).

[31] Ando, T., Fowler, A. B. & Stern, F. Electronic properties of two-dimensional systems. Reviews
of Modern Physics 54, 437–672 (1982).

[32] Sham, L. J. & Nakayama, M. Effective-mass approximation in the presence of an interface.

Physical Review B 20, 734–747 (1979).

[33] Ruskov, R., Veldhorst, M., Dzurak, A. S. & Tahan, C. Electron g-factor of valley states in realistic
silicon quantum dots. Physical Review B 98, 245424 (2018).

[34] Friesen, M., Chutia, S., Tahan, C. & Coppersmith, S. N. Valley splitting theory of SiGe/Si/SiGe
quantum wells. Physical Review B 75 (2007).

[35] Borjans, F., Zajac, D. M., Hazard, T. M. & Petta, J. R. Single-Spin Relaxation in a Synthetic
Spin-Orbit Field. Physical Review Applied 11, 044063 (2019).

[36] Zajac, D. M., Hazard, T. M., Mi, X., Wang, K. & Petta, J. R. A reconfigurable gate architecture
for Si/SiGe quantum dots. Applied Physics Letters 106, 223507 (2015).

[37] Yang, C. H. et al. Orbital and valley state spectra of a few-electron silicon quantum dot. Phys-
ical Review B 86, 115319 (2012).

[38] Yang, C. H. et al. Spin-valley lifetimes in a silicon quantum dot with tunable valley splitting.
Nature Communications 4 (2013).

[39] Petit, L. et al. Spin Lifetime and Charge Noise in Hot Silicon Quantum Dot Qubits. Physical
Review Letters 121, 076801 (2018).

[40] Tagliaferri, M. L. et al. Impact of valley phase and splitting on readout of silicon spin qubits.
Physical Review B 97, 245412 (2018).
R EFERENCES 41

[41] Wuetz, B. P. et al. Atomic fluctuations lifting the energy degeneracy in Si/SiGe quantum dots
(2021).

[42] Dehollain, J. P. et al. Nanoscale broadband transmission lines for spin qubit control. Nan-
otechnology 24, 015202 (2013).

[43] Nolting, W. & Ramakanth, A. Quantum theory of magnetism (Springer Berlin Heidelberg, 2
2009).

[44] Scappucci, G. et al. The germanium quantum information route. Nature Reviews Materials
(2020).

[45] Winkler, R. SpinOrbit Coupling Effects in Two-Dimensional Electron and Hole Systems. 6
Inversion-Asymmetry-Induced Spin Splitting. Semiconductors 129, 69–129 (2003).

[46] Luttinger, J. M. & Kohn, W. Motion of Electrons and Holes in Perturbed Periodic Fields. Phys-
ical Review 97, 869–883 (1955).

[47] Baldereschi, A. & Lipari, N. O. Spherical model of shallow acceptor states in semiconductors.
Physical Review B 8, 2697–2709 (1973).

[48] Luttinger, J. M. Quantum theory of cyclotron resonance in semiconductors: General theory.

Physical Review 102, 1030–1041 (1956).

[49] Sammak, A. et al. Shallow and Undoped Germanium Quantum Wells: A Playground for Spin
and Hybrid Quantum Technology. Advanced Functional Materials 29, 1807613 (2019).

[50] Lodari, M. et al. Low percolation density and charge noise with holes in germanium. Materi-
als for Quanutm Technology 1, 011002 (2020).

[51] Wortman, J. J. & Evans, R. A. Young’s modulus, shear modulus, and poisson’s ratio in silicon
and germanium. Journal of Applied Physics 36, 153–156 (1965).

[52] Terrazos, L. A. et al. Theory of hole-spin qubits in strained germanium quantum dots. Physi-
cal Review B 103, 125201 (2021).

[53] Bir, G. L. & Pikus, G. E. Symmetry and Strain- Induced effects in Semiconductors. Translated
from Russian by P. Shelnitz (Wiley, New York, 1974).

[54] Fischetti, M. V. & Laux, S. E. Band structure, deformation potentials, and carrier mobility in
strained Si Ge, and SiGe alloys. Journal of Applied Physics 80, 2234–2252 (1996).

[55] Dresselhaus, G. Spin-orbit coupling effects in zinc blende structures. Physical Review 100,
580–586 (1955).

[56] Rashba, E. Properties of semiconductors with an extremum loop. 1. Cyclotron and combina-
tional resonance in a magnetic field perpendicular to the plane of the loop. Sov. Phys. Solid
State 2, 1109–1122 (1960).

[57] Marcellina, E., Hamilton, A. R., Winkler, R. & Culcer, D. Spin-orbit interactions in inversion-
asymmetric two-dimensional hole systems: A variational analysis. Physical Review B 95,
075305 (2017).

[58] Lawaetz, P. Valence-band parameters in cubic semiconductors. Physical Review B 4, 3460–

3467 (1971).
 42 R EFERENCES

[59] Nichele, F. et al. Characterization of spin-orbit interactions of GaAs heavy holes using a quan-
tum point contact. Physical Review Letters 113 (2014).

[60] Bulaev, D. V. & Loss, D. Electric dipole spin resonance for heavy holes in quantum dots. Phys-
ical Review Letters 98 (2007).
2 [61] Chesi, S. & Giuliani, G. F. Exchange energy and generalized polarization in the presence of
spin-orbit coupling in two dimensions. Physical Review B 75 (2007).

[62] Fischer, J., Coish, W. A., Bulaev, D. V. & Loss, D. Spin decoherence of a heavy hole coupled to
nuclear spins in a quantum dot. Physical Review B 78, 155329 (2008).

[63] Clementi, E. & Raimondi, D. L. Atomic screening constants from SCF functions. The Journal
of Chemical Physics 38, 2686–2689 (1963).

[64] Slichter, C. P. Electric Quadrupole Effects. 485–502 (Springer, Berlin, Heidelberg,

1990). URL https://link-springer-com.tudelft.idm.oclc.org/chapter/10.1007/
978-3-662-09441-9_10.

[65] Verkhovskii, S. V. et al. Quadrupole Effects on73 Ge NMR Spectra in Isotopically Controlled
Ge Single Crystals. Zeitschrift fur Naturforschung - Section A Journal of Physical Sciences 55,
105–110 (2000).

[66] Yusa, G., Muraki, K., Takashina, K., Hashimoto, K. & Hirayama, Y. Controlled multiple quan-
tum coherence of nuclear spins in a nanometre-scale device. Nature 434, 1001–1005 (2005).

[67] Froning, F. N. et al. Ultrafast hole spin qubit with gate-tunable spin–orbit switch functionality.
Nature Nanotechnology 16, 308–312 (2021).

[68] Struck, T. et al. Low-frequency spin qubit energy splitting noise in highly purified 28Si/SiGe.
npj Quantum Information 6, 1–7 (2020).

[69] Nakajima, T. et al. Coherence of a Driven Electron Spin Qubit Actively Decoupled from Qua-
sistatic Noise. Physical Review X 10 (2020).

[70] Simmons, C. B. et al. Tunable spin loading and T1 of a silicon spin qubit measured by single-
shot readout. Physical Review Letters 106 (2011).

[71] Vukušić, L. et al. Single-Shot Readout of Hole Spins in Ge. Nano Letters 18, 7141–7145 (2018).

[72] Hu, Y., Kuemmeth, F., Lieber, C. M. & Marcus, C. M. Hole spin relaxation in Ge-Si core-shell
nanowire qubits. Nature Nanotechnology 7, 47–50 (2012).

[73] Paladino, E., Galperin, Y., Falci, G. & Altshuler, B. L. 1/ f noise: Implications for solid-state
quantum information. Reviews of Modern Physics 86, 361–418 (2014).

[74] Viola, L. & Lloyd, S. Dynamical suppression of decoherence in two-state quantum systems.
Physical Review A - Atomic, Molecular, and Optical Physics 58, 2733–2744 (1998).

[75] Cywiński, Ł., Lutchyn, R. M., Nave, C. P. & Das Sarma, S. How to enhance dephasing time in
superconducting qubits. Physical Review B 77, 174509 (2008).

[76] Uhrig, G. S. Exact results on dynamical decoupling by π pulses in quantum information

processes. New Journal of Physics 10, 083024 (2008).
R EFERENCES 43

[77] Uhrig, G. S. Keeping a quantum bit alive by optimized π-pulse sequences. Physical Review
Letters 98, 100504 (2007).

[78] Knill, E. et al. Randomized benchmarking of quantum gates. Physical Review A - Atomic,
Molecular, and Optical Physics 77 (2008).
2
 3
Q UANTUM D OT A RRAYS IN S ILICON
AND G ERMANIUM

Electrons and holes confined in quantum dots define excellent building blocks for quantum emer-
gence, simulation, and computation. Silicon and germanium are compatible with standard semi-
conductor manufacturing and contain stable isotopes with zero nuclear spin, thereby serving as
excellent host for spins with long quantum coherence. Here, we demonstrate quantum dot arrays
in silicon metal-oxide-semiconductor (SiMOS), strained silicon (Si/SiGe), and strained germanium
(Ge/SiGe). We fabricate using a multi-layer technique to achieve tightly confined quantum dots
and compare integration processes. While SiMOS can benefit from a larger temperature budget and
Ge/SiGe can make ohmic contact to metals, the overlapping gate structure to define the quantum
dots can be based on a nearly identical integration. We realize charge sensing in each platform, for
the first time in Ge/SiGe, and demonstrate fully functional linear and two-dimensional arrays where
all quantum dots can be depleted to the last charge state. In Si/SiGe, we tune a quintuple quantum
dot using the N+1 method to simultaneously reach the few electron regime for each quantum dot.
We compare capacitive cross talk and find it to be the smallest in SiMOS, relevant for the tuning of
quantum dot arrays. We put this these results into perspective for quantum technology and identify
industrial qubits, hybrid technology, automated tuning, and two-dimensional qubit arrays as four
key trajectories that, when combined, enable fault-tolerant quantum computation.

Parts of this chapter have been published in Appl. Phys. Lett. 116 (8), 080501 [1].

45
 46 3. QUANTUM D OT A RRAYS IN S ILICON AND G ERMANIUM

Quantum dots have been a leading candidate for quantum computation for more than two
decades [2]. Furthermore, they have matured recently as an excellent playground for quantum
simulation [3] and have been proposed for the design of new states of matter [4, 5]. Pioneering
studies in group III-V semiconductors led to proof-of-principles including the coherent control
of electron spins [6, 7], rudimentary quantum simulations [8], and signatures of emergent states
such as Majorana fermions [9]. The group IV semiconductors silicon and germanium have the
opportunity to advance these concepts to a practical level due to their compatibility with stan-
dard semiconductor manufacturing [10] and the availability of isotopes with zero nuclear spin,
increasing quantum coherence for single spins by four orders of magnitude [11]. Furthermore,
3 heterostructures built from silicon and germanium also offer a large parameter space in which to
engineer novel quantum electronic devices [12–14].

An initial advancement towards silicon quantum electronics [12] was the design of an inte-
gration scheme based on overlapping gates to build silicon metal-oxide-semiconductor (SiMOS)
quantum dots [15]. This technique was later adopted in strained silicon (Si/SiGe) [16] and refined
by incorporating metals with small grain size and atomic layer deposition (ALD) for layer-to-layer
isolation [17] and to enable tunable coupling between single electrons in SiMOS [18]. These devel-
opments in fabrication have led to a great body of results, including high-fidelity qubit operation
[19, 20] and two-qubit logic [21–23]. Controlling holes in silicon has been more challenging due to
type II band alignment in strained silicon, limiting experiments to SiMOS [24–26]. Strained germa-
nium on the other hand [13, 27, 28] exhibits type I band alignment and is thereby a viable platform
in which not only electrons but also holes with light effective mass [29] can be confined [30] and
coherently controlled [31].

Here, we present the fabrication and operation of quantum dots in silicon and germanium,
in linear and two-dimensional arrays. We compare integration schemes and find that while each
platform has unique aspects and opportunities, the core fabrication of overlapping gates defining
the nano-electronic devices is remarkably similar, thereby further accelerating the overall progress
in group IV semiconductor quantum dots. In each case, fabrication starts from a silicon substrate,
and integration is compatible with standard semiconductor technology. We leverage off the ohmic
contact between quantum dots in Ge/SiGe and metals [32] to avoid the need for ion implantation
and to provide means for novel hybrid systems. We show stability diagrams obtained by charge
sensing and report double quantum dots in SiMOS, Si/SiGe, and Ge/SiGe that can be depleted to
the last charge state. Fabrication is most demanding in SiMOS due to requirements on feature
size, but we also find that the resulting devices have the smallest cross capacitance, simplifying
tuning and operation. We put these results in perspective and outline a road map for quantum
technology based on group IV semiconductor platforms.

Figure 3.1a schematically shows the SiMOS, Si/SiGe, and Ge/SiGe wafer stacks used in this
study. The SiMOS 300 mm wafers are grown in an industrial complementary metal oxide semi-
conductor (CMOS) fab [14, 18, 33], while the Si/SiGe and Ge/SiGe four-inch wafers are grown us-
ing an RP-CVD reactor (ASM Epsilon 2000) [13]. Each platform is grown on a p-type natural Si
wafer. The SiMOS structure consists of 1 µm intrinsic natural silicon (i Si) followed by 100 nm 28 Si
(800 ppm purity) and 10 nm SiO2 [14]. The Si/SiGe heterostructure begins with a linearly graded
Si1−x Gex layer, where x ranges from 0 to 0.3. A relaxed Si0.7 Ge0.3 layer of 300 nm lies below the 10
nm 28 Si (800 ppm purity) quantum well which itself is separated from the 2 nm Si capping layer
by a second 30 nm relaxed Si0.7 Ge0.3 spacer layer. The Ge/SiGe wafer stack starts with 1.4 µm of
Ge and 900 nm of reverse graded Si1−x Gex where x ranges from 1 to 0.8. This lies below a 160 nm
Si0.2 Ge0.8 spacer layer, a 16 nm Ge quantum well under compressive strain, a second Si0.2 Ge0.8
layer of 22 nm and finally a thin Si cap of 1 nm [13].
3.1. M OBILITY AND S TACK BY P LATFORM 47

(a)
Si Cap 2 nm Si Cap 1 nm
SiO2 10 nm Si0.7Ge0.3 30 nm Si0.2Ge0.8 22 nm
28
Si 10 nm Ge 16 nm
28
Si 100 nm
Si0.7Ge0.3 300 nm Si0.2Ge0.8 160 nm
i
Si 1 μm Si1-xGex Si1-xGex
Ge
nat
Si nat
Si nat
Si
(b)
106
3
Ge/SiGe [12]
Si/SiGe
SiMOS [13]
Mobility (cm2/Vs)

105

104

103
1011 1012
-2
Density (cm )

Figure 3.1: Wafer stack schematics and mobility as a function of carrier density. (a) From left to right, SiMOS,
Si/SiGe, and Ge/SiGe wafers stacks. For SiMOS, a 28 Si epilayer with 10 nm thermal oxide is grown on a 1
µm intrinsic natural Si buffer layer. The Si/SiGe heterostructure consists of a 1.5 µm linearly graded SiGe
layer, a relaxed 300 nm SiGe spacer, a 10 nm 28 Si quantum well, a 30 nm SiGe spacer, and a 2 nm Si cap. The
Ge/SiGe heterostructure consists of 900 nm reverse graded SiGe layer, a relaxed 160 nm SiGe spacer, a 16 nm Ge
quantum well, a 22 nm SiGe spacer, and a 1 nm Si cap. (b) Mobility as a function of carrier density measured
in each platform. For Ge/SiGe, the peak mobility is greater than 5×105 cm2 /Vs and the critical density is
1.15×1011 cm−2 [13]. The same measurements for Si/SiGe wafers give a peak mobility of 1×105 cm2 /Vs and a
critical density of 1.2×1011 cm−2 . SiMOS data taken from [14] shows a mobility of 1×104 cm2 /Vs and a higher
critical density of 2.5×1011 cm−2 .

3.1. M OBILITY AND S TACK BY P LATFORM

Figure 3.1b shows a carrier mobility versus density characterization of the three platforms. Hall
bar structures were fabricated on coupons cut from the center of each wafer. Maximum mobility
and critical density are extracted at 1.7 K. SiMOS 300 mm processed wafers give a peak mobility
value of 1×104 cm2 /Vs, as well as a critical density of about 1.75×1011 cm−2 as shown in another
work [14]. At higher densities, SiMOS mobilities fall off due to surface roughness scattering effects
[34–36]. In Si/SiGe, we observe a lower critical density of 1.2×1011 cm−2 and a significantly higher
maximum mobility exceeding 1×105 cm2 /Vs . Similar studies conducted on natural Si/SiGe grown
in an industrial CMOS fab yielded mobilities of 4.2×105 cm2 /Vs [37]. This quality improvement
observed by moving toward industrial CMOS fab also suggests encouraging prospects for Ge/SiGe,
already exhibiting a high maximum mobility of 5×105 cm2 /Vs and critical density of 1.15×1011
cm−2 despite being grown in an academic cleanroom via RP-CVD [13].
 48 3. QUANTUM D OT A RRAYS IN S ILICON AND G ERMANIUM

Temperature SiMOS Si/SiGe Ge/SiGe

Thermal Budget 1000 oC 750 oC 500 oC

Ohmics SiMOS Si/SiGe Ge/SiGe

Dopant P+ P+ -
Metal Ti:Pt Ti:Pt Al
Annealing 1000 oC, 30 s 700 oC, 30 s 300 oC, 1h

Field Oxide SiMOS Si/SiGe Ge/SiGe

3 Oxide Window  (1) (1)

Stack All Platforms

Bondpads SiN 150 nm SiO2 SiGe SiGe
Gate Layer Ti:Pd N Si Ge
Si SiGe SiGe
Isolation Al2O3 ALD

Control SiMOS Si/SiGe Ge/SiGe

mw-Antenna   -
Micromagnet   -
Spin-Orbit (2) - 

Figure 3.2: Overview of fabrication scheme for SiMOS, Si/SiGe and Ge/SiGe quantum dots. The thermal budget
of each material prior to gate stack deposition is estimated based on the limiting mechanism of each platform
as discussed in the text. In all cases, gates are fabricated from Pd metal with a thin (3 nm) Ti adhesion layer,
with layer-to-layer isolation performed via atomic layer deposition (ALD) of Al2 O3 . These two steps can be
looped at appropriate thicknesses to form the multi-layer structure. (1) We note the possibility of such an etch
exists for the remaining platforms in the case of a Schottky gate architecture (2) We note that spin-orbit based
driving of electrons in SiMOS has been demonstrated for singlet-triplet qubits [38] and proposed for single
spin qubits [39].

3.2. I NTEGRATION S CHEME

Figure 6.1 summarises the integration scheme utilized for each platform. The thermal budget is
estimated based on the respective limiting mechanisms. For SiMOS, thermal processing is lim-
ited by the self-diffusion of natural silicon from the substrate into the 28 Si epilayer. From the self
diffusion constants measured by Bracht et al. [40], we estimate the point at which the residual
29 Si concentration within 1 nm of the Si-SiO interface increases by 1 ppm occurs at 1000 o C for
2
time scales above 1 hour, for furnace anneals in a pure argon atmosphere. Consequently, this al-
lows for extensive thermal treatment and annealing of samples. This is highly advantageous, as we
have observed that a 15 minute anneal in forming gas at 400 o C after the deposition of every gate
layer greatly improves the quality of metallic features with smallest critical dimensions (see Sup-
plementary material section Ia for detailed comparison). In addition, a final end-of-line anneal is
conducted to eliminate processing damage at 400 o C in forming gas for 30 minutes. In the cases
of Si/SiGe and Ge/SiGe, the thermal budget is limited by strain relaxation of the quantum wells,
thus the maximum processing thermal budget is given qualitatively by the temperature at which
the quantum wells were grown. This is 750 o C for strained Si and 500 o C for strained Ge [13].
The design of ohmic contacts is tailored to the specific requirements of the device. For both
Si platforms, ohmic contact is made via high fluence P ion implantation followed by evapora-
tion of Ti:Pt metallic contacts, creating n++ doped, low resistance channels. The oxide (SiO2 ) is
3.2. I NTEGRATION S CHEME 49

Ohmic Dielectric Screening Plunger Barrier

(a) (b) (c)
100 nm 100 nm 200 nm

(d) (f)
(e)
SiO2 SiGe
SiGe
Ge
Si
Si
3
Depth

SiGe SiGe

Energy

Figure 3.3: Scanning electron microscope images and corresponding device schematics with band bending
diagrams, substrate and gate stack for each of the devices. Dotted lines in (a-c) indicate the cross-section
through the quantum dot channel illustrated in (d-f) respectively, and crossed boxes indicate gates that overlap
with implanted regions to form ohmic contacts. The plunger gates (yellow), the barrier gates (blue) and the
screening gates (red) define the quantum dots. (a) SiMOS triple quantum dot linear array. Two SETs function
as charge sensors and as reservoirs for the quantum dots on either side of the array (b) Si/SiGe quintuple
quantum dot linear array. Two SETs (top) are used for charge sensing. (c) Ge/SiGe (2x2) quadruple quantum
dot array. Each quantum dot is tunnel coupled to a metallic lead (green). Measurement can be performed
in transport, or using charge-sensing by forming a sensor quantum dot under one channel to sense a double
quantum dot in the opposite channel. (d,e,f) Cross-section and bandstructure of metal, dielectric (black) and
semiconductor (d) SiMOS, (e) Si/SiGe and (f) Ge/SiGe.

etched locally directly before metal deposition using buffered hydrofluoric acid (BHF). In the case
of Si/SiGe, stray capacitance is minimized to ensure maximum power is dissipated in the variable
resistance of the sensing quantum dot for RF-readout. Germanium can make direct ohmic con-
tact to metals [32], avoiding the need for implants. We deposit Al and anneal at 300 o C for 1 hour
in vacuum to assist in Al diffusion into the quantum well. The Al ohmic contact is defined close
to the quantum dots, resulting in a very low resistance channel ideally suited for RF circuits and
enabling a tunnel contact that can even be made superconducting [41]. The implementation does
however lower the thermal budget of further processing.

Fabrication of each device utilizes a titanium-palladium (Ti:Pd) gate stack with 3 nm of Ti de-
posited for each layer to assist with adhesion. Pd makes a good gate metal due to its small grain
size [17]. Unlike the commonly used material Al, Pd does not self-oxidise and ALD can be used to
define sharp dielectric interfaces. For the SiMOS and Si/SiGe devices shown in Fig. 3.3, we utilize a
three layer gate stack that we refer to as the screening layer, the plunger layer and the barrier layer.
In order to assist climbing of overlapping gate features, the initial layer is deposited at 20 nm total
thickness, while subsequent layers at 40 nm. The layers are isolated from one another via ALD
of Al2 O3 at 7 nm thickness. We measure the breakdown electric field of the Al2 O3 to be greater
than 6 MV/cm, allowing potentials of greater than 4 V to be applied between adjacent gates. To
leverage off the high quality industrial CMOS fabrication facilities, we begin fabrication of SiMOS
devices on wafers including a 10 nm SiO2 oxide already grown. To further reduce the likelihood of
leakage from gate to substrate, we first grow a thick 10 nm Al2 O3 blanket layer over the entirety of
the substrate. Advantageously, one can etch Al2 O3 on thermally grown SiO2 selectively, allowing
the definition of a 20x20 µm2 area where the quantum dot system is defined, which we have mea-
sured to significantly reduce low-frequency drifts deduced from charge occupation stability [42]
(see Supplementary material section Ib for comparison).
 50 3. QUANTUM D OT A RRAYS IN S ILICON AND G ERMANIUM

(a) Tr (b) (c)

Cu B34
P1 P2 P3 SET
P4 P3

Cu SHT
B41 B23
SET
B12 B23 B34 B45 P1 P2
B12 B23 P1 P2 P3 P4 P5 B12
Cl
Tl Cl
-100 0 100 2.7 2.8 2.9 -40 0 40
I (pA) f (a.u.) I (pA)
1950 1580

-790 (1,0) (0,0)

(mV)
(mV)

(mV)
1900 (0,1)
3 (1,1)
(mV)

V P1V(mV)

V P2V(mV)
1550
(0,1) (1,1)
V P2VP2

P2

P2
-800
(1,1)
(0,1)
1850
(0,0) 1520
(1,0) -810
(0,0) (1,0)
1800
3200 3300 3400 3500 3600 1520 1550 1580 -960 -950 -940 -930 -920
V
VP1
P1
(mV)
(mV) (mV)
VVP1P2(mV) V P1
P1
(mV)
(mV)

Figure 3.4: Charge stability diagrams of double quantum dots depleted to the single electron/hole regime for
the three platforms. (a) SiMOS double quantum dot. Charge addition lines under P1 are not visible due to low
tunnel rate from reservoir. Map taken at 0.44 K using lock-in charge sensing. The excitation is placed on the
inter-dot gate B12 . (b) Si/SiGe double quantum dot formed under the first two plungers, sensed by the nearest
charge sensor via RF-reflectometry utilizing a resonant LC circuit at 84 MHz. Here, the plunger gate voltages
are in virtual gate space correcting for weak cross capacitive coupling. (c) Ge/SiGe depleted to the single hole
regime. A large single quantum dot is formed under P3 , B34 and P4 , by adjusting the tunnel barrier voltage B34 ,
and is used to sense a double quantum dot under P1 and P2 . The lock-in excitation is placed on the inter-dot
tunnel barrier B12 .

The final deposition step is the qubit control layer. The spin-orbit coupling for holes in ger-
manium enables qubit operation by simply applying microwave pulses to the quantum dot gates
[31, 44] and no further processing is required. In silicon, qubit driving can be realized by inte-
grating on-chip striplines [7, 11], which we fabricate using Al or NbTiN, or micromagnets [45],
which we integrate using Ti:Co. Quantum dots in Si/SiGe generally have a larger and more mobile
electron wave function as compared to SiMOS and thereby benefit most from a micromagnet in-
tegration for fast qubit driving.

A schematic of each material and associated device is shown in Fig. 3.3 and labelling of the rel-
evant gates are shown in Fig. 3.4. The SiMOS device is a three-layer, triple quantum dot structure
with dedicated plungers (P1−3 ), inter-dot barriers (B12 , B23 ) and dot-reservoir barriers (Tl , Tr ).
Two large metallic gates (Cl , Cu ) deposited in the initial layer and kept at constant potential
serve to confine the quantum dots in one lateral dimension. They also serve to screen charge noise
resulting from fluctuations near the quantum dot array.
Two single electron transistors (SETs) are positioned at either side of the quantum dot array,
and function as charge sensors for spin and charge readout. The Si/SiGe device is a quintuple
quantum dot linear array written in three layers utilizing a similar architecture to that of the SiMOS
device. The quantum dot array contains five plunger gates (P1−5 ) with inter-dot barriers (B12−45 )
and dot-reservoir barriers. Here too, the quantum dots are confined laterally and screened from
charge noise by two confinement gates. Two SETs are positioned parallel to the quantum dot chan-
nel. The Ge/SiGe device is a 2x2 quadruple quantum dot array written in two layers. Gates (P1−4 )
are positioned anti-clockwise in the array and define the potential of the quantum dots. Each pair
3.2. I NTEGRATION S CHEME 51

(a) (d) (h)

2150
-1000
1470

V P4 (mV)
V P1 (mV)
2110 -1050
V Tl (mV)

1400
2070 -1100

1550 1600 1650 -1100 -1050 -1000

1330 V P2 (mV) V P1 (mV)
(e) (i)
1550 1850 2150 1700
V P1 (mV) -1140

V P2 (mV)

V P1 (mV)
(b) 1660
-1180
3
2300
1620
-1220
V B23 (mV)

2150 1560 1600 1640 -1150 -1100 -1050

V P3 (mV) V P2 (mV)
(f) (j)
1700
2000 -1150
V P3 (mV)

V P2 (mV)
1660
-1200
1800 2100 2400
V P2 (mV)
1620 -1250
(c)
1600 1650 1690 -1250 -1200 -1150
1530
V P4 (mV) V P3 (mV)
(g) (k)
1710 -1250
V Tr (mV)

1450
V P4 (mV)

V P3 (mV)

-1300
1660

1380 -1350
1610
1550 1850 2150 1600 1640 1680 -1250 -1200 -1150
V P3 (mV) V P5 (mV) V P4 (mV)

Figure 3.5: Quantum dot arrays in SiMOS, Si/SiGe and Ge/SiGe. (a-c) SiMOS triple quantum dot device stability
diagrams. Each single quantum dot is formed under its respective plunger gate upon which an excitation is
placed for lock-in charge sensing. Each quantum dot is depleted to the single charge state. (b) Shows the
crossing of the adjacent quantum dot under P3 , through which the quantum dot is loaded. (d-g) Si/SiGe double
quantum dots tuned up sequentially using the N+1 method [43] to the single electron regime. True plunger
gate voltages are plotted, though virtual gates are swept containing small corrections to adjacent barriers and
plungers. Each double quantum dot pair is sensed using RF-reflectometry. The same SET is used for readout
in each case, as indicated by the relative signals as each double quantum dot pair is formed farther from the
charge sensor. (g) The data has been filtered to remove 50 Hz background noise for data clarity. (h-k) Ge/SiGe
2x2 array double quantum dots formed in each possible configuration. In each case, a charge sensor is formed
in the parallel channel by raising the inter-dot coupling to form a large single quantum dot with high hole
occupation. Each charge stability diagrams shows RF-sensing of double quantum dots depleted to the last
hole occupancy, in the low tunnel-coupled regime.

of adjacent quantum dots share a barrier gate (B12−41 ) capable of tuning inter-dot tunnel cou-
pling. Coupling of each quantum dot to its reservoir can be controlled via a barrier gate. This
device can be operated as a quadruple quantum dot system in transport mode, but for the present
work we intentionally tune the inter-dot barrier to form a single hole transistor (SHT) along a dot
channel that we subsequently use for charge sensing of the double quantum dot along the oppo-
site channel. For more information about device specific fabrication, see Supplementary material
 52 3. QUANTUM D OT A RRAYS IN S ILICON AND G ERMANIUM

section II.

80 nm
1
80 nm (a) P1 B12 P2 B23 P
3

0.5 24 nm 17 nm

SiMOS
0
Normalized Capacitive Coupling

3 1
P1 B 12 P2 B 23 P 3
80 nm
80 nm (b) P1 B12 P2 B
B23
2 P
3

0.5 51 nm 44 nm

Si/SiGe
0
P1 B 12 P2 B 23 P 3
200 nm
1
(c) P1 B12 P2 RB2
200 nm
0.5 43 nm 33 nm

Ge/SiGe
0
P1 B 12 P2 RB2
Gate Name

Figure 3.6: Cross capacitance to neighbouring gates of a quantum dot in the single charge occupancy regime
under gate P1 in each platform. For SiMOS (a), we observe immediate falloff of cross coupling due to the tight
quantum dot confinement present in SiMOS devices. Here the inter-dot pitches matches that of Si/SiGe at 80
nm. For Si/SiGe (b), we see significant cross-coupling between adjacent plungers and barrier gates. Here the
plunger gates are written before the barrier layer and have an inter-dot pitch of 80 nm. Ge/SiGe (c) reveals as
expected a slower fall-off of cross coupling. We attribute this to the larger plunger gate design, made possible
by lower hole effective mass. In this case, the plunger gates P1 and P2 are written in the layer above the barrier
gates B12 and RB2 , decreasing coupling to their respective quantum dots. The plunger to plunger pitch is 200
nm. Each cross-sectional cartoon shows plunger pitch and distance between each relevant gate layer to the
center of the quantum well.

3.3. S INGLY O CCUPIED Q UANTUM D OT A RRAYS

To demonstrate the success of this largely unified integration scheme, we show that we can cre-
ate stable quantum dots in each platform. Figure 3.4 shows charge stability diagrams for tunnel-
coupled double quantum dots, measured by performing charge sensing. Lock-in techniques are
used in the case of SiMOS and Ge/SiGe, where an excitation is placed on an inter-dot barrier gate
B12 in each case, and the trans-conductance of the source-drain channel is measured. We use
compensation to remain at a sensitive point of the SET/SHT Coulomb peaks [46]. In the case of
Si/SiGe, charge readout is performed using RF-reflectometry techniques. A 3 µH kinetic inductor
is bonded to the sample source which forms a resonant LC circuit when combined with parasitic
capacitance to ground. In each case we measure a charge stability diagram and show that we can
deplete down to the (0,0) electron/hole charge configuration. This is done by ensuring that the
load rate of each quantum dot is sufficiently high, and depleting to the first charge state of each
3.4. C ROSS C APACITANCE 53

quantum dot by sweeping the associated plunger gate, until no further charge transition lines are
detected (for details in tuning to the last state see our previous works [18, 47]). While operation in
the single electron regime in silicon has been routinely achieved before, this work shows the first
demonstration of the single hole regime using charge sensing of holes in Ge/SiGe. We attribute
the slight difference in slope of the first and second charge addition lines in Fig. 3.4c to a shift in
the position of the quantum dot relative to the inter-dot tunnel barrier.

In Fig 3.5. we demonstrate that quantum dots can be formed under each dedicated plunger
gate. For Fig. 3.5 (a-c), in each SiMOS quantum dot, lock-in charge sensing is performed by placing
an excitation on the respective plunger gates, while trans-conductance in the nearby SET channel 3
is measured. In each case, the first charge transition is visible. For quantum dots formed under
plungers P2−3 , electron loading is from the right SET which constitutes a reservoir. For the quan-
tum dot under P1 , loading is from the left SET via the gate Tl . The Si/SiGe quintuple quantum dot
system in Fig 5(d-g) is tuned using the N+1 strategy [43], reaching the few-electron regime simul-
taneously for all quantum dots. In Fig. 3.5 we show stability diagrams, in each of which we scan
two virtual plunger gates which allow to controllably load a single electron into each quantum
dot. Double quantum dots are formed between each set of adjacent plungers, and sensed using
RF-reflectometry like in Fig. 3.4b using the left SET for all configurations. As expected, observable
signal from charge transition lines fades as the quantum dot pairs are formed farther away from
the SET. The derivative of the reflected signal is plotted, and shows the (0,0) charge occupancy for
each charge stability diagram. For every double quantum dot, loading occurs via the left accumu-
lation gate, leading to latching effects and low tunnel rates in the quantum dots formed farther
away from the reservoir. Figure 3.5 (h-k) shows charge sensing operation of the 2x2 quantum dot
array fabricated in Ge/SiGe. In each case, a sensing quantum dot is formed in the channel parallel
to the double quantum dot by opening the inter-dot barrier such that a large single quantum dot
is formed. In the opposite channel, the inter-dot barrier is closed, forming a double quantum dot
system in the low tunnel coupled regime.

3.4. C ROSS C APACITANCE

A significant challenge for larger quantum dot arrays will manifest in tuning. The presence of large
capacitive crosstalk in GaAs has led to development of virtual gates and approaches to tune larger
systems [43, 48]. To assess the relevance of these approaches for silicon and germanium structures
we measure cross capacitance as show in Fig. 3.6. To obtain the cross coupling, we measure the
slope of the charge addition lines with respect to each gate and normalize by a cross coupling of
unity for the plunger gate associated with the respective quantum dot. Each slope is taken for the
first charge transition and in the low tunnel-coupled regime. In SiMOS, cross coupling is almost
negligible, as expected from quantum dots located only 17 nm (10 nm SiO2 and 7 nm Al2 O3 ) below
the electrostatic gates. This compares favourably to the cross coupling observed in Si/SiGe, where
falloff is significantly slower despite sharing equal gate pitch to the SiMOS array. While the cross
coupling in the Ge/SiGe system is the largest and extends over multiple neighbouring gates, it still
falls off significantly faster than quantum dots defined in GaAs [43]. For Ge/SiGe, we also observe
that the barrier gates have a relatively stronger lever arm to the quantum dots as compared to the
plunger gates, due to definition in lower layers of the multi-layer stack. Summarizing, we con-
clude that for SiMOS tuning is most straightforward considering capacitive cross talk only, while
each platform compares favourably to GaAs using an open gate layout [43].
With the ability to fabricate quantum dot devices in several group IV based platforms using a uni-
fied fabrication scheme, we look ahead to identify future challenges and opportunities related to
 54 3. QUANTUM D OT A RRAYS IN S ILICON AND G ERMANIUM

quantum information processing with semiconductor quantum dot spin qubits.

3.5. F UTURE P ERSPECTIVES

Moving forwards, we identify four focus areas for the community that through shared co-development,
will launch the field into new and practical ground.

Most quantum devices are fabricated in academic cleanrooms, where the turnaround and
feedback from measurement to design and fabrication is fast. However as designs for various
3 types of quantum dot devices converge, an opportunity exists to leverage off the excellent ma-
terial quality [14, 37, 49] and processing facilities of industrial fabrication lines. Devices fabricated
on industrially-grown 300 mm wafers have led to CMOS fab spin qubits [26], tuneable tunnel cou-
pling between single electrons in SiMOS [18] and two-qubit gate operations beyond one Kelvin
[50]. Furthering symbiotic partnerships with industry may prove highly beneficial for the develop-
ment of uniform quantum dots. The adoption of group IV based semiconductor platforms beyond
SiMOS such as strained Si and Ge as well as full 300 mm device fabrication lines would accelerate
progress in the field of semiconductor quantum dot based quantum computing, like it has in other
fields[51].

Many quantum systems have been studied as qubit candidates for quantum information pro-
cessing. It has also become clear that each of these quantum systems hold specific properties
suited to the various requirements of quantum computation[52]. As a result, emerging research
has targeted the combination of qubit implementation to leverage off specific advantages and
improve qubit quality. These hybrid directions are extensive, including the coupling of spin to
light allowing for long range interactions as has been shown on silicon based platforms[53–56],
or the coupling of spins to systems that reliably conserve the quantum state, such as topolog-
ically protected qubits[57–60]. Here, holes in Ge/SiGe make an excellent candidate for hybrid
spin-Majorana qubits, thanks to the Fermi level pinning at the valence band, allowing for tunnel-
coupled contacts to superconductors [31]. An important milestone towards demonstrating such
a hybrid qubit in Ge/SiGe will be to achieve hard gap superconductivity. This has already been
demonstrated in Ge/SiGe core shell nanowires[61, 62], thus providing scope for planar structures.
Such a hard gap would be the first step toward defining isolated zero energy states, key in many
proposals for hybrid technology [63].

As quantum devices grow in number of physical qubits, so too do the complexities related to
tuning them. As a result, a great body of work on the automated tuning of quantum devices has
emerged in the last few years in an attempt to address this concern. Due to the extremely low dis-
order of the material, these efforts were pioneered in GaAs based quantum dots, demonstrating
automated tuning to the single electron regime [48, 64] and controllable interdot tunnel coupling
[65]. However quantum dot arrays have also emerged more recently in Si/SiGe [66] and computer
automated single electron regime tune-up protocols therein[67, 68]. Moreover, with the demon-
strations of SiMOS, Si/SiGe, and Ge/SiGe quantum dots in this work, further development of au-
tomated tuning protocols will be necessary for the exploration of larger quantum dot systems. In
particular the automated tuning of interdot tunnel couplings and protocols for 2D arrays will be
critical. Furthermore, high fidelity operation of qubits in large scale quantum devices will require
precise operation at exact exchange interaction, resonance frequencies and Rabi frequencies, ac-
counting for potential drifts in these parameters over time. Tune up protocols will therefore have
to go beyond charge state control, handling qubit operation also.
3.6. C ONCLUSION 55

Scale up of the number of qubits on a quantum device requires the design and implementa-
tion of extensible two-dimensional qubit arrays. However, the wiring and fanout for each qubit
at large numbers is impractical and there is a need to engineer architectures that obey Rent’s rule
[69]. Additionally, the limited cooling power of dilution refrigerators at mK temperatures poses
a serious challenge for the scalability of quantum systems [10]. As a result, proposals for shared
control using crossbar architectures [70, 71] and on chip classical electronics [10] have been put
forward, as well as work on the operation of qubits at high temperatures[50, 72] to mitigate the
cooling power requirements of dilution refrigerators. 2D scalability will also require improved op-
eration of larger quantum devices. This includes the ability to tune all quantum dot couplings and
to shuttle spin states coherently around a lattice, placing strict requirements on the uniformity of
quantum dots. This positions Si/SiGe and Ge/SiGe as favourable platforms due to their very low
3
disorder. A milestone in 2D scalability would be the routine ability to reach single charge occu-
pancy in arbitrary quantum dots using the same cross-capacitance matrix for each quantum dot
as this would enable shared control for scalable quantum operation as is proposed in crossbar ar-
chitectures [71]. We observe Si/SiGe double quantum dots that can be tuned to the (1,1) charge
state using identical plunger gates (e.g. Fig. 3.4b), but further progress is essential to enable shared
control in large arrays. Solutions to these outstanding hurdles will be crucial to further develop ex-
tensible qubit unit cells and therefore scale quantum devices into practically useful regimes.

3.6. C ONCLUSION
In conclusion we presented a cross-platform integration scheme for multi-layer quantum dot ar-
rays in group-IV semiconductor hosts. We fabricated linear and 2D arrays of quantum dots in
the group IV platforms SiMOS, Si/SiGe and Ge/SiGe. We demonstrated single electron and hole
occupancy in double quantum dots confirmed by charge sensing. We showed stable quantum
dots under each plunger gate in a SiMOS triple quantum dot linear array, depleteable to the fi-
nal charge state. In Si/SiGe, we demonstrated tune-up of a quintuple quantum dot array utilizing
the N+1 method, successfully reaching the few electron regime in each quantum dot simultane-
ously. Moreover, we formed and sensed double quantum dots in the single hole regime in each
configuration of a 2x2 quadruple quantum dot array in Ge/SiGe. We furthermore compared the
capacitive cross talk between quantum dots and gates. We find that the cross capacitance can be
small and therefore argue that future work on strategies for the initial tuning of quantum dot arrays
should address disorder rather than capacitive cross talk, in particular for SiMOS quantum dots.
We envision that our integration scheme for fabricating quantum dot arrays in SiMOS, Si/SiGe,
and Ge/SiGe will boost collective development and enable the realization of devices capable of
simulating and computing with quantum information.
We envision that our realization of an integration scheme to build quantum dots in SiMOS,
Si/SiGe, and Ge/SiGe will boost the collective development toward large quantum dot arrays to
build, simulate, and compute with quantum information.

3.7. S I MOS FABRICATION I MPROVEMENTS

3.7.1. G ATE A NNEAL
In silicon MOS, anneals are generally used to repair damage caused by e-beam exposure and to
improve the structural quality of metal gates. We find that for SiMOS quantum dot devices, the
quality of the gates can be improved by the incorporation of a forming gas anneal at 400 ◦ C for 15
min after each gate deposition. Figure 3.7(a) shows a scanning electron microscope (SEM) image
of a device for which the anneal was implemented. Figures 3.7(b)-(g) show SEM and Atomic Force
 56 3. QUANTUM D OT A RRAYS IN S ILICON AND G ERMANIUM

Microscopy (AFM) images for two gate layers of a SiMOS device. We observe a large reduction of
surface roughness and sidewall height, which improves even further the homogeneity and yield of
the metallic gates.

(a) (b) (d) (f)

100 nm 200 nm No anneal Anneal 40 nm

3
(c) (e) (g)

0 nm

Figure 3.7: (a,b,c) SEM images taken under a 30◦ angle of (a) full device fabricated with a gate anneal. (b,c)
Separate layers after a gate anneal. (d,e,f,g) AFM images of separate gate layers before (d,e) and after (f,g) a
gate anneal. The anneal results in a smoother surface with less grains and sidewalls.

3.7.2. ALD W INDOW E TCH

Incorporating atomic layer deposition (ALD) of Al2 O3 into the gate stack introduces further sources
of charge noise [42] making it undesirable in the active region of spin qubits. On the other hand,
it is necessary to prevent inter-gate leakage when utilizing a Ti:Pd multi-layer gate stack, as well
as leakage to substrate. In the case of SiMOS, fabrication begins on a 10 nm SiO2 dielectric grown
across the substrate. While high in quality, due to the sheer area overlap of gate fan-out, there
is a non-negligible probability that a gate may overlap with a region of damaged dielectric. To
prevent leakage of gate layers to substrate in our SiMOS stack, we find an initial blanket layer of
Al2 O3 is necessary. An etching process (see section 3.8 regarding fabrication details) with a high
selectivity of Al2 O3 over SiO2 allows us to locally remove this layer in the active region. Figure 3.8
shows charge stability diagrams of two identically processed quantum dot devices in SiMOS where
Al2 O3 was present or where an oxide window was etched. Without etching, charge noise causes
significant fluctuations in the quantum dot potential, which can be observed from the constantly
shifting charge addition lines in Fig. 3.8a. Instead, when an oxide window is etched, we observe
stable transitions, see Fig. 3.8b. We attribute this stability to the removal of the ALD layer beneath
the first gate layer. We note that this behaviour is reproducible in and consistent with other SiMOS
quantum dot devices fabricated with and without the removal of the initial ALD layer in the quan-
tum dot active region.

3.8. E XTENDED FABRICATION R ECIPE

Here we present the full fabrication process for each platform, specifically for the devices fabri-
cated and studied in the work.

In the case of SiMOS, fabrication begins on a natural silicon wafer, with 1 um of intrinsic silicon
3.8. E XTENDED FABRICATION R ECIPE 57

Tr
P1

SET

B12

(a) With ALD (b) Without ALD

1500
1500

1450
1450
B12(mV)

Tr (mV)

1400 1400

1350 1350

1300 1300
2300 2500 2700 2900 3100 3300 1500 1600 1700 1800 1900 2000 2100
P1 (mV) P1 (mV)

Figure 3.8: Charge stability diagrams via lock-in charge sensing of identically processed quantum dot devices
in SiMOS, with the exception of an oxide window etch step. (a) Typical stability behaviour of a single quantum
dot with a layer of Al2 O3 beneath the screening layer. (b) Device processed with oxide window etch step.
Stability markedly improves over previous case.
 58 3. QUANTUM D OT A RRAYS IN S ILICON AND G ERMANIUM

SiMOS Si/SiGe Ge/SiGe

Deposit marker layer 1

Material M W markers EBL Etched markers Optical Litho -

Open implant windows

at energy E, fluence n E = 6 keV, n = 1x1016 cm-2 E = 20 keV, n = 5x1015 cm-2 -

3 Anneal and activate at T = 1000 C, t = 30 s T = 700 C, t = 30 s -

temperature T, time t in
an RTP, nitrogen armos-
phere

Define and write second

marker layer Ti:Pt (5:55 nm)   

HF etch, imidiately before

evaporation of Ohmics M = Ti:Pt M = Ti:Pt M = Al
(material M) t = 5:55 nm t = 5:55 nm t = 30 nm
at thickness t

1 h vacuum anneal at 300 oC t = 10 nm t = 5 nm t = 10 nm

Gate Isolation ALD Al2O3
thickness t

Transene etch ALD in active  - -

window 50 oC

Deposit gate stack in layers L

tscr = 3:17 nm tscr = 3:17 nm tbar = 3:17 nm
at thicknessess tL with gate
tplg = 3:37 nm tplg = 3:37 nm tplg = 3:37 nm
anneal T = 400 oC, 15 mins in
tbar = 3:37 nm tbar = 3:37 nm tALD= 10 nm
Forming gas. Interlayer isolation
tALD= 7 nm tALD= 7 nm
ALD thickness t.
Gate anneal  Gate anneal x Gate anneal x

Deposit qubit control layer Antenna Micromagnet -

Micromagnet or MW Antenna M = Al M = Ti:Co
Thickness t, material M t = 100 nm t = 5:200 nm

End of Line Annneal

T = 400 oC, 30 mins  - x
Forming gas.

Figure 3.9: Full fabrication recipe for the three platforms presented in this work. Ticks represent a used and
compatible process, dashes represent a not-used but not necessarily incompatible process, and crosses indi-
cate an unused and incompatible process.
3.8. E XTENDED FABRICATION R ECIPE 59

grown, followed by 100 nm epilayer 28 Si, and a 10 nm thermally grown oxide [14]. First, tungsten
(W) markers are patterned, which are used to define implant windows via electron beam lithogra-
phy (EBL). After exposure, phosphorus ions (P+ ) at 6 keV are implanted to create highly negatively
doped (n++ ) regions in each die. An activation anneal is conducted in a rapid thermal processor
(RTP) at 1000 o C for 30 seconds. A buffered hydro-fluoric (BHF) etch removes oxide in bond-pad
areas, where Ti:Pt (5:55 nm) metallic contacts are deposited, creating ohmic contacts. A second
layer of Ti:Pt markers are also written in this step. Next, a blanket Al2 O3 ALD layer of 10 nm is
grown across the entire sample. A small 20 x 20 µm2 area is exposed and etched away in the vicin-
ity of the quantum dot formation area. This improves dot stability (see Fig. 3.8). Large rounded
rectangular regions are then exposed in regions where wirebonding is expected, and 150 nm of SiN
is sputtered. These create safer bondpads with which to bond to, reducing leakage and improving
3
device yield. The SiMOS device presented in the work utilizes a three layer Ti:Pd gate stack. (3:17,
3:37, 3:37 nm). After each layer, the device is annealed in a RTP furnace for 15 minutes at 400 o C
in forming gas, then a layer of ALD is grown at 7 nm thickness. Next, the qubit control layer is
deposited. This can either be an Al or NbTiN antenna of 100 nm thickness for Electron spin res-
onance driving, or a Ti:Co micromagnet (5:195 nm) for Electron dipole spin resonance. The final
step is an end of line anneal at 400 o C for 30 minutes in forming gas in a RTP.

The Si/SiGe 5 dot linear array begins on a natural silicon substrate. A linearly graded Si1−x Gex
layer is deposited where x ranges from 0 to 0.3. A relaxed Si0.7 Ge0.3 layer of 300 nm lies below the
10 nm 28 Si (800 ppm purity) quantum well which itself is separated from the 2 nm Si capping layer
by a second 30 nm relaxed Si0.7 Ge0.3 spacer layer. The initial marker layer is written using optical
lithography and is formed by etching away the SiO2 . Next, a BHF dip removes native oxide selec-
tively where ohmic contacts of Ti:Pt (5:55 nm) are evaporated, alongside a second set of markers.
Gate stack fabrication of the device is almost identical to that of SiMOS. It is a 3-layer Ti:Pd stack
of the same thicknesses, interlayer isolated via 7 nm of Al2 O3 . However, we do not employ a gate
anneal between gate layers, despite this technically being possible within the context of thermal
budget. For control, both striplines and micromagnets are avaliable, however we prefer the elec-
trical driving option since electron wavefunctions in Si/SiGe tend to be more mobile and hence
EDSR provides a route to faster driving. We do not conduct an end of line anneal on SiGe devices.

For the fabrication of the Ge/SiGe 2x2 array, we begin with a natural silicon substrate, upon
which 1.4 µm of Ge and 900 nm of reverse graded Si1−x Gex where x ranges from 1 to 0.8 is grown.
This lies below a 160 nm Si0.2 Ge0.8 spacer layer, a 16 nm Ge quantum well under compressive
strain, a second Si0.2 Ge0.8 layer of 22 nm and finally a thin Si cap of 1 nm[13]. Ti:Pt EBL mark-
ers are then defined for future alignment. A short HF acid etch is conducted immediately before
depositing 30 nm Al on regions where ohmic contact is desired. An advantage of the Ge/SiGe
platform is the possibility of ohmic formation extremely close (within ≈ 100 nm) of the quantum
dot. Devices are then placed under vacuum for 1 h at 300 o C causing Al to diffuse through the
heterostructure into the quantum well forming ohmic contact. Atomic Layer Deposition is then
performed covering the sample in a 10 nm Al2 O3 blanket. The gate stack consists of two layers,
barrier and plunger. The barrier layer is deposited at 20 nm total thickness utilizing the Ti:Pd stack
(3:17 nm). The plunger layer is deposited at 40 nm total thickness (3:37 nm). No further processing
is required as the large intrinsic Spin-Orbit coupling of holes in Ge/SiGe provides a native electric
driving mechanism[31].
 60 R EFERENCES

R EFERENCES
[1] Lawrie, W. I. L. et al. Quantum Dot Arrays in Silicon and Germanium. Applied Physics Letters
116, 080501 (2020).

[2] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Physical Review A -
Atomic, Molecular, and Optical Physics 57, 120–126 (1998).

[3] Hensgens, T. et al. Quantum simulation of a Fermi-Hubbard model using a semiconductor

quantum dot array. Nature 548, 70–73 (2017).
3 [4] Leijnse, M. & Flensberg, K. Scheme to measure Majorana fermion lifetimes using a quantum
dot. Physical Review B 84, 140501 (2011).

[5] Sau, J. D. & Sarma, S. D. Realizing a robust practical Majorana chain in a quantum-dot-
superconductor linear array. Nature Communications 3, 964 (2012).

[6] Petta, J. R. et al. Coherent Manipulation of Electron Spins in Semiconductor Quantum Dots.
Science 309, 2180–2184 (2005).

[7] Koppens, F. H. et al. Driven coherent oscillations of a single electron spin in a quantum dot.
Nature 442, 766–771 (2006).

[8] Dehollain, J. P. et al. Nature .

[9] Mourik, V. et al. Signatures of majorana fermions in hybrid superconductor-semiconductor

nanowire devices. Science 336, 1003–1007 (2012).

[10] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors - hot, dense
and coherent. npj Quantum Information 3, 34 (2016).

[11] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nature Nanotechnology 9, 981–985 (2014).

[12] Zwanenburg, F. A. et al. Silicon quantum electronics. Reviews of Modern Physics 85, 961–1019
(2013).

[13] Sammak, A. et al. Shallow and Undoped Germanium Quantum Wells: A Playground for Spin
and Hybrid Quantum Technology. Advanced Functional Materials 29, 1807613 (2019).

[14] Sabbagh, D. et al. Quantum Transport Properties of Industrial Si 28 / Si O 2 28 . Physical

Review Applied 12 (2019).

[15] Angus, S. J., Ferguson, A. J., Dzurak, A. S. & Clark, R. G. Gate-defined quantum dots in intrinsic
silicon. Nano Letters 7, 2051–2055 (2007).

[16] Zajac, D. M., Hazard, T. M., Mi, X., Wang, K. & Petta, J. R. A reconfigurable gate architecture
for Si/SiGe quantum dots. Applied Physics Letters 106, 223507 (2015).

[17] Brauns, M., Amitonov, S. V., Spruijtenburg, P. C. & Zwanenburg, F. A. Palladium gates for
reproducible quantum dots in silicon. Scientific Reports 8, 5690 (2018).

[18] Eenink, H. G. J. et al. Tunable coupling and isolation of single electrons in silicon quantum
dots. Nano Letters 19, 8653–8657 (2019).
R EFERENCES 61

[19] Yang, C. H. et al. Silicon qubit fidelities approaching incoherent noise limits via pulse engi-
neering. Nature Electronics 2, 151–158 (2019).

[20] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nature Nanotechnology 13, 102–106 (2018).

[21] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[22] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins. Science 359, 439–442
(2018).

[23] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon. Nature 555,
3
633–637 (2018).

[24] Spruijtenburg, P. C. et al. Single-hole tunneling through a two-dimensional hole gas in intrin-
sic silicon. Applied Physics Letters 102, 192105 (2013).

[25] Liles, S. D. et al. Spin and orbital structure of the first six holes in a silicon metal-oxide-
semiconductor quantum dot. Nature Communications 9, 3255 (2018).

[26] Maurand, R. et al. A CMOS silicon spin qubit. Nature Communications 7, 13575 (2016).

[27] Failla, M. et al. Terahertz quantum Hall effect for spin-split heavy-hole gases in strained Ge
quantum wells. New Journal of Physics 18, 113036 (2016).

[28] Su, Y. H., Chuang, Y., Liu, C. Y., Li, J. Y. & Lu, T. M. Effects of surface tunneling of two-
dimensional hole gases in undoped Ge/GeSi heterostructures. Physical Review Materials 1,
044601 (2017).

[29] Lodari, M. et al. Light effective hole mass in undoped Ge/SiGe quantum wells. Physical
Review B 100, 041304 (2019).

[30] Hendrickx, N. W. et al. Gate-controlled quantum dots and superconductivity in planar ger-
manium. Nature Communications 9, 2835 (2018).

[31] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit
logic with holes in germanium (2019).

[32] Dimoulas, A., Tsipas, P., Sotiropoulos, A. & Evangelou, E. K. Fermi-level pinning and charge
neutrality level in germanium. Applied Physics Letters 89, 252110 (2006).

[33] Petit, L. et al. Spin Lifetime and Charge Noise in Hot Silicon Quantum Dot Qubits. Physical
Review Letters 121, 076801 (2018).

[34] Ando, T., Fowler, A. B. & Stern, F. Electronic properties of two-dimensional systems. Reviews
of Modern Physics 54, 437–672 (1982).

[35] Gold, A. & Dolgopolov, V. T. Temperature dependence of the conductivity for the two-
dimensional electron gas: Analytical results for low temperatures. Physical Review B 33,
1076–1084 (1986).

[36] Kruithof, G. H., Klapwijk, T. M. & Bakker, S. Temperature and interface-roughness depen-
dence of the electron mobility in high-mobility Si(100) inversion layers below 4.2 K. Physical
Review B 43, 6642–6649 (1991).
 62 R EFERENCES

[37] Wuetz, B. P. et al. Multiplexed quantum transport using commercial off-the-shelf CMOS at
sub-kelvin temperatures. arXiv 1907.11816 (2019).

[38] Jock, R. M. et al. A silicon metal-oxide-semiconductor electron spin-orbit qubit. Nature Com-
munications 9, 1768 (2018).

[39] Huang, W., Veldhorst, M., Zimmerman, N. M., Dzurak, A. S. & Culcer, D. Electrically driven
spin qubit based on valley mixing. Physical Review B 95, 075403 (2017).

[40] Bracht, H., Haller, E. E. & Clark-Phelps, R. Silicon self-diffusion in isotope heterostructures.
3 Physical Review Letters 81, 393–396 (1998).

[41] Hendrickx, N. W. et al. Ballistic supercurrent discretization and micrometer-long Josephson

coupling in germanium. Physical Review B 99, 075435 (2019).

[42] Connors, E. J., Nelson, J., Qiao, H., Edge, L. F. & Nichol, J. M. Low-frequency charge noise in
Si/SiGe quantum dots. Physical Review B 100, 165305 (2019).

[43] Volk, C. et al. Loading a quantum-dot based "Qubyte" register. npj Quantum Information 5,
29 (2019).

[44] Watzinger, H. et al. A germanium hole spin qubit. Nature Communications 9, 3902 (2018).

[45] Kawakami, E. et al. Electrical control of a long-lived spin qubit in a Si/SiGe quantum dot.
Nature Nanotechnology 9, 666–670 (2014).

[46] Yang, C. H., Lim, W. H., Zwanenburg, F. A. & Dzurak, A. S. Dynamically controlled charge
sensing of a few-electron silicon quantum dot. AIP Advances 1, 042111 (2011).

[47] Hendrickx, N. W. et al. Nature Commmunications .

[48] Baart, T. A. et al. Single-spin CCD. Nature Nanotechnology 11, 330–334 (2016).

[49] Mazzocchi, V. et al. 99.992% 28 Si CVD-grown epilayer on 300 mm substrates for large scale
integration of silicon spin qubits. Journal of Crystal Growth 509, 1–7 (2019).

[50] Petit, L. et al. Nature .

[51] Arute, F. et al. Supplementary information for "Quantum supremacy using a programmable
superconducting processor". Nature 574, 505 (2019).

[52] DiVincenzo, D. P. The physical implementation of quantum computation. Fortschritte der

Physik 48, 771–783 (2000).

[53] Mi, X. et al. A coherent spin-photon interface in silicon. Nature 555, 599–603 (2018).

[54] Samkharadze, N. et al. Strong spin-photon coupling in silicon. Science 359, 1123–1127 (2018).

[55] Landig, A. J. et al. Coherent spin–photon coupling using a resonant exchange qubit. Nature
560, 179–184 (2018).

[56] Borjans, F., Croot, X. G., Mi, X., Gullans, M. J. & Petta, J. R. Resonant microwave-mediated
interactions between distant electron spins. Nature (2019).

[57] Leijnse, M. & Flensberg, K. Hybrid topological-spin qubit systems for two-qubit-spin gates.
Physical Review B 86 (2012).
R EFERENCES 63

[58] Sarma, S. D., Freedman, M. & Nayak, C. Topologically protected qubits from a possible non-
abelian fractional quantum hall state. Physical Review Letters 94 (2005).

[59] Hyart, T. et al. Flux-controlled quantum computation with Majorana fermions. Physical
Review B 88 (2013).

[60] Hoffman, S., Schrade, C., Klinovaja, J. & Loss, D. Universal quantum computation with hybrid
spin-Majorana qubits. Physical Review B 94 (2016).

[61] Xiang, J., Vidan, A., Tinkham, M., Westervelt, R. M. & Lieber, C. M. Ge/Si nanowire meso-
scopic josephson junctions. Nature Nanotechnology 1, 208–213 (2006). 3
[62] Ridderbos, J. et al. Hard superconducting gap and diffusion-induced superconductors in Ge-
Si nanowires. Nano Letters acs.nanolett.9b03438 (2019).

[63] Alicea, J. New directions in the pursuit of Majorana fermions in solid state systems. Reports
on Progress in Physics 75 (2012).

[64] Botzem, T. et al. Tuning Methods for Semiconductor Spin Qubits. Physical Review Applied
10, 54026 (2018).

[65] Van Diepen, C. J. et al. Automated tuning of inter-dot tunnel coupling in double quantum
dots. Applied Physics Letters 113, 33101 (2018).

[66] Mills, A. R. et al. Computer-automated tuning procedures for semiconductor quantum dot
arrays. Applied Physics Letters 115, 113501 (2019).

[67] Mills, A. R. et al. Shuttling a single charge across a one-dimensional array of silicon quantum
dots. Nature Communications 10 (2019).

[68] Kalantre, S. S. et al. Machine learning techniques for state recognition and auto-tuning in
quantum dots. npj Quantum Information 5 (2019).

[69] Franke, D. P., Clarke, J. S., Vandersypen, L. M. & Veldhorst, M. Rent’s rule and extensibility in
quantum computing. Microprocessors and Microsystems 67, 1–7 (2019).

[70] Veldhorst, M., Eenink, H. G., Yang, C. H. & Dzurak, A. S. Silicon CMOS architecture for a
spin-based quantum computer. Nature Communications 8 (2017).

[71] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science Advances 4, eaar3960
(2018).

[72] Yang, C. H. et al. Operation of a silicon quantum processor unit cell above one kelvin. nature
580, 350–354 (2020).
 S ILICON
Silicon is a semiconductor material that has been the driving force behind Moore’s law - the doubling
of the number of transistors on a chip every two years - since the 1960’s. It is therefore no surprise
that it is a thoroughly studied material, with a tremendous wealth of knowledge available regarding
it’s material properties, and processing techniques. This abundance of experience makes silicon an
excellent material to investigate as a quantum computing platform.

Realizing qubits in silicon can be done in different ways. However the metal-oxide-semiconductor
format is perhaps most industrially compatible, given it is also how modern transistors are created.
Therefore realizing quantum bits in a SiMOS platform would be highly advantageous when consid-
ering the potential to bootstrap off existing industrial infrastructure.

Indeed, electron spin qubits in SiMOS quantum dots have proven highly successful qubit candi-
dates, demonstrating spin relaxation times longer than a second [1], dephasing times as high as
120 µs [2], single qubit Clifford gate fidelities as high as 99.967 % [3], and two qubit gate fidelities up
to 98 % [4]. However, some challenges, for SiMOS and for spin qubits in general remain outstanding.

Firstly, the relatively low electron mobility at the Si-SiO2 interface suggests a high interface trap
density, that can lead to unwanted quantum dot formation and highly immobile wavefunctions.
These traits can complicate qubit-qubit interactions. In other words, tunnel couple control using
dedicated barrier gates has been difficult to realise in SiMOS quantum dot devices. Chapter 4 ex-
plicitly addresses this limitation, by employing the fabrication techniques presented in Chapter 3, a
SiMOS double quantum dot device is realised with tuneable tunnel coupling.

A second major challenge pertaining to all spin qubit platforms is scale-up. On average, every qubit
in contemporary semiconductor devices require multiple electrostatic gates, and a microwave sig-
nal, which must route from the device (in a dilution refrigerator) to room temperature via wire
looms, and low loss cables capable of transmitting high frequency signals. This is a manageable
problem for a few qubits, but for the large numbers required for useful quantum computing, this
interconnect problem is completely intractable. As a result, it has been suggested that the electronics
be defined on-chip alongside arrays of qubits in an interspersed, modular fashion [5]. While help-
ing with the interconnect problem, this methodology would result in large amounts of heat being
dispersed, far beyond the cooling power of conventional dilution refrigerators. As a result, it would
be highly advantageous to operate qubits at higher temperatures, where the cooling power is much
higher. In chapter 5, we demonstrate the universal operation of a two-qubit quantum processor at
temperatures above 1 K, a significant milestone for scalable spin qubit quantum processors.

R EFERENCES
[1] Zwerver, A. M. J. et al. Qubits made by advanced semiconductor manufacturing. arXiv
2101.1265 (2021).

[2] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nature Nanotechnology 9, 981–985 (2014).

65
 66 R EFERENCES

[3] Yang, C. H. et al. Silicon qubit fidelities approaching incoherent noise limits via pulse engi-
neering. Nature Electronics 2, 151–158 (2019).

[4] Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 569, 532–536 (2019).

[5] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 1–10 (2017).

3
 4
T UNEABLE C OUPLING AND
I SOLATION OF S INGLE E LECTRONS
IN S ILICON Q UANTUM D OTS

Extremely long coherence times, excellent single-qubit gate fidelities and two-qubit logic have been
demonstrated with silicon metal-oxide-semiconductor spin qubits, making it one of the leading
platforms for quantum information processing. Despite this, a long-standing challenge has been
the demonstration of tunable tunnel coupling between single electrons. Here we overcome this hur-
dle with gate-defined quantum dots and show couplings that can be tuned on and off for quantum
operation. We use charge sensing to discriminate between the (2,0) and (1,1) charge states of a dou-
ble quantum dot and show high charge sensitivity. We demonstrate tunable coupling up to 13 GHz,
obtained by fitting charge polarization lines, and tunable tunnel rates down to below 1 Hz, deduced
from the random telegraph signal. The demonstration of tunable coupling between single electrons
in silicon provides significant scope for high-fidelity two-qubit logic toward quantum information
processing with standard manufacturing.

Parts of this chapter have been published in Nano Lett. 19 (12) 8653-8657.

67
 4. T UNEABLE C OUPLING AND I SOLATION OF S INGLE E LECTRONS IN S ILICON QUANTUM
68 D OTS

4.1. I NTRODUCTION
Quantum computation with quantum dots has been proposed using qubits defined on the spin
states of one [1], two [2] or more [3, 4] electrons. In all these proposals, a crucial element required
to realize a universal quantum gate set is the exchange interaction between electrons. The ex-
change interaction is set by the tunnel coupling and detuning, and gaining precise control over
these parameters enables to define and operate qubits at their optimal points [5–8]. Excellent
control has already been reported in GaAs [5, 6, 9], strained silicon [10, 11], and more recently in
strained germanium [12, 13]. Reaching this level of control in silicon metal-oxide-semiconductor
(SiMOS) quantum dots is highly desired as this platform has a high potential for complete inte-
gration with classical manufacturing technology [14–16]. However, current two-qubit logic with
single spins in SiMOS is based on controlling the exchange using the detuning only [17] or is exe-
cuted at fixed exchange interaction [18].
In SiMOS, a first step toward the required control to materialize architectures for large-scale
4 quantum computation [1, 19–24] has been the demonstration of tunable coupling in a double
quantum dot system operated in the many-electron regime, where gaining control is more ac-
cessible owing to the larger electron wave function [25, 26]. More recently, exchange-controlled
two-qubit operations have been shown with three-electron quantum dots [27]. However, tunnel
couplings between single electrons that can be switched off and turned on for qubit operation still
remain to be shown in SiMOS.
In this work we show a high degree of control over the tunnel coupling of single electrons
residing in two gate-defined quantum dots in a SiMOS device. The system is stable and no un-
intentional quantum dots are observed. We are able to measure charge transitions using a sensi-
tive single-electron-transistor (SET) as charge sensor, and characterize the system in the single-
electron regime. From a comparison of charge stability diagrams of weakly and strongly coupled
double quantum dots, we conclude that we control the tunnel coupling by changing quantum dot
location. We show that we can effectively decouple the double quantum dot from its reservoir
and control the inter-dot tunnel coupling of the isolated system with a dedicated barrier gate. We
quantify the tunability of the coupling by analyzing charge polarisation lines and random tele-
graph signals and find tunnel couplings up to 13 GHz and tunnel rates down to below 1 Hz.

4.2. R ESULTS AND DISCUSSION

Figure 4.1a shows a scanning electron micrograph (SEM) of a SiMOS device nominally identical
to the one measured. A high quality wafer is realized [14] with a 100 nm 28 Si epilayer, remov-
ing nuclear spin interactions to obtain spins with long quantum coherence [28], covered by 10
nm thermally grown SiO2 . Ohmic contacts are made by defining highly doped n++ regions by
phosphorus-ion implantation. We use an overlapping gate integration scheme [29] and use palla-
dium (Pd) gates, which have the beneficial property of small grain size [30]. The gates are electri-
cally isolated by an Al2 O3 layer grown by atomic layer deposition. The sample is annealed at 400
◦ C in a hydrogen atmosphere to repair e-beam induced damage to the silicon oxide and to reduce

the charge trap density [31, 32].

4.2.1. C HARGE S ENSING AND R EADOUT

Figure 4.1b shows the current through the SET, electrostatically defined using gates ST, LB and
RB, that is used as charge sensor and as an electron reservoir. The highly regular peak spacing
indicates a well defined quantum dot with a constant charging energy. We form a double quantum
dot between the confinement barriers CL and CR, using the gates P1 and P2 to tune the quantum
dot potentials. Bt and BR are used to control the tunnel coupling between the quantum dots and
from the quantum dots to the SET, respectively.
4.2. R ESULTS AND DISCUSSION 69

a b c
10 4
2.5 5
(2,0)
2 4
(1,1)
1.5 3

I dc (nA)

Counts
16 σ
1 2

7 nm Bt BR 0.5 1
P2 P1 ST
40 nm 55 nm 0
0
7 nm Al2O3 2850 3050 3250 3450 200 400 600 800
10 nm SiO2 V ST (mV) I dc (pA)

Figure 4.1: Device layout and SET characterisation. a False-colour scanning electron micrograph (SEM) of a
device identical to the one measured. Purple, yellow and blue colourings correspond to the first, second and
third metal layers respectively. Circles indicate the intended location of the quantum dots D1 and D2 and the
single-electron-transistor (SET). The quantum dots are defined using gate electrodes P1 and P2 , confined lat-
erally using CL and CR. Bt controls the tunnel coupling between the quantum dots and BR the tunnel coupling 4
to the SET. b Transport current Id c versus top gate voltage VST of the SET defined using gate electrodes ST, LB
and RB. Regular spacing of Coulomb peaks indicates a well defined quantum dot, ideal for charge sensing. c
Histogram of the charge sensor current as a response to (2,0)-(1,1) tunneling events. The counts are extracted
from 4655 single-shot traces with integration time ti = 82 µs, measurement bandwidth 0-50 kHz, and bin size
b = 5 pA. The peaks are fitted with a double Gaussian with σ(2,0) = 34.1 pA and σ(1,1) = 25.5 pA, giving a peak
spacing of over 16 σ(2,0) .

We characterize the charge readout sensitivity by recording the random telegraph signal (RTS)
originating from the tunneling of the electrons between the (2,0) and (1,1) charge states with Γc ≈
48 Hz, with Γc the inter-dot tunnel rate. The fidelity of the (2,0)-(1,1) charge readout is often lim-
ited by the sensitivity of the charge sensor to inter-dot transitions. We have designed and posi-
tioned the SET with respect to the double quantum dot in such a way that this sensitivity is maxi-
mized. Figure 4.1c shows a histogram of the readout signal obtained, using an integration time τ =
82 µs. We fit the counts with a double Gaussian curve with µ(2,0),(1,1) and σ(2,0),(1,1) the mean and
standard deviation of the Gaussian distributions corresponding to the two charge states. We find
∆µ(2,0)−(1,1) > 16 σ(2,0) corresponding to an excellent discrimination between the (2,0) and (1,1)
charge states.
To precisely measure charge transitions, we implement charge sensing using a lock-in ampli-
fier and apply a square wave excitation at f ac = 77 Hz on the gate Bt . Figure 4.2a and 4.2b show
the double quantum dot charge stability diagrams of the charge sensor response as a function of
VP 2 and VP 1 for weak (VB t = 2.9 V) and strong (VB t = 3.6 V) coupling. Horizontal and vertical
blue lines indicate the loading of an additional electron from the SET to quantum dots D1 (located
under the gate P1 ) and D2 (located under P2 ) respectively, while diagonal yellow lines indicate
electron transitions between the two quantum dots. We do not observe more charge transitions at
voltages lower than the measured range, and we conclude that the double quantum dot is in the
single electron regime. In order to highlight the difference between weak and strong coupling, Fig.
4.2c and 4.2d show higher resolution maps of the (2,0)-(1,1) anticrossing.
When we set a weak inter-dot coupling, charge addition lines of D2 are barely visible in the
charge stability diagram, because of the low tunnel rate between D2 and the reservoir. This in-
dicates that the tunnel rate is significantly smaller than the excitation frequency applied to the
gate, taking into account the increased tunnel rate caused by inelastic tunneling. Similarly, at the
(2,0)-(1,1) inter-dot transition, no transitions between the quantum dots can be observed because
of the low inter-dot coupling. The loading of the first electron in D2 can only be observed from
the shift of the D1 charge addition line, caused by the mutual capacitance E m of the two quantum
 4. T UNEABLE C OUPLING AND I SOLATION OF S INGLE E LECTRONS IN S ILICON QUANTUM
70 D OTS

a b
Vbt = 2.9 V Vbt = 3.6 V
2100 60 2100 60

40 40
V P1 (mV)

V P1 (mV)
2000 2000

I ac (pA)

I ac (pA)
20 20

0 0
1900 1900
-20 -20

2370 2420 2470 2520 200 600 1000 1400

V P2 (mV) V P2 (mV)
c d
80
150
1974 1988
60
100
1971 1985
40
V P1 (mV)

V P1 (mV)
I ac (pA)

I ac (pA)
4 1968 20 1982
50

0 0
1965 1979
-20 -50
1962 1976
2400 2420 2440 725 750 775
V P2 (mV) V P2 (mV)

Figure 4.2: Double quantum dot charge stability diagrams. a, b Charge stability diagrams of the charge sensor
response Iac as a function of voltages VP 2 and VP 1 of a double quantum dot for weak (a, VB t = 2.9 V) and
strong (b, VB t = 3.6 V) coupling. Electrons are loaded from the SET. Transitions with a tunnel rate Γ < f ac are
not visible. c, d High resolution zoom in of the (2,0)-(1,1) anticrossing for both weak (c) and strong (d) tunnel
coupling.

dots. Only in the multi-electron regime where the quantum dot wave functions are larger and have
more overlap, the coupling is sufficiently high to observe charge transition lines.
When the inter-dot coupling is strong, charge addition lines belonging to D2 are visible near
the anticrossings and at high VP 1 , where ΓR2 is increased. Additionally, t c and E m are increased
and we observe a honeycomb shaped charge stability diagram, with clearly visible inter-dot tran-
sition lines, even when only a single electron is loaded on each quantum dot.
We estimate the relative location and size of the quantum dots from the gate voltage differ-
ences ∆VP 1(2) needed to load the second electron with respect to the first electron. We addition-
ally use the cross-capacitances αr 1(2) of the plunger gates, determined by measuring the shift in
VP 1(2) of the charge transition line of the first electron in D1(2) as a function of a step in VP 2(1) ,
where αr 1(2) is the ratio between the shift and the step.
When the coupling is weak, we find ∆VP 1 ≈ 70 mV, αr 1 < 0.05 for D1 and ∆VP 2 ≈ 50 mV, αr 2 ≈
0.33 for D2 . We conclude that we have a system of two weakly coupled quantum dots located
under P1 and P2 .
We analyse how the locations of D1 and D2 change from the changes in ∆VP and αr . For D1 ,
both ∆VP 1 and αr 1 are almost independent of the coupling. For D2 , ∆VP 2 increases by a factor 11,
from ∆VP 2 ≈ 50 mV for weak coupling to ∆VP 2 ≈ 550 mV for strong coupling, while αr 2 increases
by a factor 5, from 0.3 to 1.5. The increase in αr 2 can be explained by a change in the location of
D2 toward the gate P1 , to a position partly below the gate Bt . This change of quantum dot location
will decrease the lever arm and this is likely the cause of the increase in ∆VP 2 . We conclude that
tuning from weak to strong coupling causes the location of D2 to change from a position mostly
under P2 to a position partly below Bt , while D1 is stationary under P1 .
By reducing VB R , the tunnel rate ΓR between the the SET reservoir and the quantum dots can
4.2. R ESULTS AND DISCUSSION 71

a b c d
15 60
3650 1
(1,1) 12 48
0.75

t c (GHz)
2150

t c ( eV)
9 36

V (e)
0.5 6 24
0.25 3 12
(2,0)
3400 0 0 0
2050 -400 0 400 3450 3600 3750
V P1 (mV)

V tc (mV) V tc (mV)
( eV)

e f
4.13
1950 3150
800 10 3 10 -6

( eV)
(Hz)
I dc (pA)
600 10 2 10 -7

c
1
10 10 -8

c
400
1850 10 0 10 -9
2900 200
500 1250 2000 270 370 470 0 0.5 1 2900 2915 2930
V P2 (mV) V (mV) t (s) V tc (mV)
4
Figure 4.3: Demonstration of tunable tunnel coupling. a Map of the isolated (2,0)-(1,1) and (1,1)-(0,2) anti-
crossings as a function of VP 2 and VP 1 . No additional electrons are loaded into the quantum dot islands due to
a negligible ΓR . b Map of the (2,0)-(1,1) and (1,1)-(0,2) anticrossings as a function of detuning and barrier volt-
age. The relative lever arm between Vt c and V² changes at lower barrier voltages, due to a change in quantum
dot location. The orange and purple arrows indicate the regime in which the tunnel coupling was determined
using RTS and polarisation line measurements respectively. c Polarization lines (excess charge V as a function
of detuning ²) across the anticrossing for high t c (black, Vt c = 3.85 V), intermediate t c (green, Vt c = 3.6 V) and
relatively low t c (red, Vt c = 3.4 V). d Extracted t c from polarization lines as a function of Vt c , where we find
tunable t c up to 13 GHz. e RTS for weak coupling Vt c = 2.910 V. f Extracted Γc from RTS measurements as a
function of Vt c demonstrating tunable tunnel rates down to below 1Hz.

be reduced and the loading and unloading of electrons can be prevented, resulting in an isolated
quantum dot system [27, 33]. Because the reservoir is connected to room temperature electron-
ics, decoupling the quantum dot from it may provide the advantage of reduced noise [34]. Figure
4.3a shows the (2,0)-(1,1) and (1,1)-(0,2) anticrossings as a function of VP 2 and VP 1 for strong cou-
pling. Only inter-dot transition lines are present over a wide range of voltages, much larger than
the ∆VP extracted in the previous section. This implies that no additional electrons are loaded, as
a result of a negligible coupling to the reservoir. The ability to control the inter-dot transitions of a
double quantum dot without loading additional electrons provides good prospects for the opera-
tion of quantum dot arrays that are only remotely coupled to reservoirs, as proposed in quantum
information architectures [19, 21, 22].

4.2.2. T UNNEL C OUPLE C ONTROL

We control the tunnel coupling t c with the gate BT . To compensate for the influence of VB t on
detuning ² and on-site potential U , we implement virtual gates using a cross-capacitance matrix
[35] and convert VP 2 , VP 1 and VB t to ², U and tc . Figure 4.3b shows the (2,0)-(1,1) and (1,1)-(0,2)
anticrossings as a function of the new set of virtual gates V² and Vt c . For both transitions the inter-
dot line vanishes at low Vt c , meaning that the coupling has been largely switched off. We observe
that for the (1,1)-(0,2) anticrossing, the transition line disappears at Vt c < 3.1 V, while for the (2,0)-
(1,1) anticrossing this happens for Vt c < 2.95 V. The variation may come from a small asymmetry
in the system.
We tune the double quantum dot to a significantly coupled regime and quantitatively analyze
the system by taking charge polarization lines. Figure 4.3c shows charge polarization lines at high,
intermediate and relatively low tunnel couplings within this regime. We measure the charge sensor
 72 R EFERENCES

response V as a function of detuning ² and fit the data according to a model that includes cross-talk
of ² to the charge sensor and the influence of the quantum dot charge state on the charge sensor
sensitivity [9, 36]. From the thermal broadening of the polarization line at low tunnel coupling, we
extract the lever arm of V² for the detuning axis α² ≈ 0.04 eV/mV, by assuming the electron tem-
perature to be equal to the fridge temperature of 0.44 K. Figure 4.3d shows a t c proportional to Vt c
from approximately 3 to 13 GHz, demonstrating tunable tunnel coupling in the strong coupling
regime.
At lower tunnel couplings, the thermal broadening of the polarization line prevents accurate
fitting. Instead, to obtain the inter-dot tunnel rate Γc , we measure RTS (Fig. 4.3e) at the (2,0)-(1,1)
transition and fit the counts C of a histogram of the tunnel times T to C = Ae −Γc T , where A is a
constant to normalise the counts. Furthermore we tune the system to be in the elastic tunneling
regime, with V² such that Γc(2,0)−(1,1) ≈ Γc(1,1)−(2,0) [37]. This tunnel rate is proportional to the
tunnel coupling, but in the weak coupling regime, Γc 6= t c due to localisation of the charge [38–40].
Figure 4.3f shows the obtained Γc as a function of Vt c from 1 kHz down to below 1 Hz. We note
4 that we can further reduce this tunnel rate to even smaller rates simply by further reducing this
gate voltage.

4.3. C ONCLUSIONS
We have demonstrated control over the tunnel coupling of single electrons residing in a double
quantum dot in silicon. The inter-dot coupling of the (2,0)-(1,1) charge transition can be con-
trolled by a barrier gate which changes quantum dot location. We have demonstrated control over
the tunnel coupling in the strong coupling regime from 3 to 13 GHz, as well as control over the tun-
nel rate in the weak coupling regime from 1 kHz to below 1 Hz. Achieving this degree of control in
an isolated system constitutes a crucial step toward independent control over detuning and tunnel
coupling for operation at the charge symmetry point [5, 6], and reaching the control required for
large-scale quantum computation with quantum dots [1, 19–24]. While SiMOS systems are often
said to be severely limited by disorder, the excellent control shown here provides great prospects
to operate larger arrays fabricated using conventional semiconductor technology.

R EFERENCES
[1] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Physical Review A
57, 120 (1998).

[2] Levy, J. Universal quantum computation with spin-1/2 pairs and Heisenberg exchange. Phys-
ical Review Letters 89, 147902 (2002).

[3] DiVincenzo, D. P., Bacon, D., Kempe, J., Burkard, G. & Whaley, K. B. Universal quantum com-
putation with the exchange interaction. Nature 408, 339 (2000).

[4] Shi, Z. et al. Fast hybrid silicon double-quantum-dot qubit. Physical review letters 108, 140503
(2012).

[5] Martins, F. et al. Noise suppression using symmetric exchange gates in spin qubits. Physical
review letters 116, 116801 (2016).

[6] Reed, M. D. et al. Reduced sensitivity to charge noise in semiconductor spin qubits via sym-
metric operation. Physical review letters 116, 110402 (2016).

[7] Taylor, J. M., Srinivasa, V. & Medford, J. Electrically protected resonant exchange qubits in
triple quantum dots. Physical review letters 111, 050502 (2013).
R EFERENCES 73

[8] Kim, D. et al. Quantum control and process tomography of a semiconductor quantum dot
hybrid qubit. Nature 511, 70 (2014).

[9] Hensgens, T., T.and Fujita et al. Quantum simulation of a Fermi–Hubbard model using a
semiconductor quantum dot array. Nature 548, 70 (2017).

[10] Borselli, M. G. et al. Undoped accumulation-mode Si/SiGe quantum dots. Nanotechnology

26, 375202 (2015).

[11] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins. Science 359, 439–442
(2018).

[12] Hendrickx, N. W. et al. Gate-controlled quantum dots and superconductivity in planar ger-
manium. Nature communications 9, 2835 (2018).

[13] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit 4
logic with holes in germanium. arXiv 1904.11443 (2019).

[14] Sabbagh, D. e. a. Quantum Transport Properties of Industrial Si 28 / Si O 2 28 . Physical

Review Applied 12 (2019).

[15] Mazzocchi, V. et al. 99.992% 28Si CVD-grown epilayer on 300 mm substrates for large scale
integration of silicon spin qubits. Journal of Crystal Growth 509, 1–7 (2019).

[16] Maurand, R. et al. A CMOS silicon spin qubit. Nature Communications 7, 13575 (2016).

[17] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[18] Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 569, 532–536
(2019).

[19] Veldhorst, M., Eenink, H. G. J., Yang, C. H. & Dzurak, A. S. Silicon CMOS architecture for a
spin-based quantum computer. Nature Communications 8, 1766 (2017).

[20] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3 (2017).

[21] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science advances 4, eaar3960
(2018).

[22] Taylor, J. M. et al. Fault-tolerant architecture for quantum computation using electrically
controlled semiconductor spins. Nature Physics 1, 177 (2005).

[23] Friesen, M., Biswas, A., Hu, X. & Lidar, D. Efficient multiqubit entanglement via a spin bus.
Physical review letters 98, 230503 (2007).

[24] Trauzettel, B., Bulaev, D. V., Loss, D. & Burkard, G. Spin qubits in graphene quantum dots.
Nature Physics 3, 192 (2007).

[25] Tracy, L. A. et al. Double quantum dot with tunable coupling in an enhancement-mode sil-
icon metal-oxide semiconductor device with lateral geometry. Applied Physics Letters 97,
192110 (2010).

[26] Lai, N. S. et al. Pauli spin blockade in a highly tunable silicon double quantum dot. Scientific
reports 1 (2011).
 74 R EFERENCES

[27] Yang, C. H. et al. Silicon quantum processor unit cell operation above one Kelvin.
arXiv:1902.09126 (2019).

[28] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nature Nanotechnology 9, 981–985 (2014).

[29] Angus, S. J., Ferguson, A. J., Dzurak, A. S. & Clark, R. G. Gate-defined quantum dots in intrinsic
silicon. Nano Letters 7, 2051–2055 (2007).

[30] Brauns, M., Amitonov, S. V., Spruijtenburg, P. C. & Zwanenburg, F. A. Palladium gates for
reproducible quantum dots in silicon. Scientific reports 8, 5690 (2018).

[31] Kim, J. S., Tyryshkin, A. M. & Lyon, S. A. Annealing shallow Si/SiO2 interface traps in electron-
beam irradiated high-mobility metal-oxide-silicon transistors. Applied Physics Letters 110,
123505 (2017).
4
[32] Nordberg, E. P. et al. Enhancement-mode double-top-gated metal-oxide-semiconductor
nanostructures with tunable lateral geometry. Physical Review B 80, 115331 (2009).

[33] Bertrand, B. et al. Quantum manipulation of two-electron spin states in isolated double
quantum dots. Physical review letters 115, 096801 (2015).

[34] Rossi, A., Ferrus, T. & Williams, D. A. Electron temperature in electrically isolated Si double
quantum dots. Applied Physics Letters 100, 133503 (2012).

[35] Baart, T. A. et al. Single-spin CCD. Nature Nanotechnology 11, 330 (2016).

[36] DiCarlo, L. et al. Differential charge sensing and charge delocalization in a tunable double
quantum dot. Physical review letters 92, 226801 (2004).

[37] Tarucha, S. et al. Elastic and inelastic single electron tunneling in coupled two dot system.
Microelectronic engineering 47, 101–105 (1999).

[38] Braakman, F. R., Barthelemy, P., Reichl, C., Wegscheider, W. & Vandersypen, L. M. K. Long-
distance coherent coupling in a quantum dot array. Nature Nanotechnology 8, 432 (2013).

[39] Gamble, J. K., Friesen, M., Coppersmith, S. N. & Hu, X. Two-electron dephasing in single Si
and GaAs quantum dots. Physical Review B 86, 035302 (2012).

[40] Korotkov, A. N. Continuous quantum measurement of a double dot. Physical Review B 60,
5737 (1999).
 5
U NIVERSAL QUANTUM LOGIC IN
HOT SILICON QUBITS

Quantum computation requires many qubits that can be coherently controlled and coupled to each
other [1]. Qubits that are defined using lithographic techniques are often argued to be promising
platforms for scalability, since they can be implemented using semiconductor fabrication technology
[2–5]. However, leading solid-state approaches function only at temperatures below 100 mK, where
cooling power is extremely limited, and this severely impacts the perspective for practical quantum
computation. Recent works on spins in silicon have shown steps towards a platform that can be op-
erated at higher temperatures by demonstrating long spin lifetimes [6], gate-based spin readout [7],
and coherent single-spin control [8]. However, a high-temperature two-qubit logic gate still needs to
be demonstrated. Here we show that silicon quantum dots can have sufficient thermal robustness
to enable the execution of a universal gate set above one Kelvin. We obtain single-qubit control via
electron-spin-resonance (ESR) and readout using Pauli spin blockade. We show individual coherent
control of two qubits and measure single-qubit fidelities up to 99.3 %. We demonstrate tunability of
the exchange interaction between the two spins from 0.5 up to 18 MHz and use this to execute coher-
ent two-qubit controlled rotations (CROT). The demonstration of ‘hot’ and universal quantum logic
in a semiconductor platform paves the way for quantum integrated circuits hosting the quantum
hardware and their control circuitry all on the same chip, providing a scalable approach towards
practical quantum information.

Parts of this chapter have been published in Nature 580 (7803), 355-359 and arXiv:2007.09034 [cond-mat.mes-
hall]

75
 76 5. U NIVERSAL QUANTUM LOGIC IN HOT SILICON QUBITS

5.1. H OT Q UBITS
Spin qubits based on quantum dots are among the most promising candidates for large-scale
quantum computation [2, 9, 10]. Quantum coherence can be maintained in these systems for ex-
tremely long times [11] by using isotopically enriched silicon (28 Si) as the host material [12]. This
has enabled the demonstration of single-qubit control with fidelities exceeding 99.9% [13, 14] and
the execution of two-qubit logic [15–18]. The potential to build larger systems with quantum dots
manifests in the ability to deterministically engineer and optimize qubit locations and interac-
tions using a technology that greatly resembles today’s complementary metal-oxide semiconduc-
tor (CMOS) manufacturing. Nonetheless, quantum error correction schemes predict that millions
to billions of qubits will be needed for practical quantum information [19]. Considering that to-
day’s devices make use of more than one terminal per qubit [20], wiring up such large systems
remains a formidable task. In order to avoid an interconnect bottleneck, quantum integrated cir-
cuits hosting the qubits and their electronic control on the same chip have been proposed [2, 3, 21].
While these architectures provide an elegant way to increase the qubit count to large numbers by
leveraging the success of classical integrated circuits, a key question is whether the qubits will be
robust against the thermal noise imposed by the power dissipation of the electronics. Demon-
strating a universal gate set at elevated temperatures would therefore be a milestone in the effort
towards scalable quantum systems. First steps towards this direction have already been taken in
5 an experiment by Yang et al. [8] by demonstrating a device that can be operated as a two-qubit
system at a temperature of 40 mK and continues to have good single-qubit properties when the
temperature is increased above one Kelvin.
Here, we solve this challenge and combine initialization, readout, single-qubit rotations and
two-qubit gates, to demonstrate full two-qubit logic in a quantum circuit operating at 1.1 Kelvin.
We furthermore examine the temperature dependence of the quantum coherence which we find,
unlike the relaxation process [6], to be hardly affected in a temperature range T = 0.45 K - 1.25 K.

5.2. L ARGE - SCALE QUANTUM COMPUTING IN S I MOS

Figure 5.1a conceptually displays a quantum integrated circuit. Inspired by their classical coun-
terpart where only a few control lines are needed to interact with billions of transistors, a quantum
integrated circuit hosts the quantum hardware and its electronic control on the same chip to pro-
vide a scalable solution [20]. Here we focus on the quantum hardware of such a circuit, which we
implement using silicon quantum dots.
Figure 5.1b shows the silicon quantum dot device. The qubits are realized in an isotopically
purified 28 Si epi-layer with a 29 Si residual concentration of 800 ppm. The fabrication of the quan-
tum dot device is based on an overlapping gate-scheme to allow for tightly confined quantum dots
[24, 25]. Electrons can be loaded either from the reservoir or from the single-electron-transistor
(SET) [22], which is also used for charge sensing. To allow for coherent control over the electron
spins, AC currents are applied through the on-chip aluminum microwave antenna.
Figure 5.1c shows a charge stability diagram of the double quantum dot, where the qubits Q1
and Q2 and their coupling are defined by using the gates P1, B12, and P2. Since we can freely
choose the occupancy of the two quantum dots we tune to the regime where we obtain optimal
exchange coupling, which we find with one and five electrons for Q1 and Q2 respectively. We then
operate the system close to the (5,1)-(4,2) charge anticrossing.
Single spins are often initialized via energy-selective tunneling to a nearby reservoir [26]. How-
ever, this method requires a Zeeman splitting much higher than the thermal broadening, limiting
the fidelity and making the method unpractical for high temperature operation. Instead, Pauli spin
blockade offers a convenient mechanism to perform initialization and readout [2, 7], with a rele-
vant energy scale corresponding to the singlet-triplet energy splitting, which is set by the large and
5.2. L ARGE - SCALE QUANTUM COMPUTING IN S I MOS 77

a b 200 nm
300 K Quantum integrated circuit 1-4K Br
Bt P B12 P
B23 ST 1 2

MW CONTROL MULTIPLEXER SiO2

B12
C2
ESR Bt Q1 Q2
CLASSICAL
Si
28
INTERFACE
R
DIGITAL/ANALOG

C1 Si substrate
CONVERTER

P3 P2 P1 ST LB RB
QUANTUM
0 1
HARDWARE c d Ptriplet

45 (5,2) 300

5
(5,1)

( eV)
150
U (mV)

15
DIGITAL

read
(5,0) 0
CONTROL (4,2)
(ASIC) READOUT DEMULTIPLEXER -15 (4,1) -150
(4,0)

-50 50 150 -1.5 0 1.5

(mV) f (MHz)

Figure 5.1: Large-scale approach for silicon qubits a Quantum integrated circuit as a scalable approach for
quantum computing, where the qubits and their control electronics are defined on the same chip. The control
functionality that can be integrated is strongly dependent on the available cooling power. When the qubits can
be coherently controlled above one Kelvin a broad range of electronics may be integrated, such that commu-
nication between room temperature and the coldest stage is limited to only digital signals. Additionally, long
distance spin qubit coupling mechanisms would allow to build modular architectures, where widely sparsed
qubit arrays and local electronics alternate on the same chip, further alleviating fan-out and wiring issues [2]. b
Scanning Electron Microscope (SEM) image of a quantum device identical to the one measured. Gates P1 and
P2 define the two quantum dots and the gate B12 control the inter-dot tunnel coupling. The SET is defined by
the top gate ST and the two barriers RB and LB, and it is used both as charge sensor and as reservoir [22], while
the tunnel rate is controlled by Bt. The gates C1 and C2 confine the electrons in the three quantum dots. Gates
R, Br, P3 and B23 are kept grounded during the experiment. c Electron occupancy as a function of detuning
energy between the two quantum dots ² and on-site repulsion energy U . The data have been centered at the
(4,2)-(5,1) anticrossing. The electron transitions have been measured via a lockin technique [23], by applying
an excitation of 133 Hz on gate B12. Both electrons are loaded from the SET, with Q2 having a tunneling rate
significantly lower than Q1. d Readout signal as a function of readout position ²r ead and microwave frequency
applied to Q2. When the readout level is positioned between the singlet-triplet energy splitting and the mi-
crowave frequency matches the resonance frequency of Q2, we correctly read out the transition from the state
|↓↑〉 to the blocked state |↑↑〉.
 78 5. U NIVERSAL QUANTUM LOGIC IN HOT SILICON QUBITS

a Initialization Manipulation Readout Initialization

S(4,2)

Detuning
ESR

ESR

Time
b d f
0.6
1 Q1 1 * = (2.1 ± 0.1) μs
T2(Q1)

state probability
Triplet probability

Triplet probability
0.8 0.8 0.4
0.6 0.6
0.4 0.4 0.2

0.2 0.2
0 F(Q1) = (98.7 ± 0.3) %

|
0 0

0 2 4 6 8 0 1 2 3 4 5 0 50 100
c Time ( s) e Time ( s) g Number of Cliffords
0.6
1 Q2 1 * = (2.7 ± 0.2) μs
T2(Q2)

state probability
Triplet probability

Triplet probability
0.8 0.8 0.4
0.6 0.6
0.4 0.4 0.2

0.2 0.2

5 0 F(Q2)= (99.3 ± 0.2) %

|
0 0

0 2 4 6 8 0 1 2 3 4 5 0 50 100
Time ( s) Time ( s) Number of Cliffords

Figure 5.2: Single-qubit characterization at 1.1 K. a Pulse sequence used for the experiments. Qubits Q1 and
Q2 are defined on the spin states of single-electrons, the remaining four electrons in Q2 fill the first levels and
do not contribute to the experiment. A voltage ramp allows adiabatic transitions between the (5,1) and (4,2)
charge states. Each measurement cycle consists of two of these sequences. The second cycle contains no mi-
crowave pulses and it is used as a reference to cancel low-frequency drifts during readout. b-c Rabi oscillations
for both qubits as a function of the microwave pulse duration. We extract decay time constants T2(Q1) Rabi = 8 µs

Rabi = 14 µs. d-e Decay of the Ramsey fringes for both qubits. The data correspond to the average of four
and T2(Q2)
traces where each point is obtained from 500 single-shot traces. f-g Randomized benchmarking of the single-
qubit gates for both qubits. Each data point is obtained from 500 averages of 20 Clifford sequences, for a total
of 10.000 single-shot traces. The fidelity reported refers to the primitive gates, while a Clifford-gate contains
on average 1.875 primitive gates. We have normalized the state probabilities to remove the readout errors.

tunable valley splitting energy in silicon metal-oxide-semiconductor (SiMOS) devices [27]. This
method is more robust against thermal noise and enables independent optimization of the qubit
operation frequency. We choose to set the magnetic field to B = 0.25 T, which corresponds to ad-
dressable qubits with Larmor frequencies νQ1 = 6.949 GHz and νQ2 = 6.958 GHz in the absence
of exchange interaction. This low frequency operation reduces the qubit sensitivity to electrical
noise that couples in via the spin-orbit coupling [28]. Additionally it also simplifies the demands
on the electronic control circuits and reduces the cable losses.

5.3. S INGLE Q UBIT C ONTROL

The pulse sequence used in the experiment is schematically shown in Fig. 5.2a. The sequence
starts by pulsing deep into the (4,2) charge state, where the spins quickly relax to the singlet state.
An adiabatic pulse to the (5,1) regime is applied to initialize the system in the |↓↑〉 state. At this
position in detuning energy ², single- and two-qubit gate operations are performed by applying a
microwave burst with variable frequency and duration. The sequence ends by adiabatically puls-
ing to the anticrossing where readout is performed. The antiparallel spin state with the lowest
5.4. T WO -QUBIT QUANTUM L OGIC AT 1.1 K 79

energy (which is in this experiment the state |↓↑〉) couples directly to the singlet (4,2) charge state.
The remaining antiparallel spin state (|↑↓〉) and the two parallel spin states (|↑↑〉, |↓↓〉) couple to
the three triplet (4,2) charge states. This allows to map the |↓↑〉 and the other basis states to dif-
ferent charge configurations ((4,2) or (5,1) states), which can be read out using the SET. As shown
in Fig. 5.1d, the optimal readout position can be obtained by sweeping ² and applying a π-pulse
to Q2. From the detuning lever arm of α² = 0.044 eV/V, extracted from the thermal broadening of
the polarization line, we find a readout window of 155 µeV where we can efficiently discriminate
between the singlet and triplet states.
In this high temperature operation mode, the readout visibility is mainly limited by the broad-
ening of the SET peaks. In order to maximize our sensitivity we subtract a reference signal from
each trace, then we average and normalize the resulting signal (for more details on the readout see
Fig. 5.5 and 5.6).
Figure 5.2b-g shows the single-qubit characterization of the two-qubit system. We observe
clear Rabi oscillations for both qubits (Fig. 5.2b, c) as a function of the microwave burst duration.
From the decay of the Ramsey fringes (Fig. 5.2d, e) we extract dephasing times T2(Q1) ∗ = 2.1 µs
∗
and T2(Q2) = 2.7 µs, comparable to experiments at similar high temperature [8]. These times are
significantly shorter than the longest reported times for 28 Si [11], however they are still longer than
the dephasing times for natural silicon at base temperature [16, 17]. Furthermore we measure spin
lifetimes (see Fig. 5.8) of T1(Q1) = 2.0 ms and T1(Q2) = 3.7 ms, consistent with values reported in a
5
similar device at a similar operating temperature [6].
We characterize the performance of the single-qubit gates of the two qubits by performing
randomized benchmarking [29]. In the manipulation phase we apply sequences of random gates
extracted from the Clifford group, followed by a recovery gate that brings the system to the |↓↓〉 and
|↑↑〉 states for Q1 and Q2 respectively. By fitting the decay of the readout signal as a function of the
number of applied gates to an exponential decay we extract qubit fidelities FQ1 = 98.7 ± 0.3 % and
FQ2 = 99.3 ± 0.2 %, with the second one above the fault tolerant threshold.

5.4. T WO -Q UBIT Q UANTUM L OGIC AT 1.1 K

We now turn to the two-qubit gate characterization. The ability to tune the exchange interaction
[9] is the basis to perform two-qubit operations with electrons in quantum dots. By turning on
the exchange interaction, either by controlling the detuning energy or the tunnel coupling, the
resonance frequencies of each qubit shift depending on the spin state of the other qubit. The
central inset in Fig. 6.1a shows this frequency shift for both qubits as a function of the detuning
energy between the two quantum dots, with and without a π-pulse applied to flip the spin state
of the other qubit. The full exchange spectrum is composed of the transitions f 1 (|↓↑〉 −→ |↓↓〉), f 2
(|↑↑〉 −→ |↑↓〉), f 3 (|↓↓〉 −→ |↑↓〉) and f 4 (|↓↑〉 −→ |↑↑〉). The exchange interaction J can be extracted
as the differences f 2 − f 1 and f 4 − f 3 , from which we measure tunable J in the range 0.5 - 18 MHz
(see Fig. 5.4a). At even larger exchange couplings the readout visibility drastically reduces, which
we attribute to a decrease of T2∗ (see Fig 5.4b). By fitting the exchange spectrum (see section 5.8)
we extract a tunnel coupling t c = 0.8 GHz and a Zeeman energy difference δE z = 9.1 MHz.
Having demonstrated the tunability of the exchange interaction, we use this to demonstrate
two-qubit operation. When the exchange is turned on, the resulting shift in resonance frequency
can be used to implement state selective ESR transitions (CROT), which are equivalent to a CNOT
gate up to single-qubit phases. Figure 6.1a shows controlled oscillations for both qubits, with the
control qubit set either to the spin down or spin up state, where we have set the exchange inter-
action to J = 2.5 MHz. When we prepare the state of the control qubit such that the target qubit is
in resonance with the external microwave control, we observe clear oscillations of the target qubit
as a function of the microwave burst duration, with no significant decay after multiple rotations.
 80 5. U NIVERSAL QUANTUM LOGIC IN HOT SILICON QUBITS

a b
Q1 Q1
1 1 X/2 Z(θ) X/2 2X
Triplet probability

Triplet probability
0.8 0.8 Q2 Q2
2X X/2 Z(θ) X/2
0.6 0.6
0.4 0.4
1 Q1 1 Q2
0.2 0.2

Triplet probability

Triplet probability
0.8 0.8
0 0
0.6 0.6
0 2 4 0 2 4
0.4 0.4
Time ( s) Time ( s)
30
f4 0.2 0.2
20
f (MHz)

0 0
10 f3
0 f2 0 2 0 2
(rad) (rad)
-10
f1

5 80 65
(mV)
50
c

T
d

1 1 Z/2 T X/2
1 F = (86.1 ± 0.6) %

state probability
Triplet probability

Triplet probability

0.8 0.8 C C
0.6 0.6 0.8

0.4 0.4 T X T X
0.6
0.2 0.2
C X/2

|
0 0 0.4
0 2 4 0 2 4 T X 0 5 10
Time ( s) Time ( s) Number of cliffords

Figure 5.3: Exchange and two-qubit logic at 1.1 K. a Conditional rotations on all the frequencies f i , the color
code refers to the central inset showing the full exchange diagram obtained from a gaussian fit of the data
shown in Fig. 5.7. The frequency offset is 6.948 GHz. The black lines correspond to the same transition f i ,
driven with the control qubit in the opposite state. An initialization π-pulse and recovery π-pulse are applied
to the control qubit for the sequences where either Q1 is in the spin down state or Q2 is in the spin up state. All
Rabi frequencies are set to approximately 1 MHz by adjusting the power of the microwave source to compen-
sate for the frequency dependent attenuation of the fridge line. Even when the exchange interaction is turned
on we find the resonance frequencies of both qubits to be stable over the course of several hours (see Fig. 5.9).
b Phase acquired by the control qubit during a CROT operation. A CROT gate, together with a Z-rotation of
π/2 on the control qubit is equivalent to a CNOT operation. Z gates are implemented by a software change
of the reference frame. c Primitive gates used to generate the two-qubit Clifford group (11520 gates in total).
On average, each Clifford contains 2.5694 primitive gates. Since the Z/2 gates are implemented via a software
change of the reference frame, they are not included in the gate count. All gates shown in the figure (except
for the Z/2 gate) are implemented with two π/2 controlled rotations. The compilation scheme is identical to
the one in [18]. d Decay of the |↓↑〉 state probability as a function of the number of two-qubit Cliffords applied.
A recovery gate returns the system to the |↓↑〉 state. Since we include the recovery gate in the Cliffords count,
the first data point correponds to NCliff = 2. Each data point corresponds to the average of 150 random se-
quences. The fidelity F = 86.1 ± 0.6 % corresponds to the average fidelity of the primitve gates shown in c. We
have normalized the state probabilities to remove the readout errors.
5.5. T EMPERATURE E FFECTS ON QUANTUM C OHERENCE 81

a b c J = 0.5 MHz
Q1 Q2
J = 2.5 MHz
20
3 3 3
Exchange (MHz)

15

T *2 ( s)

T *2 ( s)

T *2 ( s)
2 2 2
10

5 1 1 1

0 0 0 0
80 65 50 0 2 4 6 0.4 0.8 1.2 0.4 0.8 1.2
(mV) Exchange (MHz) Temperature (K) Temperature (K)

Figure 5.4: Dephasing dependence on temperature and exchange interaction. a Exchange energy measured
as a function of detuning. The data correspond to f 2 − f 1 and f 4 − f 3 as obtained from Fig. 6.1a. b Dephasing
time of Q2 as a function of the exchange interaction, fitted with a model taking into account gaussian quasi
static noise (see section 5.9). Similar data from Q1 are shown in Fig. 5.10. c Temperature dependence of the
dephasing time with the exhange interaction set to the minimum obtained by sweeping ² (J = 0.5 MHz) and
with the exchange interaction set to acquire the CROT operations of Fig. 6.1a (J = 2.5 MHz).

When we flip the state of the control qubit, the resonance frequency of the target qubit is shifted
and the target qubit is not driven by the microwave control.
In order to investigate the coherence of the two-qubit logic, we apply a sequence where we
5
interleave a CROT operation with duration 2π in between two π/2 single-qubit gates applied to
the control qubit with variable phase θ. As shown in Fig. 6.1b, when we invert the second π/2
pulse (θ = π) this cancels out the π phase left by the CROT operation on the control qubit and
we correctly measure transitions to the |↓↓〉 and |↑↑〉 states. This demonstrates the execution of
a coherent CROT, since the control qubit maintains its coherence even when the target qubit is
driven.
In order to show the universality of our gate set we also demonstrate two-qubit randomized
benchmarking. We apply random gates from the 11520 two-qubit Clifford group, recover the state
to the |↓↑〉 and measure how the singlet probability decays over the number of applied gates. The
decay is shown in Fig. 6.1d and the primitive gates used in 6.1c. The lower fidelity (F = 86.1 ± 0.6
%) compared to the single-qubit benchmark can be attributed to the longer time spent by the
qubits idling, which causes them to decohere faster. Possible improvements include simultaneous
driving of two transitions to reduce idling times, optimized pulse shaping to reduce accidental
excitations of nearby transitions and operation at the symmetry point [30, 31].

5.5. T EMPERATURE E FFECTS ON Q UANTUM C OHERENCE

To further investigate the quantum coherence of the system we measure the decay of the Ramsey
fringes for different values of the exchange interaction, see Fig. 5.4b. We find that by increasing the
exchange interaction the coherence is reduced, which we explain by the increased qubit sensitiv-
ity to electrical noise. We can fit the data with a model (see section 5.9) that includes quasi-static
electrical noise coupling in via the exchange interaction and via the Zeeman energy difference
between the two qubits. From the fit we extract the fluctuation
p amplitudes δ² = 21 µeV (corre-
sponding to a power spectrum at 1 Hz of A ² ≈ 6 µeV/ Hz) and δE Z = 400 kHz. The noise in ²
is comparable with values extracted at fridge base temperature [32], and consistent with charge
noise values extracted from current fluctuation measurements of SETs (see section 5.9 for further
information) [6, 33].
To analyze the thermal impact, we characterize the temperature dependence of T2∗ for two
values of exchange (J = 0.5 MHz and J = 2.5 MHz) and we find it to be approximately stable in
the range T = 0.45 K - 1.25 K (see Fig. 5.4c). While weak dependencies of T2∗ have been reported
 82 5. U NIVERSAL QUANTUM LOGIC IN HOT SILICON QUBITS

in other single-qubit experiments [8], we observe here that the weak temperature dependence
is maintained even when the exchange interaction is set to an appreciable value where we can
perform two-qubit logic.
The origin of the electrical noise limiting T2∗ can potentially come from extrinsic or intrinsic
sources. Although we cannot rule out all extrinsic noise sources, we have confirmed that atten-
uating the transmission lines does not affect the T2∗ and we thus rule out a direct impact of the
waveform generator and the microwave source. When intrinsic charge noise is the dominant con-
tribution, a simple model based on an infinite number of two-level fluctuators (TLFs) predicts a
square root dependence of the dephasing rate on the temperature [34]. However, this model as-
sumes a constant activation energy distribution of the TLFs. Deviations from this assumption have
been observed in SET measurements, leading to anomalous temperature dependencies [35]. The
small size of quantum dots, in particular SiMOS qubits, may lead to only a few TLFs being relevant
for the dephasing and these may explain the observed weak temperature dependence (see section
5.10 for more details).
Importantly, the weak dependence of T2∗ on temperature makes silicon qubits remarkably ro-
bust against temperature, enabling to execute a universal quantum gate set. The ability to operate
lithographically defined qubits above one Kelvin resolves one of the key challenges toward the in-
tegration of quantum hardware and control electronics on the same chip. This integration can re-
5 duce the number of lines going from room temperature to the device and, at the same time, greatly
simplify on-chip wiring, facilitating the realization of quantum integrated circuits for large-scale
quantum computation.

5.6. M ETHODS
5.6.1. E XPERIMENTAL SETUP
All measurements are performed in a Bluefors dry dilution refrigerator with a base temperature
of Tbase ≈ 0.45 K, operated at T = 1.1 K. DC-voltages are applied using battery-powered voltage
sources and AC voltages are applied through bias-tee on the sample PCB with a cut-off frequency
of 3 Hz. The pulse sequences are generated by an arbitrary waveform generator (AWG) Tektronix
AWG5014C, combined with a microwave signal generated by a Keysight PSG8267D vector source.
ESR currents are delivered via the PSG8267D using the internal IQ-mixer, driven by two output
channels of the AWG. Both qubits can be addressed by setting the vector source to an intermediate
frequency and IQ-mixing the signal with a (co)sine wave generated on channels 3 and 4 of the
AWG. For the two-qubit randomized benchmarking experiment the pulse sequences are generated
by an arbitrary waveform generator Keysight M3202A, that allows for faster waveform uploads. We
apply a source-drain bias voltage of VSD = 0.5 mV to the single-electron transistor and measure
the current using an in-house built transimpedance amplifier.

5.6.2. S INGLE - QUBIT RANDOMIZED BENCHMARKING

The single-qubit Clifford group C 1 consists of 24 rotations. We implement the group using X and
Y rotations, using the primitive gates: {I , ±X /2, ±Y /2, ±X , ±Y }. On average one Clifford gate con-
tains 1.875 primitive gates. We implement the gates using only frequencies f 1 and f 4 for Q1 and
Q2 respectively. The complete list of gates is given in the following table:
5.6. M ETHODS 83

Single-qubit Cliffords
I
X
Y
Y, X
X/2, Y/2
X/2, -Y/2
-X/2, Y/2
-X/2, -Y/2
Y/2, X/2
Y/2, -X/2
-Y/2, X/2
-Y/2, -X/2
X/2
-X/2
Y/2
-Y/2
-X/2, Y/2, X/2 5
-X/2, -Y/2, X/2
X, Y/2
X, -Y/2
Y, X/2
Y, -X/2
X/2, Y/2, X/2
-X/2, Y/2, -X/2

The phase control needed to implement X and Y rotations is achieved using the internal I-Q
mixer of the microwave source. The fidelity reported in Figure 5.2 refers to the average fidelity of
the gates in the generator group. All error bars are 1 standard deviation from the mean.

5.6.3. T WO - QUBIT RANDOMIZED BENCHMARKING

The two-qubit Clifford group C 2 consist of 11520 elements c 2 with properties c 2† P c 2 ∈ ±P where
P are the Pauli operators. We generate the Clifford gates in our experiment using the set of con-
ditional rotations in Fig. 6.1a where two subsequent conditional rotations implement a primitive
gate. We compile the Clifford gates from the set of primitive gates together with virtual Z/2 gates on
both qubits and search for combinations with the minimal amount of gates. The resulting average
Clifford gate consists of 2.5694 primitive gates which are calibrated such that each conditional ro-
tation takes exactly 330 ns, with the exchange interaction set to 3 MHz. To minimize cross-talk, the
timing and the exchange interaction are chosen such that the off-resonant pulse is synchronized
with the resonant pulse. All error bars are 1 standard deviation from the mean.
 84 5. U NIVERSAL QUANTUM LOGIC IN HOT SILICON QUBITS

5.7. E XTENDED D ATA

a b
T = 0.5 K T = 1.1 K T = 1.1 K
250 250

200 200 0.6

Triplet probability
150 150
Counts

Counts
0.5
100 100

50 50 0.4
0 0
-100 -50 0 50 100 -100 0 100 0 2 4 6 8
Current (pA) Current (pA) Time ( s)

Figure 5.5: Charge readout and visibility. a Histograms of the readout signal for the singlet and triplet T−
state for two operating temperatures. The sensitivity is reduced at higher temperatures mainly because of the
thermal broadening of the Coulomb peaks. The readout signal is obtained by subtracting a reference line,
obtained from a sequence where no microwave pulse is applied. The integration time corresponds to 40 µs.
The readout fidelity may be improved by optimizing the charge sensing [36], and by using an RF-reflectometry
or dispersive measurement scheme, as shown in [7]. b Rabi oscillations of Q1 (see also Fig. 5.2b), obtained
by assigning to each single-shot trace the state spin up or spin down, using a threshold obtained from the
5 histograms in a. From the data we can extract the visibility, which we find to be V ≈ 0.2 at T = 1.1 K.

a
Triplet Conversion Probability

1
0.8
0.6
0.4
0.2
0

DU DD UU UD
States

Figure 5.6: Spin to charge conversion. a Normalized probability that the four two-electron spin states are
detected as a triplet state. The probability that the triplet antiparallel spin state is correctly identified as a
triplet can be lowered by the non perfect adiabaticity of the pulse and by a faster triplet-singlet relaxation.

a b

30 30

20 20

f (MHz)

f (MHz)

10 10

0 0

-10 -10

1785 1770 1755 1785 1770 1755

(mV)
(mV)

Figure 5.7: Exchange interaction. a-b Resonance frequency of both qubits as a function of the detuning en-
ergy. In a we show the transitions f 1 and f 4 , while in b we show the transitions f 2 and f 3 . We measure the
excited states by ESR controlled spin flips applied to the control qubit.
5.7. E XTENDED D ATA 85

a Q1 b Q2

1 1

Triplet probability

Triplet probability
0.8 0.8
0.6 0.6
0.4 0.4
0.2 T1 = (2.0 ± 0.3) ms 0.2 T1 = (3.7 ± 0.9) ms
0 0

0 2 4 6 8 0 5 10 15
Time (ms) Time (ms)

Figure 5.8: Relaxation times. a-b Single-spin relaxation times of Q1 and Q2. The measurements are performed
by fitting the decay of the states |↓↑〉 and |↑↓〉 to the |↓↓〉 state. We extract T1(Q1) = 2.0 ms and T1(Q2) = 3.7 ms,
consistent with [6]. Triplet probabilities have been normalized to remove readout errors.

a b
Q1

200 200 5
f 1 (KHz)

f 1 (KHz)
0 0
-200 -200
-400
-600 -600
-400 0 5
(mV)

0 1 2 3 4 5 0 50 100
Time (hours) Counts
readout

-5
Q2

200 200
-10
f 4 (KHz)

f 4 (KHz)

0 0
-200 -200
-400 -400 -15
-600 -600 0 1 2 3 4 5
0 1 2 3 4 5 0 50 100 Time (hours)
Time (hours) Counts

Figure 5.9: Time dependence of resonance frequencies and readout point. a Time dependences of the reso-
nance frequencies f 1 and f 4 of Q1 and Q2 respectively. The exchange interaction is set to 2.5 MHz. The data
have been offset by 6.9491 GHz and 6.9620 GHz for f 1 and f 4 respectively. b Time dependence of the readout
point obtained by taking a sweep along the detuning axis in a measurement identical to the one shown in Fig.
5.1d. The best readout point is returned with a gaussian fit of the visibility peak.

3
T *2 ( s)

0
1 3 5
Exchange (MHz)

Figure 5.10: Dephasing times for Q1 and Q2 as a function of exchange interaction. a Dephasing times of Q1
and Q2 as a function of exchange interaction, fitted with the model discussed in section 5.9. Due to a different
tuning configuration, the dephasing times are slightly longer than the ones reported in Figure 5.2. In this new
configuration we measure a tunnel couping t c = 0.8 GHz and a Zeeman energy difference δE Z = 10.6 MHz.
 86 5. U NIVERSAL QUANTUM LOGIC IN HOT SILICON QUBITS

5.8. F IT OF THE FULL EXCHANGE SPECTRUM

We operate our device in the (5,1) charge occupation regime, close to the (4,2) transition. Con-
sidering that the four lowest-energy electrons in the right dot completely freeze out the s-orbital,
they do not contribute to the dynamics and thus can be neglected. In the (5,1) charge occupation
regime there are four relevant spin states in the lowest valley, |↑↑〉 , |↑↓〉 , |↓↑〉 , |↓↓〉 which are coupled
via tunneling to the (4,2) singlet states |s(4, 2)〉. Therefore, our system can be well-described by the
Hubbard Hamiltonian [37]

Ez ∗ /2
E x,2 ∗ /2
E x,1 0 0

 E x,2 /2 ∆E z /2 0 ∗ /2
E x,2 t 0∗ 
 
H =

E x,1 /2 0 −∆E z /2 ∗ /2
E x,1 −t 0∗ ,

(5.1)
 
 0 E x,2 /2 E x,1 /2 −E z 0 
0 t0 −t 0 0 ε −U

where E z = µB (g 1 B z,1 + g 2 B z )/2, ∆E z = µB (g 1 B z,1 − g 2 B z,2 ), and E x,i = µB g i (B x,i + i B y,i ) with
B i = (B x,i , B y,i , B z,i )T and g i being the magnetic field and g-factor of quantum dot i = 1, 2. Fur-
thermore, t 0 describes the tunnel coupling between the dots, ε is defined as the difference of
chemical potentials in quantum dot 1 and 2, and U describes the charging energy of quantum dot
5 1. The microwave antenna allows
¢ us to apply local time-dependent transverse magnetic fields,
ac cos 2π f ac t + φ , giving rise to electron spin resonance. For large detuning |² −
¡
E x,i = E x,i + E x,i
U | À t 0 the system can be approximated by a Heisenberg Hamiltonian [37]

H = J (S 1 · S 2 − 1/4) + µB g 1 B 1 · S 1 + µB g 2 B 2 · S 2 (5.2)

with the spin operators S = (S x , S y , S z )T and the exchange interaction

J ≈ 2t 02 /(U − ε). (5.3)

The four resonances observed in the experiment, f 1 (|↓↑〉 −→ |↓↓〉), f 2 (|↑↑〉 −→ |↑↓〉), f 3 (|↓↓〉 −→
|↑↓〉) and f 4 (|↓↑〉 −→ |↑↑〉), are then approximately given by
q
h f1 = E z − J 2 + ∆E z2 /2 − J /2 (5.4)
q
h f2 = E z − J 2 + ∆E z2 /2 + J /2 (5.5)
q
h f3 = E z + J 2 + ∆E z2 /2 − J /2 (5.6)
q
h f4 = E z + J 2 + ∆E z2 /2 + J /2. (5.7)

In order to extract parameters

q from the experiment we simultaneously fit f 2 − f 1 = f 4 − f 3 = J /h and
f 4 − f 2 = + f 3 − f 1 = J 2 + ∆E z2 /h to extract exchange as a function of detuning, the Zeeman field
difference, and the tunneling amplitude. Our best fits yield t 0 = 0.8 ± 0.05 GHz and ∆E z = 9.1 ± 0.1
MHz.

5.9. N OISE MODEL AND NOISE FITTING

In this section we derive analytical expressions for the pure dephasing time T2∗ which are then
used to extract the noise amplitude. In a Ramsey experiment the qubits are initialized in the |↓↑〉 p
superposition state different for the two qubits: |Q2+〉 = (|↓↑〉 + |↑↑〉)/ 2
state, then brought into a p
and |Q1+〉 = (|↓↑〉 + |↓↓〉)/ 2, then mapped back after some waiting time, and finally measured.
5.10. T EMPERATURE DEPENDENCE OF THE DEPHASING TIME 87

T *2 ( s)
2

0
0 2 4 6
Exchange (MHz)

Figure 5.11: Dependence of the dephasing times on the exchange interaction. a Same data as shown in Fig.
5.4b, with fits accounting for fully correlated (blue line) and fully uncorrelated (black line) noise sources.

The phase e −2πiRΦ(τ) the two qubits Racquire during their free evolution is given by their energy dif-
ference Φ(τ) = 0τ d t f 4 and Φ(τ) = 0τ d t f 1 . We assume that there are two dominating channels
where noise can couple to the qubits, which is assumed to be longitudinal. We assume the noise
comes from electrostatic fluctuations, which can couple via the detuning energy through the ex-
change interaction and through the g-factor modulation and spin-orbit coupling to the difference
in Zeeman energy fields. The acquired phase for the two qubits are then given by 5
ΦQ1,(Q2) (t ) = f 4,(1) + [D² δ²(t ) + D∆E z δ∆E z (t )], (5.8)

where D² = ∂ f 4 /∂J × ∂J /∂² and D∆E z = ∂ f 4 /∂∆E z . The envelope of the Ramsey experiment is then
given by the free induction decay [38]
h i
2 f (τ) = 1 + exp −πi 〈Φ(τ)2 〉 (5.9)
h ³ ´i
1 + exp −πt 2 D²2 σ2² + D∆E
2
σ2∆E + κ2 D² D∆E z σ² σ∆E z (5.10)
z z

under the assumption of Gaussian distributed noise Rand zero mean 〈Φ(τ)〉 = 0. For the second line
∞
we assumed quasi-static noise with dispersion σ² = −∞ S ² (ω)d ω, where
S(ω) = −∞ 〈δ²(t )2 〉 e −i ωt d t is the power spectral density
R∞
R ∞ of the noise. Similar expressions hold
for σ∆E z . The correlation coefficient is defined κ = −∞ K ²,∆E z (ω)d ω/(σ² σ∆E z ) with the cross-
〈δ²(t )δ∆E z (t )〉 e −i ωt d t . The dephasing time is then given by
R∞
spectral density K ²,∆E z = −∞
¡ ∗ ¢−1 p q 2 2 2 σ2
T2 = π D² σ² + D∆E ∆E
+ κ2 D² D∆E z σ² σ∆E z . (5.11)
z z

We fit the dephasing times as a function of exchange with either a fully uncorrelated noise κ = 0
or a fully correlated κ = 1 ansatz. The fits can be seen in Fig. 5.11a. Our best fit yields σ² = 21 µeV,
σ∆E z = 400 kHz. Assuming that the origin of the noise is 1/ f and knowing our measurement time,
we can convert σ² to the value ofpits relative power spectrum at
¢ 1 Hz, a metric often reported in
literature. We obtain A ² ≈ 6 µeV/ Hz with σ2² = A 2² log f uv / f rf [38]. The lower and higher cutoff
¡

f rf ∼ 10−2 Hz and f uv ∼ 103 Hz are set by the experiment in the quasistatic approximation.

5.10. T EMPERATURE DEPENDENCE OF THE DEPHASING TIME

In our experiment we observe a weak temperature dependence of T2∗ . This is consistent with the
assumption of random telegraphic noise coming from two level fluctuators (TLFs). The power
spectral density of a single TLF is given by [34]
A ν
S(ω, ν) = . (5.12)
2 cosh2 [E /(2k B T )] ν2 + ω2
 88 R EFERENCES

Here, A is the coupling strength of the fluctuations, E the (activation) energy gap between the two
states of the TLF, and ν the switching rate. An explanation for the weak temperature dependence of
T2∗ arises from the fact that Eq. 5.12 saturates if k B T À E . Assuming that only a few TLFs couple to
our system there is only a small probability to find a TLF which has an activation energy E exactly
in the temperature range between 0.4 K till 1.2 K. The same arguments hold if instead of a two-level
fluctuator an Anderson impurity is the origin of charge noise [39].
On the other hand, if we assume a large ensemble of TLFs a linear temperature dependence [6]
is expected assuming that the number of ‘activated’ TLFs increases with temperature. For a large
ensemble the noise spectral density reads [34]
Z 2π f uv
S(ω) ∝ P (ν, T )S(ω, ν) d ν, (5.13)
2π f rf

where P (ν, T ) describes the contribution of the process with a switching rate between ν and
ν + d ν, thus, the probability density of a TLFs that contributes to the dephasing process. Assum-
ing a temperature dependent switching rate ν = ν0 e −E /(kB T ) leads to P (ν, T ) = P (E , T )|∂ν/∂E |−1
and one finds P (ν, T ) = P (E , T )k B T /ν [34]. Assuming a constant distribution of activation en-
ergies, P (E , T ) = const and inserting this into Eq. 5.13 the characteristic 1/ f noise with a linear
temperature dependence can be reproduced [32],
5
2π A ²
S(ω) ≈ P (E , T )k B T ≡ (5.14)
ω ω
for 2π f rf ≤ ω ≤ 2π f uv . However, recent work shows that the assumption of a constant distribution
of activation energies P (E , T ) is not entirely valid [35], and this can lead to anomalous temperature
dependencies.

R EFERENCES
[1] Ladd, T. D. et al. Quantum computers. Nature 464, 45–53 (2010).

[2] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 34 (2017).

[3] Veldhorst, M., Eenink, H. G. J., Yang, C. H. & Dzurak, A. S. Silicon CMOS architecture for a
spin-based quantum computer. Nat. Commun. 8, 1766 (2017).

[4] Devoret, M. H. & Schoelkopf, R. J. Superconducting circuits for quantum information: an

outlook. Science 339, 1169–1174 (2013).

[5] Neill, C. et al. A blueprint for demonstrating quantum supremacy with superconducting
qubits. Science 360, 195–199 (2018).

[6] Petit, L. et al. Spin lifetime and charge noise in hot silicon quantum dot qubits. Physical
Review Letters 121, 076801 (2018).

[7] Urdampilleta, M. et al. Gate-based high fidelity spin read-out in a cmos device. Nature Nan-
otechnology 14, 737–741 (2019).

[8] Yang, C. et al. Silicon quantum processor unit cell operation above one kelvin.
arXiv:1902.09126 (2019).

[9] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Phys. Rev. A 57,
120–126 (1998).
R EFERENCES 89

[10] Zwanenburg, F. A. et al. Silicon quantum electronics. Rev. Mod. Phys. 85, 961–1019 (2013).

[11] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nat. Nanotech. 9, 981–985 (2014).

[12] Itoh, K. M. & Watanabe, H. Isotope engineering of silicon and diamond for quantum com-
puting and sensing applications. MRS Commun. 4, 143–157 (2014).

[13] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nature Nanotechnology 13, 102–106 (2017).

[14] Yang, C. et al. Silicon qubit fidelities approaching incoherent noise limits via pulse engineer-
ing. Nature Electronics 2, 151–158 (2019).

[15] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[16] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins. Science 359, 439–442
(2018).

[17] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon. Nature 555,
633–637 (2018).
5
[18] Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 569, 532–536
(2019).

[19] Fowler, A. G., Mariantoni, M., Martinis, J. M. & Cleland, A. N. Surface codes: Towards practical
large-scale quantum computation. Physical Review A 86, 032324 (2012).

[20] Franke, D. P., Clarke, J. S., Vandersypen, L. M. & Veldhorst, M. Rent’s rule and extensibility in
quantum computing. Microprocessors and Microsystems 67, 1–7 (2019).

[21] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science advances 4, eaar3960
(2018).

[22] Morello, A. et al. Single-shot readout of an electron spin in silicon. Nature 467, 687–691
(2010).

[23] Yang, C. H., Lim, W. H., Zwanenburg, F. A. & Dzurak, A. S. Dynamically controlled charge
sensing of a few-electron silicon quantum dot. AIP Advances 1, 042111 (2011).

[24] Lawrie, W. et al. Quantum dot arrays in silicon and germanium. arXiv:1909.06575 (2019).

[25] Angus, S. J., Ferguson, A. J., Dzurak, A. S. & Clark, R. G. Gate-Defined Quantum Dots in In-
trinsic Silicon. Nano Lett. 7, 2051–2055 (2007).

[26] Elzerman, J. et al. Single-shot read-out of an individual electron spin in a quantum dot. Na-
ture 430, 431–435 (2004).

[27] Yang, C. H. et al. Spin-valley lifetimes in a silicon quantum dot with tunable valley splitting.
Nat. Commun. 4, 2069 (2013).

[28] Ruskov, R., Veldhorst, M., Dzurak, A. S. & Tahan, C. Electron g-factor of valley states in realistic
silicon quantum dots. Physical Review B 98, 245424 (2018).
 90 R EFERENCES

[29] Magesan, E., Gambetta, J. M. & Emerson, J. Scalable and robust randomized benchmarking
of quantum processes. Physical Review Letters 106, 180504 (2011).

[30] Reed, M. et al. Reduced sensitivity to charge noise in semiconductor spin qubits via symmet-
ric operation. Physical review letters 116, 110402 (2016).

[31] Martins, F. et al. Noise Suppression Using Symmetric Exchange Gates in Spin Qubits. Phys.
Rev. Lett. 116, 116801 (2016).

[32] Güngördü, U. & Kestner, J. Pulse sequence designed for robust c-phase gates in simos and
si/sige double quantum dots. Physical Review B 98, 165301 (2018).

[33] Freeman, B. M., Schoenfield, J. S. & Jiang, H. Comparison of low frequency charge noise in
identically patterned si/sio2 and si/sige quantum dots. Applied Physics Letters 108, 253108
(2016).

[34] Paladino, E., Galperin, Y., Falci, G. & Altshuler, B. 1/f noise: Implications for solid-state quan-
tum information. Reviews of Modern Physics 86, 361–418 (2014).

[35] Connors, E. J., Nelson, J., Qiao, H., Edge, L. F. & Nichol, J. M. Low-frequency charge noise in
5 si/sige quantum dots. arXiv:1907.07549 (2019).

[36] Eenink, H. et al. Tunable coupling and isolation of single electrons in silicon metal-oxide-
semiconductor quantum dots. Nano letters 19, 8653–8657 (2019).

[37] Burkard, G., Loss, D. & DiVincenzo, D. P. Coupled quantum dots as quantum gates. Physical
Review B 59, 2070 (1999).

[38] Ithier, G. et al. Decoherence in a superconducting quantum bit circuit. Physical Review B 72,
134519 (2005).

[39] Beaudoin, F. & Coish, W. A. Microscopic models for charge-noise-induced dephasing of solid-
state qubits. Physical Review B 91, 165432 (2015).
 6
D ESIGN AND INTEGRATION OF
SINGLE - QUBIT ROTATIONS AND
TWO - QUBIT GATES IN SILICON
ABOVE ONE K ELVIN

Spin qubits in quantum dots define an attractive platform for quantum information because of their
compatibility with semiconductor manufacturing, their long coherence times, and the ability to op-
erate above one Kelvin. However, despite demonstrations of SWAP oscillations, the integration of this
two-qubit gate together with single-qubit control to create a universal gate set as originally proposed
for single spins in quantum dots has remained elusive. Here, we show that we can overcome these
limitations and execute a multitude of native two-qubit gates, together with single-qubit control, in
a single device, reducing the operation overhead to perform quantum algorithms. We demonstrate
single-qubit rotations, together with the two-qubit gates CROT, CPHASE and SWAP, on a silicon dou-
ble quantum dot. Furthermore, we introduce adiabatic and diabatic composite sequences that al-
low the execution of CPHASE and SWAP gates on the same device, despite the finite Zeeman energy
difference. Both two-qubit gates can be executed in less than 100 ns and, by theoretically analyzing
the experimental noise sources, we predict control fidelities exceeding 99%, even for operation above
one Kelvin.

Parts of this chapter have been published in arXiv:2007.09034 [cond-mat.mes-hall]

91
 6. D ESIGN AND INTEGRATION OF SINGLE - QUBIT ROTATIONS AND TWO - QUBIT GATES IN
92 SILICON ABOVE ONE K ELVIN

6.1. I NTRODUCTION
Two-qubit gates are at the heart of quantum information science, as they may be used to create
entangled states with a complexity beyond what is classically simulatable [1], and ultimately may
enable the execution of practically relevant quantum algorithms [2]. Optimizing two-qubit gates is
therefore a central aspect across all qubit platforms [3]. In quantum dot systems, two-qubit gates
can be naturally implemented using the exchange interaction between spin qubits in neighbour-
ing quantum dots [4]. Pulsing the interaction drives SWAP oscillations, where the spin states in the
quantum dots are being exchanged, when the exchange energy is much larger than the Zeeman
energy difference of the qubits [4–6], while it results in controlled-phase (CPHASE) oscillations,
where only the phase information is exchanged, when the Zeeman energy difference is much
larger than the exchange energy [7]. Single-qubit gates need also to be implemented to access
the full two-qubit Hilbert space, and this requires distinguishability between the qubits. This is
commonly obtained through the spin-orbit coupling [8, 9] or by integrating nanomagnets [10, 11],
causing significant Zeeman energy differences. Realizing a high-fidelity SWAP-gate in this sce-
nario would require extremely large values of exchange interaction. For this reason, the CPHASE
operation has been the native gate in experimental demonstrations of two-qubit logic when the
exchange interaction is pulsed [12–14]. An alternative implementation of two-qubit logic can be
realized by driven rotations, which become state dependent in the presence of exchange interac-
tion and can be used to realize controlled-rotation (CROT) operations [15–19]. Driving rotations
can also be used to realize a resonant SWAP gate [20], which can be used to perform state swap-
ping. While universal quantum logic can be obtained by combinations of single-qubit rotations
and an entangling two-qubit operation [21], the ability to directly execute a multitude of two-qubit
6 gates would reduce the number of operations required to execute practical algorithms.
Here, we demonstrate on the same device the implementation of the CROT, SWAP, and CPHASE,
which are all essential gates in quantum computing and error correction applications. SWAP oper-
ations can in particular be useful in large quantum dot arrays, providing a mean to achieve beyond
nearest-neighbor connectivity. We overcome the limitations imposed by the finite Zeeman energy
difference between the qubits by introducing control sequences, which also allows the execution
of the CPHASE and the SWAP in short time scales and a predicted high-fidelity. Moreover, we
demonstrate these operations at temperatures exceeding one Kelvin. The cooling power at these
elevated temperatures is much larger and thereby more compatible with the operation of classical
electronics, such that quantum integrated circuits based on standard semiconductor technology
become feasible [22–24].

6.2. R ESULTS AND DISCUSSION

6.2.1. S ILICON QUANTUM DOT DEVICE
The experimental two-qubit system is based on electron spin states confined in a silicon double
quantum dot as schematically shown in Fig. 6.1a. The silicon double quantum dot is fabricated us-
ing an overlapping gate architecture on a silicon wafer with an isotopically enriched 28 Si epilayer
of 800 ppm residual concentration of 29 Si [19, 25]. In order to obtain an optimal exchange cou-
pling between the electrons, qubits Q1 and Q2 are defined with NQ1 = 1 and NQ2 = 5, where N is
the charge occupancy. Spin readout is performed at the (1,5)-(2,4) charge anticrossing, where the
|↑↓〉 tunnels to the singlet (2,4) charge state, while the other sp in states are blocked because of the
Pauli exclusion principle. By using an adiabatic pulse from the (2,4) to the (1,5) region, we initial-
ize the system in the |↓↑〉 state. Because of the limited sensitivity of the single-electron-transistor
(SET) that we use for charge readout, we average the single-shot readout traces and subtract a
reference signal. We therefore obtain a current signal, proportional to the probability to have a
blocked state. More details about the readout scheme can be found in [19]. We note that the read-
6.2. R ESULTS AND DISCUSSION 93

a b c
10 2
ESR
Bt 15 f4
P1 P2

Coherence Time ( s)

f1

Current (pA)
Energy
10
f3 10 1
Bac f2
5 Ramsey
Ez
є 0
10 0
є
6.943 6.95 6.957 6.964 10 0 10 1
Frequency (GHz) Number of
pulses
Qcontrol Qcontrol
d e
3 3
y y
x x

f 3 mw time ( s)

f 4 mw time ( s)
Current (pA)

Current (pA)
2 2
Qtarget Qtarget y
x
1 1

y y
x x 0 0
0 1 2 3 0 1 2 3
CROT CPHASE SWAP f 2 mw time ( s) f 1 mw time ( s)

Figure 6.1: Two-qubit gates and quantum coherence of silicon spin qubits operated at a T = 1.05 K. a
Schematic representation of the double quantum dot system. The device is the same as used in [19]. Two
plunger gates (P 1 and P 2 ) and one barrier gate (B t ) are used to control the detuning energy ² and the tun-
nel coupling t between the quantum dots. Spin manipulation occurs via electron-spin-resonance (ESR) using
an on-chip microwave line. The energy diagram displays the four electron spin states as a function of ². We
exploit both driven rotations and pulsed exchange for coherent control. Controlled rotations (CROTs) can in
principle be executed at all points where J 6= 0, given that gate times are appropriately set. CPHASE gates are
conveniently executed when the exchange interaction is much smaller than the Zeeman energy difference be-
tween the qubits, while SWAP oscillations can be realized when the exchange interaction is much larger. b
Using ESR control we find the four resonance frequencies of the two-qubit system. Here, the exchange inter-
action is tuned to 3 MHz. The spectrum is composed of the frequencies: f 1 (|↑↓〉 −→ |↓↓〉), f 2 (|↓↓〉 −→ |↓↑〉),
6
f 3 (|↑↑〉 −→ |↓↑〉) and f 4 (|↑↓〉 −→ |↑↑〉). c Coherence times as a function of the number of refocusing π pulses.
Here, the exchange is set to 2 MHz. The plot includes the dephasing times measured through a Ramsey ex-
periment to allow comparison. d-e Realization of CROT operations. Rabi oscillations of the target qubit are
controlled by the spin state of the control qubit. We find controlled rotations on all the four resonance fre-
quencies f 1 , f 2 , f 3 , f 4 .

out fidelity can be further improved, even at these higher temperatures [26], but here we focus
on the coherent control (details on the reconstruction are in section 6.4). We perform spin ma-
nipulation via electron spin resonance (ESR) using an on-chip aluminum microwave antenna. All
measurements have been performed in a dilution refrigerator at a temperature of Tfridge = 1.05 K
and with an external magnetic field of B ext = 250 mT.
Readout on Pauli spin blockade is relatively insensitive to temperature, since it does not rely on
any external reservoir. However, a finite temperature can still affect qubit readout in the form of an
enhanced relaxation [27]. Furthermore, the initialization fidelity can also be lowered due to a non
zero population of the excited valley states in the (2,4) charge configuration. By taking into account
the two singlet and the three triplet states and estimating a valley splitting of E vs = 300 µeV from
previous works [19, 27], we compute a total population of the ground singlet (2,4) charge state of
87%. This initialization fidelity can be pushed beyond 99% with a valley splitting E vs > 550 µeV (see
section 6.5). Similar valley splitting values have already been measured in Si-MOS samples [9].

6.2.2. Q UBIT CHARACTERIZATION

We control the exchange interaction J via the detuning ² between the two quantum dots and we
measure couplings from J = 2 MHz up to J = 45 MHz, as shown in Fig. 6.4a. By fitting the exchange
spectrum we extract a Zeeman energy difference between the two qubits ∆E z = 11 MHz, which
originates from the the electron g-factor variations due to spin orbit coupling. This frequency
 6. D ESIGN AND INTEGRATION OF SINGLE - QUBIT ROTATIONS AND TWO - QUBIT GATES IN
94 SILICON ABOVE ONE K ELVIN

difference is large enough to have a negligible impact on qubit control fidelities. The fitting sug-
gests a negligible dependence of ∆E z on detuning, further supported by the small magnetic field
applied and the absence of external magnetic gradients. Figure 6.1b shows the four resonance fre-
quencies of the two-qubit system when J = 3 MHz. At this value of exchange interaction we tune
the π-rotation times to be t CROT = 660 ns such that we synchronize the Rabi oscillations of the
target transition with the closest off-resonant transition in order to suppress crosstalk [28]. From
Ramsey experiments on frequencies f 1 and f 4 we measure dephasing times T2,Q1 ∗ = 2.3 µs and
∗
T2,Q2 = 2.9 µs. The Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence can extend the coherence
times, by filtering out the low frequency noise. As shown in Fig. 6.1c, we measure a maximum
T2,Q1 = 63 µs and T2,Q2 = 44 µs when 15 refocusing pulses are applied, setting benchmarks for the
coherence time of quantum dot spin qubits at temperatures above one Kelvin.
When the exchange interaction is set to a non-zero value, it is possible to realize the CROT
via driven rotations since the resonance frequency of one qubit depends on the state of the other
qubit. This CROT gate is a universal two-qubit gate and equivalent to a CNOT gate up to single
qubit phases [19]. Figures 6.1d-e show controlled rotations by setting both configurations of target
and control qubits.

6.2.3. H OT, FAST, AND HIGH - FIDELITY CPHASE GATES

An alternative way to achieve a universal gate set is through the implementation of the CPHASE
gate. Moving in detuning energy toward the (1,5)-(2,4) charge anticrossing lowers the energy of
the antiparallel |↓↑〉 and |↑↓〉 states with respect to the parallel |↓↓〉 and |↑↑〉 spin states. Therefore,
pulsing the detuning for a time t results in a phase gate on the target qubit conditional on the
6 spin state of the control qubit. When the total phase φ = φ|↓↑〉 + φ|↑↓〉 = (2n + 1)π with n integer,
a CPHASE gate is realized [7]. A high-fidelity implementation of such a gate requires a Zeeman
energy difference between the two qubits much larger than the exchange interaction, in order to
suppress the evolution of the exchange gate [4]. This condition is conveniently met in devices with
micromagnets [13–15], where the CPHASE is the most natural choice as native two-qubit gate.
In our system, ∆E z is comparable in magnitude to the accessible J (see section 6.6), due to
the small B ext applied. This means that a detuning pulse will also cause the |↓↑〉 and |↑↓〉 states to
undergo SWAP rotations. While these rotations occur along a tilted angle due to the non-zero ∆E z ,
they can still reduce the fidelity of the CPHASE gate. In order to avoid unwanted SWAP rotations
we implement an adiabatic detuning pulse, by ramping ² to the desired value instead of changing
it instantaneously (see schematic in Fig. 6.2j). In this way, a high-fidelity CPHASE gate can still
be realized with an arbitrarily small ∆E z at the cost of a longer gate time. In Fig. 6.2a-c and 6.2d-f
we change the duration of a detuning pulse in between a Ramsey-like experiment on Q1, with and
without a π pulse applied to Q2. The frequency of the oscillations of Q1 depends strongly on the
spin state of Q2, thereby demonstrating a controlled phase operation. Because of the finite Zee-
man energy difference, the antiparallel |↓↑〉 state shifts significantly more in energy than the |↑↓〉
state. Consequently, the oscillations in Fig. 6.2c are significantly faster than in Fig. 6.2f. Similarly,
the decay time in Fig. 6.2e is significantly longer than in Fig. 6.2b because of the lower sensitivity
to electrical noise. In Fig. 6.2g-i the pulse time is calibrated such that the total phase φ = 3π. We
measure this in a Ramsey-like experiment where we probe the phase acquired by the target qubit
for different control qubit states. From Fig. 6.2h-i we can observe that the resulting oscillations
are nicely out-of-phase, which demonstrates the CPHASE gate. We achieve a gate time t CPHASE =
152 ns, which is mostly limited by the adiabatic ramps which take t r = 60 ns. From a comparison
with simulations we find that the contribution of both ramps to the total phase φ is approximately
1.7π.
This gate time can be significantly sped up with the implementation of a geometric CPHASE
gate, that does not require adiabaticity [29]. For the implementation of this gate we synchronize
6.2. R ESULTS AND DISCUSSION 95

a d g j
Q1 Q1 T J=27.5 MHz
X/2 X/2 X/2 X/2 X/2 X/2(θ)

Detuning
J=10 MHz x
UCZ(t) UCZ(t) CZ
Q2 Q2 C
X ( X ( y
67 ns Time
b e h k
15 20 Q1
T*2,C = (397 ± 24) ns T*2,C = (3.9 ± 0.6) µs 1
1

Spin up Probability

States Probability
Current (pA)

Current (pA)
10 15 UU
DU
0.5 0.5
5 10 UD
DD
0 5
c f 0 0
0 200 400 600 0 1000 2000 3000
85 85
Time (ns) Time (
s) i 55 60 65 70 75
0
/2
3
/2 2
l
Q2 Phase (rad) (mV)
80 80 1 f

Spin up Probability
Current (pA)

Current (pA)

(mV)

(mV)

3 /2
75 75

Phase (rad)
0.5
70 70
0 /2
65 65
0 200 400 600 0 1000 2000 3000 0
/2
3
/2 2
65 70 75
Time (ns) Time (ns) Phase (rad) (mV)

Figure 6.2: Adiabatic and diabatic CPHASE operation at T = 1.05 K. b-c,e-f Conditional phase oscillations by
adiabatically pulsing the detuning energy ² to increase the exchange interaction J , measured using the quan-
tum circuit in a,d. The antiparallel spin states acquire a phase with respect to the parallel states, resulting
in coherent oscillations as a function of the duration of the detuning pulse. At smaller detuning values, the
exchange interaction increases resulting in faster oscillations. Due to the exchange interaction, the energy dif-
ference E ↓↑ − E ↓↓ (measured in b-c) is smaller than E ↑↑ − E ↑↓ (measured in e-f), resulting in an acquired phase
on the target qubit (T) that is dependent on the state of the control qubit (C). g Schematic of the quantum
circuit to verify CPHASE operation. The adiabatic detuning pulse of the CPHASE gate is tuned such that the
antiparallel spin states acquire a total phase of 3π. The exchange is increased to J = 27.5 MHz using a ramp
t r = 60 ns and the total gate time is t CPHASE = 152 ns. We verify CPHASE operation by measuring the normal-
ized spin-up probability, obtained through conversion of the readout current, and observe clear antiparallel
6
oscillations (measured in h-i). d Schematic representation of an adiabatic (dashed black and shown in g-i) and
a diabatic (solid blue) CPHASE. The diabatic CPHASE is optimized by changing the amplitude of ² and mea-
suring probabilities of the four possible spin states in k. Due to the finite Zeeman difference (∆E z = 11 MHz)
SWAP-interactions are not negligible. However, the exchange can be tuned such that the states undergo rota-
tions of 2π. We tune and optimize this by measuring the phase, projected to the spin states through a π/2-pulse
on the target qubit. We obtain a diabatic CPHASE for t CPHASE = 67 ns (measured in k-l).

the unwanted exchange oscillations with the total gate duration, i.e. our gate performs a CPHASE
evolution while the exchange oscillations performs a complete cycle. For a perfectly diabatic pulse
the condition for the exchange interaction is:
q
J = (4 J res + 3∆E z2 + 4J res
2 )/3, (6.1)

where J res is the residual exchange interaction at the point where we perform CROT gates (see
section 6.7).
Figures 6.2k-l show the experimental implementation of the geometric CPHASE gate. We
sweep the amplitude of the detuning pulse and monitor the spin state probabilities (see section
6.4) during exchange oscillations, and the total phase acquired by the antiparallel spin states. We
notice that, when ² ≈ 68 mV, the antiparallel spin states execute a 2π rotation, while acquiring a to-
tal phase shift of π. At this value of detuning we measure J ≈ 10 MHz (see section 6.6) and therefore
in agreement with Eq.6.1. The total gate time is reduced here to t CPHASE = 67 ns.

6.2.4. H OT, FAST AND HIGH - FIDELITY SWAP OPERATION

We now turn to the implementation of a SWAP gate, the originally proposed quantum gate for
quantum dots [4]. Despite the experimental demonstration of exchange oscillations [5, 6, 30], its
 6. D ESIGN AND INTEGRATION OF SINGLE - QUBIT ROTATIONS AND TWO - QUBIT GATES IN
96 SILICON ABOVE ONE K ELVIN

a b e

x x x
60 60

Exchange interaction
Ramp time (ns)

Current (pA)
y y y
40 40 27 MHz

20 20
3 MHz 2.4 MHz
0 tSWAP tcorr tSWAP
0
0 100 200 300 400 0 100 200 300 400
Time (ns) Time (ns) Time

c d f g
1 1 1 1 Pflip= 93 %
0.8 0.8 0.8
Pflip= 64 %

Probabiliy
Probability

Probability
Probabiliy

0.6 0.6 0.6

0.5
0.4 0.4 0.4

0.2 0.2 0.2

0 0 0 0

0 5 10 15 20 25 UU
UU DU
DU UD
UD DD
DD 58 60 62 64 66 68 UU DU UD DD
UU DU UD DD
Time (ns) States t corr (ns) States

Figure 6.3: Pulsed SWAP and composite exchange pulse for high-fidelity SWAP at T = 1.05 K. a-b SWAP oscil-
lations as a function of the ramp time for a detuning pulse such that J = 23 MHz. When the pulsing becomes
adiabatic with respect to variations in J , the exchange oscillations are suppressed. In order to maximize the
readout signal we project the |↑↓〉 to the |↑↑〉 with a π pulse on f2 . Traces in b correspond to ramp times 0, 16,
33, 49 and 67 ns. We do not consider in these timings the finite bandwidth of the setup. Each trace has been
offset by 15 pA for clarity. c-d Probabilities of the four spin states as a function of the SWAP interaction time.
The states |↑↑〉 and |↓↓〉 are not affected, while the states |↓↑〉 and |↑↓〉 oscillate. Due to the finite Zeeman differ-
ence we achieve a maximum |↑↓〉 state probability of 64 % for t SWAP = 18 ns. The exchange interaction is set to
J = 27 MHz. e Pulse sequence of the composite SWAP gate to correct for errors coming from the finite Zeeman
6 energy difference. The Bloch spheres on top show the time evolution when starting in the |↓↑〉 state, with the
Bloch vector depicted in nanosecond time steps. We first diabatically pulse the exchange to J = 27 MHz, in
order to bring the state on the equator of the singlet-triplet Bloch sphere. Then we correct for the phase offset
with an adiabatic exchange pulse to J = 2.4 MHz. We complete the state flip with another exchange pulse to
J = 27 MHz. f Spin state probability after applying the composite SWAP and as a function of the adiabatic pulse
time t corr , from which we find the optimum t corr = 62 ns. g Spin state probability after executing the composite
SWAP sequence starting from the initial state |↓↑〉. Compared to the detuning pulse as shown in d we find a
clear improvement in the spin flip SWAP probability.

implementation together with single-qubit gates is rather challenging because of the requirement
of a negligible Zeeman difference between the qubits. In the following we will discuss a protocol
that can overcome this problem and allow for a high-fidelity SWAP gate, even in the presence of a
finite ∆E z .
In order to observe SWAP oscillations, we implement a sequence where we initialize in the |↓↑〉
state and pulse ² for a time t . Clear exchange oscillations between the |↓↑〉 and the |↑↓〉 state are vis-
ible when qthe detuning pulse is diabatic (see Fig. 6.3a and 6.3b), where the oscillation frequency is
f SWAP = J 2 + ∆E z2 . As we make the pulse more adiabatic by ramping ², the oscillations disappear
and the regime becomes suitable for a CPHASE implementation as discussed before. Even when
the detuning pulse is perfectly diabatic, we do not obtain a perfect SWAP due to the finite ∆E z .
Instead, the spin states rotate in the Bloch sphere around the tilted axis of rotation r = (J , 0, ∆E z )T ,
similar to what happens for off-resonant driving. Figure 6.3c and 6.3d show that when starting in
the |↓↑〉 state, a maximum |↑↓〉 state probability of 64% is obtained in t SWAP = 18 ns, which is in
agreement with our simulated predictions (see section 6.7).
Composite pulse sequences [36, 37] can correct for the tilted axis of rotation. It is possible
to achieve full population transfer with an exchange sequence consisting of alternating diabatic
and adiabatic exchange pulses. The corresponding time evolution operators in the odd parity
subspace are:
6.2. R ESULTS AND DISCUSSION 97

Gate time (ns) F ideal (%) F noise (%)

CROT 660 99.4 89.0
CPHASE 152 99.9 97.8
Diabatic CPHASE 67 99.9 99.4
SWAP 19 84.3 84.2
Composite SWAP 89 99.9 99.4

Table 6.1: Gate times and simulated fidelities for silicon qubits at T = 1.05 K. Gate times and simulated fi-
delities for all the two-qubit gates discussed in section 6.2.3, where F ideal represent the fidelity in the absence
of noise and F noise takes into account the experimental noise at 1.05 Kelvin. We find high-fidelity two-qubit
gates can be obtained in silicon above one Kelvin, by using diabatic CPHASE or composite SWAP sequences.
The CROT fidelity is calculated as a conditional π-flip for better comparison. Good agreement is obtained
with previous experiments [19], confirming that the simulated noise is an accurate estimate of the real noise.
Further improvement in the fidelities of the CROT and the CPHASE may be obtained by incorporating pulse
shaping [31–35].

Ur = e iΦr e i θr r ·σ (6.2)
iΦz i θz Ẑ
Uz = e e (6.3)

for a diabatic and an adiabatic pulse respectively (see section 6.7). Here σ = ( X̂ , Ŷ , Ẑ ) is the
vector consisting of the Pauli matrices, Φr,z = J t r,z /2 the accumulated entangling phase during
6
q
the pulse, and θr,z = t r,z J 2 + ∆E z2 /2 the angle of rotation. The condition for a SWAP gate is
Utot = UrUzUrUz2 Ur2 · · · ≡ X̂ . The number of necessary pulses depends on the angle of rotation;
obviously a minimal pulse sequence requires |∆E z | ≤ J . In the typical regime of operation for de-
vices with micromagnets, where J < ∆E z , a multi-step sequence is required. In the limit J ¿ ∆E z
many steps are necessary and the pulse sequence becomes gradually an ac signal giving rise to the
ac-SWAP gate [20]. Furthermore, it is essential to include the global phase which corresponds to a
conditional phase evolution in the full two-qubit space and needs to vanish when implementing
a SWAP
p gate. This protocol is highly versatile and can also produce maximally entangling gates,
i.e., SWAP if Utot ≡ i X̂ /2 and i SWAP for Utot ≡ i X̂ . While finding an optimal sequence for such a
composition can be done in general following the procedure of [37], here we extend these consid-
erations into a multi-qubit space, which gives rise to additional constraints.
A possible minimal length solution for a SWAP gate is sketched in Fig. 6.3e and the trajectory
of the qubit state is seen in the inset. In the experiment, we calibrate the exchange interaction at all
stages of the pulse, fix the time of the diabatic pulses to 12 ns and sweep the length of the adiabatic
pulse t corr in order to find the best point. Figure 6.3f shows how the four spin probabilities change
when sweeping t corr . We find an optimal t corr = 62 ns and the four spin state probabilities for a
total pulse duration t SWAP = 88 ns are plotted in Fig. 6.3g. The SWAP probability exceeds 90%,
where the remaining error is dominated by miscalibrations, inaccuracies in the gates needed to
reconstruct the spin state probabilities, and state-preparation-and-measurement (SPAM) errors.
We note here that constructing a SWAP gate out of CPHASEs and CROTs would result in a gate
time significantly slower than the sequence discussed here. A SWAP gate can be compiled using
3 consecutive CROT gates, which would give a total SWAP time of ≈ 2 µs. A SWAP gate compiled
from the much faster CPHASE gate requires 11 primitive operations [38], which include 8 single-
qubit gates and would therefore give an even larger overhead. Therefore, the composite exchange
sequence can improve the gate time by more than one order of magnitude.
 6. D ESIGN AND INTEGRATION OF SINGLE - QUBIT ROTATIONS AND TWO - QUBIT GATES IN
98 SILICON ABOVE ONE K ELVIN

6.2.5. H OT TWO - QUBIT GATE PERFORMANCE

In order to assess the performance of our two-qubit gates, we perform time dependent simula-
tions of the Heisenberg hamiltonian, based on the exchange-detuning curve that me measure
experimentally (see section 6.6). Additionally, we also take into account the effects of the finite
bandwidth of the setup (300 MHz) on the pulse generation. Table 6.1 shows the fidelities asso-
ciated with the two-qubit gates CROT, CPHASE, and SWAP. Here, F ideal represents the simulated
fidelities taking into account the relevant parameters, but neglecting any decoherence. We find
F ideal > 99% for all gates except the SWAP, which is limited in fidelity by the finite ∆E z . At the el-
evated temperatures discussed in this work, thermal noise can impact the gate performances in
the form of a larger charge noise amplitude [27], which couple to the spin as an effective magnetic
noise via the exchange interaction. Therefore, we have also modelled the decoherence by adding
stochastic fluctuations of the detuning, sampled from a 1/ f noise spectrum (see section 6.7). By
fitting the experimental data in Fig 6.2b and 6.2e, we conclude that our model is able to reproduce
the decoherence with good agreement. Based on these simulations we determine F noise . The fi-
delity of the CROT and the CPHASE gate are significantly affected by the noise, due to the relatively
long gate times, and we find that the predicted CROT fidelity F noise = 89 % is close to the exper-
imentally measured fidelity F = 86 % [19]. The SWAP, diabatic CPHASE and composite SWAP are
less affected by the noise and in particular we predict that both the diabatic CPHASE and compos-
ite SWAP can be executed with fidelities above 99 %. While experimental data will be needed to
validate these predictions, these results showcases how a multitude of native two-qubit gates can
be executed with high-fidelities and remarkable gate speeds. The limiting factor to the fidelities
is the charge noise, as we have to significantly pulse the detuning to control the exchange inter-
6 action. Significant improvements can be expected by keeping the detuning at zero and instead
pulsing the tunnel coupling, as this scheme is to first order insensitive to charge noise.
The ability to execute a diverse set of high-fidelity two-qubit gates define silicon quantum
dots as a versatile platform for quantum information. The low magnetic field operation and the
small Zeeman energy difference between qubits is furthermore beneficial for the realization of
scalable qubit tiles, as it supports high-fidelity shuttlers and on-chip resonators for long-distance
qubit links. Moreover, the ability to execute quantum logic at temperatures exceeding one Kelvin
provides a pathway to quantum integrated circuits that host both the qubits and their control
circuitry for scalable quantum hardware.

6.3. M ETHODS
The experiments have been performed in a Bluefors refrigerator with a base temperature Tbase ≈
0.45K with a 3 tesla magnet. In the experiments we make use of d.c. voltages and a.c. voltages.
The d.c. voltages are supplied via battery-powered voltage sources and filtered through Cu-powder
filters and 30 Hz and 150 kHz filters. The a.c. voltages are supplied through a bias-tee that is on the
sample printed circuit board with a cut-off frequency of 3 Hz. Pulses are generated by an arbitrary
wave form generator (Keysigh M3202A) with 14 bit resolution and 1 GS/s. Microwave signals are
applied via a Keysight E8267D.

6.4. R ECONSTRUCTION OF THE SPIN STATE PROBABILITIES

In order to readout the spin states we average all single-shot readout traces and subtract a ref-
erence sequence in which no gates are performed. The corresponding readout signal is there-
fore a current that is proportional to the probability of having a blocked state. In order to be
able to reconstruct the four probability amplitudes |A|2 , |B |2 , |C |2 , |D|2 of an arbitrary state ψ =
A |↑↑〉 + B |↑↓〉 + C |↓↑〉 + D |↓↓〉 it is necessary to know the current signal of the four spin states
{α −→ |↓↑〉 , β −→ |↑↑〉 , γ −→ |↓↓〉 , δ −→ |↑↓〉}. By initializing the |↓↑〉 state and then using the four
6.5. T EMPERATURE EFFECTS ON READOUT AND INITIALIZATION 99

frequencies f1 , f2 , f3 , f4 we can reach all spin states and therefore measure the parameters α, β, γ, δ.
Once we measure the current signals for all spin states we need to gain information about the
state ψ. We therefore apply the following sequences and measure the parameters φ0 , φ1 , φ2 , φ3 :

• Sequence: prepare state ψ and then measure. We measure a current φ0 equal to:

φ0 = |A|2 β + |B |2 δ + |C |2 α + |D|2 γ (6.4)

• Sequence: prepare state ψ, apply a π pulse on f4, and then measure. We measure a current
φ1 equal to:

φ1 = |C |2 β + |B |2 δ + |A|2 α + |D|2 γ (6.5)

• Sequence: prepare state ψ, apply a π pulse on f1, and then measure. We measure a current
φ2 equal to:

φ2 = |A|2 β + |B |2 δ + |D|2 α + |C |2 γ (6.6)

• Sequence: prepare state ψ, apply a π pulse on f2, apply a π pulse on f4, and then measure.
We measure a current φ3 equal to:

φ3 = |C |2 β + |A|2 δ + |B |2 α + |D|2 γ (6.7)

Therefore we have the following system of equations, that we want to solve for the probabilities
|A|2 , |B |2 , |C |2 , |D|2 :
6
  2  
β δ α γ |A| φ0

α  2  
δ β γ |B |  φ1 
· = (6.8)
α  |C |2  φ2 

β δ γ

δ α β γ |D| 2 φ3
We solve the system by inverting the matrix:
−1  
|A|2
  
β δ α γ φ0
 2 
 |B |  α δ β γ · φ1 
 
 2 =  (6.9)
 |C |  β δ γ α φ2 
|D|2 δ α β γ φ3
Finally the resulting amplitudes are normalized. The extracted probabilities correspond to
the diagonal parts of the density matrix ρ which may violate i ρ i i = 1 due to measurement and
P

gate errors. In order to ensure the physicality of our results we perform a maximum-likelihood
estimation [39] using the diagonal elements of the density matrix.

6.5. T EMPERATURE EFFECTS ON READOUT AND INITIALIZATION

To account for the effect for temperature on the initial and final state we consider that the sys-
tem is coupled to a thermal bath with temperature T = 1.05 K. Under the assumption of thermal
equilibrium the initial and final states are then given by the Gibbs state

ρ th = 1/Z exp −βH(0,2)

¡ ¢
(6.10)

with the partition function Z = Tr exp −βH(0,2) and the inverse temperature β = (k B T )−1 =
© ¡ ¢ª

1/(89 µeV). Since readout and initialization is performed deep in the (0, 2) charge regime, the
relevant Hamiltonian H(0,2) consists of two electrons in a double quantum dot in the presence
 6. D ESIGN AND INTEGRATION OF SINGLE - QUBIT ROTATIONS AND TWO - QUBIT GATES IN
100 SILICON ABOVE ONE K ELVIN

a b c T*2,C = (3.9 ± 0.1) µs

T*2,C = (290 ± 3) ns
50 1 1
Fit
40 JSWAP 0.8 0.8

Triplet probability

Triplet probability
(MHz)
JQ1
JQ2

J (MHz)
30 0.6 0.6

Exchange
20 0.4 0.4

10 0.2 0.2

0 0 0
40 60 80 100 0 200 400 600 0 1000 2000 3000
ε (mV)
Detuning (mV) Time (ns) Time (ns)

Figure 6.4: Exchange interaction a Exchange energy as a function of detuning energy. The data points J SWAP
have been extracted from the frequency of SWAP oscillations, J Q1 and J Q2 are obtained from the energy differ-
ence between f 1 and f 2 , and f 3 and f 4 respectively. The fit (solid black line) is used for simulations of detuning
noise. We observe frequency jumps in the first five points of the data set, and these are therefore neglected in
the fitting. b-c Simulated data for the sequence used in Fig. 6.2a and 6.2b.

of spin and valley [40]. The two-electron eigenstates in this regime are (approximately) anti-
symmetric combinations of spin and valley singlet-triplet states labeled as |spin, valley〉. In the
basis {|S, T+ 〉 , |T+ , S〉 , |S, T0 〉 , |T0 , S〉 , |T− , S〉 , |S, T− 〉} and neglecting electron-electron interactions
the eigenenergies are

H(0,2) = diag(2E v , E v + E z , E v , E v , E v − E z , 0), (6.11)

where E z ≈ 29 µeV is the Zeeman energy in the doubly occupied QD and E v is the valley splitting.
The valley splitting can be estimated from previous works [19, 27], E v ≈ 300 µeV, giving rise to a
6 ground state probability, Eq. (6.10), of P GS ≈ 87 %. A thermal ground state probability of >90 %
(>99 %) can be reached at this elevated temperature with a valley splitting of E v & 327 µeV (E v &
542 µeV) [9].

6.6. E XCHANGE SPECTRUM

Figure 6.4a shows the exchange spectrum as a function of detuning. When J < 10 MHz we measure
the exchange interaction by mapping f 1 , f 2 , f 3 , f 4 as a function of the detuning energy ² and then
extract J as the difference between f 1 and f 2 , and f 3 and f 4 respectively. When Jq> 10 MHz we
extract the exchange from the frequency of the SWAP oscillations given by f SWAP = J 2 + ∆E z2 .
The best agreement to our data set is obtained by fitting with an exponential dependence [41,
42] J (²) = exp(2α(² − V0 )) and constant ∆E z (²) = ∆E z with parameters α = −0.022 ln(MHz)/mV,
V0 = 120 mV, and ∆E z = 11 MHz.

6.7. T WO - QUBIT GATE SIMULATIONS

6.7.1. P HYSICAL MODEL
For all the two-qubit gates we obtain the fidelities by numerically solving the time-dependent
Schrodinger equation i ħΨ̇(t ) = H (t )Ψ(t ) using a step size t = 50 ps. We have confirmed that a
faster sampling rate does not change the simulation results. In order to resemble real setups and
to avoid sampling problems we filter all time-dependent signals using a high-pass filter with a
cut-off frequency of 300 MHz. All simulations are performed using the Heisenberg Hamiltonian

H (t ) = ∆E z (²)(S z,1 − S z,2 )/2 + J (²)(S 1 · S 2 − 1/4), (6.12)

where S i = (S x,i , S y,i , S z,i ) is the spin operator of the electron in quantum dot i . The parameters
∆E z (²) and J (²) are the qubit frequency difference and the exchange interaction which both de-
pend on the detuning ².
6.7. T WO - QUBIT GATE SIMULATIONS 101

To include the effect of decoherence we add stochastic fluctuations of the detuning, ² → ² +

δ²(t ) to each run of the time evolution and average the resulting density matrix. To emulate the
effect of 1/ f charge noise we generate time-dependent fluctuations δ²(t ) following a S(ω) = A ² / f
spectral density using the method described in Ref. [43, 44]. The amplitude of the noise A ² is
set such that it reproduces the decay time of the exchange oscillations measured experimentally.
Using the same value of noise we then simulate the Ramsey sequences of Fig. 6.2a and 6.2b and
we show the results in Fig. 6.4b and 6.4c. The decay times that we extract are in agreement with
the measured decays T2,C↓∗ ∗
= (397 ± 24) ns and T2,C↑ = (3.9 ± 0.6) µs, where C refers to the control
qubit.
For benchmarking the performance of the simulated gate operations we use the average gate
fidelity from QuTiP [45].

6.7.2. A DIABATIC AND DIABATIC CPHASE GATES

Using the Hamiltonian in Eq. 6.12 it is possible to implement a controlled-phase (CPHASE) gate.
The conventional approach for the CPHASE implementation is to use a smooth exchange pulse
such that the system undergoes a adiabatic time evolution and it accumulates a phase difference
between the even {|↑↑〉 , |↓↓〉} and odd parity subspace {|↑↓〉 , |↓↑〉}. The adiabatic time evolution of
the system described by the Hamiltonian in Eq. 6.12 is the unitary

Uad = e −i Φ(S z,1 S z,2 −1/4) × e −i (θad −θres )(S z,1 −S z,2 )/2 (6.13)
Rt
with the entangling phase Φ = 0 (J (t 0 ) − J res )d t 0 /ħ and the single-qubit phases
6
R q q
θad = 0t ∆E z2 + J 2 (t 0 )d t 0 /ħ and θres = ∆E z2 + J res 2 t /ħ. Here we explicitly took into consider-

ation that J (t = 0) = J res . An adiabatic CPHASE gate is then obtained for Φ = (2n + 1)π with n
being an integer number. Virtual single-qubit z-gates are then applied to compensate for the
single-qubit phase evolution θad . Non-adiabatic (diabatic) errors due to a violation of the adia-
batic condition enter via spin-flip terms S x,1 S x,2 + S y,1 S y,2 and S x,1 S y,2 − S y,1 S z,2 and give rise to
SWAP oscillations. They can be mitigated using a decoupling scheme [28] or sufficiently long ramp
times (see Fig. 6.3a).
Faster CPHASE gates can be realized using diabatic pulses where we synchronize the SWAP
oscillation with the CPHASE evolution [29]. The diabatic time evolution of a system described by
Eq. 6.12 is given by:

¢r ·σ
µ ¶
Udia = e −i Φ(S z,1 S z,2 −1/4) × e i θres (S z,1 −S z,2 )/2 cos θdia − i sin θdia
¡ ¢ ¡
, (6.14)
|r |
q
where r = (∆E z ·J −∆E z ·J res , 0, ∆E z2 +J ·J res ) describes the tilted SWAP rotation, θdia = ∆E z2 + J 2 t /ħ
the angle of the rotation, and σ = ( X̂ , Ŷ , Ẑ ) is the vector consisting of Pauli matrices in the odd par-
ity space

X̂ = |↑↓〉 〈↓↑| + |↓↑〉 〈↑↓| (6.15)

Ŷ = −i |↑↓〉 〈↓↑| + i |↓↑〉 〈↑↓| (6.16)
Ẑ = |↑↓〉 〈↑↓| − |↓↑〉 〈↓↑| = S z,1 − S z,2 . (6.17)

Note, that we again took into consideration that J (t = 0) = J res and assumed a piece-wise constant
exchange pulse of the form
(
J if t > 0 and t ≤ t end
J (t ) = . (6.18)
J res else
 6. D ESIGN AND INTEGRATION OF SINGLE - QUBIT ROTATIONS AND TWO - QUBIT GATES IN
102 SILICON ABOVE ONE K ELVIN

A diabatic CPHASE gate is then given for Φ = (2n + 1)π and θdia = mπ with m, n being integers. A
generic solution for J , which satisfies both constraints, is given by

q
2 m 2 + 4J m 2
−(2n + 1)2 ∆E z2 (2n + 1)2 − 4m 2 − 4J res
¡ ¡ ¢ ¢
res
J= . (6.19)
4m 2 − (2n + 1)2

The formula in equation 6.1 is obtained for n = 0 and m = 1, while the results in Ref. [29] are
obtained for J res = 0.

6.7.3. SWAP GATE

For ∆E z = J res = 0 the time-evolution of Eq. 6.14 gives rise to oscillations between the |↑↓〉
p and
|↓↑〉 states which can be used to implement the classical SWAP and maximally entangling SWAP
gate [4]. This is not the case for a finite ∆E z 6= 0 and J res > 0 and the maximum flip probability is
limited to

¯ ¯2 ∆E z2 (J − J res )2
P flip = ¯〈↑↓|Udia |↓↑〉¯ = (6.20)
(∆E z2 + J 2 )(∆E z2 + J res )

heavily reducing the maximum fidelity. When J res = 3 MHz, ∆E z = 11 MHz and J = 27 MHz we
6 obtain a maximum flip probability P flip = 63 % which is a close match to the observed value in
Figure 6.3c. From the formula above it is clear that a high-fidelity implementation of a SWAP
gate with J res = 3 MHz and ∆E z = 11 MHz would require an extremely large exchange interac-
tion, usually not achieved in state-of-the-art experiments
p in silicon. Furthermore, it is also im-
possible
p pto implement a high-fidelity entangling SWAP in such a way due to the requirement
SWAP SWAP = SWAP.
In order to overcome these issues we decompose our desired SWAP rotation in the odd parity
subspace in terms of tilted r rotations from the diabatic time evolution and z rotations from the
adiabatic phase evolution. We then construct a composite pulse sequence by alternating diabatic
and adiabatic pulses which yields our desired operation Utot

Utot = Udia (t 1 , J 1 )Uad (t 2 , J 2 )Udia (t 3 , J 3 )Uad (t 4 , J 4 ) · · · (6.21)

Here, the tuple (t i , J i ) defines a diabatic or adiabatic pulse jwithk time t i and exchange J i . The
total number of pulses for a SWAP are estimated to be N = 2 90° β
, where b·c is the floor function
and β denotes the angle between the vectors r and z = (0, 0, 1). Note, that the final Uad operation
only needs to correct single qubit phases which in our experiment can be done alternatively using
virtual single qubit z-gates.
Enforcing the constraint Φtot = Φ1 + Φ2 + Φ3 + Φ4 + · · · = mπ, with integer m and Φi being the
entangling phase of the pulse Udia,ad (t i , J i ), ensures that the composite pulse yields a SWAP gate.
Using instead Φtot = Φ1 + Φ2 + Φ3 + Φ4 + · · · = mπ/2 yields the universal i SWAP gate.
In our experiment we choose a minimal length symmetric composite pulse consisting of two
identical diabatic pulses Udia (t 1 , J 1 ) = Udia (t 3 , J 3 ) = Ur (t SWAP , J ) and an adiabatic pulse Uad (t 2 , J 2 ) =
Uz (t corr , J corr ) in between, UrUrUr = X̂ . The first diabatic pulse executes half of the desired SWAP
rotation, the adiabatic pulse realizes a rotation in the X − Y -plane, and the second diabatic pulses
R EFERENCES 103

finalizes the SWAP rotation. The pulse times are given by the conditions
µ ¶
2ħ sin−1 p 1
2P flip
t SWAP = q , (6.22)
∆E z2 + J 2
 q 
Z t
SWAP +t corr
q (J 2 − ∆E z2 )(∆E z2 − J res
2 ) − 4∆E 2 J J
z res
2 2 0 −1 
∆E z + J corr (t )d t = 2ħ cos
0 −  (6.23)
t SWAP ∆E z (J − J res )

and the condition for the entangling phase Φtot reads

Z t
SWAP +t corr
2 ħJ t SWAP + ħ J corr (t 0 )d t 0 = mπ. (6.24)
t SWAP

Using a numerical search we find that the pulse sequence

t

2.37 MHz × tramp

 if t > 0 and t ≤ t ramp

2.37 MHz if t > t ramp and t SWAP − t ramp

J corr (t ) = t −t SWAP +t ramp (6.25)


 2.37 MHz × t ramp if t > t SWAP − t ramp and t ≤ t SWAP


0 else


allows to achieve gate fidelities F > 99.9 %.

6
R EFERENCES
[1] Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Na-
ture 574, 505–510 (2019).

[2] Reiher, M., Wiebe, N., Svore, K. M., Wecker, D. & Troyer, M. Elucidating reaction mechanisms
on quantum computers. Proceedings of the National Academy of Sciences 114, 7555–7560
(2017).

[3] Ladd, T. D. et al. Quantum computers. Nature 464, 45–53 (2010).

[4] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Physical Review A
57, 120–126 (1998).

[5] Petta, J. R. et al. Coherent Manipulation of Coupled Electron Spins in Semiconductor Quan-
tum Dots. Science 309, 2180–2184 (2005).

[6] He, Y. et al. A two-qubit gate between phosphorus donor electrons in silicon. Nature 571,
371–375 (2019).

[7] Meunier, T., Calado, V. E. & Vandersypen, L. M. K. Efficient controlled-phase gate for single-
spin qubits in quantum dots. Physical Review B 83, 121403 (2011).

[8] Nowack, K. C., Koppens, F., Nazarov, Y. V. & Vandersypen, L. Coherent control of a single
electron spin with electric fields. Science 318, 1430–1433 (2007).

[9] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nature Nanotechnology 9, 981–985 (2014).
 104 R EFERENCES

[10] Kawakami, E. et al. Electrical control of a long-lived spin qubit in a Si/SiGe quantum dot.
Nature Nanotechnology 9, 666–670 (2014).

[11] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nature Nanotechnology 13, 102 (2018).

[12] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[13] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon. Nature 555,
633–637 (2018).

[14] Xue, X. et al. Benchmarking gate fidelities in a si/sige two-qubit device. Physical Review X 9,
021011 (2019).

[15] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins. Science 359, 439–442
(2018).

[16] Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 569, 532–536
(2019).

[17] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit
logic with holes in germanium. Nature 577, 487–491 (2020).

[18] Yang, C. H. et al. Operation of a silicon quantum processor unit cell above one kelvin. Nature
580, 350–354 (2020).
6
[19] Petit, L. et al. Universal quantum logic in hot silicon qubits. Nature 580, 355–359 (2020).

[20] Sigillito, A. J., Gullans, M. J., Edge, L. F., Borselli, M. & Petta, J. R. Coherent transfer of quan-
tum information in a silicon double quantum dot using resonant swap gates. npj Quantum
Information 5, 1–7 (2019).

[21] Barenco, A. et al. Elementary gates for quantum computation. Physical Review A 52, 3457
(1995).

[22] Veldhorst, M., Eenink, H. G. J., Yang, C. H. & Dzurak, A. S. Silicon CMOS architecture for a
spin-based quantum computer. Nature Communication 8, 1766 (2017).

[23] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 34 (2017).

[24] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science Advances 4, eaar3960
(2018).

[25] Lawrie, W. I. L. et al. Quantum dot arrays in silicon and germanium. Applied Physics Letters
116, 080501 (2020).

[26] Urdampilleta, M. et al. Gate-based high fidelity spin read-out in a cmos device. Nature Nan-
otechnology 14, 737–741 (2019).

[27] Petit, L. et al. Spin lifetime and charge noise in hot silicon quantum dot qubits. Physical
Review Letters 121, 076801 (2018).

[28] Russ, M. et al. High-fidelity quantum gates in si/sige double quantum dots. Physical Review
B 97, 085421 (2018).
R EFERENCES 105

[29] Burkard, G., Loss, D., DiVincenzo, D. P. & Smolin, J. A. Physical optimization of quantum error
correction circuits. Physical Review B 60, 11404 (1999).

[30] Maune, B. M. et al. Coherent singlet-triplet oscillations in a silicon-based double quantum

dot. Nature 481, 344 (2012).

[31] Martinis, J. M. & Geller, M. R. Fast adiabatic qubit gates using only σ z control. Physical
Review A 90, 022307 (2014).

[32] Güngördü, U. & Kestner, J. Pulse sequence designed for robust c-phase gates in simos and
si/sige double quantum dots. Physical Review B 98, 165301 (2018).

[33] Calderon-Vargas, F. et al. Fast high-fidelity entangling gates for spin qubits in si double quan-
tum dots. Physical Review B 100, 035304 (2019).

[34] Güngördü, U. & Kestner, J. Analytically parametrized solutions for robust quantum control
using smooth pulses. Physical Review A 100, 062310 (2019).

[35] Güngördü, U. & Kestner, J. Robust implementation of quantum gates despite always-on ex-
change coupling in silicon double quantum dots. Physical Review B 101, 155301 (2020).

[36] Vandersypen, L. M. K. & Chuang, I. L. Nmr techniques for quantum control and computation.
Reviews of Modern Physics 76, 1037 (2005).

[37] Zhang, X.-M., Li, J., Wang, X. & Yung, M.-H. Minimal nonorthogonal gate decomposition for
qubits with limited control. Phys. Rev. A 99, 052339 (2019). 6
[38] Lee, S. et al. The cost of quantum gate primitives. Journal of Multiple-Valued Logic & Soft
Computing 12 (2006).

[39] James, D. F. V., Kwiat, P. G., Munro, W. J. & White, A. G. Measurement of qubits. Physical Review
A 64, 052312 (2001).

[40] Jiang, L. et al. Coulomb interaction and valley-orbit coupling in si quantum dots. Phys. Rev.
B 88, 085311 (2013).

[41] Dial, O. E. et al. Charge noise spectroscopy using coherent exchange oscillations in a singlet-
triplet qubit. Physical Review Letters 110, 146804 (2013).

[42] Cerfontaine, P., Botzem, T., DiVincenzo, D. P. & Bluhm, H. High-fidelity single-qubit gates for
two-electron spin qubits in gaas. Physical Review Letters 113, 150501 (2014).

[43] Koski, J. V. et al. Strong photon coupling to the quadrupole moment of an electron in a solid-
state qubit. Nature Physics 16, 642–646 (2020).

[44] Yang, Y.-C., Coppersmith, S. N. & Friesen, M. Achieving high-fidelity single-qubit gates in a
strongly driven charge qubit with 1/f charge noise. npj Quantum Information 5, 1–6 (2019).

[45] Johansson, J., Nation, P. & Nori, F. Qutip 2: A python framework for the dynamics of open
quantum systems. Computer Physics Communications 184, 1234 (2013).
 G ERMANIUM
Not to be confused with Geranium
Wikipedia page on Germanium

We shift focus now to different quantum computing platform: Hole qubits in strained planar ger-
manium quantum wells (Ge/SiGe) [1]. It has been predicted for a long time that the spin states of
holes in quantum dots could be a promising candidate for quantum computing. The strong spin-
orbit interaction facilitates fast, electrical manipulation of qubits [2], removing the need for bulky
structures such as micromagnets or co-planar waveguides. In germanium specifically, the Fermi-
level is pinned to the valence band for virtually any metallic contact [3], allowing for the creation of
p-type Ohmic contact without the need for implantation or annealing. Additionally, the impressive
hole mobility and percolation density in germanium quantum wells (See Chapter 3) suggests a very
uniform material, particularly when compared to that of electrons in Silicon-MOS [4]. These quali-
ties position hole states in Ge/SiGe as promising qubit candidates for a scalable quantum computing
architecture.

Indeed, exciting and rapid progress in the field has been made possible by the combination of these
favourable qualities, and pioneering work in silicon platforms. The field of hole spins in planar ger-
manium progressed from the first demonstration of a quantum dot in Ge/SiGe [5] to two qubit logic
in transport mode operated many-hole spin qubits in only two years [6].

The following chapters progress the field of quantum computing with hole states in planar ger-
manium quantum wells. We begin, in chapter 7, by demonstrating the first single-hole spin qubit
defined in a quantum dot. This experiment revealed that single hole spin qubits can be coherently
controlled with Rabi oscillations approaching 100 MHz. It was also shown that the g -factor of such
hole spins could be tuned significantly by altering the electrostatic environment around the spins.
Chapter 8 addresses the spin-relaxation lifetime of quantum dots, finding that spin lifetime can be
extended as high as 32 ms, by limiting dot-reservoir coupling. Further demonstration of an electri-
cally susceptible g-factor is also shown by measuring the change in resonance frequency of each qubit
as a function of their plunger gate potentials. The effect of local electric field on resonance frequency
is expected as a result of the spin-orbit coupling present for holes in germanium, and is evidence
that charge noise is the limiting mechanism for quantum coherence. By adapting the Ge/SiGe het-
erostrucutre such that the quantum wells were positioned deeper, we drastically reduced the charge
noise experienced by the hole spin qubits, facilitating the work of the following chapters. In Chapter
9, we perform four-qubit logic on a 2x2 array of hole spin qubits, demonstrating controlled rotations
on all neighbouring qubit pairs, a native three-qubit i-Toffoli gate, a novel four-qubit gate resem-
bling a CCCNOT form, and perform a four-qubit algorithm in the form of a Greenberger-Horne-
Zeilinger state generation. In chapter 10, we characterise the fidelities of single qubit gates on the
same 2x2 quantum dot array using randomized benchmarking. We find that the native single qubit
gate fidelities can be as high as 99.990(2)%. We also investigate how classical cross-talk affects single
qubit fidelities by presenting a novel benchmarking technique called N -copy benchmarking, that
separates classical cross-talk noise from quantum cross-talk noise. We find that single-qubit ex-
change insensitive fidelities for a two-copy benchmarking pair of qubits can be as high as 99.967 %,

107
 108 R EFERENCES

while those of four-copy experiments are limited to around 99 %. In chapter 11, we investigate the co-
herence properties of germanium hole spin qubits, analysing their dependence on charge noise and
hyperfine interactions. We conclude, based on dephasing time behaviour as a function of applied
magnetic field, that charge noise limits coherence in hole spin qubits in the investigated magnetic
field regimes. However, we also observe decohering effects on the hole spin due to hyperfine inter-
actions with the residual 73 Ge nuclear spins in the quantum well. We study the strengths of these
interactions and find that at even at appreciable magnetic field strengths, the hyperfine interaction
will limit spin dephasing within an order of magnitude of the charge noise limited values. These
findings demonstrate the necessity of isotopic purification of germanium quantum wells, particu-
larly given the present trends of Ge/SiGe material improvement.

R EFERENCES
[1] Scappucci, G. et al. The germanium quantum information route. Nat. Rev. Mater. (2020).

[2] Bulaev, D. V. & Loss, D. Electric dipole spin resonance for heavy holes in quantum dots. Phys.
Rev. Lett. 98, 097202 (2007).

[3] Dimoulas, A., Tsipas, P., Sotiropoulos, A. & Evangelou, E. K. Fermi-level pinning and charge
neutrality level in germanium. Appl. Phys. Lett. 89, 252110 (2006).

[4] Lawrie, W. I. L. et al. Quantum Dot Arrays in Silicon and Germanium. Appl. Phys. Lett. 116,
080501 (2020).

6 [5] Hendrickx, N. W. et al. Gate-controlled quantum dots and superconductivity in planar germa-
nium. Nat. Commun. 9, 2835 (2018).

[6] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit logic
with holes in germanium. Nature 577, 487–491 (2020).
 7
A S INGLE -H OLE S PIN Q UBIT
Qubits based on quantum dots have excellent prospects for scalable quantum technology due to
their compatibility with standard semiconductor manufacturing. While early research mostly fo-
cused on the simpler electron system, recent demonstrations using many-hole quantum dots illus-
trated the favourable properties holes can offer for fast and scalable quantum control. Here, we es-
tablish a single-hole spin qubit in germanium and demonstrate the integration of single-shot read-
out and quantum control. We deplete a planar germanium double quantum dot to the last hole,
confirmed by radio-frequency reflectrometry charge sensing, and achieve single-shot spin readout.
To demonstrate the integration of the readout and qubit operation, we show Rabi driving on both
qubits and find remarkable electric control over their resonance frequencies. Finally, we analyse the
spin relaxation time, which we find to exceed one millisecond, setting the benchmark for hole-based
spin qubits. The ability to coherently manipulate a single hole spin underpins the quality of strained
germanium and defines an excellent starting point for the construction of novel quantum hardware.

Parts of this chapter have been published in Nature Communications 11 (1), 1-6

109
 110 7. A S INGLE -H OLE S PIN QUBIT

7.1. I NTRODUCTION
Group-IV semiconductor spin qubits [1] are promising candidates to form the main building block
of a quantum computer due to their high potential for scalability towards large 2D-arrays [2–5] and
the abundance of net-zero nuclear spin isotopes for long quantum coherence [6, 7]. Over the past
decade, all prerequisites for quantum computation were demonstrated on electron spin qubits in
silicon, such as single-shot readout of a single electron [8], high-fidelity single-qubit gates [9, 10]
and the operation of a two-qubit gate [11–14]. However, hole spins may offer several advantages
[15, 16], such as a strong spin-orbit coupling (SOC) and a large excited state energy. Early research
demonstrated the feasibility of using the SOC for all-electric driving [17, 18], but these experiments
were limited by nuclear spins and the coherent driving of a single hole spin remained an open
challenge. More recently, hole spins in group-IV materials have gained attention as a platform for
quantum information processing [19–21]. In particular hole states in germanium can provide a
high degree of quantum dot tunability [22–24], fast and all-electrical driving [20, 21] and Ohmic
contacts to superconductors for hybrids [25, 26]. These experiments culminated in the recent
demonstration of full two-qubit logic [21]. While hole spins have been read out in single-shot
mode using the Elzerman technique [27], these experiments require magnetic fields impractical
for hole qubit operation due to the strongly anisotropic g -factor of hole spins in germanium [28].
Pauli spin blockade readout allows for spin readout independent of the Zeeman splitting of the
qubit, leveraging the large excited state energy purely defined by the orbital energy for holes in
germanium. Furthermore, achieving these assets on a single-hole spin demonstrates full control
over the materials system and allows to tune the quantum dot occupancy at will, optimising the
different qubit properties. Moreover, the ability to study a platform at the single-particle level
would provide great insight into its physical nature, crucial for holes which originate from a more
complicated band structure than electrons [29, 30].
In this work, we make this step and demonstrate single-shot readout and operation of a single
hole spin qubit. We grow undoped strained germanium quantum wells [31] and fabricate devices
7 using standard manufacturing techniques [2]. The high mobility and low effective mass [32] allow
us to define quantum dots of relatively large size, alleviating the restraints on fabrication. We de-
plete the quantum dots to their last hole, confirmed by charge sensing using a nearby single hole
transistor (SHT). The use of radio-frequency (RF) reflectometry [33–35] enables a good discrimi-
nation of the charge state, while maintaining a high measurement bandwidth to allow for fast spin
readout. We make use of Pauli spin blockade to perform the spin-to-charge conversion [36], max-
imally taking advantage of the large excited state energy splitting of E ST = 0.85 meV and obtain
single-shot spin readout. Finally, we demonstrate the integration of readout and qubit operation
by performing all-electrically driven Rabi rotations on both qubits. Studying the control of a single
hole qubit, we find a remarkably strong dependence of the resonance frequency on electric field
and show a tunability of almost 1 GHz using only small electric potential variations.

7.2. R ESULTS
7.2.1. S INGLE HOLE QUANTUM DOT AND PAULI SPIN BLOCKADE
A false-coloured SEM picture of the quantum dot device is depicted in Figure 7.1a. The device
consists of a quadruple quantum dot system in a two-by-two array [2]. We tune the top two quan-
tum dots into the many-hole regime, such that they can be operated as a single hole transistor.
In order to perform high-bandwidth measurements of the sensor impedance, we make use of RF-
reflectometry, where the SHT is part of a resonant LCR-circuit further consisting of an off-chip
superconducting resonator together with the parasitic device capacitance. We apply a microwave
signal to the tank circuit and measure the amplitude of the signal reflected by the LCR-circuit (see
Figure 7.1a). The amplitude of the reflected signal |S 21 | depends on the matching of the tank cir-
7.2. R ESULTS 111

a b 0 0.05
| S 21 | (a.u.)

-880 (0,2) (0,1) (0,0)

V P1 (mV)
-900
(1,2) (1,1) (1,0)
-920
P1
RB2 -940
(2,2) (2,1) (2,0)
RB1 B12 P2 -960
-1200 -1100
V P2 (mV)

Figure 7.1: Fabrication and operation of a planar germanium double quantum dot. a False-coloured scan-
ning electron microscope image of the quadruple quantum dot device. Ohmic contacts are indicated in yellow,
a first layer of electrostatic barrier gates is indicated in green and the second layer of plunger gates is coloured
in purple (for details, see Methods). The scale bar corresponds to 100 nm. We use the double quantum dot in
the top channel as a single hole transistor (SHT) to sense changes in the charge occupation of the quantum
dots formed under plunger gates P1 and P2. A schematic illustration of the electrostatic potential defining the
two single-hole quantum dots is depicted above the figure. The charge sensor impedance is measured using
reflectometry on a resonant circuit consisting of a superconducting resonator and the parasitic device capaci-
tance. Barrier gates RB1 and RB2 can be used to control the tunnel rate of each quantum dot to its respective
reservoir and gate B12 controls the interdot tunnel coupling. b Charge stability diagram of the double quantum
dot system, where depletion of both quantum dots up to the last hole can be observed.

cuit impedance with the measurement setup and is therefore modulated by a change in the charge
sensor impedance caused by the movement of a nearby charge.
We make use of the RF sensor to map out the charge stability diagram of the double quantum
7
dot system defined by plunger gates P1 and P2. The tunnel coupling of the quantum dots to their
reservoirs, as well as the interdot tunnel coupling can be tuned by gates RB1, RB2 an B12 respec-
tively. Next, we tune the device to the single hole regime for both quantum dots (Fig. 7.1b and
Figure 7.5), where (N1 ,N2 ) indicates the charge occupation, with N1 (N2 ) the hole number in the
dot under P1 (P2). In our previous work [2], we further detail that we can deplete all four quantum
dots in this device down to their last hole. In order to perform readout of the spin states, we make
use of Pauli spin blockade (PSB), which is expected to be observed both at the (1,1)-(0,2) and (1,1)-
(2,0) charge transitions. We define the virtual gates [37] detuning V² and energy VU (see Fig. 7.2a
and Methods) and sweep across the (1,1)-(2,0) and (1,1)-(0,2) transitions in this gate space. As a
result of its triplet character, the |↓↓〉 state has a negligible coupling to the S(2,0) or S(0,2) singlet
charge states (Fig. 7.2b). When pulsing across the (1,1)-S(2,0) or (1,1)-S(0,2) anti-crossings, PSB
prevents charge movement when the system is in the |↓↓〉 ground state. However, when the system
resides in the singlet-like lower antiparallel spin state (in this case |↓↑〉, with Q2 being the qubit
with lower Zeeman energy), charge movement to a doubly occupied quantum dot state is possi-
ble, therefore leaving the system in a (0,2) or (2,0) charge state. This results in a spin-to-charge
conversion, which in turn can be picked up in the reflectometry signal from the SHT.
Indeed, we find that by sweeping the detuning voltage across the interdot transition from the
(1,1) into the (0,2) charge region (Fig. 7.2d), tunnelling is blocked up to the reservoir transitions
(indicated in white) when the system is initialised in the |↓↓〉 state. In this case we rely on the fast
diabatic return sweep combined with fast spin relaxation compared to the sweep rate to prepare
the system in the blocking |↓↓〉 state. When we inverse the sweeping direction, the system remains
 112 7. A S INGLE -H OLE S PIN QUBIT

a c 0 1 e g -0.9 -0.6
S21,n (a.u.) (2,0) (R) (2,0) (1,1) (R) S21(a.u.)
(0,2) ΔVε 0
ΔVU ΔVε π ↑↑
(0,1)
VU

RF

trace
4 50
(0,1) 1
2 0
Vε

V U (mV)

V U (mV)
(1,2) (1,1) (1,0) -1 100
(2,0) (1,1) (R)
M 0 (0,2)

S 21 (a.u.)
(1,1) (2,0) -0.6
ΔVε ↓↑
R 1
VP1 -2 R 0 RF
(1,2) (1,1)
(2,1) (2,0) -1 -0.8
VP2 -4
(1,0) (1,1) (R) 0 25 50
ΔVε
b d ΔVU
f h ( s)
4 -0.5 -0.5
1
2 0
V U (mV)

V U (mV)

S 21 (a.u.)

S 21 (a.u.)
-1
0 -0.7 ↓↑ -0.7
1 ↑↑
-2 R 0
S(2,0) -1 T 1,RO = 26 s
-4 -0.9 -0.9
M R -4 -2 0 2 4 -2 0 2 0 25 50 0 250 500
detuning, ε
V (mV) V (mV) ( s) counts

Figure 7.2: Single-shot spin readout of a single hole. a Schematic of a typical hole charge stability diagram
with both possible regions of readout indicated in blue and red. The typical manipulation (M) and readout
(R) points are indicated in green. b Two-hole energy diagram, with the five lowest lying energy states around
7 the (1,1)-S(2,0) anticrossing. c Colour map of the normalised sensor response (normalisation in Methods) as
a function of the applied gate voltages VU and V² . In the large panel, we linearly sweep V² and step VU , as
indicated in the inset above the figure. The smaller panels on the right show the same effect for the (1,1)-(0,2)
anticrossing (top, red box in a), and the (1,1)-(2,0) anticrossing (bottom, blue box in a), now using a two-level
voltage pulse (details in Methods). d Similar colour map as in c, but with a reversed sweeping direction from
(1,1) into the (0,2) region. The triangular spin blockade window is indicated by the dashed white line. The
smaller panels on the right again demonstrate the same effect for both the (1,1)-(0,2) (top) as well as the (1,1)-
(2,0) (bottom) anti-crossings, by first loading a random spin in one of the dots (details in Methods). e Schematic
illustration of the three-level pulses used in panels f-h, indicating the detuning voltage ∆V² in blue and red,
and the RF-pulses in orange. f The averaged charge sensor response as a function of measurement time τ at R
for |↑↑〉 initialisation (red) and |↓↑〉 initialisation (blue). The gray shaded area indicates the integration window
for the threshold detection. g A sample of 100 single-shot traces (top), averaged for 3 µs per data point, with
τ = 0 the start of the readout phase. The bottom panel shows two single traces, where the purple (yellow)
trace corresponds to the readout of a blocked (not blocked) spin state. Dashed lines correspond to the sensor
signal for the different charge states. h Histogram of 5000 single-shot traces, integrating the signal for 5.5 µs as
indicated in f. The blue (red) histogram corresponds to an initialisation in the |↓↑〉 (|↑↑〉) state. The dashed line
corresponds to the optimised threshold for readout.
7.2. R ESULTS 113

a (2,0) (1,1) (R) b (2,0) (1,1) (R)

Δε Δε tp
RF twait RF

1
T 1 = 1.2 ms f R = 57 MHz
0.6

0.8
0.4
P

0.6
0.2

0.4 0
10 -5 10 0 0 100 200
t wait (s) t p (ns)

Figure 7.3: Spin relaxation and coherent driving of a single hole. a The system is initialised in the |↓↑〉 state
after which the qubits idle at the measurement point. The spin-up fraction P ↑ of Q2 is measured as a function
of waiting time t wait and shows a typical T1 -decay with a relaxation time of T1 = 1.2 ms. b Driving of the single
hole qubit Q2 shows coherent oscillations in P ↑ as a function of the microwave pulse length t p . The coherent
operation of Q1 is shown in Figure 7.8.

in the (0,2) charge states at the same values of V² and VU (Fig. 7.2c). After optimising the different
tunnel rates in the device, we confirm the Pauli spin blockade at both the (1,1)-(2,0) and (1,1)-
(0,2) anticrossings by loading a random spin before performing the readout, thereby not relying
on a relaxation process for the initialisation (small panels of Fig. 7.2c,d). The diamond-shaped
window of differential signal allows for a singlet/triplet readout of the system spin state and we
select readout point R (see Figure 7.6). We note that the interdot transition line is shifted slightly 7
towards positive detuning with respect to the reservoir transition lines. This is the direct result of
a small voltage offset present across the device Ohmics, resulting in the unusual diamond-shaped
spin readout window, but not limiting the readout. As holes in germanium do not have any valley
states, the T(2,0) state is expected to be defined by the next quantum dot orbital. By increasing
the bias voltage across the two quantum dots, we shift the interdot transition line further. At large
enough bias, the Pauli spin blockade window is capped as a result of the T(2,0) state becoming
available in energy and from this we extract an excited state energy of E ST = 0.85 meV, using a
lever arm of α² = 0.21 as extracted from polarisation line measurements (Supplementary Figure
7.7).

7.2.2. Q UBIT OPERATION

To coherently control the qubits, we implement a three-level voltage pulsing scheme (Fig. 7.2e)
and operate at an external magnetic field of B = 0.67 T. We initialise the system by pulsing deep
into the (2,0) region (α² V² > E ST ), where the spins quickly relax into the (2,0) singlet state. Next,
we ramp adiabatically into the (1,1) region, preparing the system into the |↓↑〉 state. At this point
(M) we perform the qubit operations by applying microwave pulses to gate P1, taking advantage of
the SOC-mediated EDSR. Rotating Q1 (Q2) will bring the system into the |↑↑〉 (|↓↓〉) state. Finally,
the spin-state is read out by pulsing adiabatically into the readout window. Only the |↓↑〉 will allow
a direct tunnelling into the (2,0) charge state, while tunnelling is blocked for all other states due to
PSB.
Fig. 7.2f displays the charge sensor signal throughout the readout period, both for a |↓↑〉 ini-
 114 7. A S INGLE -H OLE S PIN QUBIT

tialisation (blue) as well as a |↑↑〉 initialisation (red) by applying a π-pulse to Q1. When no pulse is
applied and the system is prepared in the |↓↑〉 state, a fast transition into the (1,1)-charge state, cor-
responding to a sensor signal of S 21 ≈ −0.6 can be observed. The remaining decay (Tdecay = 2 µs)
in this case can be attributed to the response of the SHT-signal to the voltage pulses on the gates.
However, when the system is prepared in the |↑↑〉 state, a significantly slower relaxation into the
(1,1) state is observed, due to the spin blockade combined with the slow T+ (1,1)-S(2,0) relaxation.
By fitting a double exponential decay, accounting for the SHT response, we extract a spin relax-
ation at the readout point of T1,RO = 26 µs. A sample of 100 single-shot traces is plotted in Fig.
7.2g, together with two individual traces using a post-processing integration time of 3 µs. A clear
distinction of the (1,1) and (2,0) charge states can be observed from the sensor response. To de-
termine the spin state of the qubits, we perform a threshold detection of the single-shot signal
integrated from τ0 = 1.0 µs up to τmeas = 6.0 µs for maximised visibility, discarding the initial sta-
bilisation of the SHT and optimising between the charge discrimination and spin relaxation. A his-
togram of 5000 single-shot events illustrates the clear distinction between the singlet (S 21 > −0.72)
and the triplet (S 21 < −0.72) spin state readout (Fig. 7.2h). We find a spin readout visibility of
v = 56% as obtained from the difference in spin-up fraction between the two prepared states. A
large part of this reduced visibility is caused by relaxation of the blocked triplet state during the
measurement, expected to amount to a signal reduction of P relax = 1 − e −τmeas /T1,RO = 0.21. This
gives good prospects for increasing the readout fidelity by optimising the spin relaxation, for in-
stance by optimising the reservoir tunnel rates and moving to latched PSB readout mechanisms
[38, 39]. Alternatively, by using high-Q on-chip resonators [40] the signal-to-noise ratio could be
significantly improved, thereby lowering the required integration time and reducing the effective
relaxation. The remaining triplet fraction of 0.11 that can be observed for the readout of the |↓↑〉
state could be attributed to an anadiabaticity of the pulsing or a small coupling between the T(1,1)
and S(2,0) states due to the SOC. This could be mitigated by further optimising the readout pulse
sequence.

7 Now we probe the single spin relaxation time by initialising the system in the |↓↑〉 state and
letting the system evolve at a detuning voltage ∆V² = −7 mV from the (1,1)-(2,0) anticrossing. Fig.
7.3a shows the spin-up fraction as a function of the waiting time t wait , from which a single spin
relaxation time of T1,Q2 = 1.2 ms can be extracted. This is substantially longer than reported be-
fore in planar germanium heterostructures [21], most likely as a result of the more isolated single
hole spins as compared to the transport measurements with high reservoir couplings, and is also
longer than all relevant time scales for qubit operation. Moreover, this relaxation time compares
favourably to results obtained for holes in Ge nanowires [41], Ge hutwires [27] and other hole spins
[42, 43].
To demonstrate coherent control of a single hole, we modulate the length of the driving mi-
crowave pulse and measure the spin-up fraction (Fig. 7.3b). A clear sinusoidal Rabi oscillation can
be observed on Q2, with a Rabi frequency of f R = 57 MHz (coherent operation of Q1 in Figure 7.8).
We probe the phase coherence of both qubits by performing a Ramsey sequence in which we apply
two π/2-pulses, separated by a time τ in which we let the qubit freely evolve and precess at a fre-
quency offset of ∆ f = 7.4 MHz and ∆ f = 23.7 MHz respectively. In Fig. 7.4b the Ramsey decay for
∗
Q1 and Q2 are plotted and we extract coherence times of T2,Q1 ∗
= 380 ns and T2,Q2 = 140 ns. These
coherence times are of comparable order, but slightly lower than previously reported numbers in
the same heterostructere for a many-hole quantum dot [21]. In order to explain the origin of this,
we measure the resonance frequency of both qubits as a function of the detuning voltage ∆V² . We
find a very strong dependence of the resonance frequency of both qubits on the detuning voltage
over the entire range of voltages measured, with the g -factor varying between g Q1 = 0.27 − 0.3 and
g Q2 = 0.21 − 0.29. This strong electric field dependence of the resonance frequency will increase
the coupling of charge noise to the qubit spin states, which in turn will reduce phase coherence
7.3. D ISCUSSION 115

a (2,0) (1,1) (R) b (2,0) (1,1) (R)

Δε π Δε π/2 π/2
RF ΔVε RF τ

2.8 0.6
T *2 = 380 ns
0.4
2.6

P
Q1
f resonance (GHz)

0.2

2.4 0 200 400 600

0.6 *
T 2 = 140 ns
2.2 Q2 0.4

P
B = 0.67 T 0.2
2
-40 -20 0 0 100 200
V (mV) (ns)

Figure 7.4: Electric g-factor modulation and phase coherence of the qubit resonances. a The resonance
frequency of both qubits shows a strong modulation as a function of the detuning voltage ∆V² . b We perform
a Ramsey experiment on both qubits to probe the phase coherence times, with T2,Q1 ∗ ∗
= 380 ns and T2,Q2 =
140 ns. The comparatively short phase coherence can be attributed to the strong dependence of f resonance to
electric fields, coupling charge noise to the spin state, leading to increased decoherence.

[21]. The ratio in local slopes of the resonance frequency δ f Q2 /δ f Q1 = 2 is similar to the ratio
in phase coherence of both qubits T2,Q1 ∗ /T ∗ = 2.5, consistent with charge-noise limited coher-
2,Q2
ence. The strong modulation of the qubit resonance frequency by electric field could be explained
from the strong SOC present [44, 45]. This is further supported by the Rabi frequency changing
as a function of detuning voltage (see Figure 7.9), as is predicted to be a result of the strong SOC
7
[44, 45]. We attribute the slightly different resonance frequency of Q1 and Q2 to an asymmetry in
the potential landscape of the two dots. Although the strong g -factor modulation seems mainly a
cause of decoherence in this case, careful optimisation of the electric field landscape could render
a situation in which the qubit Zeeman splitting is well controllable, while maintaining a zero local
slope for high coherence [45].

7.3. D ISCUSSION
The demonstration that single hole spins can be coherently controlled and read out in single-shot
mode, together with the spin relaxation times T1 > 1 ms, defines planar germanium as a mature
quantum platform. These aspects are demonstrated on a two-dimensional quantum dot array,
further highlighting the advancement of germanium quantum dots. Moreover, controlling a sin-
gle hole spin represents an important step towards reproducible quantum hardware for scalable
quantum information processing.

M ETHODS
7.3.1. FABRICATION PROCESS
We grow strained Ge/SiGe heterostructures in an Epsilon 2000 (ASMI) reduced-pressure chemical
vapor deposition reactor on a 100 mm n-type Si(001) substrate. The growth sequence comprises a
1.6-µm-thick relaxed Ge layer; a 1-µm-thick step-graded Si1−x Gex layer with final Ge composition
 116 7. A S INGLE -H OLE S PIN QUBIT

x=0.8; a 500-nm-thick strain-relaxed Si0.2 Ge0.8 buffer layer; a 16-nm-thick strained Ge quantum
well; a 22-nm-thick strain-relaxed Si0.2 Ge0.8 barrier; a sacrificial Si cap layer <2 nm thick. Further
details on the heterostructure are discussed in Ref. [31]. Ohmic contacts are defined by electron
beam lithography, electron beam evaporation and lift-off of a 30-nm-thick Al layer. Electrostatic
gates consist of a Ti/Pd layer with a thickness of 20 and 40 nm respectively for the barrier and
plunger gate layer. Both layers are separated from the substrate and each other by 10 nm of ALD-
grown Al2 O3 .

7.3.2. E XPERIMENTAL SETUP

We use a Bluefors dry dilution refrigerator with a base temperature of Tbath ≈ 20 mK to perform
the measurements. Battery-powered voltage sources are used to supply DC-voltages on the gates.
Additionally, AC-voltages generated by a Tektronix AWG5014C arbitrary waveform generator can
be supplied to the gates through a bias-tee with a cut-off frequency of ≈ 10 Hz. Similarly, we can
also apply a microwave signal generated by a Keysight PSG8267D vector source to gate P1 for qubit
driving. Driving both qubits at the same power on gate P2, we observe significantly slower Rabi
oscillations in Q1. From this we assume Q1 to be located under P1, and thus Q2 under P2, in
correspondence with the trend in Rabi frequencies observed in a previous work [21].
We use an in-house built RF generator to supply the reflectometry signal. The signal is attenu-
ated by 84 dB and applied to one of the sensor Ohmics via a Mini-Circuits ZEDC-15-2B directional
coupler. The reflected signal is amplified by a Caltech CIRLF3 SiGe-amplifier at the 4K-stage of
our fridge and an in-house built RF-amplifier at room temperature, and finally IQ-demodulated
to give a measure of S 21 .

7.3.3. M EASUREMENT DETAILS

The large panels of Fig. 7.2c and Fig. 7.2d are measured by continuously sweeping ε and stepping
U , while measuring the sensor response. The smaller panels in Fig. 7.2c show the sensor response
7 after applying a two-level voltage pulse to load the (2,0) or (0,2) charge configuration and vary the
readout point across the map. The signal is then integrated for 10 µs at each pixel. The smaller
panels in Fig. 7.2d show the sensor response after applying a three-level voltage pulse to first
randomly load a spin in the second dot by pulsing across the (1,0)-(1,1) reservoir transition. Next,
we pulse across the (1,1)-(0,2) or (1,1)-(2,0) interdot transition to perform the spin readout. The
colour scale of the signal in Fig. 7.2c left and Fig. 7.2d left panels is normalised by S 21,n = 10S 21 +
3.5. The top right panel in Fig. 7.2c is normalised by S 21,n = 12.5S 21 +5.375. The bottom right panel
in Fig. 7.2c is normalised by S 21,n = 12.5S 21 + 5. The top right panel in Fig. 7.2d is normalised by
S 21,n = 20S 21 + 16.8. The bottom right panel in Fig. 7.2d is normalised by S 21,n = 25S 21 + 20.75,
with S 21,n the normalised sensor signal as plotted in Fig. 7.2 and S 21,n the demodulated sensor
signal strength in mV.
For the measurements in Fig. 7.2f-h and Fig. 7.3 and 7.4, typical adiabatic ramp times of
Tramp ≈ 1µs are used.

7.3.4. V IRTUAL GATES

In order to allow independent control over the detuning and energy of the quantum dots more
easily, we define the virtual gate axes of Vε = VP2 − 0.5VP1 and VU = 0.5VP2 + VP1 .
7.4. E XTENDED D ATA 117

7.4. E XTENDED D ATA

0 0.01 0.02
| S 12 | (a.u.)

-1060

-1080
V P1 (mV)
-1100

-1120

-1140
-1280 -1260 -1240 -1220 -1200
V P2 (mV)

Figure 7.5: Depletion of a hole double quantum dot in germanium. Colour map of the sensor signal as a
function of the voltages on plunger gates P1 and P2. No extra addition lines can be observed beyond VP1 ≈
−1100 and VP2 ≈ −1260, indicating the double quantum dot is fully depleted. This is the same anti-crossing as
observed in Fig. 7.1b and the slight decrease of plunger gate voltages can be attributed to a time-dependent
hysteretic drift as a result of extensive gate voltage sweeping.

7
 118 7. A S INGLE -H OLE S PIN QUBIT

0 0.5
v

-1260.5

V U (mV)
-1261

-1261.5

-1262
-664 -662 -660
V (mV)

Figure 7.6: Optimisation of the readout-point. Colour map of the visibility of the readout v, as defined by
v = P blocked,π − P blocked,0 , with P blocked,π being the probability of measuring a blocked signal after applying
a resonant π-pulse to Q1 and P blocked,0 the probability of measuring a blocked signal without applying any
microwave pulses. A clear optimal spot for readout can be observed at VU = −1260.7 mV, V² = −661.0 mV.

7 -0.32 = 0.21 eV/V

2
-0.34
V U (mV)
S 21 (a.u.)

-0.36 0

-0.38
-2
-0.4
-4
-0.42
-500 0 500 -0.5 0 0.5 1
Detuning (uV) V (meV)

Figure 7.7: Lever arm and excited state energy of the quantum dot. a Polarisation line measurement of the
(1,1)-(0,2) anticrossing. We fit the thermally limited polarisation line to a model including cross-talk to the
charge sensor and the effect of the charge state on the sensor sensitivity [46]. Assuming a hole temperature
of 100 mK as measured previously [47], we find a lever arm of α² = 0.21 eV/V, in good agreement with results
obtained on similar devices. b We measure the excited state energy by applying a DC bias across the quantum
dot ohmics, shifting the anti-crossing towards the negative detuning voltage. For large enough bias, the read-
out window is capped off, as a result of the excited state becoming available in energy. From this we deduce an
excited state splitting of E ST = 0.85 meV.
7.4. E XTENDED D ATA 119

f R = 93 MHz
0.6

0.4
P

0.2

0
0 50 100 150 200
t p (ns)

Figure 7.8: Coherent operation of Q1. We rotate Q1 by applying a resonant microwave pulse to gate P1 and
observe fast Rabi oscillations, with a frequency of f R = 93 MHz.

P blocked
0 0.2 0.4 0.6 7
2.8
f resonance (GHz)

2.6

2.4

2.2

-45 -40 -35 -30 -25 -20 -15 -10 -5 0

V (mV)

Figure 7.9: Detuning dependence of the resonance frequency of Q1 and Q2. Colour map indicating a blocked
state fraction P blocked as a function of the detuning voltage ∆V² of the manipulation point. We apply a mi-
crowave pulse to gate P2 with a duration of t p = 105 ns, corresponding to an approximate 3π-pulse on Q2 and
a π-pulse on Q1 at ∆V² = −5 mV. The resonance line corresponding to Q2, can be observed to split and recom-
bine throughout the map, as a direct result of the Rabi frequency changing resulting in rotations exceeding a
3π-pulse.
 120 R EFERENCES

R EFERENCES
[1] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Phys. Rev. A 57,
120–126 (1998).

[2] Lawrie, W. I. L. et al. Quantum dot arrays in silicon and germanium. Appl. Phys. Lett. 116,
080501 (2020).

[3] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 34 (2017).

[4] Veldhorst, M., Eenink, H. G. J., Yang, C. H. & Dzurak, A. S. Silicon CMOS architecture for a
spin-based quantum computer. Nature Communications 8, 1766 (2017).

[5] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science Advances 4, eaar3960
(2018).

[6] Itoh, K. M. & Watanabe, H. Isotope engineering of silicon and diamond for quantum com-
puting and sensing applications. MRS Commun. 4, 143–157 (2014).

[7] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nat. Nanotech. 9, 981–985 (2014).

[8] Morello, A. et al. Single-shot readout of an electron spin in silicon. Nature 467, 687–691
(2010).

[9] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nat. Nanotech. 102–106 (2017).

[10] Yang, C. H. et al. Silicon qubit fidelities approaching incoherent noise limits via pulse engi-
7 neering. Nat Electron 2, 151–158 (2019).

[11] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[12] Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 1 (2019).

[13] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon. Nature 555,
633–637 (2018).

[14] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins. Science 359, 439–442
(2018).

[15] Bulaev, D. V. & Loss, D. Electric Dipole Spin Resonance for Heavy Holes in Quantum Dots.
Phys. Rev. Lett. 98, 097202 (2007).

[16] Bulaev, D. V. & Loss, D. Spin Relaxation and Decoherence of Holes in Quantum Dots. Phys.
Rev. Lett. 95, 076805 (2005).

[17] Pribiag, V. S. et al. Electrical control of single hole spins in nanowire quantum dots. Nature
Nanotechnology 8, 170–174 (2013).

[18] Nowack, K. C., Koppens, F. H. L., Nazarov, Y. V. & Vandersypen, L. M. K. Coherent Control of a
Single Electron Spin with Electric Fields. Science 318, 1430–1433 (2007).

[19] Maurand, R. et al. A CMOS silicon spin qubit. Nature Communications 7, 13575 (2016).
R EFERENCES 121

[20] Watzinger, H. et al. A germanium hole spin qubit. Nature Communications 9, 3902 (2018).

[21] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit
logic with holes in germanium. Nature 1–5 (2020).

[22] Hendrickx, N. W. et al. Gate-controlled quantum dots and superconductivity in planar ger-
manium. Nature Communications 9, 2835 (2018).

[23] Hardy, W. J. et al. Single and double hole quantum dots in strained Ge/SiGe quantum wells.
Nanotechnology 30, 215202 (2019).

[24] Froning, F. N. M. et al. Single, double, and triple quantum dots in Ge/Si nanowires. Appl.
Phys. Lett. 113, 073102 (2018).

[25] Hendrickx, N. W. et al. Ballistic supercurrent discretization and micrometer-long Josephson

coupling in germanium. Phys. Rev. B 99, 075435 (2019).

[26] Vigneau, F. et al. Germanium Quantum-Well Josephson Field-Effect Transistors and Interfer-
ometers. Nano Lett. 19, 1023–1027 (2019).

[27] Vukušić, L. et al. Single-Shot Readout of Hole Spins in Ge. Nano Lett. 18, 7141–7145 (2018).

[28] Mizokuchi, R., Maurand, R., Vigneau, F., Myronov, M. & De Franceschi, S. Ballistic One-
Dimensional Holes with Strong g-Factor Anisotropy in Germanium. Nano Lett. 18, 4861–4865
(2018).

[29] He, L., Bester, G. & Zunger, A. Electronic Phase Diagrams of Carriers in Self-Assembled Quan-
tum Dots: Violation of Hund’s Rule and the Aufbau Principle for Holes. Phys. Rev. Lett. 95,
246804 (2005).

[30] Liles, S. D. et al. Spin and orbital structure of the first six holes in a silicon metal-oxide-
7
semiconductor quantum dot. Nat. Comm. 9 (2018).

[31] Sammak, A. et al. Shallow and Undoped Germanium Quantum Wells: A Playground for Spin
and Hybrid Quantum Technology. Advanced Functional Materials 29, 1807613 (2019).

[32] Lodari, M. et al. Light effective hole mass in undoped Ge/SiGe quantum wells. Phys. Rev. B
100, 041304 (2019).

[33] Schoelkopf, R. J., Wahlgren, P., Kozhevnikov, A. A., Delsing, P. & Prober, D. E. The Radio-
Frequency Single-Electron Transistor (RF-SET): A Fast and Ultrasensitive Electrometer. Sci-
ence 280, 1238–1242 (1998).

[34] Reilly, D. J., Marcus, C. M., Hanson, M. P. & Gossard, A. C. Fast single-charge sensing with a rf
quantum point contact. Appl. Phys. Lett. 91, 162101 (2007).

[35] Barthel, C. et al. Fast sensing of double-dot charge arrangement and spin state with a radio-
frequency sensor quantum dot. Phys. Rev. B 81, 161308 (2010).

[36] Ono, K., Austing, D. G., Tokura, Y. & Tarucha, S. Current Rectification by Pauli Exclusion in a
Weakly Coupled Double Quantum Dot System. Science 297, 1313–1317 (2002).

[37] Hensgens, T. et al. Quantum simulation of a Fermi–Hubbard model using a semiconductor

quantum dot array. Nature 548, 70–73 (2017).
 122 R EFERENCES

[38] Studenikin, S. A. et al. Enhanced charge detection of spin qubit readout via an intermediate
state. Appl. Phys. Lett. 101, 233101 (2012).

[39] Harvey-Collard, P. et al. High-Fidelity Single-Shot Readout for a Spin Qubit via an Enhanced
Latching Mechanism. Phys. Rev. X 8, 021046 (2018).

[40] Zheng, G. et al. Rapid gate-based spin read-out in silicon using an on-chip resonator. Nature
Nanotechnology 14, 742–746 (2019).

[41] Hu, Y., Kuemmeth, F., Lieber, C. M. & Marcus, C. M. Hole spin relaxation in Ge–Si core–shell
nanowire qubits. Nat. Nanotech. 7, 47–50 (2012).

[42] Bogan, A. et al. Single hole spin relaxation probed by fast single-shot latched charge sensing.
Communications Physics 2, 1–8 (2019).

[43] Gerardot, B. D. et al. Optical pumping of a single hole spin in a quantum dot. Nature 451,
441–444 (2008).

[44] Terrazos, L. A. et al. Theory of Hole-Spin Qubits in Strained Germanium Quantum Dots.
(2020).

[45] Wang, Z. et al. Suppressing charge-noise sensitivity in high-speed Ge hole spin-orbit qubits.
(2019).

[46] DiCarlo, L. et al. Differential Charge Sensing and Charge Delocalization in a Tunable Double
Quantum Dot. Phys. Rev. Lett. 92, 226801 (2004).

[47] Petit, L. et al. Spin Lifetime and Charge Noise in Hot Silicon Quantum Dot Qubits. Phys. Rev.
Lett. 121, 076801 (2018).
7
 8
S PIN R ELAXATION B ENCHMARKS
AND I NDIVIDUAL Q UBIT
A DDRESSABILITY FOR H OLES IN
Q UANTUM D OTS
We investigate hole spin relaxation in the single- and multi-hole regime in a 2x2 germanium quan-
tum dot array. We find spin relaxation times T1 as high as 32 ms and 1.2 ms for quantum dots with
single- and five-hole occupations respectively, setting benchmarks for spin relaxation times for hole
quantum dots. Furthermore, we investigate qubit addressability and electric field sensitivity by mea-
suring resonance frequency dependence of each qubit on gate voltages. We can tune the resonance
frequency over a large range for both single and multi-hole qubits, while simultaneously finding that
the resonance frequencies are only weakly dependent on neighbouring gates. In particular, the five-
hole qubit resonance frequency is more than twenty times as sensitive to its corresponding plunger
gate. Excellent individual qubit tunability and long spin relaxation times make holes in germa-
nium promising for addressable and high-fidelity spin qubits in dense two-dimensional quantum
dot arrays for large-scale quantum information.

Parts of this chapter have been published in Nano. Lett. 20 (10), 7237-7242 [1].

123
 8. S PIN R ELAXATION B ENCHMARKS AND I NDIVIDUAL QUBIT A DDRESSABILITY FOR H OLES
124 IN Q UANTUM D OTS

Qubits based on spin states are well established candidates for quantum information pro-
cessing [2]. Pioneering studies were conducted on low-disorder gallium arsenide heterostructures
[3, 4], but quantum coherence remained limited due to hyperfine interaction with nuclear spins.
These interactions can be eliminated by using isotopically enriched group IV semiconductors as
the host material [5]. In silicon this has led to landmark achievements, such as extremely long
quantum coherence [6] and relaxation times [7], single qubit gates with fidelities beyond 99.9%
[8, 9], execution of two-qubit gates [10, 11], quantum algorithms [12], and the operation of sin-
gle qubit rotations [13] and two-qubit logic [14] above one Kelvin as a key step toward quantum
integrated circuits [15–17].
In it’s natural form, germanium contains only 7.76% isotopes with non-zero nuclear spin and,
like silicon, can be isotopically enriched [18] to eliminate nuclear spin dephasing. Recent advances
in materials science enabled high mobility strained planar germanium (Ge/SiGe) heterostructures
[19] for the fabrication of stable gate-defined quantum dots that can confine holes [20], which are
predicted to have a multitude of favourable properties for quantum control [21, 22]. The inherent
strong spin-orbit coupling of holes allows for fast qubit control [23–25] without integrating ex-
ternal components that complicate scalability, such as nano-magnets and microwave antennas.
Moreover, holes do not suffer from valley degeneracy and their small effective mass of m∗ h
= 0.05
m∗ e [26] gives rise to large orbital splittings at the band center. These beneficial aspects thereby
position holes in germanium as a promising material for quantum information [27].
While it has been demonstrated that both single- and multi-hole qubits can be coherently
controlled and read out in planar germanium [25, 28], an open question remains which hole oc-
cupancy is most advantageous for quantum operation. Electron spin qubits in silicon have been
operated with quantum dots containing one, three and even more electrons, with more electrons
typically performing favourably in terms of driving speed when driven electrically due to greater
wave function mobility [10, 29]. Here, we focus on single and multi-hole spin qubit operation
in germanium and concentrate on two critical elements for quantum information with quantum
dots: the spin relaxation time and the qubit addressability. We find that both the spin relaxation
times of the single-hole (T1,|n=1〉 ) and five-hole (T1,|n=5〉 ) qubits are long, with the longest relax-
ation time for single-holes measured to be T1,|n=1〉 = 32 ms. Furthermore, we observe that sin-
gle and multi-hole qubits exhibit a strong but comparable resonance frequency dependence on
electric gate voltage. Interestingly, we find that while the qubit resonance frequency can be sig-
8 nificantly tuned with the corresponding plunger gate, it is only weakly dependent on neighbour
plunger gates. We thereby conclude that hole spin qubits can be locally addressed, crucial for the
operation of dense qubit arrays.

8.1. D EVICE AND S ETUP

The experiments are performed on a two-dimensional 2x2 quantum dot array fabricated using
a multi-layer gate stack [30] (See Fig. 8.1a). Four plunger gates P1−4 define four quantum dots,
whose interdot tunnel couplings are controllable via barrier gates B12−41 . Four metallic reservoirs
O1−4 can be controllably coupled to each quantum dot via their respective barrier gates RB1−4 .
We operate in a configuration whereby electrostatic gates P1 , B12 and P2 define one large quantum
dot, serving as single-hole transistor (SHT) for charge sensing, shown in Fig. 8.1b. By connecting
an inductor of L ≈ 2 µH inline with the ohmic O1 , we form a resonant tank circuit at a frequency
of f O1 = 150.7 MHz used for fast rf charge sensing [31, 32]. To optimise the reflection signal and
minimise the inline resistance, we use a superconducting inductor made out of NbTiN. A second
inline inductance connected to ohmic O3 makes the device reconfigurable for any of the four dou-
8.1. D EVICE AND S ETUP 125

(a)
CO1 P1 B12 RB2 O2
O1
CO2
RB1
100 nm P2

B41 B23

P4
RB3
CO4
O4 RB4 B34 P3 CO3 O3

(b) (c)

-0.1 -0.16
|S11 | (a.u.)

|S11 | (a.u.)

-0.15 -0.18
-0.2 -0.2
-0.25 -0.22
-0.3 -0.24
140 145 150 155 -1380 -1360 -1340
f (MHz) P 1 (mV)

Figure 8.1: (a) Coloured scanning electron microscope image of a nominally identical 2x2 quantum dot array.
Each quantum dot is defined by a plunger gate, P1−4 (yellow) and barrier gates B12−41 (blue) are used to set
the tunnel coupling. In addition, each quantum dot is coupled to a reservoir, O1−4 (green), via a barrier gate
RB1−4 . A cut off gate, CO1−4 , is present for good confinement of the quantum dots. Ohmics 1 and 3 are bonded
to an inductor to create a tank circuit with the parasitic capacitance of the device to ground. A radiofrequency
tone is applied to the ohmics and the reflected signal |∆S 11 | returns via a directional coupler and is read out.
(b) Reflected signal of the two tank circuits. Two clear resonances occur at f O1 = 150.7 MHz and f O3 = 143.3
MHz for tank circuits connected to ohmics O1 and O3 respectively. (c) Single-hole transistor (SHT) Coulomb 8
oscillations measured in the tank circuit response by applying a microwave tone of 150.7 MHz. A sensing
quantum dot is formed underneath the plunger gates P1 and P2 , by opening the interdot barrier gate B12 .

ble quantum dot-sensor combinations, with a second tank circuit resonance at f O3 = 143.3 MHz.
The reflected signal response to rf power delivered to the sample in a frequency range encom-
passing these two resonances is shown in Fig. 8.1b. Modulation of the channel resistance due to
Coulomb oscillations in the SHT is shown in Fig. 8.1c. Next, we apply voltages to the plunger gates
P3 and P4 to form quantum dots that load via the reservoir barriers RB3 and RB4 , respectively.
By applying sawtooth wave pulses to the plunger gates and simultaneously applying the inverse
pulses to the SHT plunger gates P1 and P2 , we can tune up the device to a double quantum dot of
arbitrary occupancy while compensating the charge sensor in real time.
 8. S PIN R ELAXATION B ENCHMARKS AND I NDIVIDUAL QUBIT A DDRESSABILITY FOR H OLES
126 IN Q UANTUM D OTS

(a) -0.4 -0.2 0 (b) (c) 2 (h) (j) S1,5 L

2
|S11| (a.u.) (0,5) T- L R
VU (mV)

0 (1,1)

U
0 (1,0) (0,1) (0,2)
(0,6) S0,6 R
(1,5)

VU
N0,5 I S0,6 I
30 -2 -2
-1.5 0 1.5 -1.5 0 1.5 (1,6) ϵ
20 (k)
I
(d) 2 (e) Vϵ
2

VU
10
L R I
VU (mV)
ΔP4 (mV)

0 (1,2) (0,3) 0 (1,3) (0,4)

P 4 (mV)

0
-2 (i) (l) Vϵ
-2 tc = 2.5 GHz
-10 -0.34 TST = 103 μs
-1.5 0 1.5 -1.5 0 1.5

-20 -0.36 0
|S11 | (a.u.)

S 11 | (a.u.)
(f) 2 (g) 2
-0.38 -0.02
-30
VU (mV)

0 (1,4) (0,5) 0 (1,5) (0,6) -0.4 -0.04

-40
-0.42 -0.06 L R

|
-2 -2
-100 0 100 -1.5 0 1.5 -1.5 0 1.5 0 1 2 0 200 400
P 3 (mV) Vϵ (mV) Vϵ (mV) V (mV) tR ( s)

Figure 8.2: (a) Double quantum dot charge stability diagram obtained by rf-charge sensing in a reconfigurable
quadruple quantum dot. A double quantum dot is formed under P3 (blue) and P4 (red) with controllable inter-
dot tunnel coupling by tuning interdot barrier gate B34 and using rf-charge sensing we can clearly monitor the
charge occupancy. The map is taken with 2000 averages. (b-g) Charge stability diagrams of charge anticross-
ings (1,N-1)⇐⇒(0,N) for integer steps in N from 1 to 6. When N is odd (b,d,f), no spin blockade is present in
the transitions. For even N (c,e,g), we observe spin blockade evidenced by triangular extensions of the charge
addition lines (white lines are drawn in as a visual guide). (h-i) Interdot tunnel coupling measurement for the
8 (1,5)⇐⇒(0,6) transition. Sweep direction is negative in detuning axis to avoid spin blockade artifacts in spec-
trum. A tunnel coupling of t c = 2.5 GHz is obtained and kept within 200 MHz of this value for all measurements
in the work. (j) Cartoon of the initialization (I), loading (L) and readout (R) points of our pulse sequences in de-
tuning ² and energy U space. The blue sequence loads the |S(1,5) 〉 singlet state which is not blocked on readout.
The red sequence initializes in the (0,5) charge state N0,5 and loads a randomly oriented spin resulting in the
singlet state or triplet state with equal probability (red), the latter of which results in a blocked signal. (k) Pulse
sequences as a function of virtual detuning (Ve ) and energy (VU ) gates, for the case of loading a singlet state
(blue) and triplet/singlet state with equal probability (red). (l) Readout traces showing the signal difference
between the spin blocked (red) and unblocked (blue) signals as a function of time spent in the read phase (tR ).
Each trace is averaged 1000 times. The small deviation in the blue trace around tR = 0 µs is due to ringing in
the applied AWG pulse. Loading a random spin under P3 (red) leads to a singlet-triplet decay time of TST = 103
µs.
8.2. S PIN B LOCKADE FOR H OLE F ILLINGS 127

8.2. S PIN B LOCKADE FOR H OLE F ILLINGS

Figure 8.2a shows the charge stability diagram for a sweep of gates P3 versus P4 over the first few
charge addition lines in each quantum dot. We focus on the set of anticrossings for the first charge
addition line of the quantum dot under P3 of the form (NP 3 , NP 4 ) ⇐⇒ (NP 3 -1, NP 4 + 1), where
NP 3(4) is the charge occupation of the quantum dot formed under P3(4) , from double dot occupa-
tion (1,0) ⇐⇒ (0,1) to (1,5) ⇐⇒ (0,6). We define a virtual gate space in detuning V² and energy
VU through a linear transformation of the gate voltages on P3 and P4 . We apply sawtooth wave
pulses that sweep V² from -2 mV to +2 mV and steps VU from -2.5 mV to +2.5 mV with respect to
the anticrossing. Figures 8.2b-g shows the resulting stability diagrams.
Pauli spin blockade is observed for (Nodd ,Nodd ) ⇐⇒ (Neven ,Neven ) type transitions up to the
sixth occupancy. This is consistent with a Fock-Darwin level filling observed for electrons in gal-
lium arsenide [33] and holes in silicon [34] until the same charge occupancy. Working now with
the (1,5) ⇐⇒ (0,6) anticrossing, we extract a lever arm α = 0.18 from the thermally broadened po-
larization line using a hole temperature of 100 mK, which allows us to extract the tunnel couplings
[35]. Figures 8.2h,i show the used pulse scheme and measured trace. The interdot coupling is kept
constant within t c = 2.5 ± 0.2 GHz for all measurements. In Figures 8.2j, 8.2k, and 8.2l, we show
the relevant pulse sequences in both energy-detuning and virtual gate space, as well as the result-
ing traces for Pauli spin blockade (PSB) readout respectively. Each pulse contains an initialization
(I), load (L) and read (R) phase. Here we compare the partially blocked (red) and unblocked (dark
blue) signals, allowing us to distinguish between the spin up and spin down states in the lower
energy quantum dot. The small decay at the beginning of the unblocked trace is due to ringing in
the AWG pulses. By loading a hole in P3 with a random spin state, we expect to observe a blocked
signal approximately half of the time. Monitoring the readout signal in the charge sensor as a func-
tion of time provides us with the spin relaxation at the readout position, which we find to be T ST =
103 µs.

8.3. S PIN R ELAXATION

We now assess the spin relaxation of the single- and five-hole qubits. All experiments were per-
formed at a magnetic field B = 0.67 T, allowing for a comparison with previous germanium hole
8
spin qubit experiments [25, 28] in a similar magnetic field regime. The Zeeman energy difference
between our qubits manifests from spin-orbit induced changes in the g-factor of each qubit, lift-
ing the antiparallel spin state energy degeneracy when an external magnetic field is applied. Figure
8.3a shows the pulse sequences used to measure the spin relaxation times of the hole spins in each
quantum dot. Each pulse sequence consists of an initialization (I), load (L), and read (R) phase,
with two ramps between the I and L phases (tIL ), and the L and R phases (tLR ). Using the first two
sequences (red and blue in Fig. 8.3a) a randomly orientated spin is loaded into the quantum dot
defined under P4 or P3 respectively. The Red and blue pulse sequences also contain an extra wait
(W) step before initialization in the (1,5) charge state to relax the spins to their ground state. This
allows deterministic probing of the spin relaxation time of each dot, by varying the load wait time
tL .
The third pulse sequence (yellow in Fig. 8.3a) initializes the system in the singlet state with
charge configuration (NP 3 ,NP 4 ) = (0,6) (|S(0,6) 〉). The system is then tuned to the charge configu-
ration (NP 3 ,NP 4 ) = (1,5) (|S(1,5) 〉). We pulse with a ramp time tI L = 100 ns, resulting in a diabatic
movement through the charge anticrossing, and through fast charge relaxation we expect to ini-
tialize the | ↑, ↓〉 and | ↓, ↑〉 states randomly with equal probability. This initialization then allows
us to efficiently measure both spin relaxation times in a single measurement, since the readout
 8. S PIN R ELAXATION B ENCHMARKS AND I NDIVIDUAL QUBIT A DDRESSABILITY FOR H OLES
128 IN Q UANTUM D OTS

(a) (b)
W I t IL L t LR R

ϵ 0.6

(1,5) Fraction
W/L R
U
0.4
U
I T1,|n=1>= 4.2 ms
t 0.2
ϵ T1,|n=5>= 1.0 ms

I 0
ϵ 0 5 10
tL (ms)
W/L R
U
(c)
U 1
t
ϵ T1,|n=1>= 32 ms
| S 11 | (a.u.)

I t IL L t LR R T1,|n=5>= 1.2 ms
0.5
ϵ

U L R I

U 0
0 50
ϵ t
tL (ms)
8
Figure 8.3: (a) Pulse sequences utilized for different loading protocols. Red loads a random spin in P4 , blue
loads a random spin in P3 , and yellow loads either the |S(1,5) 〉 or a mixture of |S(1,5) 〉 and |T0 〉 depending on the
adiabaticity of the IL pulse. (b) Deterministic loading of a single-hole in P4 and P3 (red, blue respectively) and
mixed state loading (yellow). We extract spin relaxation times of T1,|n=5> = 1.0 ms and T1,|n=1> 4.23 ms for
each deterministically loaded quantum dot, which we fit as a double exponential for the mixed loading case.
(c) Longest spin relaxation trace taken after minimizing reservoir-dot tunnel coupling. We extract two spin
relaxation times of T1,|n=1> = 32 ms and T1,|n=5> = 1.2 ms.
8.4. QUBIT A DDRESSABILITY 129

signal is a linear combination of both spin relaxation decays, and is useful since it allows for fast
measurements even when the quantum dot-reservoir couplings are low. Upon readout, the inher-
ent spin orbit coupling in our system results in an avoided crossing between the |T− 〉 and |S(0,6) 〉
states, potentially limiting our readout fidelity. We therefore minimize the ramp time and operate
with tLR = 100 ns.
In Fig. 8.3b, we show the spin relaxation times of the quantum dots using the three sequences.
We find T1,|n=5〉 = 1.0 ms and T1,|n=1〉 = 4.23 ms by fitting exponential decays to the individual
measurements. The measurement corresponding to the sequence with randomly preparing a spin
up state in one of the two quantum dots is fitted with a double exponential curve using the time
constants of the individual decays, and we have left the amplitudes and asymptotes as free fitting
parameters. We find approximately equal amplitudes for each decay, in correspondence with an
equal loading of both anti-parallel spin states.
We can further increase the single-hole relaxation time by reducing the quantum dot-reservoir
coupling. Using the barrier gate RB3 , we tune the quantum dot-reservoir coupling of the single-
hole quantum dot from 81.43 KHz to 27.45 KHz (see section 8.6.1). We note that these dot-reservoir
couplings do not represent the actual tunnelling times at the point of measurement, which are ex-
pected to be orders of magnitude longer. The spin relaxation decay shown in Fig. 8.3c has been
analysed using the above mentioned double exponential fit and we find an significantly increased
single-hole spin relaxation time T1,|n=1〉 = 32 ms. By limiting the dot-reservoir tunnel coupling, we
have demonstrated spin relaxation lifetimes significantly longer than results previously reported
for planar germanium quantum dots (T 1,|n=1〉 = 1.2 ms [28]), hut wires (T 1 = 90 µs [36]), nanowires
(T 1 = 600 µs [37]), and even holes in gallium arsenide (T 1 = 60 µs [38]) and silicon (T 1 = 8.3 µs [39])
at similar magnetic fields. We expect that the main cause for the observation of longer spin re-
laxation times for single-hole spins compared to many hole spins originates from the tighter con-
finement of the quantum dot under P3 . This leads to larger energy splittings to the excited states,
a smaller degree of capacitive coupling to electrical fluctuations, and a smaller dot-reservoir cou-
pling. Additionally, further reduction of the dot-reservoir tunnel coupling would likely improve
the multi-hole spin relaxation time. Further investigation into the magnetic field dependence of
T1 could produce information about the spin relaxation mechanisms present and yield insights
into means of further optimizing T1 . However, our demonstration of T1 up to 32 ms, shows en-
couraging spin relaxation times for quantum information processing, and that spin states in pla-
nar germanium define the benchmark for spin relaxation in hole based quantum dots. 8

8.4. Q UBIT A DDRESSABILITY

The presence of spin-orbit coupling allows for electrical and coherent control of the spin states
without the need for additional structures such as striplines or micromagnets [21, 22, 24, 25]. We
investigate the individual tunability and addressability of the single- and multi-hole qubits. In Fig.
8.4a, we show results where we have applied a microwave tone of length t mw = 400 ns to the gate
P4 . We observe two resonance frequencies at 3.33 GHz and 3.53 GHz in Fig. 8.4a, corresponding
to an in-plane Zeeman energy difference ∆E z = 200 MHz, as a result of the spin-orbit induced
g-factor difference between each qubit. The difference in peak width is due to the delivery of
microwave power via plunger gate P4 , resulting in a power broadened resonance linewidth with
respect to that of the spin under P3 . Figures 8.4b,c show the dependence of each resonance fre-
quency on the electrostatic gate voltages on the two relevant plunger gates P3 and P4 . We initialize
in the |S(0,6) 〉 singlet state, then load in different points in the (1,5) charge state by changing the
potentials applied to P3 and P4 . We then manipulate the spins by applying a microwave tone to
 8. S PIN R ELAXATION B ENCHMARKS AND I NDIVIDUAL QUBIT A DDRESSABILITY FOR H OLES
130 IN Q UANTUM D OTS

(a) P4 P3

0.4

(0,6) Fraction
0.3

0.2

0.1
3.3 3.35 3.4 3.45 3.5 3.55
Frequency (GHz)
(b) f1 (MHz) (c) f2 (MHz)
10 20 30 20 60 100

-1235 -1235
P4 (mV)

P4 (mV)
-1240 -1240
-1245 -1245

-1110 -1100 -1110 -1100

P3 (mV) P3 (mV)

Figure 8.4: (a) Qubit resonance frequency of the five-hole (3.33GHz) and single-hole qubit (3.53 GHz). The
magnetic field is set to B = 667 mT. A microwave tone is applied to the gate P4 , which drives both hole spins, as
indicated by the cartoon inset. We extract in-plane g −factors of g |n=1〉 = 0.362 and g |n=5〉 = 0.383 for plunger
gate values P3 = 1098 mV, P4 = 1236 mV. (b) Five-hole (f 1 ) and (c) single-hole (f 2 ) resonance frequency de-
pendence on gate voltage. We find a strong dependence of the resonance frequency on the respective plunger
gate, but a significantly reduced dependence on the neighbouring plunger gate voltage.

P4 and read out in the PSB window. The mechanism by which our resonance frequency changes
is due to the modulation of the spin-orbit interaction via change in local electric fields, leading to
8 a modulation of the g-factor of each qubit. The resonance frequency dependence on gate voltage
is approximately linear. For the five-hole qubit we find a dependence on its plunger gate voltage
df 1 /dP4 = -4.78 MHz/mV and we find df 1 /dP3 = -0.155 MHz/mV. For the single-hole qubit we
find a slightly stronger dependence on its plunger gate voltage df 2 /dP3 = 6.78 MHz/mV and we
find a cross talk, df 2 /dP4 = -1.79 MHz/mV. These values are comparable to those measured in a
previous work on the same device under different electrostatic tuning parameters, and therefore
we expect the coherence times of each qubit to be on the order of 300 ns as measured earlier [28].
This corresponds to a cross talk ratio of about 1/30 for the five-hole qubit and about 1/4 for the
single-hole qubit. The cross talk for the single-hole qubit is comparable to the lever arm ratio (see
section 8.6.2) αP 3 /P 4 ( f 2 ) = 0.11. Remarkably, the five-hole qubit has a lever arm ratio αP 4 /P 3 ( f 1 ) =
0.07, significantly larger then the resonance frequency cross talk ratio.

8.5. C ONCLUSION
In summary, we have demonstrated benchmarks for spin relaxation in hole quantum dots and
found T1,|n=1〉 = 32 ms for a single-hole qubit and T1,|n=5〉 = 1.2 ms for a five-hole qubit and con-
8.5. C ONCLUSION 131

clude that spin relaxation is not a bottleneck for quantum computation with holes. We have shown
the presence of Pauli-spin blockade at different hole fillings and found it to be consistent with a
Fock-Darwin spectrum that only involves spin degeneracy. We find that both the single-hole and
multi-hole qubit resonance frequency can be tuned over a large range. We find that the reso-
nance frequencies are only weakly dependent on neighbouring gates, which results in good local
addressability. The observation of the sign difference in the resonance frequency dependence on
gate voltage and the strength of the cross talk ratio of the resonance frequencies may provide in-
sights in the nature of the driving mechanism of holes in planar germanium. This is relevant for
future work and a possible scenario is that the reduced cross talk of the five-hole qubit originates
from an increased heavy-hole light-hole mixing. Such a change may affect the qubit resonance
frequency dependence on the amplitude and orientation of the electric field, but further research
is needed to investigate this. The long spin lifetimes and excellent individual qubit addressability
are encouraging for the operation of hole qubits positioned in large two-dimensional arrays.

8
 8. S PIN R ELAXATION B ENCHMARKS AND I NDIVIDUAL QUBIT A DDRESSABILITY FOR H OLES
132 IN Q UANTUM D OTS

8.6. M ETHODS
8.6.1. T UNNEL R ATE A NALYSIS
We measure the dot-reservoir tunnel coupling of the quantum dot under plunger gate P 3 . Before
each measurement of spin relaxation in Figure 8.3, we pulse from the (0,5) charge state to the mea-
surement point (1,5) charge state, and measure the sensor response. We observe an exponential
decay, the time constant of which determines our dot reservoir tunnel coupling. We pulse the
virtual energy gate U from the (0,5) to the (1,5) charge state. Figures 8.5a-b show the resulting
sensor response as a function of time in the (1,5) charge state for the spin relaxation measured in
figures 8.3b and 8.3c respectively. Due to the imperfect charge sensor compensation, we observe a
short initial transient in the first few microseconds, followed by the actual charge state transient to
which we fit an exponential decay, shown in the inset. We extract dot-reservoir load rates of 81.43
kHz to 27.45 kHz for the spin relaxation times measured in figures 8.3b and 8.3c respectively.

(a)
-0.1
(0,5) -0.396
| S 11 | (a.u.)

-0.2 -0.398

-0.4

-0.3
-0.402
20 40 60 80 100

-0.4 (1,5)

-20 0 20 40 60 80 100 120

tR ( s)
(b)
(0,5)
| S 11 | (a.u.)

-0.512

-0.4 -0.514

8 (1,5)
-0.516

100 120 140 160 180 200

-0.5
100 120 140 160 180 200 220
tR ( s)

Figure 8.5: Dot-Reservoir coupling between the singly occupied quantum dot defined under the plunger gate
P3 . (a) A dot-reservoir coupling of 81.43 kHz is extracted from the dot loading transient for the spin relaxation
measurement in figure 8.3b, and (b) 27.45 kHz for figure 8.33c.

8.6.2. R ELATIVE L EVER A RM

We extract the relative lever arms of the plunger gates P3 and P4 to the quantum dot potentials un-
derneath them. We take these values from the charge addition line slopes in the stability diagram
in figure 8.2g. Here, the figure is taken using a virtual gate matrix space of detuning and energy
(²,U) which are defined as a linear combination of the voltages on gates P3 and P4 (VP 3 ,VP 4 ):

²
µ ¶ µ ¶ µ ¶
1 −1.05 V
= ∗ P3
U 1.05 1 VP 4
8.6. M ETHODS 133

By calculating the gradient of the single and multi hole qubit charge addition lines in figure 8.2g,
we can solve for the changes in the plunger gate voltage space, and calculate the ratios for each
quantum dot, giving αP 3 /P 4 = 0.11 and αP 4 /P 3 = 0.07.

8
 134 R EFERENCES

R EFERENCES
[1] Lawrie, W. I. L. et al. Spin Relaxation Benchmarks and Individual Qubit Addressability for
Holes in Quantum Dots. Nano Lett. 20, 7237–7242 (2020).

[2] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Physical Review A -
Atomic, Molecular, and Optical Physics 57, 120–126 (1998).

[3] Petta, J. R. et al. Coherent Manipulation of Electron Spins in Semiconductor Quantum Dots.
Science 309, 2180–2184 (2005).

[4] Koppens, F. H. et al. Driven coherent oscillations of a single electron spin in a quantum dot.
Nature 442, 766–771 (2006).

[5] Itoh, K. M. & Watanabe, H. Isotope engineering of silicon and diamond for quantum com-
puting and sensing applications (2014).

[6] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nature Nanotechnology 9, 981–985 (2014).

[7] Yang, C. H. et al. Spin-valley lifetimes in a silicon quantum dot with tunable valley splitting.
Nature Communications 4, 1–8 (2013).

[8] Yang, C. H. et al. Silicon qubit fidelities approaching incoherent noise limits via pulse engi-
neering. Nature Electronics 2, 151–158 (2019).

[9] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nature Nanotechnology 13, 102–106 (2018).

[10] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[11] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins. Science 359, 439–442
(2018).

[12] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon. Nature 555,
8 633–637 (2018).

[13] Yang, C. H. et al. Silicon quantum processor unit cell operation above one Kelvin. arXiv
1902.09126 (2019).

[14] Petit, L. et al. Universal quantum logic in hot silicon qubits. Nature 580, 355–359 (2019).

[15] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors - hot, dense
and coherent. npj Quantum Information 3, 34 (2016).

[16] Veldhorst, M., Eenink, H. G., Yang, C. H. & Dzurak, A. S. Silicon CMOS architecture for a
spin-based quantum computer. Nature Communications 8 (2017).

[17] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science Advances 4, eaar3960
(2018).

[18] Itoh, K. et al. High purity isotopically enriched 70Ge and 74Ge single crystals: Isotope sepa-
ration, growth, and properties. Journal of Materials Research 8, 1341–1347 (1993).

[19] Sammak, A. et al. Shallow and Undoped Germanium Quantum Wells: A Playground for Spin
and Hybrid Quantum Technology. Advanced Functional Materials 29, 1807613 (2019).
R EFERENCES 135

[20] Hendrickx, N. W. et al. Gate-controlled quantum dots and superconductivity in planar ger-
manium. Nature Communications 9, 2835 (2018).

[21] Bulaev, D. V. & Loss, D. Spin relaxation and decoherence of holes in quantum dots. Physical
Review Letters 95, 076805 (2005).

[22] Bulaev, D. V. & Loss, D. Electric dipole spin resonance for heavy holes in quantum dots. Phys-
ical Review Letters 98, 097202 (2007).

[23] Maurand, R. et al. A CMOS silicon spin qubit. Nature Communications 7, 13575 (2016).

[24] Watzinger, H. et al. A germanium hole spin qubit. Nature Communications 9, 3902 (2018).

[25] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit
logic with holes in germanium. Nature 577, 487–491 (2020).

[26] Lodari, M. et al. Light effective hole mass in undoped Ge/SiGe quantum wells. Physical
Review B 100, 041304 (2019).

[27] Scappucci, G. et al. The germanium quantum information route. arXiv 2004.08133 (2020).

[28] Hendrickx, N. W. et al. A single-hole spin qubit. arXiv 1912.10426 (2019).

[29] Leon, R. C. et al. Coherent spin control of s-, p-, d- and f-electrons in a silicon quantum dot.
Nature Communications 11, 1–7 (2020).

[30] Lawrie, W. I. L. et al. Quantum Dot Arrays in Silicon and Germanium. Applied Physics Letters
116, 080501 (2020).

[31] Schoelkopf, R. J., Wahlgren, P., Kozhevnikov, A. A., Delsing, P. & Prober, D. E. The radio-
frequency single-electron transistor (RF-SET): A fast and ultrasensitive electrometer. Science
280, 1238–1242 (1998).

[32] Vukušić, L., Kukučka, J., Watzinger, H. & Katsaros, G. Fast Hole Tunneling Times in Germa-
nium Hut Wires Probed by Single-Shot Reflectometry. Nano Letters 17, 5706–5710 (2017).
8
[33] Tarucha, S., Austing, D. G., Honda, T., van der Hage, R. J. & Kouwenhoven, L. P. Shell filling
and spin effects in a few electron quantum dot. Physical Review Letters 77, 3613–3616 (1996).

[34] Liles, S. D. et al. Spin and orbital structure of the first six holes in a silicon metal-oxide-
semiconductor quantum dot. Nature Communications 9, 1–7 (2018).

[35] DiCarlo, L. et al. Differential charge sensing and charge delocalization in a tunable double
quantum dot. Physical Review Letters 92, 226801 (2004).

[36] Vukušić, L. et al. Single-Shot Readout of Hole Spins in Ge. Nano Letters 18, 7141–7145 (2018).

[37] Hu, Y., Kuemmeth, F., Lieber, C. M. & Marcus, C. M. Hole spin relaxation in Ge-Si core-shell
nanowire qubits. Nature Nanotechnology 7, 47–50 (2012).

[38] Bogan, A. et al. Single hole spin relaxation probed by fast single-shot latched charge sensing.
Communications Physics 2, 1–8 (2019).

[39] Bohuslavskyi, H. et al. Pauli blockade in a few-hole PMOS double quantum dot limited by
spin-orbit interaction. Applied Physics Letters 109, 193101 (2016).
 9
A F OUR -Q UBIT G ERMANIUM
Q UANTUM P ROCESSOR
The prospect of building quantum circuits [1, 2] using advanced semiconductor manufacturing po-
sitions quantum dots as an attractive platform for quantum information processing [3, 4]. Extensive
studies on various materials have led to demonstrations of two-qubit logic in gallium arsenide [5],
silicon [6–12], and germanium [13]. However, interconnecting larger numbers of qubits in semicon-
ductor devices has remained an outstanding challenge. Here, we demonstrate a four-qubit quantum
processor based on hole spins in germanium quantum dots. Furthermore, we define the quantum
dots in a two-by-two array and obtain controllable coupling along both directions. Qubit logic is im-
plemented all-electrically and the exchange interaction can be pulsed to freely program one-qubit,
two-qubit, three-qubit, and four-qubit operations, resulting in a compact and high-connectivity cir-
cuit. We execute a quantum logic circuit that generates a four-qubit Greenberger-Horne-Zeilinger
state and we obtain coherent evolution by incorporating dynamical decoupling. These results are
an important step towards quantum error correction and quantum simulation with quantum dots

Parts of this chapter have been published in Nature 591 (7851), 580-585

137
 138 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

9.1. I NTRODUCTION
Fault-tolerant quantum computers utilizing quantum error correction [1] to solve relevant prob-
lems [2] will rely on the integration of millions of qubits. Solid-state implementations of phys-
ical qubits have intrinsic advantages to accomplish this formidable challenge and remarkable
progress has been made using qubits based on superconducting circuits [14]. While the devel-
opment of quantum dot qubits has been at a more fundamental stage, their resemblance to the
transistors that constitute the building block of virtually all our electronic hardware promises ex-
cellent scalability to realize large-scale quantum circuits [3, 4]. Fundamental concepts for quan-
tum information, such as the coherent rotation of individual spins [15] and the coherent coupling
of spins residing in neighbouring quantum dots [16], were first implemented in gallium arsenide
heterostructures. The low disorder in the quantum well allowed the construction of larger arrays
of quantum dots and to realize two-qubit logic using two singlet-triplet qubits [5]. However, spin
qubits in group III-V semiconductors suffer from hyperfine interactions with nuclear spins that
severely limit their quantum coherence. Group IV materials naturally contain higher concentra-
tions of isotopes with a net-zero nuclear spin and can furthermore be isotopically enriched [17] to
contain only these isotopes. In silicon electron spin qubits, quantum coherence can therefore be
sustained for a long time [18, 19] and single qubit logic can be implemented with fidelities exceed-
ing 99.9 % [20, 21]. By exploiting the exchange interaction between two spin qubits in adjoining
quantum dots or closely separated donor spins, two-qubit logic could be demonstrated [6–12]. Sil-
icon, however, suffers from a large effective mass and valley degeneracy [22], which has hampered
progress beyond two-qubit demonstrations.
Holes in germanium are emerging as a promising alternative [23] that combine favourable
properties such as a host material with a natural abundance of zero nuclear spin isotopes that can
furthermore be enriched for long quantum coherence [24, 25], low effective mass and the absence
of low-energy valley states [26] for relaxed requirements on device design, low charge noise for
a quiet qubit environment [27], and low disorder for reproducible and well controlled quantum
dots [28, 29]. In addition, strained germanium quantum wells defined on silicon substrates are
compatible with semiconductor manufacturing [30]. Furthermore, hole states in general can ex-
hibit strong spin-orbit coupling that allows for all-electric operation [13, 31–33] and that removes
the need for microscopic components such as microwave striplines [6, 9, 12, 15] or nanomagnets
[7, 8, 34, 35], which is particularly beneficial for the fabrication and operation of two-dimensional
qubit arrays. The realization of strained germanium quantum wells in undoped heterostructures
[36] has led to remarkable progress. In two year’s time, germanium has progressed from the for-
mation of stable quantum dots and quantum dot arrays [28, 29, 37], to demonstrations of single
qubit logic [38], long spin lifetimes [39], and the realization of fast two-qubit logic in germanium
9 double quantum dots [13].

9.2. A T WO - BY-T WO Q UBIT Q UANTUM D OT A RRAY

Here, we advance semiconductor quantum dots beyond two-qubit logic and execute a four-qubit
quantum circuit using a two-dimensional array of quantum dots. We achieve this by defining the
four-qubit system on the spin states of holes in gate-defined germanium quantum dots. Fig. 9.1a
shows a scanning-electron-microscopy (SEM) image of the germanium quantum processor. The
quantum dots are defined in a strained germanium quantum well on a silicon substrate [27]
(Fig. 9.1b) using a double layer of electrostatic gates and contacted by aluminium ohmic contacts.
A negative potential on plunger gates P1-P4 accumulates a hole quantum dot underneath that
hosts qubits Q1-Q4, which can be coupled to neighbouring quantum dots through dedicated bar-
rier gates. In addition, two quantum dots are placed to the side of the two-by-two array, and the to-
tal system comprises six quantum dots. Via an external tank circuit, we configure these additional
9.3. S INGLE QUBIT C ONTROL 139

two quantum dots as radio frequency (rf) charge sensors for rapid charge detection. Using the
combined signal of both charge sensors [37], we measure the four quantum dot stability diagram
as shown in Fig. 9.1c. Making use of two virtual gate axes, we arrange the reservoir addition lines of
the four quantum dots to have different relative slopes of approximately −1, +1, −0.75, 0.75 mV/mV
for Q1, Q2, Q3, and Q4 respectively. Well defined charge regions (indicated as (Q1,Q2,Q3,Q4) in the
white boxes) are observed, with vertical anticrossings marking the different interdot transitions.
For the qubit readout we make use of Pauli-spin blockade to convert the spin states into a
charge signal that can be detected by the sensors. In germanium, however, the spin-orbit coupling
can significantly lower the spin lifetime during the readout process, in particular when the spin-
orbit field is perpendicular to the external magnetic field, reducing the readout fidelity [38, 40].
Here, we overcome this effect by making use of a latched readout process [41, 42]. During the
readout process, shown in Fig. 9.1d and e, a hole can tunnel spin-selectively to the reservoir as
a result of different tunnel rates of both quantum dots to the reservoir. After this process, the
system is locked in this charge state for the (long) reservoir tunnel time Tin (details in Methods
section). The high level of control in germanium allows tuning Tin to arbitrarily long time scales
by changing the potential applied to the corresponding reservoir barrier gate. We set Tin, Q2 =
200 µs and Tin, Q4 = 2.4 ms (Fig. 9.8), both significantly longer than the signal integration time
Tint = 10 µs. Furthermore, we project all qubit measurements on the Q1Q2 and Q3Q4 readout
pairs, such that the spin-orbit field is oriented along the direction of the external magnetic field
B 0 = 1.05 T to minimize spin relaxation.

9.3. S INGLE Q UBIT C ONTROL

Coherent rotations can be implemented by applying electric microwave signals to the plunger
gates that define the qubits, exploiting the spin-orbit coupling for fast driving [31, 33]. We ini-
tialize the system in the |↓↓↓↓〉 state by sequentially pulsing both the Q1Q2 and Q3Q4 double
quantum dot systems from their respective (0, 2)S states adiabatically into their (1, 1)T− states.
We then perform the qubit manipulations, after which we perform the spin readout as described
above. We observe qubit resonances at f Q1 = 2.304 GHz, f Q2 = 3.529 GHz, f Q3 = 3.520 GHz, and
f Q4 = 3.882 GHz, corresponding to effective g -factors of g Q1 = 0.16, g Q2 = 0.24, g Q3 = 0.24, and
g Q4 = 0.26. We note that these g -factors can be electrically modulated using nearby gates as a
means to ensure individual qubit addressability [39], as can also be observed in Fig. 9.3. Fig. 9.1f
shows the single-shot spin-up probability P ↑ for each of the four qubits after applying an on-
resonant microwave burst with increasing time duration t p , resulting in coherent Rabi oscillations.
To quantify the quality of the single qubit gates, we perform benchmarking of the Clifford
group [43] (Fig. 9.9a) and find single qubit gate fidelities exceeding 99 % for all qubits. The fidelity
of Q3 reaches to 99.9 %, thereby comparing to benchmarks for quantum dot qubits in isotopically 9
purified silicon [20, 21]. We find spin lifetimes between T1 = 1 − 16 ms (Fig. 9.9b), comparable to
values reported before for holes in planar germanium [39]. Furthermore, we observe T2∗ to be be-
tween 150-400 ns for the different qubits (Fig. 9.10a), but are able to extend phase coherence up to
T2CPMG = 100 µs by performing Carr-Purcell-Meiboom-Gill (CPMG) refocusing pulses (Fig. 9.10c),
more than two orders of magnitude larger than previously reported for hole quantum dot qubits
[13, 32, 33]. This indicates the qubit phase coherence is mostly limited by low-frequency noise,
which is confirmed by the predominantly 1/ f α noise spectrum we observe by Ramsey and dy-
namical decoupling noise spectroscopy (Fig. 9.11).

9.4. O NE , T WO AND T HREEFOLD C ONDITIONAL R OTATIONS

Universal quantum logic can be accomplished by combining the single qubit rotations with a two-
qubit entangling gate. We implement a conditional rotation (CROT) gate [7–9, 12], where the
 140 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

a c d
(0,2)T
P2 (1,1)AP
S1
20 (1,1)P
(0,0,0,0) (0,1)
P1 P3
(0,2)S
10 slow Tin
S2 (0,0,0,0) fast

U 1234 (mV)
P4 0 0.4 0.8
(1,0,1,0) e visibility
0 (0,1,0,1)
(0,1,1,1)
5
(1,1,1,1)

U 12 (mV)
b -10

(0,2,1,1)
0

(1,1,1,1)
SiGe (2,1,2,1) -5
-20 (1,2,1,2)
Ge
SiGe -10 (1,2,1,1)
-10 0 10 -10 -5 0 5
12,34
(mV) 12
(mV)

f
(0,2)T
(1,1)AP 0.8
0.8
(1,1)P 0.6
0.6
(0,2)S 0.4 0.4
P

P
slow Tin
0.2 0.2
0 0.4 0.8 Q1 Q2
visibility 0 0
(0,1,1,1) 0.8 Q4 Q3
0.8
0.6
0.6
(0,2,1,1)

0.4
P

0.4
0.2
0.2
(1,2,1,1) 0
0 5 0 200 400 600 800 0 200 400 600 800 1000
12
(mV) tp (ns) tp (ns)

Figure 9.1: Four germanium hole spin qubits. a, Scanning electron microscope image of the four quantum dot
device. We define qubits underneath the four plunger gates indicated by P1-P4. The qubits can be measured
using the two charge sensors S1 and S2. The scale bar corresponds to 100 nm. b, Schematic drawing of the
9 Ge/SiGe heterostructure. Starting from a silicon wafer, a germanium quantum well is grown in between two
Si0.2 Ge0.8 layers at a depth of 55 nm from the semiconductor/dielectric interface. c, Four quantum dot charge
stability diagram as a function of two virtual gates. At the vertical and diagonal bright lines a hole can tunnel
between two quantum dots or a quantum dot and its reservoir respectively. As a result of the virtual axes
²12,34 and U1234 , the addition lines of the different quantum dots have different slopes, allowing for an easy
distinction of the different charge occupations indicated in the white boxes as (Q1, Q2, Q3, Q4). d, Energy
diagram illustrating the latched Pauli spin blockade readout. When pulsing from the (1,1) charge state to the
(0,2) charge state, only the polarized triplet states allow the holes to move into the same quantum dot, leaving
an (0,2) charge state (green). Interdot tunnelling is blocked for the two antiparallel spin states and as a result
the hole on the first quantum dot will subsequently tunnel to the reservoir leaving an (0,1) charge state (red),
locking the different spin states into different charge states. e, Readout visibility as defined by the difference
in readout between either applying no rotation and a π-rotation to Q2. The readout point is moved around
the (1,1)-(0,2) anticrossing of the Q1Q2 system and a clear readout window can be observed bounded by the
different (extended) reservoir transition lines indicated by the dotted lines. f, The qubits can be rotated by
applying a microwave tone resonant with the Zeeman splitting of the qubit. Coherent Rabi rotations can be
observed as a function of the microwave pulse length t p for all qubits Q1-Q4.
9.4. O NE , T WO AND T HREEFOLD C ONDITIONAL R OTATIONS 141

B0
Pblocked
0.9
(rad)

3 0.5
2 0.1
control

flow

0
(rad)

3
2
control

fhigh

0
(rad)

3
flow 2
control

0
(rad)

3
fhigh 2
control

0
θ 0 2 3 0 2 3 0 2 3 0 2 3 0 2 3 0 2 3 0 2 3 0 2 3
ϕ Q1
(rad) Q1
(rad) Q2
(rad) Q2
(rad) Q3
(rad) Q3
(rad) Q4
(rad) Q4
(rad)

Figure 9.2: Controlled rotations between all nearest-neighbour qubit pairs. By selectively enabling the ex-
change interaction between each pair of qubits, we can implement two-qubit controlled rotations (CROTs).
The pulse sequence consists of a single preparation gate with length θ on the control qubit (labelled green),
followed by a controlled rotation on one of the resonance lines of the target qubit (labelled in red). Both qubit
pairs Q1Q2 and Q3Q4 are read out in single-shot mode and the position of the eye on top of each column in-
dicates the respective readout pair. Each of the four main columns corresponds to conditional rotations on a
different qubit as indicated by the red dot. Rows one and two show the results for the horizontal interaction
(dark green), while rows three and four show the two-qubit interaction for the vertical direction (light green)
with respect to the external magnetic field, as indicated in the top left. Rows one and three correspond to
9
driving the lower frequency f low conditional resonance line, while rows two and four show driving of the other
resonance line f high .
 142 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

resonance frequency of the target qubit depends on the state of the control qubit, mediated by
the exchange interaction J between the two quantum dots. The exchange interaction between
the quantum dots is controlled using a virtual barrier gate (details in Methods), coupling the two
quantum dots while keeping the detuning and on-site energy of the quantum dots constant and
close to the charge-symmetry point. We demonstrate CROT gates between all four pairs of quan-
tum dots in Fig. 9.2, proving that spin qubits can be coupled in two dimensions. Because the
target qubit resonance frequency depends on the control qubit state, the conditional rotation is
characterized by the fading in and out of the target qubit rotations as a function of the control
qubit pulse length. For driving the two separate transitions, the pattern is shifted by a π rotation
on the control qubit. When the control qubit is in a different readout pair as the target qubit (rows
3 and 4), we can independently observe the single qubit control, and two-qubit target qubit rota-
tions in the two readout systems. By setting the pulse length equal to φQ = π, a fast CX gate can be
obtained within approximately t p = 100 ns between all of the four qubit pairs.
The low effective mass and high uniformity in the material allow full control over the interdot
coupling by dedicated tunnel barrier gates. To demonstrate this, we measure the qubit resonance
frequency as a function of the eight possible permutations of the different basis states of the other
three qubits, as illustrated in Fig. 9.3a,b. Without any exchange present, the resonance frequency
of the target qubit should be independent on the preparation of the other three qubits, as schemat-
ically depicted in Fig. 9.3c. When the exchange interaction with one of the neighbouring quantum
dots is enabled, the resonance line splits in two (Fig. 9.3de), allowing for the operation of the CROT
gate. When both barriers to the nearest-neighbours are pulsed open at the same time, we observe
the expected fourfold splitting of the resonance line (Fig. 9.3f-i). This allows performing a res-
onant i -Toffoli three-qubit gate (Fig. 9.3k and Fig. 9.12), which has theoretically been proposed
as an efficient manner to create the Toffoli, Deutsch, and Fredkin gates [44]. We observe a differ-
ence in the efficiency at which the different conditional rotations can be driven, as can also be
seen from the width of the resonance peaks in Fig. 9.3f-i. This is expected to happen when the
exchange energy is comparable to the difference in Zeeman splitting and is caused by the mixing
of the basis states due to the exchange interaction between the holes [45] (details in Methods).
Finally, we open three of the four virtual barriers and observe the resonance line splitting in eight,
corresponding to all eight permutations of the control-qubit preparation states (Fig. 9.3j). This
enables us to execute a resonant four-qubit gate and in Fig. 9.3l we show the coherent operation
of a three-fold conditional rotation (see Fig. 9.12 for the coherent operation of the other resonance
lines). The good control over the interdot coupling thus enables a demonstration of the localized
nature of the exchange interaction [3], coupling the different spins by electric gate pulses.

9 9.5. C ONDITIONAL P HASE G ATES B ETWEEN N EIGHBOURING

Q UBIT PAIRS
While the demonstration of these conditional rotations can be beneficial for the simulation of
larger coupled spin systems, the ability to dynamically control the exchange interaction allows
for faster two-qubit operations. We efficiently implement controlled phase (CPHASE) gates [6, 8,
12] between the different qubit pairs by adiabatically pulsing the exchange interaction with the
respective virtual barrier gate. We control the length and size of the voltage pulse (Fig. 9.13) to
acquire a CZ gate, in which the antiparallel spin states accumulate a phase of exactly θ = π with
respect to the parallel spin states. We demonstrate this in Fig. 9.4a,b, where we employ a Ramsey
sequence to measure the conditional phase. After the exchange pulse UCZ , we apply a software Z
gate to both the target and control qubits to correct the individual single qubit phases. The virtual
barrier pulses enable fast CZ gates between all neighbouring qubit pairs, all executed well within
10 ns (details in Table 9.2).
9.5. C ONDITIONAL P HASE G ATES B ETWEEN N EIGHBOURING QUBIT PAIRS 143

a
π,fq π,fq π,fq π,fq
R R
R R
R R

b 1Q (c) 2Q (d,e) 3Q (f-i) 4Q (j) g j

↑↑↑
frequency

↓↑↑
↑↓↑ 0.6 0.6
↓↓↑
↑↑↓
P up

P up
↓↑↓ 0.2
↑↓↓ 0.2
↓↓↓
2.6 2.65 2.7 3.66 3.7 3.74
c e h k fq (GHz)
0.8
0.8
0.6 0.6 0.6
0.5
0.4 0.4
P up
P up

P up

P up
0.2 0.2 0.2 0.2
3.45 3.5 3.36 3.4 3.44 3.48 3.35 3.4 3.45 0 200 400
d f i l
0.8 0.6
0.6 0.6
0.6
0.4 0.4
P up

P up
P up

P up

0.2 0.2 0.2 0.2

3.7 3.75 3.8 3.65 3.7 3.75 4.1 4.15 4.2 0 100 200
fq (GHz) fq (GHz) fq (GHz) tp (ns)

Figure 9.3: Resonant one, two, three, and four-qubit gates. a, Circuit diagram of the experiment performed
in panels c-l. All eight permutations of the three control qubit eigenstates are prepared, with R being either no
pulse or a π-pulse on the respective qubit. Next, the resonance frequency of the target qubit is probed using
a π-rotation with varying frequency f q . Finally, the prepared qubits are projected back and the target qubit
state is measured. By changing the different interdot couplings J , we can switch between resonant single, two,
three, and four-qubit gates as indicated in the dashed boxes. b, Turning on the exchange interaction between
the different qubit pairs splits the resonance frequency in two, four, and eight for 1, 2 and 3 enabled pairs
respectively. The colours of the line segments correspond to the colours in panels c-l. c, By turning all exchange
interactions off, the qubit resonance frequency of Q2 is independent of the prepared state of the other three 9
qubits, resulting in an effective single-qubit rotation. d-e, By turning on a single exchange interaction J 12 (d) or
J 23 (e), the resonance line splits in two. The additional offset of the resonance frequencies is caused by electric
modulation of the hole g -factor. f-i, Turning on both exchange interactions to the neighbouring quantum dots
results in the resonance line splitting in four, for Q2 (f), Q1 (g), Q3 (h), Q4 (i) respectively. j, Turning on the
exchange interactions between three pairs of quantum dots J 12 , J 23 , J 41 splits the resonance line in eight. k-l,
Resonant driving of the three-qubit gate (k) and the four-qubit gate (l) with Q2 being the target qubit, shows
Rabi driving as a function of pulse length t p , demonstrating the coherent evolution of the operation.
 144 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

a c
Qtarget X Z(θ) X Q3 X Y2 X
UCZ UCZ
J on (t) (t)
Qcontrol X2 X2 Q2 Y2

Z(φ1) tCZ < 10 ns Q3 X X Y2 X X

Vbarrier idle idle
= UCZ UCZ = J off (t) (t)
Z(φ2)
time Q2 X Y2 X

b d
0.8 1 0.6
0.8 τon = 130 ns
0.6 0.6
P up

P up

Q1Q2 Q2Q3 0.5 0.4

0.4 0.4
0.2 τoff = 200 ns
0.2 0.2
0 2 4 6 0 2 4 6 0 0.5 0 0.5
1 1
0.8 0.8 τon, echo = 220 ns τoff, echo = 2100 ns
0.6 0.6 0.8
P up

P up

0.4 Q3Q4 Q4Q1 0.5

0.4
0.2 0.6
0.2
0 2 4 6 0 2 4 6 0 0.5 0 5 10
(rad) (rad) 2t ( s) 2t ( s)

Figure 9.4: Controlled phase gate and dynamical decoupling. a, Circuit diagram of the experiment performed
in panel b. The controlled phase gate is probed by performing a Ramsey sequence on the target qubit for both
basis states of the control qubit. The phase of the second π/2 (X) gate is swept by performing an update of
the microwave phase through quadrature modulation. Additionally, a phase update is performed on both the
target and control qubit to compensate for any single qubit phases picked up as a result of the gate pulsing to
achieve a controlled-Z (CZ) gate. b, The spin-up probability of the target qubit (in bold) as a function of the
phase θ of the second X gate for the control qubit initialized in the |↓〉 (blue) and |↑〉 (red) state. Measurements
9 for the inverted target and control qubits in Fig. 9.13c. By applying an exchange pulse and single qubit phase
updates, we achieve a CZ gate at θ = 0 rad. c, Circuit diagrams of the experiment performed in panel d. The
phase coherence throughout the two-qubit experiment is probed using a Ramsey sequence, both for the case
with J on (top) and off (bottom) and both with (orange) and without (blue) applying an echo pulse. d, Spin-up
probability as a function of the experiment length, for the situation with exchange on (left, triangles) and off
(right, circles). From the decay data we extract characteristic decay times τ of τon = 130 ns, τon, echo = 220 ns,
τoff = 200 ns, and τoff, echo = 2100 ns (details in Methods).
9.6. D EMONSTRATION OF A F OUR QUBIT G REENBERGER-H ORNE -Z EILINGER S TATE 145

To prepare our system for quantum algorithms, we implement decoupling pulses into the
multi-qubit sequences to extend phase coherence [8], as demonstrated in Fig. 9.4c,d. We per-
form a CPHASE gate of length t between qubits Q2 and Q3 (Fig. 9.4d, left, triangles) and compare
the decay of the resulting exchange oscillations for the situations with (orange) and without (blue)
a Y2 echo pulse. We observe an increased decay time of τ = 220 ns for the decoupled CPHASE gate,
as compared to τ = 130 ns for a standard CPHASE gate. Next, we entangle Q2 and Q3 by forming
the |Ψ+ 〉 Bell state and let the system evolve for time 2t (Fig. 9.4d, right, circles). We then disen-
tangle the system and measure the spin-up probability of Q3 as a function of the evolution time.
Without the decoupling pulse, we observe a loss of the two-qubit coherence after a characteristic
time τ = 200 ns. However, by applying an additional Y2 pulse to both Q2 and Q3, we can extend
this time scale beyond 2 µs, sufficient to perform a series of single and multi-qubit gates, owing to
our short operation times.

9.6. D EMONSTRATION OF A F OUR Q UBIT G REENBERGER -H ORNE -

Z EILINGER S TATE
a
Q3 X(tprep) Y2 X

Q2 X X Y2 X X

Q1 X X Y2 X X

Q4 X X Y2 X X

I II III IV V VI VII VIII IX

b
1
0.8
0.6
Pnorm

0.4
0.2
0
Q3Q4 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
tprep (ns) tprep (ns) tprep (ns) tprep (ns) tprep (ns) tprep (ns) tprep (ns) tprep (ns) tprep (ns)
Q1Q2

Figure 9.5: Coherent generation of a four-qubit Greenberger-Horne-Zeilinger (GHZ) state. a-b, A four-qubit
GHZ state is created by applying three sequential two-qubit gates, each consisting of an X-CZ-X gate circuit.
Next, a Y2 decoupling pulse is applied, after which we disentangle the GHZ state again (circuit diagram in a).
Pulses pictured in the same column are applied simultaneously. The initial state of Q3 is varied by applying a
preparation rotation of length t . For different stages throughout the algorithm (dashed lines), we measure the
non-blocked state probability as a function of t for both the Q1Q2 and Q3Q4 readout system, normalized to
their respective readout visibility. At the end of the algorithm the qubit states correspond to the initial single
9
qubit rotation, and the clear oscillations confirm the coherent evolution of the algorithm from isolated qubit
states to a four-qubit GHZ state. (b).

We show this by coherently generating and disentangling a four-qubit Greenberger-Horne-

Zeilinger (GHZ) state (Fig. 9.5). Making use of the fast two-qubit CZ gates, as well as a decoupling
pulse on all qubits, we can maintain phase coherence throughout the experiment. We perform
parity readout on both the Q1Q2 (red) and Q3Q4 (green) qubit system at different stages of the
algorithm (I-IX). Both qubit systems are sequentially readout after each experiment and the ob-
served blocked state fraction is normalized to the readout visibility. We prepare a varying initial
state by applying a microwave pulse of length t to Q3, as visible at point I. After applying CZ gates
between all four qubits, the system resides in an entangled GHZ type state at point IV/V, for a π/2
preparation pulse on Q3. The effective spin state oscillates between the antiparallel |1010〉 and
|0101〉 states as a function of t prep , resulting in a high readout signal for all t . The small oscillation
 146 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

that can still be observed for the Q1Q2 system, is caused by a small difference in readout visibility
for the two distinct antiparallel spin states. Next, we deploy a Y2 decoupling pulse to echo out all
single qubit phase fluctuations during the experiment (Fig. 9.14). After disentangling the system
again, we project the Q3 qubit state by applying a final X(π/2) gate, and indeed recover the initial
Rabi rotation as a demonstration of the coherent evolution of a multi-spin entangled state (see Fig.
9.15).
The demonstration of a two-by-two array of four qubits shows that quantum dot qubits can
be coupled in two-dimensions and multi-qubit logic can be executed. The hole states used are
subject to strong spin-orbit coupling, enabling all-electrical driving of the spin state, beneficial
for scaling up to even larger systems. In future experiments the performance of the qubit gates
can be further optimized, by making use of tailored pulses and quantifying their performance
using benchmarking sequences. The ability to freely couple one, two, three and four spins using
electric gate pulses has great prospects both for performing high-fidelity quantum gates as well
as studying exotic spin systems using analog quantum simulations. Furthermore, we envision
that the low-disorder in planar germanium and the potential to leverage advanced semiconductor
manufacturing will be beneficial for the realization of scalable qubit tiles [46–48] for fault-tolerant
quantum processors.

9.7. M ETHODS
9.7.1. D EVICE FABRICATION
The device was fabricated on a Ge/SiGe heterostructure with a 55-nm-deep buried quantum well,
grown in an industrial reactor by reduced vapour deposition, as detailed in [27, 36]. Ge quantum
wells are fully compatible with a 300 mm semiconductor foundry line [23, 30]. Starting from a Si
wafer, the heterostructure comprises a 1.6 µm relaxed Ge layer; a 1 µm step graded Si1−x Gex layer
with a final Ge composition of x = 0.8; a 500 nm relaxed Si0.2 Ge0.8 buffer layer; the 16-nm-thick
compressively strained Ge quantum well; a 55 nm Si0.2 Ge0.8 spacer layer and finally a sacrificial
Si cap layer (< 2 nm). We define ohmic contacts by electron beam lithography and subsequent
etching of the oxidized Si cap layer and deposition of a 30 nm Al contact layer [28]. Electrostatic
gates are defined in two layers (20 nm and 40 nm Ti/Pd respectively), separated from both the
substrate and each other by 7 nm of ALD-grown Al2 O3 .

9.7.2. D EVICE SCREENING

After the devices are fabricated they undergo visual inspection as a first screening step. Devices
that pass this step are wire bonded onto a printed circuit board (PCB) and cooled down to T = 4.2 K
9 in a liquid helium cryostat. Channel turn-on and gate leakage is tested for channels through the
north sensor, through the south sensor, and from the north to the south sensor (through the device
area). In a fully automated transport measurement, all gates relevant to the respective channel are
swept down to Vgate = −1.5 V, after which they are swept back up and down to Vgate = 0 V. These
measurements are checked to confirm if (i) all channels turn on and (ii) all gates modulate the
current as expected (exemplary data in Fig. 9.6b). Devices that pass all of these screenings are
selected for cool down in a dilution refrigerator (see for device yield and test results Fig. 9.6a).

9.7.3. E XPERIMENTAL SETUP

Measurements are performed in a Bluefors dilution refrigerator with a base temperature of Tbase =
20 mK. We use battery powered voltage sources to bias the electrostatic gates. Additionally, coaxial
lines are connected to all plunger and barrier gates through an on-PCB bias-tee, which can be
pulsed using a Keysight M3202A arbitrary waveform generator. Plunger gates P2, P3, and P4 are
furthermore connected to a vector source (Rohde&Schwarz SGS100A for P3/P4 and Keysight PSG
9.7. M ETHODS 147

a Visual Dilution b
inspection 4K testing refrigerator
10 10 4K testing
Passed 2 SHT north full device
1 full channel
8 turn on

I (nA)
5 turn off
Gate leakage
Batch SQ20_111

5
6 indiv. gates
Passed 15 0 turn on
16
turn off
8 SHT south 4
Untested 6

I (nA)
8 4 2
2
1 0 0
Failed -1000 0 -1000 0
V gate (mV) V gate (mV)

Figure 9.6: Screening of qubit devices. a, The qubit devices undergo a visual screening as well as a transport
screening at a temperature of T = 4.2 K. Out of the full batch of 16 nominally identical devices, 15 passed visual
inspection. Seven of these devices were tested at T = 4.2 K and two devices were found to pass all testing, of
which one (the device presented in this work) was mounted in a dilution refrigerator. b, 4.2 Kelvin transport
data of the device presented in this work. Three different channels are turned on by sweeping all gates down
to Vgate = −1500 mV (black), thereby accumulating charge in the undoped strained Ge quantum well. Then
the effect of the individual gates is tested by sweeping them up and down to Vgate = 0 V (coloured lines). All
channels turn on and the gates affect the transport current as expected from the device layout.

E8257D for P2) through room-temperature diplexers with a stop band of f = 400 − 1500 MHz. We
modulate the qubit driving pulses using the quadrature modulation inputs and use multiple vector
sources to be able to drive all qubit resonance lines to overcome the limited output bandwidth of
400 MHz of the AWGs. Qubits Q2 and Q4 are driven using the vector source connected to P3, Q1 is
driven from gate P4, and Q3 is driven from gate P2.

The charge sensors are connected to a resonant tank circuit consisting of a in-house made
niobium-titanium-nitride (NbTiN) kinetic inductor with an expected inductance of L = 2 µH. We
apply a resonant rf-tone at f = 147.3 MHz and f = 139.9 MHz for sensor S1 and S2 respectively.
The reflected signal is split using a directional coupler mounted to the mK-plate of the fridge and
amplified by a cryogenic amplifier at the 4 K stage. Next, the signal is demodulated using an in-
house build demodulation setup and measured using a Keysight M3102A digitizer card. This is
9
further detailed in Fig. 9.7 below. Both sensors can differentiate the different charge states in both
the Q1Q2 and Q3Q4 qubit system. In this work, we only show the data of the sensor closest to the
respective qubit pair.

The data in Fig. 9.5 are normalized with respect to the readout visibility as obtained from a
Rabi measurement. We find P Q1Q2, not blocked = 0.15, P Q1Q2, blocked = 0.78, P Q3Q4, not blocked =
0.10, and P Q3Q4, blocked = 0.93.

9.7.4. V IRTUAL GATE MATRICES

In order to map out the transition lines of all four quantum dots in a single measurement, we
define the following virtual gates [49] as linear combination of the physical gates P1 -P4 , as well as
 148 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

computer

Keysight PXIe chassis (pulsing) SPI rack (reflectometry)

AWG AWG AWG AWG AWG AWG digi split RF RF RF RF IQ IQ
M3202A M3202A M3202A M3202A M3202A M3202A M3102A 50 Ω src src cmb amp mixer mixer
FPGA
in ref ref in in ref ref
V(t)
SLP-30
out trig trig in out in in

out out out out out out out

12x coax dc-block 27.5 MHz

6 dB 6 dB 6 dB ferrite choke
VLFX-470 VLFX-470
SPI rack
45 dB
I Q PM I Q PM I Q PM (dc, isolated)
PSG E8267D SGS100A SGS100A ferrite choke ferrite choke
out out out DAC (x26)

26x dc-loom
signal types
12 dB 15 dB 20 dB 26 dB 22 dB 14 dB dc
control signals
RF LF RF LF RF LF
Diplexers

data signals
rf (pulsing)
rf (readout)
out out out rf (microwave)
P2 P3 P4
RT

50 K 4x superconducting coax 8x graphite coax

P1-P4 6 dB 20 dB

4K 6 dB 6 dB 20 dB +35 dB

20 mK 3 dB 50 Hz

PCB
Bias-tees (x12)

Bias-tees (x2)

RF LF LF RF
typical pulsing
100p

100p
100n

100n
100k

5k1
1M

out out
gates
I (IF)
~ 100 MHz
Q (IF)
50 ns

NbTiN inductor (x2) PM

9 MW (RF)
~ 3 GHz

Figure 9.7: Schematic of the measurement set-up. Signals to gates are colour coded by signal type. Typical
pulse lengths and shapes are shown in the bottom right inset.

the sensor plunger gates PS1 and PS2 . We write:

   
P1 1 1
 P   −1 1 
 2  µ
1  ∆²12,34
¶
 P3   0.75
   
=
 P4  −0.75 1  ∆U1234
 
   
PS1   −0.1 −0.4 
PS2 −0.05 −0.51
9.7. M ETHODS 149

with ∆²12,34 and ∆U1234 the virtual gates used in Fig. 9.1c.
In addition, we define a virtual gate system to allow independent control of the different inter-
dot couplings and quantum dot detuning and on-site energy and write:
   
P1 1.26 0.74 0.31 −0.17 −0.55 0 0 −0.49
 ²12
 P  −1.39 0.61 −0.36 −0.36 −1.03 0 −0.60 0 
 
 2 
 P3   0.28 −0.28 1.39 0.61 0 −0.47 −0.60 0   U12 
   

 P4  −0.30 −0.30 −1.39 0.61 0 0 ²34 
   
−0.91 −0.92 
    
B12   0 0 0 0 1.00 0 0 0 

U34 

 = 
B   0 0 0 0 0 1.00 0 0  vB12 
 
 34  

B   0  
 23   0 0 0 0 0 1.00 0 
 vB34 
 
B41   0 0 0 0 0 0 0 1.00  vB23 
  
   
PS1  −0.09 −0.15 0.01 −0.03 0 0 0 0  vB41
PS2 0 0 −0.09 −0.15 0 0 0 0

with ²mn the detuning voltage and Umn the voltage controlling the on-site energy of quantum
dots m and n, vBmn the virtual barrier gate controlling the coupling between quantum dots m
and n, and Pn , Bmn and PS1−2 the various physical gates.

9.7.5. L ATCHED READOUT MECHANISM

a Q1Q2 b Q3Q4 c d e f
14 60 1 ↓↓
Tin = 0.19 ms Tin = 2.4 ms Q1Q2 Q1Q2 Q3Q4 Q3Q4
↓↑
12 50
0.8 ↑↓
sensor signal (a.u.)

sensor signal (a.u.)

10 ↑↑
40
8 0.6
P blocked

30
6 0.4
20
4
0.2
2 10

0 0 0
0.5 1 1.5 0 2 4 0 100 200 0 500 1000 0 100 200 0 500 1000
t (ms) t (ms) tramp (ns) tramp (ns) tramp (ns) tramp (ns)

Figure 9.8: Readout characteristics. a,b, We measure the difference in charge sensor signal between the
blocked and non-blocked states as a function of the measurement time at the readout point. An exponen-
tial decay can be observed related to the tunnel time Tin of Q2 (Q4) to the reservoir for the Q1Q2 (a) and Q3Q4
(b) readout system respectively. c-f, We vary the ramp time between the manipulation phase and the readout
phase and measure the blocked state probability of the four different two qubit basis states by applying prepa-
ration π pulses to the relevant qubits, both for the Q1Q2 readout system (a,b) and the Q3Q4 readout system
9
(c,d). By increasing the interdot coupling during the readout and elongating the ramp between the manipu-
lation and readout point, we can switch between a parity readout (a,c) and a single state readout (b,d). The
dashed line corresponds to the optimized readout ramp time used for the measurements in this manuscript.

To reduce readout infidelity as a result of spin relaxation, we make use of charge latching
through the reservoir [41, 42]. We achieve this effect by pulsing into the area in the (0,2) charge
region bounded by the extended (1,1)-(0,1) (fast) and the extended (1,1)-(1,2) (slow) transitions
(dotted lines in Fig. 9.1e). When the interdot tunnelling into the (0,2) charge state is blocked, the
hole in the first quantum dot will quickly tunnel into the reservoir. This locks the spin state in the
metastable (0,1) charge state, with the decay to the (0,2) ground state governed by the slow tunnel
rate Tin between the second quantum dot and the reservoir.
We operate in a parity readout mode where we observe both antiparallel spin states to be
blocked (Fig. 9.8c,d), opposite to conventional parity Pauli spin blockade readout [50]. This may
 150 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

be explained by the strong spin-orbit coupling mixing the parallel (1,1) states with the (0,2) state
and causing strong relaxation of the upper parallel spin state. We note that both singlet-triplet
readout for single state discrimination and parity readout are compatible with the execution of
quantum algorithms [50]. However, by both increasing the interdot coupling and elongating the
ramp between the manipulation and readout point, we can transition into a state selective read-
out where only the |↓↑〉 state results in spin blockade (Fig. 9.8e,f), with a slightly reduced readout
visibility. Optimal parity readout is obtained for a ramp time of t ramp ≈ 20 ns, while single state
readout is optimal at t ramp ≈ 800 ns.
Each charge sensor can detect transitions in both qubit pairs, but is most sensitive to their
respective nearby quantum dots. We maximize the readout visibility as defined by the difference
between the readout of a spin-up and spin-down state by scanning the readout level around the
relevant anticrossing. This is illustrated for the Q1Q2 pair in Fig. 9.1e, where a clear readout win-
dow with maximum visibility can be observed bounded between the (extended) reservoir transi-
tions of the two quantum dots.

a b
FQ1 = 99.85(2) % FQ2 = 99.69(7) % 1 T1Q1 = 0.84(6) ms 1 T1Q2 = 7.6(5) ms
0.8 0.8 P up, normalized
P up

0.5 0.5
0.7
0.6 0 0
0.6
0 500 0 100 200 300 0 5 10 0 50
FQ3 = 99.88(3) % 0.8 FQ4 = 99.4(2) % 1 T1Q3 = 16.1(8) ms 1 T1Q4 = 11.5(5) ms
P up, normalized

0.8
P up

0.7 0.5 0.5

0.6
0.6
0 0
0 500 0 100 200 0 50 100 0 50 100
NCliffords NCliffords t wait (ms) t wait (ms)

Figure 9.9: Randomized benchmarking of the Clifford group and spin relaxation times of the different
qubits. a, We quantify the quality of the single qubit gates by performing randomized benchmarking of the
single-qubit Clifford group [43]. The decay curve of the qubit state is measured as a function of the number of
Clifford gates applied. Each data point consists of 1000 single shots
¡ for 30 different
¢ randomly selected Clifford
sequences of length NCliffords . The decay is fitted to P up = a exp −(NCliffords /m) + y 0 , with a the initial spin-
m
up probability, m the decay parameter, and y 0 an offset. F = 1− 2·1.875 is extracted based on the average single
qubit gate length of 1/1.875 Clifford gates. Error margins correspond to 1σ. b, The spin relaxation time T1 is
9 measured at the manipulation point by applying a π X -pulse separated by a waiting time t wait from¡the readout ¢
phase, as illustrated in the schematic on top. By fitting the normalized spin-up fraction to P = exp −t wait /T1 ,
Q1 Q2 Q3 Q4
we find spin relaxation times of T1 = 0.84(6) ms, T1 = 7.6(5) ms, T1 = 16.1(8) ms, and T1 = 11.5(5) ms.
Error margins correspond to 1σ.

9.7.6. C ONDITIONAL ROTATION GATES

We perform a series of qubit pulses, consisting of a single qubit control pulse (green) and a target
qubit two-qubit pulse (red), as indicated in the diagram in Fig. 9.2. The length of both the control
pulse θcontrol as well as the target qubit pulse φQ1-Q4 are varied, with t p (φ = π) = 50 − 110 ns (de-
tails in Table 9.1). The conditional rotations are performed on all four target qubits (four double
columns) for both the horizontally interacting qubits (rows 1 and 2), as well as the vertically inter-
acting qubits (rows 3 and 4), by driving the |↓↓〉-|↑↓〉 transitions with f low (rows 1 and 3), as well
as the inverse |↓↑〉-|↑↑〉 transitions with f high (rows 2 and 4), with |Qtarget Qcontrol 〉. We then per-
9.7. M ETHODS 151

a τ
b τ τ
c N
τ/2 τ τ/2
π/2 π/2 π/2 π π/2 πX/2 πY πY πX/2

1 1 1 10 2

Normalized echo amplitude

P blocked

P blocked
0.75 0.75 0.75 1
0.75

( s)
0.5 0.5
0.5
0 200 400 0 200 400 0 5 10 0 5 10 10 1 0.5

T CPMG
Nπ
0.75 0.75

2
Q1 1
P blocked

P blocked
0.75 0.75
0.5 0.5 Q2 10
Q3 0 150
0.25 0.5 0.5 Q4 1500
0.25
10 0
0 500 0 200 400 0 5 10 0 5 10 10 0 10 2 10 0
(ns) (ns) 2 ( s) 2 ( s) N Total evolution time ( s)

Figure 9.10: Ramsey, Hahn echo and Carr-Purcell-Meiboom-Gill (CPMG) measurements on the different
qubits. a, The phase coherence time T2∗ is measured using a Ramsey sequence consisting of two X(π/2)-
pulses
¡ separated ¢ by ¡a waiting ¢time τ as illustrated in the schematic on top. By fitting the data to P =
cos 2π∆ f τ + φ0 exp −(τ/T2∗ )α , with ∆ f the frequency detuning, φ0 a phase offset and α the power of the
∗
decay, we find spin dephasing times of T2,Q1 ∗
= 201 ns, T2,Q2 ∗
= 146 ns, T2,Q3 ∗
= 445 ns, and T2,Q4 = 150 ns
for Q1-Q4 respectively. b, Using an additional X(π)-pulse, low-frequency fluctuations of the qubit reso-
nance frequency can be echoed out, allowing to probe the Hahn-echo decay time T2Hahn . Fitting the data
to P = exp −(τ/T2Hahn )α , we find Hahn echo times of T2,Q1
³ ´
Hahn = 4.3µs, T Hahn = 5.5µs, T Hahn = 3.8µs, and
2,Q2 2,Q3
Hahn = 2.9µs. c, Using a CPMG sequence of repeated Y(π) pulses, we can increase the echo bandwidth and
T2,Q4
CPMG > 100 µs. The phase coherence can be observed to increase with
extend the phase coherence to over T2,Q1
the amount of refocusing pulses (left), with exemplary decay traces for Q1 plotted in the right panel.

form a measurement on both readout pairs by sequentially pulsing the Q1Q2 (left sub-columns),
and the Q3Q4 qubit pairs (right sub-columns) to their respective readout points. When driving
the |↓↓〉-|↑↓〉 transition of the qubit pairs used for readout (row 1), we apply an additional single-
qubit π-pulse to the preparation qubit to preserve symmetry with the other measurements, as the
control qubit also serves as the readout ancillary qubit.

9.7.7. S HIFT IN R ABI FREQUENCY DUE TO EXCHANGE

When exchange interaction is present in the quantum dot system, the energy levels of the qubit
are modified and the qubit basis states hybridize. Conditional driving of the spin states is a di-
rect consequence of the shift of the energy levels, which allows for controlled multi-qubit gates as
previously employed to drive two-qubit gates [6–8, 54]. The hybridization of the qubit states on
the other hand gives rise to conditional Rabi frequencies that also depend on the strength of the
exchange interaction [54]. Furthermore, the exchange interaction can become anisotropic in the 9
presence of spin-orbit coupling [45]. When only considering the exchange interaction between
neighbouring quantum dots, a general Hamiltonian can be written for the four quantum dots in
the (1,1,1,1) charge regime as:

4 ¡
Si · J i j Si + B + B ac cos 2π f t + φ · S i ,
X X ¡ ¢¢
H= (9.1)
〈i , j 〉 i =1

where the first sum runs along every neighbouring quantum dot pair 〈i , j 〉 with the corresponding
tensorial exchange interaction J i j . We note that the term B consists of both the Zeeman effect
due to the external magnetic field, and the contribution due to the spin-orbit interaction. We also
explicitly separate the static Zeeman interaction from the field induced by the electric driving.
We take D to be the unitary matrix which diagonalizes Hamiltonian (9.1) for B ac = 0, e.g.,
D † H (B ac = 0)D = 1. Now, the effective Rabi amplitude between the eigenstates of the undriven
 152 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

Q1 Q2 Ramsey

S(f) (Hz 2 /Hz)

10 10 10 10 CPMG
1/f 0.78 1/f 0.65

10 5 1/f 0.50 10 5 1/f 0.45

1/f 1.08 1/f 1.18
10 -4 10 0 10 4 10 -4 10 0 10 4

Q3 Q4

S(f) (Hz 2 /Hz)

10 10 10 10
1/f 0.58 1/f 0.73

10 5 1/f 0.64 10 5 1/f 1.04

1/f 0.68 1/f 0.56
10 -4 10 0 10 4 10 -4 10 0 10 4
f (Hz) f (Hz)

Figure 9.11: Noise spectroscopy using Ramsey and CPMG measurements. We measure the effective noise
spectrum acting on the qubit, both tracing the resonance frequency using repeated Ramsey measurements
[20] (in blue), as well as by using the filter function of a dynamical decoupling measurement [18, 51] (in red).
Dashed blue and red lines are fits to the Ramsey and CPMG data respectively. The black line is a fit to the com-
bined data set, where the weight of both sets is normalized for the amount of data points. The effective noise
can be observed to increase towards low frequencies, consistent with the upwards trend of T2CPMG observed
p
in Fig. 9.7c. The effective charge noise measured in this heterostructure is S cn ( f ) = 6 µV/ Hz at 1 Hz [27].
Combining this with a typical resonance frequency slope of d f /dV = 5 MHz/mV [39], results in an effective
resonance frequency noise power of S( f ) = 9 · 108 Hz2 /Hz, comparable to what is observed experimentally,
suggesting coherence is limited by charge noise in our system. The effect of charge noise could be mitigated by
careful optimization of the electric field environment [52] or moving to a multi-hole charge occupancy, screen-
ing the influence of charge impurities [53], potentially enabling even higher fidelity operations. Alternatively,
noise could originate in the nuclear spin bath present in natural germanium, which could be overcome by
isotopically enriching the material.

Hamiltonian |ξ〉 and |ζ〉 in the adiabatic limit of exchange is given by:
1
Ω|ξ〉→|ζ〉 = 〈ξ| D † B ac D |ζ〉 , (9.2)
4
where the prefactor 1/4 is coming from the spin and the rotating wave approximation. There-
fore, the Rabi amplitude depends on the exact form of the exchange interaction, as well as which
transition is driven.

9 9.7.8. F ITTING OF THE TWO - QUBIT DECAY DATA

In order to extract the decay time scale in ¢Fig. 9.4d, we fit the exchange interaction data to the
model function P = A cos 2π( f 0 + t δ)t + φ0 exp(−t /τ)+ y 0 , with amplitude A, frequency f 0 , phase
¡

offset φ0 , and offset y 0 . We note that we allow for a small linear shift of the precession frequency δ,
typically of size δ = 10 MHz/µs, as a result of pulse imperfections in these relatively large and ex-
tended exchange pulses. We observe a small creep towards the final pulse amplitude to be present,
most likely caused by the skin effect in the coaxial lines, explaining the small observed frequency
shift throughout the experiment. The data for the situation with no exchange present is fitted to
the exponential decay P = exp(−t /τ) + y 0 , from which we deduce the decay time scale τ.

9.7.9. S PECIFICATION OF THE CROT AND CZ GATE TIMES

The different two-qubit gate times are listed below both for the CROT (Table 1) as well as the CZ
(Table 2) gates.
9.7. M ETHODS 153

target transition control t π,control (ns) t π,target (ns)

1 |↓↓〉-|↓↑〉 2 67 85
1 |↓↑〉-|↑↑〉 2 67 95
1 |↓↓〉-|↓↑〉 4 61 104
1 |↓↑〉-|↑↑〉 4 61 108
2 |↓↓〉-|↓↑〉 1 45 105
2 |↓↑〉-|↑↑〉 1 41 105
2 |↓↓〉-|↓↑〉 3 38 113
2 |↓↑〉-|↑↑〉 3 38 100
3 |↓↓〉-|↓↑〉 2 65 53
3 |↓↑〉-|↑↑〉 2 65 83
3 |↓↓〉-|↓↑〉 4 49 83
3 |↓↑〉-|↑↑〉 4 45 68
4 |↓↓〉-|↓↑〉 1 45 105
4 |↓↑〉-|↑↑〉 1 45 120
4 |↓↓〉-|↓↑〉 2 38 68
4 |↓↑〉-|↑↑〉 2 38 74

Table 9.1: CROT driving times used in Fig. 9.2.

9
two-qubit system t ramp (ns) t gate (ns)
Q1Q2 3 6
Q2Q3 10 4
Q3Q4 10 5
Q4Q1 3 6

Table 9.2: CZ gate details.

 154 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

a b
0.8
f1 0.6 f1 0.6 f5
0.5
P up

P up
P up

0.2 0.2 0.2

0.80 200 400 0 f 100 200 0 f 100 200
f 2 0.6 2 0.6 6
0.5
P up

P up
P up

0.2 0.2 0.2

0.80 200 400 0 100 200 0 100 200
f 3 0.6 f3 0.6 f7
0.5
P up

P up
P up

0.2 0.2 0.2

0.80 200 400 0 f 100 200 0 f 100 200
f 4 0.6 4 0.6 8
0.5
P up

P up
P up

0.2 0.2 0.2

0 200 400 0 100 200 0 100 200

tp (ns) tp (ns) tp (ns)

Figure 9.12: Driving of all resonance lines of the coupled three and four qubit system. a, Both the coupling
between Q2 and Q1 as well as Q2 and Q3 are enabled, using the respective virtual barrier gates. This splits
the resonance line in four, as shown in Fig. 9.3. Driving each of the separate lines, results in the conditional
rotation of Q2 depending on the states of Q1 and Q3. We measure the spin up probability after driving each
of the four resonance lines for time t p , for all four permutations of the Q1 and Q3 basis states as initial state,
following the colour scheme of Fig. 9.3. The driving power is adjusted for each of the transitions to synchronize
the π-rotation times, with a f 1 = 330 mV, a f 2 = 500 mV, a f 3 = 280 mV, and a f 4 = 400 mV, for f 1−4 from low to
9 high. b, Similarly, by additionally opening up the coupling between Q3 and Q4 as well, the resonance line splits
in four and we can drive all separate lines individually. The eight lines are driven using the same microwave
power in this figure and a strong difference in rotation frequencies can be observed for the different transitions
f 1−8 from low to high. This also results in a small off-resonant driving effect for some of the lines.
9.7. M ETHODS 155

a b c
Qtarget X Z(ϕ) X π + ϕ0, control = up
ϕ0, control = down
Qcontrol X2 X2 CZ-gate
P up 0.6 0.8
8 4
-94 0.6
(rad)

P up
0.8 6 3 0.4 Q1Q2 0.4 Q2Q3
-92
0.7 4 0.2 0.2
2
V B12 (mV)

-90 0.6 -75 -70 -65 -75 -70 -65 0 2 4 6 0 2 4 6

V B12 (mV) V B23 (mV)
0.5
-88 0.8 0.6
6 3
0.4 0.6
P up

2.5 Q3Q4 Q4Q1

(rad)

-86 4 0.4
0.3 0.4
2
-84 0.2 2 0.2 0.2
1.5
0 2 4 6 0 2 4 6 -95 -90 -85 -80 -75 -70 0 2 4 6 0 2 4 6
(rad) (rad) V B34 (mV) V B41 (mV) (rad) (rad)

Figure 9.13: Tuning of the CZ-gates. a,b, The CZ-gates between all four qubit pairs are tuned using a Ramsey
sequence (analogous to Fig. 9.4), where the spin-up probability is measured as a function of the phase φ of
the final π/2 pulse as well as the depth of the exchange pulse VBmn , with m and n the relevant qubits (a). We
choose to tune the height of the voltage pulse rather than its length, due to the limited temporal¡resolution
¢ of
the exchange pulses (1 ns). The acquired phase φ0 is obtained by fitting each line to P = A cos φ + φ0 + y 0 ,
with A the visibility and y 0 an offset. A CZ-gate is achieved when the difference in acquired phase is exactly
∆φ = π, for the situation where the control qubit is |↓〉 (blue) compared to |↑〉 (orange). The barrier gate voltage
at which this occurs is obtained from the intersection of two locally linear fits to the extracted acquired phase
(b). c, The CZ-gates between all four qubit pairs probed by a Ramsey sequence using the inverse qubits as
target and control as compared to the data in Fig. 9.4b. The target qubit is marked in bold.

9
 156 9. A F OUR-QUBIT G ERMANIUM QUANTUM P ROCESSOR

a
twait twait
Q3 X(t) Y2

Q2 Y2
GHZ GHZ-1
Q1 Y 2

Q4 Y2

b c
400 400

300 300
2twait (ns)

2twait (ns)
200 200

100 100

0 0
0 50 100 150 200 0 50 100 150 200
t prep (ns) t prep (ns)

Figure 9.14: Time evolution of the four-qubit GHZ state. a, Circuit diagram of the experiments performed in
panels b,c. We first apply a preparation pulse to Q3 and then generate a four qubit GHZ-state analogous to
Fig. 9.5. Next we let the entangled system evolve for time t wait , then apply an optional Y2 decoupling pulse
and finally disentangle the GHZ-state again. b,c, We vary both the waiting time and preparation time t prep and
plot the spin-up fraction of Q3 in the case without (b) and with (c) decoupling pulse. It can be clearly observed
that without the echo pulse, the system has fully decohered at the end of the algorithm, However, by applying
the decoupling pulse, the coherence of the entangled system can be maintained for a prolonged time scale,
with a characteristic decay time of τ = 390 ns.

Q1Q2
Q3Q4 I II III IV V VI VII VIII IX
1 1 1 1 1 1 1 1 1
P norm

No dephasing 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5

0 0 0 0 0 0 0 0 0
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
1 1 1 1 1 1 1 1 1
9
P norm

Fit dephasing
0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
parameter
0 0 0 0 0 0 0 0 0
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
1 1 1 1 1 1 1 1 1
P norm

Complete
0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
dephasing
0 0 0 0 0 0 0 0 0
0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200 0 100 200
t prep (ns) t prep (ns) t prep (ns) t prep (ns) t prep (ns) t prep (ns) t prep (ns) t prep (ns) t prep (ns)

Figure 9.15: Dephasing of the four-qubit GHZ state. We model the quantum circuit performed in Fig. 9.5 and
account for qubit decoherence by applying a depolarizing channel Λλ (ρ) = λρ + 1−λ d
1. We plot the expected
measurement outcomes for qubit pairs Q1Q2 (blue) and Q3Q4 (orange). The top row corresponds to the case
of perfect coherence in each panel. In the center row, we fit the depolarization parameter to the measurement
data. The finite rotations visible in panels 4 and 5 in Fig. 9.5 can be reproduced by including gate errors in
the model. Finally, the bottom row corresponds to a full depolarization of the state. If the qubit system is
completely dephased at any point in time, no recovery of the signal can be observed in panel IX.
R EFERENCES 157

R EFERENCES
[1] Terhal, B. M. Quantum error correction for quantum memories. Rev. Mod. Phys. 87, 307–346
(2015).

[2] Reiher, M., Wiebe, N., Svore, K. M., Wecker, D. & Troyer, M. Elucidating reaction mechanisms
on quantum computers. PNAS 114, 7555–7560 (2017).

[3] Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Phys. Rev. A 57,
120–126 (1998).

[4] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 34 (2017).

[5] Shulman, M. D. et al. Demonstration of Entanglement of Electrostatically Coupled Singlet-

Triplet Qubits. Science 336, 202–205 (2012).

[6] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[7] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins. Science 359, 439–442
(2018).

[8] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon. Nature 555,
633–637 (2018).

[9] Huang, W. et al. Fidelity benchmarks for two-qubit gates in silicon. Nature 569, 532–536
(2019).

[10] He, Y. et al. A two-qubit gate between phosphorus donor electrons in silicon. Nature 571,
371–375 (2019).

[11] Madzik,
˛ M. T. et al. Conditional quantum operation of two exchange-coupled single-donor
spin qubits in a MOS-compatible silicon device. Nature Communications 12, 181 (2021).

[12] Petit, L. et al. Universal quantum logic in hot silicon qubits. Nature 580, 355–359 (2020).

[13] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit
logic with holes in germanium. Nature 577, 487–491 (2020).

[14] Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Na-
ture 574, 505–510 (2019). 9
[15] Koppens, F. H. L. et al. Driven coherent oscillations of a single electron spin in a quantum
dot. Nature 442, 766–771 (2006).

[16] Petta, J. R. et al. Coherent Manipulation of Coupled Electron Spins in Semiconductor Quan-
tum Dots. Science 309, 2180–2184 (2005).

[17] Itoh, K. M. & Watanabe, H. Isotope engineering of silicon and diamond for quantum com-
puting and sensing applications. MRS Commun. 4, 143–157 (2014).

[18] Muhonen, J. T. et al. Storing quantum information for 30 seconds in a nanoelectronic device.
Nat. Nanotech. 9, 986–991 (2014).

[19] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control-fidelity.
Nat. Nanotech. 9, 981–985 (2014).
 158 R EFERENCES

[20] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nature Nanotechnology 13, 102–106 (2018).

[21] Yang, C. H. et al. Silicon qubit fidelities approaching incoherent noise limits via pulse engi-
neering. Nat Electron 2, 151–158 (2019).

[22] Zwanenburg, F. A. et al. Silicon quantum electronics. Rev. Mod. Phys. 85, 961–1019 (2013).

[23] Scappucci, G. et al. The germanium quantum information route. Nature Reviews Materials
1–18 (2020).

[24] Itoh, K. et al. High purity isotopically enriched 70-Ge and 74-Ge single crystals: Isotope sep-
aration, growth, and properties. J. Mater. Res. 8, 1341–1347 (1993).

[25] Bulaev, D. V. & Loss, D. Spin Relaxation and Decoherence of Holes in Quantum Dots. Phys.
Rev. Lett. 95, 076805 (2005).

[26] Lodari, M. et al. Light effective hole mass in undoped Ge/SiGe quantum wells. Phys. Rev. B
100, 041304 (2019).

[27] Lodari, M. et al. Low percolation density and charge noise with holes in germanium. Mater.
Quantum Technol. 1, 011002 (2021).

[28] Hendrickx, N. W. et al. Gate-controlled quantum dots and superconductivity in planar ger-
manium. Nature Communications 9, 2835 (2018).

[29] Lawrie, W. I. L. et al. Quantum dot arrays in silicon and germanium. Appl. Phys. Lett. 116,
080501 (2020).

[30] Pillarisetty, R. Academic and industry research progress in germanium nanodevices. Nature
479, 324–328 (2011).

[31] Bulaev, D. V. & Loss, D. Electric Dipole Spin Resonance for Heavy Holes in Quantum Dots.
Phys. Rev. Lett. 98, 097202 (2007).

[32] Maurand, R. et al. A CMOS silicon spin qubit. Nature Communications 7, 13575 (2016).

[33] Watzinger, H. et al. A germanium hole spin qubit. Nature Communications 9, 3902 (2018).

9 [34] Pioro-Ladrière, M. et al. Electrically driven single-electron spin resonance in a slanting Zee-
man field. Nature Physics 4, 776–779 (2008).

[35] Tokura, Y., van der Wiel, W. G., Obata, T. & Tarucha, S. Coherent Single Electron Spin Control
in a Slanting Zeeman Field. Phys. Rev. Lett. 96, 047202 (2006).

[36] Sammak, A. et al. Shallow and Undoped Germanium Quantum Wells: A Playground for Spin
and Hybrid Quantum Technology. Advanced Functional Materials 29, 1807613 (2019).

[37] van Riggelen, F. et al. A two-dimensional array of single-hole quantum dots. Appl. Phys. Lett.
118, 044002 (2021).

[38] Hendrickx, N. W. et al. A single-hole spin qubit. Nature Communications 11, 3478 (2020).

[39] Lawrie, W. I. L. et al. Spin Relaxation Benchmarks and Individual Qubit Addressability for
Holes in Quantum Dots. Nano Lett. 20, 7237–7242 (2020).
R EFERENCES 159

[40] Danon, J. & Nazarov, Y. V. Pauli spin blockade in the presence of strong spin-orbit coupling.
Phys. Rev. B 80, 041301 (2009).

[41] Yang, C. H. et al. Charge state hysteresis in semiconductor quantum dots. Appl. Phys. Lett.
105, 183505 (2014).

[42] Harvey-Collard, P. et al. High-Fidelity Single-Shot Readout for a Spin Qubit via an Enhanced
Latching Mechanism. Phys. Rev. X 8, 021046 (2018).

[43] Knill, E. et al. Randomized benchmarking of quantum gates. Phys. Rev. A 77, 012307 (2008).

[44] Gullans, M. J. & Petta, J. R. Protocol for a resonantly driven three-qubit Toffoli gate with silicon
spin qubits. Phys. Rev. B 100, 085419 (2019).

[45] Hetényi, B., Kloeffel, C. & Loss, D. Exchange interaction of hole-spin qubits in double quan-
tum dots in highly anisotropic semiconductors. Phys. Rev. Research 2, 033036 (2020).

[46] Taylor, J. M. et al. Fault-tolerant architecture for quantum computation using electrically
controlled semiconductor spins. Nature Physics 1, 177–183 (2005).

[47] Veldhorst, M., Eenink, H. G. J., Yang, C. H. & Dzurak, A. S. Silicon CMOS architecture for a
spin-based quantum computer. Nat. Commun. 8, 1766 (2017).

[48] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science Advances 4, eaar3960
(2018).

[49] Hensgens, T. et al. Quantum simulation of a Fermi–Hubbard model using a semiconductor

quantum dot array. Nature 548, 70–73 (2017).

[50] Seedhouse, A. E. et al. Pauli Blockade in Silicon Quantum Dots with Spin-Orbit Control. PRX
Quantum 2, 010303 (2021).

[51] Chan, K. W. et al. Assessment of a Silicon Quantum Dot Spin Qubit Environment via Noise
Spectroscopy. Phys. Rev. Applied 10, 044017 (2018).

[52] Wang, Z. et al. Suppressing charge-noise sensitivity in high-speed Ge hole spin-orbit qubits.
(2019).

[53] Barnes, E., Kestner, J. P., Nguyen, N. T. T. & Das Sarma, S. Screening of charged impurities with
multielectron singlet-triplet spin qubits in quantum dots. Phys. Rev. B 84, 235309 (2011). 9
[54] Russ, M. et al. High-fidelity quantum gates in Si/SiGe double quantum dots. Phys. Rev. B 97,
085421 (2018).
 10
S IMULTANEOUS DRIVING OF
SEMICONDUCTOR SPIN QUBITS AT
THE FAULT- TOLERANT THRESHOLD

Quantum computers will require the ability to perform quantum operations on large numbers of
qubits with error rates below the fault tolerant threshold required for quantum error correction
schemes [1]. Quantum dots define a promising platform due to their compatibility with semicon-
ductor manufacturing, and high-fidelity operations beyond 99.9% have been realized with individ-
ual qubits [2], though investigations into the effect of operation of qubits simultaneously do not
reach comparable values [3]. Here we present single qubit randomized benchmarking results in a
two dimensional array of spin qubits, finding native gate fidelities as high as 99.990(1) %. We also
characterize qubit performance during the simultaneous operation of N qubits, for which we de-
fine and utilize a novel benchmarking protocol called ’N -Copy benchmarking’. We find two- and
four-copy fidelities of 99.905(8) % and 99.34(4) % respectively. We also find that benchmarking
next-nearest neighbour pairs can return fidelities within the error margin of their single qubit cases,
indicating that cross talk can be highly local in the absence of an exchange interaction. These char-
acterizations of the single-qubit gate quality and the ability to operate simultaneously are crucial
aspects for scaling up germanium based quantum information technology.

Parts of this chapter have been published in arXiv:2109.07837 [cond-mat.mes-hall]

161
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
162 THRESHOLD

10.1. I NTRODUCTION
Spin qubit approaches to quantum computing have made remarkable progress in the last few
years. In both silicon and germanium [4, 5], two-dimensional approaches to scale up quantum
dots have been the focus of multiple recent efforts [6, 7], as they provide greater qubit connectivity
and enable the use of surface code architectures for quantum error correction [8]. For electrons
in silicon, scaling qubits in the second dimension is challenging due to the need for components
such as striplines and nanomagnets to enable qubit drive, and qubit realizations have been lim-
ited to linear arrays [9–11]. Instead, spin qubits based on holes in germanium can be driven all-
electrically through the intrinsic strong spin-orbit coupling [12–14]. Furthermore, advances in
strained germanium (Ge/SiGe) have yielded low charge noise and percolation density [15] and
high hole mobility [16], indicative of a highly uniform platform. These advantages have resulted
in the Ge/SiGe platform maturing rapidly over the last few years and led to demonstrations of
long spin relaxation times [17], single hole qubits and singlet triplet qubits [14, 18] and universal
operation on a 2x2 qubit array [19].
However, as spin qubits expand into two dimensions, the growing number of possible qubit
cross talk interactions motivates careful characterization. In the present work, we make use of a
2x2 quantum dot array of hole spin qubits, to characterize the single qubit fidelities of our system,
as well as characterizing their quality while driving two and four qubits simultaneously. We per-
form randomized benchmarking in the single qubit Clifford space, and investigate the dependence
of fidelity on qubit Rabi period, finding that single qubit fidelities can be as high as 99.990(1)%. We
then investigate the individual single qubit performance using a novel benchmarking technique
we call N -copy benchmarking (See section 10.7). This technique applies the same Clifford opera-
tion to N qubits simultaneously, and results in a good approximation to simultaneous randomized
benchmarking (SRB) [20] in the limit of low exchange interaction (See section 10.6). We find for
N =2 and N =4, elementary gate fidelities of F 2Qπ/2 = 99.905(8)% and F π/2 = 99.34(4)% respectively.
4Q
We compare these experiments at two magnetic fields B = 1 T and B = 0.65 T and find that while
individual qubit operation performs best at the lower magnetic field, multi-qubit operation per-
forms better at a higher magnetic field, due to the relevance of qubit addressability.

10.2. R ESULTS
Figure 10.1a shows a false-coloured scanning electron microscope (SEM) image of the device used
in the experiment [19]. It consists of two gate layers and an ohmic layer, where ohmic contacts
to the quantum well are created by diffused Al [21]. By applying potentials to the plunger gates
P1 -P4 , we can define four quantum dots, each filled with a single hole spin such that we operate
in the (1,1,1,1) charge regime (see section 10.5a). The tunnel couplings between these quantum
dots can be tuned with the dedicated interdot barrier gates B12-41. We also define two larger
quantum dots using P S12 and P S34 in the multiple hole regime, which we utilize as charge sensors.
By applying a rf tone to the ohmic gates O1 and O3 via two off-chip inductors bonded in-line, we
10 form a resonant tank circuit allowing us to perform fast rf charge sensing of our quantum dots to
read out the spin states of our qubits, we perform spin-to-charge conversion in the form of Pauli
spin blockade (PSB). Figure 1b depicts the two PSB readout pairs in our system, with the Q1Q2
system comprising of the two qubits Q1 and Q2 (orange, yellow), and the Q3Q4 system containing
qubits Q3 and Q4 (purple, green). We make use of a latched readout mechanism [19, 22], whereby
the dot-reservoir tunnel rate is limited significantly for one quantum dot per readout pair, which
is depicted by a dashed arrow. We tune the tunnel rates of quantum dots Q2 and Q4 to be ΓQ2 = 5
kHz and ΓQ4 = 0.416 kHz respectively, and read out in an integration time of 10 µs such that we are
well within the dot-reservoir limiting timescales. We apply an in-plane magnetic field Bext to split
the spin states. Two different magnetic fields were used in this experiment, of Bext = 1 T and Bext =
10.2. R ESULTS 163

(a) (b) (c)

2 2
O2 B12 P2 0.65 T
PS12 Fast 1.00 T
100 nm Slow 1 1
O1
ΓQ2
B23
Q1 Q2 0 0

PBlocked
P3 Q1Q2 (1,1)AP 0 500 0 500

U
P1 2 2
Q4 Q3 Q3Q4 (1,1)P
B41 O3 ΓQ4 1 1
PS34 (0,1)
P4 B34 O4 Bext Γ (0,2)S
0 0
0 500 0 500
τ (ns)

Figure 10.1: Heavy hole spin qubit array in germanium. (a) Scanning electron microscope (SEM) image of the
device. The device is comprised of two lithographically defined layers of Ti:Pd constituting the plunger (blue)
and barrier (red) gates, as well as an Al ohmic layer (green) forming low resistance ohmic contacts with the
quantum well directly. An external magnetic field Bext is applied in-plane with respect to the quantum well.
(b) Latched spin blockade readout mechanism. We consider two separate readout systems Q1Q2 (red) and
Q3Q4 (blue), each containing a double quantum dot pair and single hole transistor. By reducing the reservoir
tunnel coupling to quantum dot Q2(4) ΓQ2(4) , we are able to suppress the (1,1)T to (0,2)S transition longer than
the typical singlet triplet relaxation rates [14], facilitating a readout integration window of 10 µs. (c) Ramsey
sequences on qubits Q1-4 respectively, at magnetic field B = 1 T (circles, lower) and 0.65 T (triangles, upper).
Extracted spin dephasing times at 1 T for qubits Q1 (orange), Q2 (yellow), Q3 (purple) and Q4 (green) are T2∗Q1
=
186 ± 19 ns, T2∗
Q2
= 119 ± 14 ns, T2∗
Q3
= 323 ± 52 ns and T2∗
Q4
= 147 ± 26 ns. At 0.65 T the dephasing times increase
to T2∗
Q1
= 276 ± 22 ns, T2∗
Q2
= 166 ± 14 ns, T2∗
Q3
= 472 ± 31 ns and T2∗
Q4
= 228 ± 15 ns. Data for different magnetic
fields is offset by unity for clarity.

0.65 T (Fig. 10.1c). These fields are chosen to provide a comparison between the different regimes
of coherence, qubit drive response, and qubit resonance frequency spacing.

10.2.1. S INGLE Q UBIT B ENCHMARKING

In order to get a baseline reading of the individual qubit fidelities, we performed randomized
benchmarking to estimate the average gate fidelity of each qubit [23]. Here, operations randomly
selected from the Clifford group are applied to each single qubit, initialised in a known state, after
which a final recovery Clifford C−1 is applied to bring the qubit back to its original state. Imper-
fections in the applied gates, and gradual qubit decoherence result in a decay of the recovered
state probability as the number of applied Clifford operators is increased, allowing the extraction
of a fidelity by fitting the decay [23]. Each element of the Clifford group can be constructed from
a variety of generator gates. We construct our Clifford group from a minimal generator set Gi ∈
{X π/2 , Yπ/2 }. We find this set advantageous since it contains on average 3.217 qubit π/2 rotations
(generators) per Clifford, that differ only by a software phase shift. This means the estimated Clif-
ford fidelity is a direct indicator of the generator fidelity, by equally weighting the generators of the
same length [24].
10
Figures 10.2a-d show the randomized benchmarking sequences for Q1-4 respectively. A red
(blue) measurement window indicates PSB readout on the Q1Q2 (Q3Q4) double quantum dot pair.
Each qubit is initialised in the spin down state. For each sequence length N, 32 random permuta-
tions of N Cliffords are averaged to give the final trace, each of which is composed of 1500 single
shot measurements. An exponential decay is fit to the resulting trace (see section 10.4.3), from
which an average generator fidelity FQ π/2 can be extracted for each qubit.
i
Holes in germanium allow for very fast electrical driving, with Rabi frequencies exceeding hun-
dreds of MHz [25, 26]. However, rapid qubit manipulation is not always optimal for coherent qubit
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
164 THRESHOLD

(a) N-1
(b) (c) (d)
C C-1 N-1

C C -1
N-1

C C-1 N-1
C C-1
(e) (f) (g) (h)
-1 -1 -1 -1
10 10 10 10

10-2 10-2 10
-2
10
-2
/2

/2

/2

/2
1-F

1-F

1-F

1-F
10-3 10-3 10-3 10-3

10-4 10-4 10
-4
10
-4

0 100 200 300 50 100 150 200 50 100 150 20 40 60 80 100 120
t (ns) t (ns) t (ns) t (ns)
(i) (j) (k) (l)
1 1 1 1

0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6

Blocked

Blocked

P Blocked

P Blocked
0.4 0.4 0.4 0.4
P

0.2 0.2 0.2 0.2

FQ1
π/2
= 99.980(4) % FQ2
π/2
= 99.936(9) % FQ3
π/2
= 99.990(1) % FQ4
π/2
= 99.8(4) %
0 0 0 0
100 105 100 105 100 105 100 105
NC NC NC NC

Figure 10.2: Single qubit randomized benchmarking at 0.65 T. (a-d) Random Clifford sequences applied to
each qubit. Each qubit in the array is prepared in the spin down state. N-1 randomly selected Cliffords are
applied to a single qubit, after which a recovery Clifford (C−1 ) is applied bringing the system back to the | ↓↓>
state. Each sequence is repeated 32 times with different random permutations of Cliffords. Readout occurs
via PSB on one of the two readout pairs Q1Q2 (red) or Q3Q4 (blue). (e-h) Dependence of qubit fidelity on ap-
plied microwave power. An optimal power tuning occurs due to a trade-off between decoherence (low power)
and errors introduced at high power including gate calibration errors and driving non-linearities. Error bars
reflect the statistical error of the fitting. (i-l) Best single qubit benchmarks for each qubit. All native π/2 fideli-
ties except Fπ/2
Q4
exceed 99.9 %, with Fπ/2 Q3
approaching four nines. These traces correspond to the respective
highlighted points in (e-h).
10
10.2. R ESULTS 165

control, with high powers leading to enhanced systematic errors in qubit operation arising from
effects such as sample heating or pulse imperfections. Indeed we find a strong dependence of the
single qubit fidelities on the drive speed. Figures 10.2e-h show the generator infidelities (1-FQ π/2 )
i
as a function of qubit drive speed. Despite being able to drive qubit rotations in as fast as 10 ns,
we find that the associated single qubit fidelity suffers as a result, visible by a sharp decrease in the
fidelity for qubits Q1 and Q3. Fidelity in these cases could be limited by a number of mechanisms,
such as quantum dot anharmonicties [27, 28] or systematic Pauli errors due to gate tuning.
Figures 10.2i-l show the randomized benchmarking data for the optimal Rabi periods. We ex-
tract generator fidelities above 99 % for each qubit in the array, with qubit Q3 performing the best
Q3
with Fπ/2 = 99.990(1) %. For single qubit randomized benchmarking, we expect a fully decohered
state to exhibit a read out spin down probability of about 1/2. However in the presence of finite
exchange and classical cross-talk between the active qubit and the readout qubit in the spin block-
ade pair, state leakage can occur to all four states in the two-qubit subspace, resulting in a readout
signal of about PBlocked = 0.329 (see section 10.5). We find that for the case of qubit Q4, the plateau
of the spin blocked probability approaches the expected value of the fully depolarized two-qubit
subspace for all driving powers.
To account for state leakage, an extra exponential is added for fitting randomized benchmark-
ing traces for qubit Q4, yielding two characteristic decay constants (see section 10.4.3). From this
π/2 = 99.8(4)%, and a leakage rate of Lπ/2 = 0.07(2)% per
analysis, we calculate a qubit fidelity FQ4 Q4
generator.

10.2.2. N - COPY R ANDOMIZED B ENCHMARKING

The operation of practical quantum computers will require multiple qubits to be controlled simul-
taneously [29]. Doing so however can affect performance to the detriment of the qubits due to both
classical and quantum cross-talk [3]. We now turn to the characterization of single qubit fidelities
using N -copy randomized benchmarking. Here, the same generation procedure described in the
single qubit case is used to create a random sequence of single qubit Cliffords, which is applied
to all qubits in the experiment. The recovery Clifford always brings the system back to the |↓↓〉 or
|↓↓↓↓〉 state for two- or four-copy experiments respectively. It is important to distinguish between
experiments wherein the driven qubits occupy the same measurement system (ES) ie. qubits Q1
and Q2 or Q3 and Q4, or different systems (ED) such as qubits Q1 and Q3. The measurement op-
erators for ES and ED experiments will differ, leading to different decay behaviours (see section
10.7.2). We perform two-copy randomized benchmarking (2CRB) experiments with initial state
|↓↓〉, since the fitting procedure is simplified and provides a good estimate of the true fidelity in
the regime of low exchange (see section 10.8.2, and 10.9). Figures 10.3a-b show a complete char-
acterization of the single, two- and four-copy benchmarking at a consistent Rabi period of tπ =
96 ns and magnetic field Bext = 0.65 T. From this dataset, we can approximate the relative loss in
fidelity in the ED case for each qubit due to the additional driving of another qubit by defining the
quantity δQ i |Q j = ²Q i |Q j - ²Q i where ²Q i = 1 - FQ i is the infidelity of qubit Qi (Full numerical data
available in section 10.10). For ES-2CRB, the resulting fidelity is related to the combined fideli- 10
ties of both qubits in the system. We observe that qubits driven simultaneously with their nearest
neighbours typically result in relatively larger error rates, while next nearest neighbour pairs can
give error rates that are very low. For example, the relative loss of qubit Q3 due to the driving of
qubit Q2 δQ 3 |Q 2 = 0.35% is significant, while for the non-adjacent qubit pair Q3 and Q1, δQ 3 |Q 1 is
immeasurable within the error bar and gave a simultaneous fidelity of FQ3 π/2 ≈ Fπ/2 = 99.964(3)%.
Q3|Q1
The same is true for the opposite configuration, where δQ 1 |Q 3 again is immeasurable within the
error bar, and FQ1π/2 ≈ Fπ/2 = 99.97(1)%. Optimum Rabi periods for simultaneous driving also ex-
Q1|Q3
ist due to a trade-off between qubit coherence and frequency crowding. The fidelity of the Q3Q4
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
166 THRESHOLD

(a) (b) (c) (e)

8 8 1

7 7 10-2

PBlocked
0.5
6 6
Fπ/2 = 99.905(8) %
5 5 0
Q3Q4

100 102 104

P Blocked
P Blocked

/2
4 4 NC 10-3

1-F
(d)
1
3 3

2 2
PBlocked
0.5
1 1
10-4 1.00 T
Fπ/2 = 99.0(1) % 0.65 T
Q3Q4|Q1Q2
0 0 0
100 105 100 105 0 50 100 1 2 4
NC NC NC NQ

Figure 10.3: N -copy randomized benchmarking in a four qubit array. (a-b) Benchmarking characterization at
tπ = 96 ns for single qubit randomzied benchmarking, two- and four-copy randomized benchmarking exper-
iments. (a) and (b) correspond to readout systems Q1Q2 and Q3Q4 respectively. For a pair of qubits, N − 1
randomly selected Clifford sequences are applied simultaneously followed by a recovery pulse, and read out
via PSB. Line and marker color, as well as the corresponding cartoons indicate which qubits are being bench-
marked, and which PSB readout pair is being probed. The additional coloured dot indicates which system is
being read out, such that a red(blue) dot indicates readout of the Q1Q2(Q3Q4) system. (c) Decay for a ES 2CRB
π/2
trace at 0.650 T. The readout system is Q3Q4. Rabi period is set to tRabi = 61 ns for the qubits. A fidelity of FQ3Q4
= 99.905(8) % is extracted. (d) Four-copy randomized benchmarking at 0.650 T. The readout system is Q3Q4.
π/2
The Rabi period for all qubits is set to tRabi = 150 ns. A fidelity of FQ3Q4|Q1Q2 = 99.0(1) % is extracted. (e) Sum-
mary of the fidelities, reported as the native average π/2 generator infidelity 1-Fπ/2 as a function of number
of driven qubits NQ , such that NQ =1, NQ = 2 and NQ = 4 represent single qubit, ES two-copy and four-copy
randomized benchmarking fidelities respectively. Data points for B = 0.65 T (triangles) and B = 1 T (circles) is
compared. Colours are indicative of the qubits involved in the experiment. For NQ = 2, orange/yellow markers
symbolize qubits Q1 and Q2, while purple/green markers symbolize qubits Q3 and Q4. Red and blue markers
for NQ = 4 qubits represent all four qubits, read out in system Q1Q2 and Q3Q4 respectively.

10
10.3. C ONCLUSION 167

π/2
system in a four-copy experiment, FQ3Q4|Q1Q2 = 99.0(1)%, is optimal for a Rabi period of 150 ns.
Further increase of the Rabi period does not yield better fidelities as the coherence of qubit Q4 is
limiting this value. We find we can also improve the ES 2CRB fidelity of the adjacent qubit pair
π/2 = 99.905(8)% by decreasing the Rabi period to 61 ns, (see fig. 10.3c), indicating that deco-
FQ3Q4
herence may have be the limiting process. The observation of high fidelity single qubit gates for
ES and ED 2CRB, as well as four-copy benchmarking (4CRB) fidelities above 99%, shows promise
for qubit operation in large scale spin qubit arrays.
Two different magnetic field settings were applied in this work to clarify the importance of
spin coherence on the gate fidelities (see Fig. 10.1c). At lower magnetic fields, the effect of the
spin-orbit coupling via the g-factor modulation of the qubits during driving is reduced, decreas-
ing the susceptibility of the resonance frequency to charge noise [30]. Since the Rabi frequency
is linear in magnetic field, the drive power has to be increased to compensate [31]. This leads to
increased qubit cross-talk when driving qubits simultaneously and suggests the existence of an
optimal magnetic field strength. Figure 10.3e summarises the randomized benchmarking results
of this work, providing detailed information on the extracted average generator as a function of
number of driven qubits NQ . On average, the individual single qubit fidelities are higher at lower
field, as are the fidelities for 2CRB. The power required to drive the systems at their optimal Rabi
period however, was found to be inversely proportional with Bz as expected, and as such were ob-
served to shift to longer Rabi periods. We observed that at the stronger magnetic field of 1 T, higher
fidelities can be achieved for the four-copy case. We attribute this result to the faster achievable
Rabi frequencies as well as larger qubit frequency splittings outweighing the relative loss in coher-
ence. We expect that the magnitude of these classical cross talk effects can be reduced through
pulse engineering techniques [32, 33], facilitating shorter Rabi periods when driving qubits. Ad-
ditionally, the high g-factor tuneability of hole states observed in Ge/SiGe [14] and Si [34] could
be exploited to reduce the cross talk effect by maximising the separation of the qubits’ resonance
frequencies [35].

10.3. C ONCLUSION
We have characterized qubit operational fidelities in a two-dimensional spin qubit array. Sin-
gle qubit fidelities of up to 99.990(1)% estimated using single qubit randomized benchmarking,
are strongly dependent on Rabi period, leading to an optimal driving speed configuration in our
qubits. We expect that pulse shaping will further increase this optimal drive speed if care is taken
to reduce classical cross talk between qubits. The large g-factor tuneability of the platform could
facilitate better resonant frequency spacings which could also lead to reduced cross talk. However,
we still achieve ES 2CRB fidelities exceeding 99.9% for two adjacent qubits, and 99% for four-copy
randomized benchmarking, defining the state-of-the-art for single qubit fidelities in quantum dot
based spin qubits.

10.4. M ETHODS 10
10.4.1. D EVICE FABRICATION
The Ge/SiGe wafer is fabricated on a silicon substrate. We use reduced-pressure chemical vapour
deposition to grow a 1.6 µm strain-relaxed Ge layer, a 1 µm reverse graded Si1−x Gex (x varies
from 1 to 0.8), a 500 nm constant composition Si0.2 Ge0.8 , and a 16 nm compressively strained Ge
quantum well. Finally on the quantum well we grow a 55 nm Si0.2 Ge0.8 barrier followed by an
oxidised Si cap layer (<2 nm). An ohmic contact layer is created by first defining it using electron
beam lithography, etching away the oxidised cap, then depositing 30 nm of Al. This layer is then
covered in 7 nm of Al2 O3 via atomic layer deposition at 300 o C. The gate stack is in two overlapping
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
168 THRESHOLD

layers of Ti:Pd (3/37 nm), separated by 7 nm of Al2 O3 .

10.4.2. G ENERATION OF THE S INGLE -Q UBIT C LIFFORD S ET

We quantify the quality of the single-qubit gates on all qubits by performing randomized bench-
marking using the Clifford group C 1 = {C n ∈ U1 |C n PC n† = P } with the single-qubit Pauli group P =
Q
{I , X , Y , Z }. All 24 elements of the Clifford group are generated from a minimal set, C n = g i ∈G g i
with G = {Xπ/2 , Yπ/2 }. The average number of elementary gates per Clifford is 3.217. All 24 Clif-
ford gates are provided in section 10.6. While not the most efficient choice for elementary gates,
this particular choice of set leads to uniform Clifford gates, whose elementary gates vary only by
a θ = π/2 phase-shift in the MW signal and are identical otherwise, beneficial for comparison and
performing 2CRB protocols.

10.4.3. F ITTING
Standard fitting of randomized benchmarking decays assumes a single exponential decay of the
form PBlocked = aF x + c where c should be equal to the average signal of the |↓, ↓〉 and |↑, ↓〉 sub-
space (where the first index corresponds to the qubit being benchmarked). Here F represents the
C = 1 − (1 − F )/2 and native gate fidelity
circuit level fidelity, from which the Clifford fidelity FQi Qi
π/2 = 1 − (1 − F )/(2 × 3.217) can be calculated. However, for qubit Q4, and for the high power
FQi Qi
regime of qubits Q1-3, we observe that the blocked signal plateaus to values corresponding to the
fully decohered two-qubit subspace, indicating state leakage. In this case, we fit with two expo-
nentials PBlocked = a 1 F 1x + a 2 F 2x +c, where c is set to the average signal of the four two-qubit states
in the readout pair. Here, F 1 and F 2 correspond to the two circuit level fidelities, representing a
leakage rate and an actual qubit fidelity. In this case, we assign the lower of the two values to the
qubit fidelity.
For ES 2CRB, we fit a single exponential, giving a result that is representative of the average fidelity
of both qubits in the system. For ED 2CRB, we fit a single exponential when the plateau corre-
sponds to a fully depolarized single-qubit subspace, and a double exponential when it decays to a
fully depolarized two-qubit subspace representing leakage. In the latter case, the fidelity is always
calculated from the faster of the two exponents. For 4CRB, a single exponential is fit. For all fits,
we assign confidence intervals on the extracted gate fidelities and leakage at the 95% confidence
level of the variance.

10.4.4. S IMULTANEOUS E LEMENTARY G ATE T UNING

In order to tune the system for the case of two-simultaneously driven qubits, we first determine
the non-simultaneously driven resonance frequency and Rabi frequency by applying a Xπ pulse to
each qubit separately and perform readout out via PSB. We then tune the values of tπ within 1 ns of
each other and then set the desired simultaneous t π as the average of the two single-qubit values of
t π . This constitutes a rough calibration of the qubits. To fine tune, we apply a X2π rotation on both
qubit simultaneously, and sweep the frequency of both qubits. Fitting a two-dimensional Gaussian
10 then gives the first iteration correction to the two qubits’ resonance frequencies. We then perform
the same experiment, but varying the amplitude of the MW power of both qubits. When driving
four qubits simultaneously, the tuning procedure is extended to include both systems Q1Q2 and
Q3Q4 in the tuning iteration.
10.5. I NITIALIZATION AND R EADOUT 169

10.5. I NITIALIZATION AND R EADOUT

(a) (b) (c) (d) (e) X(t)

30
- X X
Xπ π/2
π/2

10 2 40

∆VPS12 (mV)
(0,1,1,1) 1
0 20

P Blocked (a.u.)
20 0

V U 12 (mV)
(1,1,1,1)

Counts
-2 0 0.5

(0
-10 0 0.1 0.2 0.02 0.06

,2,
(0,0,0,0)

1,1
2 40

∆VPS34 (mV)
)
(1,2,1,1)
10 -20
0 20 0
-20 -10 0 10 0 100 200 300
-2 0
V 12 (mV) -0.2 0 -0.2 -0.15 tpulse (ns)
V U 12,34 (mV)

Sensor Val (a.u.) Sensor Val (a.u.)

0
(f) (g) (h) (i) X(t)
- X X
Xπ π/2 π/2

2
∆VPS34 (mV)

(1,1,1,1) 10 (1,1,0,1) 40 1
-10
0 20

P Blocked (a.u.)
0
V U 34 (mV)

-2 0
Counts
-0.4 0 -0.1 0 0.5
-10 (1,1,1,1)
-20 2 30
(1,

∆VPS12 (mV)
1,0

-20 20
,2)

(1,1,1,2) 0 0
10
-20 -10 0 10 0 100 200 300
-2 0
-30 V 34 (mV) 0.1 0.2 0.06 0.08 tpulse (ns)
-20 0 20
V (mV) Sensor Val (a.u.) Sensor Val (a.u.)
12,34

Figure 10.4: (a) Charge stability diagram of the 2x2 quantum dot array charge filling. The DC voltage point of
the Digital-Analogue converters (DACs) is positioned within the (1,1,1,1) charge filling regime. Anticrossings
between the (1,1,1,1)/(0,2,1,1) and (1,1,1,1)/(1,1,0,2) charge transitions are shown in red and blue respectively.
We use these anticrossings for spin to charge conversion and thus for readout of the qubit state. The axes
²12,34 and U12,34 are virtual gates describing the combined detuning and energy of the two qubit pairs Q1Q2
and Q3Q4. (b) Readout calibration of the Q1Q2 system. While in the (1,1,1,1) charge state, we prepare the qubit
system in the | ↓↓↓↓> state and apply a π-pulse to qubit Q2, bringing the system to the | ↓↑↓↓> state. We then
sweep the virtual detuning (V² 12 ) and on-site energy (VU 12 ) around the (1,1,1,1)/(0,2,1,1) anticrossing. We
then repeat the experiment without applying the X π -pulse to Q2, and plot the difference of the two measure-
ments, and pick the point of greatest visibility as the readout position. (c-d) Single shot readout calibration.
(c) Sensor value plot as a function of virtual sensor plunger gate potential (VPS12 ) for sensors one (upper, red)
and two (lower, blue), for the Q1Q2 system. Qubit Q2 is brought into a superposition state by applying a X π/2 -
pulse, and the system is pulsed to the readout point from (b). We repeat 1000 single shot measurements, for
each value of VPS12 and plot the resulting histogram of sensor response. A double-gaussian is fitted to the line
cuts to find the sensor response to the blocked and non-blocked signals, and the plunger gate voltage for max-
imal separation between the gaussian peaks, is set. (d) The same experiment performed at the given VPS12 .
A double gaussian is fit to the resulting histogram, and the single-shot threshold is calculated as the point of
equal separation between the gaussian peaks. (e) Visibility of the two qubit subspace of the Q1Q2 system. Rabi
oscillations are performed on Q1 with Q2 prepared either in the down state (red trace) or the up state (yellow
10
trace). There is a small blocked fraction difference between the | ↓↑> and | ↑↓> states, and the | ↑↑> state is less
blocked than the | ↓↑>. (f-i) The same procedure as (b-e), but pertaining to the Q3Q4 system, (1,1,1,1)/(1,1,0,2)
anticrossing, with X π - and X π/2 -pulses on qubit Q4, instead.
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
170 THRESHOLD

10.6. C LIFFORD G ATE S ET

Table 10.1: single-qubit Clifford sequences and their composition via the minimal generator set. We bench-
mark by selecting a random sequence of Cliffords from the table below excluding C1 , and calculate the recovery
Clifford that projects the system back into it’s original state. We only use a gate set containing π/2 rotations
around the Bloch Sphere, so the gates Xπ/2 and Yπ/2 are explicitly referring to a rotation of π/2 around the x-
axis and y-axis of the Bloch sphere of a single-qubit respectively. A Yπ/2 rotation is just an Xπ/2 rotation with a
π/2 software phase correction. There are on average 3.217 generators per Clifford composition. The extracted
fidelity then corresponds exactly to the π/2 rotation fidelity.

Clifford Composition
C1 I
C2 Yπ/2
C3 Xπ/2
C4 Yπ/2 Xπ/2
C5 Xπ/2 Yπ/2
C6 Yπ/2 Yπ/2
C7 Xπ/2 Xπ/2
C8 Yπ/2 Xπ/2 Xπ/2
C9 Yπ/2 Yπ/2 Xπ/2
C10 Yπ/2 Xπ/2 Yπ/2
C11 Yπ/2 Yπ/2 Yπ/2
C12 Xπ/2 Xπ/2 Xπ/2
C13 Xπ/2 Xπ/2 Yπ/2
C14 Xπ/2 Yπ/2 Yπ/2
C15 Yπ/2 Xπ/2 Yπ/2 Yπ/2
C16 Yπ/2 Yπ/2 Xπ/2 Yπ/2
C17 Xπ/2 Xπ/2 Xπ/2 Yπ/2
C18 Yπ/2 Xπ/2 Xπ/2 Xπ/2
C19 Yπ/2 Yπ/2 Xπ/2 Xπ/2
C20 Xπ/2 Yπ/2 Yπ/2 Yπ/2
C21 Yπ/2 Yπ/2 Yπ/2 Xπ/2
C22 Yπ/2 Yπ/2 Yπ/2 Xπ/2 Yπ/2
C23 Yπ/2 Xπ/2 Xπ/2 Xπ/2 Yπ/2
C24 Yπ/2 Xπ/2 Yπ/2 Yπ/2 Yπ/2

10
10.7. T WO -C OPY R ANDOMIZED B ENCHMARKING 171

10.7. T WO -C OPY R ANDOMIZED B ENCHMARKING

Here we derive the functional forms of two-copy randomzied benchmarking. This involves ran-
domized benchmarking by applying the same randomly selected Clifford to two qubits in the same
readout system, simultaneously. We are therefore interested in the representation structure of the
representation:
⊗2
R 2 (c)(ρ) = Uc⊗2 ρUc† (10.1)
where c indexes a Clifford operation, ρ is an input state density matrix defining a two-qubit system,
and Uc is the unitary representation of a given Clifford sequence indexed by c. The irreducible
subrepresentations of this representation are as follows:
n1 o
Vtr,1 = Span II (10.2)
2
n 1 o
Vtr,2 = Span p (X X + Y Y + Z Z ) (10.3)
2 3
n1 1 1 o
Vadj,1 = Span I X, IY, I Z (10.4)
2 2 2
n1 1 1 o
Vadj,2 = Span X I, Y I, Z I (10.5)
2 2 2
n 1 1 1 o
Vadj,3 = Span p (X Y − Y X ), p (Z Y − Y Z ), p (X Z − Z X ) (10.6)
2 2 2 2 2 2
n 1 1 o
V3 = Span p (X X − Y Y ), p (X X + Y Y − 2Z Z ) (10.7)
2 2 2 6
n 1 1 1 o
V4 = Span p (X Y + Y X ), p (Z Y + Y Z ), p (X Z + Z X ) (10.8)
2 2 2 2 2 2

Here, the subscripts t r , and ad j abbreviate trivial and adjoint respectively, as per the convention
in ref. [], while the subscripts 3 and 4 simply enumerate two more non-equivalent representations.
From these four representations, we can write down the functional form of such a protocol to be
n o
mª m
+ A 3 f 3m + A 4 f 4m
©
p(m) ≈ Tr A tr M tr + Tr A adj M adj (10.9)

where m is the length of a sequence, A tr , M tr are 2 × 2 real matrices (associated to the trivial repre-
sentations), A adj , M adj are 3×3 real matrices (associated to the adjoint representations), and A 3 , f 3
and A 4 , f 4 are real number associated to the third and fourth irreducible sub-representations re-
spectively. It follows from standard RB theory that the conventional average fidelity of the channel
Λ is given as:
4F avg (Λ) − 1 1 n o
= ((Tr{M tr } − 1) + 3 Tr M adj + 2 f 3 + 3 f 4 ) (10.10)
3 15
The prefactors A rep have matrix elements and real values given by 10
n o
[A rep ]i , j = |Vrep |−1 Tr P rep,i (E ) Tr P rep, j (ρ)
© ª
(10.11)
© ª
A 3 = (1/2) Tr{P 3 (E )} Tr P 3 (ρ) (10.12)
© ª
A 4 = (1/3) Tr{P 4 (E )} Tr P 4 (ρ) (10.13)

for i , j indexing the number of copies of a given representation rep, ρ ≈ |00〉 〈00| = (1/4)(I I + I Z +
Z I + Z Z ) is the prepared initial state, and we observe the POVM element E ≈ |00〉 〈00| = (1/4)(I I +
I Z + Z I + Z Z ).
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
172 THRESHOLD

10.7.1. S TRUCTURE OF THE MATRIX DECAYS

Two-copy RB differs from standard RB [23] and simultaneous RB [20] by the presence of matrix
valued decay rates (encoded by the matrices M tr and M adj ). To estimate the average fidelity we
are interested in learning the trace of these matrices. This can be done in practice by assuming
that the matrices M tr , M adj are diagonalisable (note that this is a rather weak assumption). Then
the matrix valued signal decomposes into a linear combination of exponentials with decay rates
set by the eigenvalues. This is worth addressing in more detail. We begin by treating M tr . If we
assume that Λ is a trace preserving quantum channel, then certain entries of M tr are fixed a priori.
We concretely get that
µ ¶
1 0
M tr = (10.14)
α f tr
where α and f 1 are real parameters. Note that only the diagonal parameter f 1 contributes to the
average fidelity. The case of M adj is more sophisticated, and has some novel features not earlier
observed in benchmarking experiments. By construction M adj is a 3 × 3 real matrix, which means
it will have at least 1 real eigenvalue f adj and two more eigenvalues that are either both real, or
complex numbers that are each others complex conjugate. The telltale sign of the appearance of
such a complex eigenvalue pair would be the appearance of damped oscillations in the 2CRB data.
Note from figure 10.5 that we do in fact observe such oscillations in certain 2CRB experiments.
Hence we will from now on assume that we are indeed in this scenario with two complex conjugate
eigenvalues λadj , λ̄adj and one real eigenvalue f adj .
To gain further insight into M adj it is worth making a natural extra assumption on the noise
channel Λ. In particular we will assume that Λ is invariant under the interchange of two qubits.
Examples of such channels include a global depolarizing channel and crosstalk due to exchange
interaction (on which more later). Through a straightforward calculation
p we see that if Λ is invari-
ant under the interchange of the two qubits, then v = (1, 1, 0)/ 2 is an eigenvector of M adj . Since
this eigenvector is real, thepassociated eigenvector must also be. Complementary, the subspace
spanned by w 1 = (1, −1, 0)/ 2, w 2 = (0, 0, 1) carries the pair of complex eigenvalues.

It is worth noting that this decomposition is robust against violations of the symmetry as-
sumption we made above, as long as the complex eigenvalues are sufficiently far from being real.
By this we mean that "the eigenvector carrying the real eigenvalue" will perturb continuously even
under a symmetry breaking perturbation. This means the analysis above holds broadly even for
quantum channels that do not respect the symmetry (as long as there are two complex eigenval-
ues). This is important as we expect the symmetry condition not to hold exactly in practice (for
instance if the two qubits have different local error rates).

To summarise, under some mild assumptions the average fidelity takes the form:

4F avg (Λ) − 1 1
= ( f tr + 3( f adj + λadj + λ̄adj ) + +2 f 3 + 3 f 4 ). (10.15)
3 15
10 Where all parameters can in principle be extracted from decay experiments. However, what decays
are visible depends on the choice of input state and output measurement (note that this not the
case in standard RB, making 2CRB more difficult to execute and analyse even beyond the complex-
ity of the representation structure). We have implemented several types of 2CRB, distinguished by
different input states and measurements. We now discuss these in greater detail.

10.7.2. D IFFERENT TYPES OF TWO - COPY RB

With the information collected above we can further discuss the types of 2CRB experiments dis-
cussed in the main text. We distinguish three types of 2CRB, each with different input states ρ and
10.7. T WO -C OPY R ANDOMIZED B ENCHMARKING 173

measurement operators E :

ρ ≈ |00〉〈00| = (1/4)(I I + I Z + Z I + Z Z ), E ≈ |00〉〈00| = (1/4)(I I + I Z + Z I + Z Z )

(even, same-system (ES))
ρ ≈ |00〉〈00| = (1/4)(I I + I Z + Z I + Z Z ), E ≈ |0〉〈0| ⊗ I = (1/2)(I I + Z I )
(even, different-system (ED))
ρ ≈ |01〉〈01| = (1/4)(I I − I Z + Z I − Z Z ), E ≈ |01〉〈10| = (1/4)(I I − I Z + Z I + Z Z ).
(odd, same-system (OS))

The logical fourth experiment (odd, different-system (OD)) was not performed, so we will not
analyse it here. With these expressions for E , ρ we can directly calculate what the pre-factors A
in equation 10.9 are in the ideal case for the three executed types of 2CRB:

p ¶

1 1 0

µ
1 1 1/ 3 1  1
A tr = p , A adj = 1 1 0 , A 3 = , A4 = 0 (ES)
4 1/ 3 1/3 12 12
0 0 0

1 0 0

µ ¶
1 1 0 1 
A tr = , A adj = 1 0 0 , A 3 = 0, A 4 = 0 (ED)
4 0 0 12
0 0 0
p ¶

1 −1 0

µ
1 1 −1/ 3 1  1
A tr = p , A adj = −1 1 0 , A 3 = , A4 = 0 (OS)
4 −1/ 3 1/3 12 12
0 0 0

The first thing to note here is that in all three experiments A 4 = 0 in the absence of SPAM errors.
Hence we do not expect to be able to reliably extract f 4 from the performed experiments. We will
now discuss each experiment in more detail:

10.7.3. E VEN SAME - SYSTEM (ES) TWO - COPY RB

As can be seen above, in the ES experiment the matrices A tr and A adj are rank one. Hence it is
not a priori clear we can reliably observe the full matrix decays (i.e. observe the eigenvalues of
the matrix M ). This is worth addressing in more detail. We begin by treating M tr . Recall that
under the assumption of gate-dependent trace preserving noise this matrix has a lower-triangular
form (given by equation 10.14) Note that only the diagonal parameter f 1 contributes to the average
fidelity. Using the expression for A tr for the ES experiment we see that

m¢ 1 ¡
p
= 1 + (1/3) f trm + α(1 − f trm )/( 3(1 − f tr ) = a f trm + b
¡ ¢
tr A tr M tr (10.16)
4

for some real constants a, b. Hence the trivial "sector" contributes only a single exponential decay
to the model in equation 10.9. Moreover, in the high fidelity limit the constant a will be approxi- 10
mately 1/12 for the
n ES experiment,
o which means that f 1 will typically be observable.
m . Under the symmetry assumption made above we know that v =
Next consider tr A adj M adj
p
(1, 1, 0)/ 2 is an eigenvector of M adj with real eigenvalue f adj . Note also that A adj = v T v in the
n o
m = f m . Hence in the absence of SPAM errors the ES
ES experiment, and thus that tr A adj M adj adj
experiment perfectly detects the one real decay in M adj , and does not detect the two conjugate
complex eigenvalues at all. Hence the presence of of oscillations in the even experiment are only
due to SPAM errors.
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
174 THRESHOLD

To summarize, in the ES 2CRB experiment, we expect the data to take the functional form

p(m)even ≈ a tr f trm + b tr + A 3 f 3m + a adj f adj

m
(10.17)

where a tr , b tr , A 3 , a adj are real numbers encoding SPAM.

10.7.4. O DD SAME - SYSTEM (OS) TWO - COPY RB

The discussion of the odd same-system (OS) 2CRB experiment is largely analogous. Considering
M tr first, we see that

m¢ 1 ¡
p
= 1 + (1/3) f trm − α(1 − f trm )/( 3(1 − f tr ) = a f trm + b
¡ ¢
tr A tr M tr (10.18)
4
for some real constants a, b. Again in the high fidelity limit a will be approximately 1/12 and we
thus expect good visibility for f tr . For M adj we note that in the OS type experiment we have A adj =
p
w 1T w 1 with w 1 = (1, −1, 0)/ 2. This means the OS experiment does not observe the real eigenvalue
f ad (as the corresponding eigenvector is orthogonal). Instead we have that (by the realness of the
overal signal) n o
m
tr A adj M adj = a adj (λm
adj + λ̄adj ) (10.19)
for some real parameter a adj . Thus in this experiment we expect a quite strong oscillating compo-
nent in the signal. This is borne out in practice, see [10.5].
To summarise, in the OS two-copy experiment we expect the data to take the functional form

p(m)odd ≈ a tr f trm + b tr + A 3 f 3m + a adj (λm m

adj + λ̄adj ) (10.20)

where a tr , b tr , A 3 , a adj are real numbers encoding SPAM.

10.7.5. E VEN DIFFERENT- SYSTEM (DS) TWO - COPY RB

Finally we discuss the case of ED 2CRB. This experiment has a significantly simpler decay structure
(but as a consequence also collects less information about error processes). Note first that
½ µ ¶µ ¶¾
© mª 1 1 0 1 0 1
tr A tr M tr = tr = (10.21)
4 0 0 α f tr 4

which means f tr is not visible in this experiment. Furthermore, we see that A adj ∝ (1, 0, 0)T (1, 1, 0),
and hence that n o
m m
tr A adj M adj = a adj f adj (10.22)
for some real parameter a adj .
To summarise, in the ED two-copy experiment we expect the data to take the functional form
m
p(m)odd ≈ b tr + a adj f adj (10.23)
10
which means the ED experiment leads, under mild assumptions on the noise channel and SPAM,
to a standard single exponential decay with offset.

10.8. T WO - COPY RB IN PRACTICE

In the previous section we discussed the general functional form of the different types of 2CRB,
and how the extracted decay rates relate to the fidelity. In this section we will discuss in more de-
tail how to extract these decay rates and also how to interpret them. There are three problems to
deal with here. The first is the question of extracting the decay rates from the data p(m). In the
10.8. T WO - COPY RB IN PRACTICE 175

next subsection we will discuss how to deal with this problem. The second problem is that none
of the three performed types of 2CRB give access to the decay rate f 4 . This means that even if we
perform all three experiments on the same pair of qubits we can not evaluate that average fidelity
directly. We can partially deal with this question by arguing that (which we do in section 10.8.1)
if one has access to all other decay rates ( f tr , f 3 , f adj , λadj , λ̄adj ) one can derive a very sharp lower
bound on the average fidelity proper. This allows us to give rigorous lower bounds on the average
fidelity in the case where both the ES and OS two-copy experiments were performed (see 10.5).

Finally we deal with the question of how to interpret the decay rates themselves in the scenario
where no rigorous statement on the average fidelity can be given (i.e. for qubit pairs where only
the ES or ED experiment have been performed).

10.8.1. L OWER B OUND ON AVERAGE F IDELITY

Here we will give a proof of the claim that we can give a lower bound on the average fidelity, despite
not being sensitive to the decay f 4 from Equation 10.15. This follows from the complete positivity
of the channel Λ. Concretely we assume we have access (through experiment), to the decay rates
f tr , f 3 , f adj , λadj , λ̄adj (i.e. only f 4 is missing). We will argue that f 4 is sharply lower bounded by the
average infidelity r (Λ) = 1 − F avg (Λ). Key to this is the following general lemma, which might be of
broader interest.

Theorem 1. Let P,Q be commuting n-qubit Pauli

ª operators and consider for a quantum channel
Λ the Pauli transfer matrix element 2n Tr P Λ(Q) . We have
©

n
¯δP,Q − 2−n Tr P Λ(Q) ¯ ≤ 2 2 + 1 r (Λ).
¯ © ª¯
(10.24)
2n

Proof. Since P,Q commute, they are diagonal in a joint basis {|x〉}x∈{0,1}n . Since the average fidelity
is invariant under unitary conjugation we can, without loss of generality, take this basis to be the
computational basis. Writing P |x〉 = p(x) |x〉 and Q |x〉 = q(x) |x〉 for the eigenvalues of P,Q and
noting that Tr{|x〉〈x| Λ(|x̂〉〈x̂|)} ∈ [0, 1]

Tr P Λ(Q) = p(x)q(x) Tr{|x〉〈x| Λ(|x〉〈x|)} + p(x)q(x̂) Tr{|x〉〈x| Λ(|x̂〉〈x̂|)}

© ª X X X

x∈{0,1}n x∈{0,1}n x̂∈{0,1}n , x̂6=x

(10.25)

We begin by bounding the second term. We have

¯ ¯ ¯ ¯
p(x)q(x̂) Tr{|x〉〈x| Λ(|x̂〉〈x̂|)}¯ ≤ ¯ Tr{|x̂〉〈x̂| Λ(|x〉〈x|)}¯
X X X
(10.26)
¯ ¯ ¯ ¯
¯
x∈{0,1}n x̂∈{0,1}n , x̂6=x x̂∈{0,1}n , x̂6=x
1 − Tr{|x〉〈x| Λ(|x〉〈x|)} ,
X ¡ ¢
= (10.27)
x̂∈{0,1}n
10
where we used that x̂ Tr{|x̂〉〈x̂| Λ(|x〉〈x|)} = 1. We can bound this directly directly by using Theo-
P

rem 1 from [36], which states that

1 − Tr{|x〉〈x| Λ(|x〉〈x|)} ≤ (2n + 1)r (Λ)

X ¡ ¢
(10.28)
x∈{0,1}n

For the first term we have two cases. If P = Q then p(x) = q(x) and thus p(x)q(x) = 1, which means
we can lower bound the first term by 2n − (2n + 1)r (Λ). On the other hand if P 6= Q we know that
there are exactly 2n−1 values of x s.t p(x)q(x) = 1 and 2n−1 values of x s.t p(x)q(x) = −1 (since P,Q
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
176 THRESHOLD

must now be trace orthogonal). Labeling these two sets X 1 , X −1 we see that

p(x)q(x) Tr{|x〉〈x| Λ(|x〉〈x|)} = Tr{|x〉〈x| Λ(|x〉〈x|)} − Tr{|x〉〈x| Λ(|x〉〈x|)}

X X X

x∈{0,1}n x∈X 1 x∈X −1

(10.29)
1 − Tr{|x〉〈x| Λ(|x〉〈x|)} − 1 − Tr{|x〉〈x| Λ(|x〉〈x|)}
X ¡ ¢ X ¡ ¢
= (10.30)
x∈X −1 x∈X 1
1 − Tr{|x〉〈x| Λ(|x〉〈x|)} + 1 − Tr{|x〉〈x| Λ(|x〉〈x|)}
X ¡ ¢ X ¡ ¢
≤ (10.31)
x∈X −1 x∈X 1
n
≤(2 + 1)r (Λ) (10.32)

Putting these two cases together we obtain the lemma statement.

With this lemma we can directly lower bound f 4 in terms of the average infidelity r (Λ):

1 ³ ´
f4 = Tr{(X Y + Y X )Λ(X Y + Y X )} + Tr{(X Z + Z X )Λ(X Z + Z X )} + Tr{(Z Y + Y Z )Λ(Z Y + Y Z )}
24
(10.33)
1³ 5 5 ´
≥ 6(1 − r (Λ)) − 6 r (Λ) (10.34)
6 2 2
≥ 1 − 5r (Λ) (10.35)

And thus by equation 10.15 we see that

4 1 ¡
f tr + 2 f 3 + 3( f adj + λadj + λ̄adj ) + 3 − 15r (Λ)
¢
1 − r (Λ) ≥ (10.36)
3 15
which after a rearranging of terms gives
³ 1 ¡ ¢´
r (Λ) ≤ 3 1 − f tr + 2 f 3 + 3( f adj + λadj + λ̄adj) + 3 . (10.37)
15
Note that this bound is asymptotically accurate, in the sense that if f tr = f 2 = f adj = λadj = 1 then
r (Λ) = 0.

10.8.2. S YMMETRIC AND A NTI - SYMMETRIC INPUT STATES FOR TWO - COPY
RB
We perform ES 2CRB on systems Q1Q2 (a) and Q3Q4 (b) for symmetric |↓↓〉 and anti-symmetric
|↑↓〉 input states. The resulting decays give us access to every eigenvalue required to estimate the
average fidelity of the system, bar one. However, as shown in section 10.8.1, we can place a rig-
orous lower bound on the average fidelity using Equation 10.37, assuming we have experimental
access to the remaining eigenvalues. Figures 10.5 a and b show 2CRB experiments on combined
10 systems Q1Q2 and Q3Q4 respectively. In these experiments, we perform two-copy randomzied
benchmarking for a symmetric input state |↓↓〉 (Yellow trace in (a), Cyan trace in (b)) and an anti-
symmetric input state |↑↓〉 (red in (a), blue in (b)). We find that fitting a single exponent with one
real decay rate to the symmetric input traces is sufficient, while for the anti-symmtric case, we
fit three exponentials comprising one real decay and two complex conjugate decays, that give us
access to decay rates λadj and λ̄adj from equation 10.15. In the case of the real decay rates, the
form of the symmetric input states in equation 10.17 should access all real decay rates, while for
the anti-symmetric case, there will be no contribution to the decay from f adj . We also have no a-
priori reason to know which decay is represented by which eigenvalue. However, by choosing the
lower of the real decays for substitution into equation 10.37, we can be certain that we are lower
10.9. D IRECT COMPARISON OF TWO - COPY BENCHMARKING AND TWO - QUBIT
SIMULTANEOUS RANDOMIZED BENCHMARKING 177

(a) (b)

π/2 π/2
Fπ/2= 99.43 % Fπ/2= 99.81 %
FL = 98.03 % FL = 99.01 %
N-1 N-1

X2 C C-1 X2 X2 C C-1 X2
N-1 N-1

C C-1 C C-1

Figure 10.5: ES 2CRB of qubits Q1 and Q2 (a) and qubits Q3 and Q4 (b) at Bext = 0.65 T for symmetric input
state |↓↓〉 ((a),yellow and (b),cyan) and anti-symmetric input state |↑↓〉 ((a),red and (b),Blue). To prepare the
antisymmetric space, we apply a X π pulse to flip Q1(Q3) in the Q1Q2(Q3Q4) system before performing ran-
domized benchmarking, then applying a second X π pulse to the same qubit to project the final system back
to the | ↓↓> state. A rigorous lower bound on the fidelity Fπ/2
L
can be obtained from the extracted decay rates
of both symmetric and anti-symmetric experiments. The value Fπ/2 represents the average native gate fidelity
calculated using 10.15 assuming f 4 = 0.

bounding the fidelity. Tables 10.2 and 10.3 show the extracted decay rates for systems Q1Q2 and
Q3Q4 respectively. By taking the lower of the real decays extracted from the symmetric and anti-
π/2 ≥ 98.03%
symmetric starting state cases, we find lower bounds on the average fidelities of FQ1Q2
π/2 ≥ 99.01%.
and FQ3Q4

10.9. D IRECT COMPARISON OF TWO - COPY BENCHMARKING AND

TWO - QUBIT SIMULTANEOUS RANDOMIZED BENCHMARK -
ING
10
In this section we provide numerical evidence, beyond the formal bounds given above, that the
decays numbers extracted from ES and ED 2CRB are representative of the average fidelity. We do
this by directly comparing 2CRB (see section 10.7 to standard simultaneous randomized bench-
marking (SRB) [20].

One drawback of SRB is the requirement that all Clifford gates are required to be identical
in duration in order to avoid synchronization problems of simultaneous Clifford operations. In
general these problems can be avoided by extending short Clifford gate sequences with an idling
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
178 THRESHOLD

Input State Decay Value

|↓↓〉 f t r , f ad j , f 3 0.9722
|↑↓〉 f t r , f3 0.9798
|↑↓〉 λad j 0.9200+0.2779i
|↑↓〉 λ̄ad j 0.9200-0.2779i

Table 10.2: Measured decay rates for system Q1Q2, for symmetric and anti-symmetric input states in the two-
copy benchmarking experiment presented in Figure 10.5a.

Input State Decay Value

|↓↓〉 f t r , f ad j , f 3 0.9709
|↑↓〉 f t r , f3 0.9770
|↑↓〉 λad j 0.9715+0.1110i
|↑↓〉 λ̄ad j 0.9715-0.1110i

Table 10.3: Measured decay rates for system Q3Q4, for symmetric and anti-symmetric input states in the two-
copy benchmarking experiment presented in Figure 10.5b.

time such that the duration of all Clifford gate sequences are equal. Fortunately, elementary gates
based on single-qubit π/2-microwave pulses with individual phase control allow the construction
of all 24 single-qubit Clifford gates using two "real" and a up to two "virtual" Z -gates (see Ref. [3]
and table 10.4). We use this compilation strategy for a direct comparison of 2CRB and standard
SRB. For standard two-qubit SRB, random single-qubit Cliffords are chosen with the important
caveat that each Clifford gate is chosen independently of the other qubit. We note here that the
discussion regarding the simulation is independent of the Clifford compilation as derived above.

10.9.1. S IMULATION OF THE C LIFFORD GATE SUPEROPERATOR

In order to efficiently yet accurately simulate the two protocols, we first compute the full time-
evolution superoperator for each combination of single-qubit Clifford gates on two qubits assum-
ing quasistatic noise. Specifically, we solve the time-dependent Liouville–von Neumann equation
for the density matrix ρ

d
|ρ〉 = H (t )ρ − ρH (t ) ,
£ ¤
iħ (10.38)
dt
where ħ = h/(2π) is the reduced Planck constant. Here, the system consisting of two germanium
hole spin qubits is described by the Hamiltonian following [19, 37]

H (t ) = µB B 1 · S 1 + µB B 2 · S 2 + S 1 · J S 2 . (10.39)
10 Here, J is the tensorial form of the exchange interaction between qubits Q1 and Q2, µB is the
Bohr magneton, h the Planck constant and S j = (σx , σ y , σz )T /2 is the vector consisting of the
spin matrices acting on Qj. The effective magnetic field experienced by Qj is given by

B 1 = (bQ1, f Q1 cos 2π f Q1 t + φQ1 + b Q1, f Q2 cos 2π f Q2 t + φQ2 , 0, g Q1 B z )T

¡ ¢ ¡ ¢
(10.40)
T
B 2 = (bQ2, f Q1 cos 2π f Q1 t + φQ1 + b Q2, f Q2 cos 2π f Q2 t + φQ2 , 0, g Q2 B z )
¡ ¢ ¡ ¢
(10.41)

which consists of the external magnetic field B z multiplied by the in-plane g-factor g Qj as well
as contributions due to the spin-orbit interaction. Assuming an isotropic exchange interaction,
10.9. D IRECT COMPARISON OF TWO - COPY BENCHMARKING AND TWO - QUBIT
SIMULTANEOUS RANDOMIZED BENCHMARKING 179

µB g Qj
J j k = J j k 13 and driving frequencies f Qj = h B z À J the dominating contribution of the ex-
change interaction is of the form H J = +S 1 · J S 2 ≈ − 4J ZZ(see Ref. [38]). Single-qubit rotations
µB g Qj µ
are implemented via resonant driving f Qj = h B z with a Rabi frequency ΩQj = 2hB bQ j , f Qj . To
utilize relevant parameters to the present experiment, we set ΩQ1 = ΩQ2 = 10.6 MHz. We sim-
ulate crosstalk such that each qubit is affected by both driving fields (one resonant and one is
off-resonant) via the amplitudes bQ j , f Q1 and bQ j , f Q2 . To simplify our simulations and maximize
the crosstalk effect we set bQ j , f Qk = 10.6 MHz for j , k = 1, 2.
We solve Eq. (10.38) by iteratively computing the unitary propagator according to
i
U (t + ∆t ) = e − ħ H (t +∆t )U (t ). (10.42)

We discretize H (t +∆t ) into N segments of length ∆t such that H (t ) is constant in the time-interval
[t , t + ∆t ]. For all simulations, we choose ∆t = 10 ps. Furthermore, we include quasistatic noise
in the simulations by introducing classical fluctuations of fQ1 → fQ1 + δ fQ1 , fQ2 → fQ2 + δ fQ2
which are assumed to be static on the time-scale of a single Clifford gate. The fluctuations follow
a Gaussian distribution with mean µ = 0, standard deviation σδ fQ j and are estimated from the
corresponding measured dephasing times 10.1 T2,Qj ? = (2pπσ −1 if not otherwise mentioned.
δ fQ j )
To estimate the noisy density matrix we perform 250 noise implementations for each Clifford gate.
In order to find the noisy superoperator, we repeat the simulations for a full basis set of initial
density matrices ρ init . The noisy superoperator is then constructed via the linear system

ρ final,1 ρ init,1
   
ρ ρ
 init,2 
 
 final,2 

 .. 
 = L  ..  ,
  (10.43)
 .   . 
ρ final,16 ρ init,16

where L is the 16 × 16-dimensional superoperator. The noisy superoperator in Liouville form is

then constructed by transforming into the basis ρ 1 = |0〉 〈0| , ρ 2 = |0〉 〈1| , ρ 3 = |0〉 〈2| , · · · , ρ 16 = |4〉 〈4|
following row ordering.

10.9.2. S IMULATION OF THE CLIFFORD EXPERIMENTS

With the superoperator in Liouville form of each Clifford gate, we can now model a randomized
benchmarking experiment. We model each randomized benchmarking protocol using the follow-
ing procedure.
1. Create a sequence of N -Cliffords chosen randomly on both qubits, and add the recovery
Clifford to the sequence. For the standard two-qubit SRB protocol, Cliffords are picked at
random and are uncorrelated between qubits. For the 2CRB protocol, the same Clifford is
applied to both qubits. Repeat this K times.
2. Compute the final superoperator via multiplication of the corresponding two-qubit super-
operator in Liouville form. This allows us to efficiently project onto the initialization and 10
measurement at a later stage.
3. Chose an initial state and a measurement POVM to compute the expectation value for each
sequence. Average over all K implementations.
4. Fit the final data.

10.9.3. C OMPARISON BETWEEN THE RESULTS

Figures 10.6a,b and 10.6c,d show results for the simulated randomized benchmarking experiments
for ES two-copy and simultaneous randomized benchmarking. It is clear from the study varying
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
180 THRESHOLD

(a) (b)

x10--2

J = 1 kHz

J = 10 kHz
P|↓↓〉

J = 100 kHz

J = 1000 kHz

J = 10000 kHz

Nc
(c) (d)

Data, SimRB
Data, 2CRB
T2,RB
= 5 µs
Fit, SimRB
Fit, 2CRB

T2,RB = 2 µs
P|↓↓〉

T2,RB = 1 µs

T2,RB = 0.5 µs

Nc

Figure 10.6: Full Hamiltonian simulations comparing two-copy randomzied benchmkaring and simultane-
10 ous Randomzied benchmarking. (a) Effect of varying exchange for 2CRB (red) and SRB (black). Exchange is
varied between 1 kHz and 10 MHz. For J = 10 MHz, single qubit gates become controlled rotations, resulting in
a deviation from the characteristic decay form. (b) Summary of exponential fit Clifford fidelity for (a). For low
values of J, 2QRB and SRB give fidelities within error bars of each other, while at higher J, 2CRB returns higher
fidelities. (c) Effectively varying cross talk levels by changing the qubit dephasing times in the simulation, for
J = 260 kHz, reflecting the expected residual exchange for qubit system Q1Q2. (d) Comparison of extracted
Clifford fidelities for 2CRB and SRB at different dephasing times. While always returning slightly lower error
rates, 2CRB appears to provide a good estimate of the average system fidelity.
10.9. D IRECT COMPARISON OF TWO - COPY BENCHMARKING AND TWO - QUBIT
SIMULTANEOUS RANDOMIZED BENCHMARKING 181

classical cross talk noise via effective qubit coherence T 2,RB in Figures 10.6c,d that ES 2CRB re-
turns a higher fidelity than SRB as expected. However, in the limit of low exchange interaction,
these fidelities can be very close, making ES 2CRB a good approximation to the fidelity of the sys-
tem in our experiments.

Table 10.4: Single-qubit clifford compilation for a direct comparison between two-copy randomized bench-
marking and simultaneous randomized benchmarking without idling times. We simulate both benchmarks by
selecting a random sequence of Cliffords from the table below excluding C1 , and calculate the recovery Clifford
that projects the system back into it’s original state. We use here a gate set containing ±π/2 rotations around
the Bloch Sphere, so the gates X±π/2 and Y±π/2 are referring to explicit rotations of ±π/2 around the x-axis and
y-axis of the Bloch sphere of a single-qubit. Z±π/2 are rotations around the z-axis and implemented "virtually"
via software correction of the signal generator. In this respect a Yπ/2 rotation is implemented identical to a
Xπ/2 rotation with a π/2 software phase correction.

Clifford Composition
C1 Xπ/2 X−π/2
C2 Xπ/2 Xπ/2
C3 Y−π/2 Y−π/2
C4 Xπ/2 X−π/2 Zπ/2 Zπ/2
C5 Xπ/2 Y−π/2
C6 Xπ/2 Yπ/2
C7 X−π/2 Y−π/2
C8 X−π/2 Yπ/2
C9 Y−π/2 Xπ/2
C10 Y−π/2 X−π/2
C11 Yπ/2 Xπ/2
C12 Yπ/2 X−π/2
C13 Y−π/2 Xπ/2 Z−π/2
C14 Y−π/2 X−π/2 Zπ/2
C15 Xπ/2 Y−π/2 Zπ/2
C16 Xπ/2 Yπ/2 Z−π/2
C17 X−π/2 Xπ/2 Zπ/2
C18 X−π/2 Xπ/2 Z−π/2
C19 Xπ/2 Y−π/2 Z−π/2
C20 Xπ/2 Yπ/2 Zπ/2
C21 Y−π/2 Xπ/2 Zπ/2
C22 Y−π/2 X−π/2 Z−π/2
C23 Xπ/2 Xπ/2 Zπ/2
C24 X−π/2 X−π/2 Z−π/2 10
 10. S IMULTANEOUS DRIVING OF SEMICONDUCTOR SPIN QUBITS AT THE FAULT- TOLERANT
182 THRESHOLD

10.10. S UMMARY OF N - COPY BENCHMARKING DATA

Table 10.5: Extracted fidelities from figure 10.3a-b, a 2CRB experiment as well as single qubit RB and four
qubit simultaneous driving error rates and leakage rates (when applicable) reported (in %). Qubits measured
simultaneously in the same readout system (ES, see 10.7.2) have different measurement operators to those
measured in different systems (DS), so the reported number is a combined system fidelity (FQ1Q2 = FQ2Q1 and
FQ3Q4 = FQ4Q3 in the table). For this dataset, tπ = 96 ns. Leakage rates LQ i |Q j are reported in the case of single
qubit RB or DS 2CRB when the decay plateau approaches the expected value of the fully depolarized two-qubit
subspace, and are taken to be the longer of the two decay constants (see methods 10.4.3). The quoted errors
correspond to the fitting error of the traces.

Q1 Q2 Q3 Q4
1-F1Q 0.0329 ± 0.0042 0.3155 ± 0.0310 0.0308 ± 0.0019 0.3224 ± 0.0446
L1Q - - - 0.0153 ± 0.0041
1-FQ i |Q1 - 0.4018 ± 0.0130 0.0249 ± 0.0022 0.3087 ± 0.0690
1-FQ i |Q2 0.4018 ± 0.0130 - 0.3791 ± 0.0300 0.4692 ± 0.0978
1-FQ i |Q3 0.0301 ± 0.0048 0.3445 ± 0.0294 - 0.2778 ± 0.0298
1-FQ i |Q4 0.2676 ± 0.0565 0.7324 ± 0.0388 0.2778 ± 0.0298 -
LQ i |Q1 - - - 0.0175 ± 0.0065
LQ i |Q2 - - 0.0210 ± 0.0034 0.0741 ± 0.0213
LQ i |Q3 - 0.0195 ± 0.0068 - -
LQ i |Q4 0.0273 ± 0.0092 - - -
1-F4Q 9.8565 ± 0.9031 2.0007 ± 0.0428

(a) (b)

10-3 10-3

7 4

3.5
6
3
5
2.5
1-Fπ/2

4 2

1.5
3
1
2
0.5

1 0

10 Figure 10.7: Table of infidelities and relative fidelity losses of the 2CRB experiment reported in Table 10.5 and
Figure 10.3a-b. The plots can be interpreted by considering the qubit in question to be the column, and the
qubit to be driven with to be the row. Qubits are represented diagrammatically by a coloured sphere, while the
plunger gate from which they are driven is depicted above them by a coloured cylinder. (a) Absolute infidelity
of the simultaneously driven qubits. Values on the diagonals represent the single-qubit fidelity in the non-
simultaneously driven space. (b) relative fidelity loss δ of each qubit pair. Diagonals have no significance and
are set to 0.
R EFERENCES 183

R EFERENCES
[1] Terhal, B. M. Quantum error correction for quantum memories. Reviews of Modern Physics
87, 307–346 (2015).

[2] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise and fidelity
higher than 99.9%. Nature Nanotechnology 13, 102–106 (2018).

[3] Xue, X. et al. Benchmarking Gate Fidelities in a Si/SiGe Two-Qubit Device. Physical Review X
9, 021011 (2019).

[4] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 1–10 (2017).

[5] Scappucci, G. et al. The germanium quantum information route. Nature Reviews Materials
(2020).

[6] Van Riggelen, F. et al. A two-dimensional array of single-hole quantum dots. Applied Physics
Letters 118, 44002 (2021).

[7] Ansaloni, F. et al. Single-electron operations in a foundry-fabricated array of quantum dots.

Nature Communications 11, 1–7 (2020).

[8] Fowler, A. G., Mariantoni, M., Martinis, J. M. & Cleland, A. N. Surface codes: Towards practical
large-scale quantum computation. Physical Review A - Atomic, Molecular, and Optical Physics
86, 032324 (2012).

[9] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 (2015).

[10] Takeda, K. et al. Quantum tomography of an entangled three-spin state in silicon. arXiv
2010.10316 (2020).

[11] Sigillito, A. J. et al. Site-Selective Quantum Control in an Isotopically Enriched Si 28

/Si0.7Ge0.3 Quadruple Quantum Dot. Physical Review Applied 11, 061006 (2019).

[12] Watzinger, H. et al. Ge hole spin qubit. Nature Communications 9, 1–6 (2018).

[13] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit
logic with holes in germanium. Nature 577, 487–491 (2020).

[14] Hendrickx, N. W. et al. A single-hole spin qubit. Nature Communications 11, 1–6 (2020).

[15] Lodari, M. et al. Low percolation density and charge noise with holes in germanium (2020).
URL https://doi.org/10.1088/2633-4356/abcd82.

[16] Sammak, A. et al. Shallow and Undoped Germanium Quantum Wells: A Playground for Spin
10
and Hybrid Quantum Technology. Advanced Functional Materials 29, 1807613 (2019).

[17] Lawrie, W. I. et al. Spin Relaxation Benchmarks and Individual Qubit Addressability for Holes
in Quantum Dots. Nano Letters 20, 7237–7242 (2020).

[18] Jirovec, D. et al. A singlet triplet hole spin qubit in planar Ge. arXiv 2011.13755 (2020).

[19] Hendrickx, N. W. et al. A four-qubit germanium quantum processor. Nature 591, 580–585
(2021).
 184 R EFERENCES

[20] Gambetta, J. M. et al. Characterization of addressability by simultaneous randomized bench-

marking. Physical Review Letters 109 (2012).

[21] Lawrie, W. I. L. et al. Quantum Dot Arrays in Silicon and Germanium. Applied Physics Letters
116, 080501 (2020).

[22] Harvey-Collard, P. et al. High-Fidelity Single-Shot Readout for a Spin Qubit via an Enhanced
Latching Mechanism. Physical Review X 8, 021046 (2018).

[23] Knill, E. et al. Randomized benchmarking of quantum gates. Physical Review A - Atomic,
Molecular, and Optical Physics 77 (2008).

[24] Cerfontaine, P. et al. Closed-loop control of a GaAs-based singlet-triplet spin qubit with 99.5%
gate fidelity and low leakage. Nature Communications 11 (2020).

[25] Wang, K. et al. Ultrafast Operations of a Hole Spin Qubit in Ge Quantum Dot. arXiv
2006.12340 (2020).

[26] Froning, F. N. et al. Ultrafast hole spin qubit with gate-tunable spin–orbit switch functionality.
Nature Nanotechnology 16, 308–312 (2021).

[27] Crippa, A. et al. Electrical Spin Driving by g -Matrix Modulation in Spin-Orbit Qubits. Physical
Review Letters 120, 137702 (2018).

[28] Scarlino, P. et al. Second-Harmonic Coherent Driving of a Spin Qubit in a Si/SiGe Quantum
Dot. Physical Review Letters 115, 106802 (2015).

[29] Li, R. et al. A crossbar network for silicon quantum dot qubits. Science Advances 4, 3960–3966
(2018).

[30] Bulaev, D. V. & Loss, D. Electric dipole spin resonance for heavy holes in quantum dots. Phys-
ical Review Letters 98, 097202 (2007).

[31] Terrazos, L. A. et al. Theory of hole-spin qubits in strained germanium quantum dots. Physi-
cal Review B 103, 125201 (2021).

[32] Yang, C. H. et al. Silicon qubit fidelities approaching incoherent noise limits via pulse engi-
neering. Nature Electronics 2, 151–158 (2019).

[33] Barends, R. et al. Superconducting quantum circuits at the surface code threshold for fault
tolerance. Nature 508, 500–503 (2014).

[34] Liles, S. D. et al. Electrical control of the g -tensor of the first hole in a silicon MOS quantum
dot. arXiv 2012.04985 (2020).

10 [35] Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Na-
ture 574, 505–510 (2019).

[36] Wallman, J. J. Bounding experimental quantum error rates relative to fault-tolerant thresh-
olds (2015).

[37] Hetényi, B., Kloeffel, C. & Loss, D. Exchange interaction of hole-spin qubits in double quan-
tum dots in highly anisotropic semiconductors. Phys. Rev. Research 2, 033036 (2020).

[38] Meunier, T., Calado, V. E. & Vandersypen, L. M. K. Efficient controlled-phase gate for single-
spin qubits in quantum dots. Phys. Rev. B 83, 121403 (2011).
 11
Q UANTUM C OHERENCE IN
G ERMANIUM H OLE S PIN Q UBITS
"...and so do all who live to see such times. But that is not for them to decide. All we have to decide
is what to do with the time that is given us."
Lord of the Rings: The Fellowship of the Ring

To operate a quantum computer, quantum information needs to be stored reliably [1]. This ability
is compromised when qubits exchange energy with unaccounted for external systems. Quantum
information loss can be quantified by two timescales: relaxation (T1 ) and dephasing (T2 ) rates of
a spin state. For hole spin qubits in planar germanium (Ge/SiGe) [2], T1 is not currently a limiting
timescale, with relaxation rates exceeding 32 ms [3]. However, pure dephasing times (T∗ 2 ) have never
exceeded a microsecond [4–6]. Here we study the magnetic field dependence of T∗ 2 for hole spin qubits
between 0.1 T and 1 T. We find charge noise to be the dominant decohering process in this regime,
and observe T∗ 2 as high as 1.9(1) µs at low magnetic field, extendable to 504(10) µs using refocussing
techniques. Furthermore, we study the effect of the 73 Ge nuclear spin population, discovering that
hole spin qubits in our device are sensitive to the nuclear spin bath. By developing a model to explain
the decohering behaviour, we extract a nuclear quadrupole splitting for the 73 Ge nuclei of 4.17 kHz,
and total hyperfine coupling strength of 34.4 kHz that we predict to limit spin dephasing times to
T∗2,hf
= 6.54 µs. These results simultaneously represent a milestone in achievable spin coherence in
hole spin qubits, while also demonstrating their sensitivity to hyperfine coupling, providing concrete
evidence of the necessity of isotopic purification.

185
 186 11. QUANTUM C OHERENCE IN G ERMANIUM H OLE S PIN QUBITS

11.1. I NTRODUCTION
Hole spin qubits in Ge/SiGe are leading candidates for quantum information processing [2]. Crit-
ical improvements in material growth [7], fabrication [8, 9] and tuning [10] have led to four-qubit
quantum logic [4] and single qubit fidelities approaching 99.99% [11], marking the current state of
the art for spin qubits in quantum dots. Hole spin qubits are attractive due to their strong spin-
orbit interaction, facilitating a fast, all-electrical driving mechanism [12–14] that circumvents the
requirement of microwave striplines or nanomagnets [15]. The implication of this relaxed req-
uisite, alongside the low effective mass [16], absence of valley degeneracy, high material qual-
ity [17] and compatibility with existing complimentary semiconductor metal oxide technology,
is that hole spin qubits in Ge/SiGe have an inherent scalability advantage over contemporary spin
qubit platforms.
However, as with any mechanism that couples spin to charge, the spin-orbit interaction has
consequences for coherence. The motion of charge present in all semiconductor devices can cou-
ple to the hole spin, resulting in dephasing. Furthermore, the effect of this noise becomes stronger
at higher magnetic fields [12, 18], resulting in a trade-off between enhanced single qubit drive
speeds and dephasing [11]. For hole spins in III-V heterostructure hosts however, nuclear spin
dynamics dominate hole spin coherences up until very high magnetic fields [19, 20], and limit de-
phasing times to tens of ns [21, 22]. In contrast, since the concentration of spin-free isotopes in
natural Ge is 92.24%, the decoherence due to nuclear spins is expected to be significantly weaker.
Additionally, the hyperfine interaction of hole spins is highly anisotropic, resulting in an Ising-type
coupling in the growth direction [23, 24]. Therefore, applying an in-plane magnetic field to break
spin degeneracy strongly suppresses hyperfine interactions. This allows for lower magnetic field
operation which reduces coupling to charge noise and increases coherence [6].
In the present work, we explore the quantum coherences of hole spin qubits defined in a Ge/SiGe
heterostructure. We observe a clear inverse relationship between T∗ 2 and magnetic field strength,
indicating that charge noise is limiting T2∗ . We measure unprecedented dephasing times for hole
spin qubits of T2∗ = 1.9(1) µs at the lowest magnetic field strength. We extend these coherences
up to T2CPMG = 504(10) µs using Carr-Purcell-Meiboom-Gill (CPMG) pulses, We also use CPMG
pulse sequences as filter functions [25, 26], to sense the 73 Ge Lamor procession. Furthermore,
a model is developed that characterizes the decohering behaviour, from which we estimate the
hyperfine coupling strength as well as the nuclear quadrupolar splitting of the 73 Ge nuclear spin.
An effective hyperfine limited dephasing time is estimated to be remarkably close to the longest
measured T2∗ . This demonstrates the necessity of isotopic purification, marking a turning point
for germanium hole spin qubits.

11.2. R ESULTS
The device shown in Figure 11.1a is a 2x2 quantum dot array of heavy hole spin qubits (see sec-
tion 11.4.1). We operate in the (1,1,1,1) charge state such that exactly one hole is present in each
quantum dot. The charge state of the device is measured by monitoring the electrostatic poten-
tial using two radio-frequency single hole transistors (rf-SHT). Spin readout is performed via a
spin-to-charge conversion technique based on Pauli spin blockade (PSB). Single-shot readout is
performed by defining a threshold discriminating between the blocked and unblocked PSB read-
out signal detected by the rf-SHT (see section 11.4.3). The natural germanium quantum well (Fig.
11 11.1b) contains a 7.76% 73 Ge isotopic content, with nuclear spin number I = 9/2. Unlike elec-
trons, hole wavefunctions confined within a quantum well are modulated by Bloch wavefunctions
with p-type symmetry, which have nodes at nuclear sites. This results in a vanishing contact hy-
perfine interaction term. Additional hyperfine interactions arise from the dipole-dipole coupling
between the hole spin (S) and nuclear spin (I ), and from coupling between the hole angular mo-
11.2. R ESULTS 187

(a) (b) (c) (d) (e)

P2

[001]
Si0.2Ge0.8
Ω
Q1 Q2 Ω
P3

B[001]

B[001]
P1 Ge ωres (t) = Ω + δΩ(t)

ω
Q4 Q3
Bnuc
Bext
Si0.2Ge0.8 θ’ Bnuc, Ω
P4 θ
[100] Bext B[100] B[100]

Figure 11.1: Device and hole spin noise mechanisms (a) False coloured scanning electron microscope image
of the device used in this study. Plungers P1-P4 (blue) tune the potential such that there is a single hole oc-
cupying each quantum dot, Each hole constitutes a qubit, labelled Q1 (orange), Q2 (yellow) Q3 (purple) and
Q4 (green). Scale is 100 nm. (b) Cross sectional sketch of the heterostructure. The blue rectangle represents
a plunger gate, above a purple Si0.2 Ge0.8 spacer layer. The red region indicates the Ge quantum well. A po-
tential well created by applying a negative voltage to the plunger gate traps a single hole spin (blue). Ge in its
natural form contains 7.76 % abundance of non-zero nuclear spins. (c) Schematic (not to scale) illustrating the
effect of an out-of-plane misalignment of the magnetic field Bext of angle θ. An angle θ 0 represents an effective
misalignment angle, reflecting the large anisotropy of the g -factor. Ω represents the effective quantization axis
due to the misaligned magnetic field. (d) Fluctuations of the Overhauser field Bnuc give rise to Zeeman noise
via the projection of the out-of-plane component Bnuc,⊥ onto Ω, Bnuc,Ω . (e) Electric field fluctuations due to
local charge noise couples to the g-tensor via the spin-orbit interaction.

mentum (L) and I . Furthermore, random flip-flop events between nuclei lead to a fluctuating
Overhauser field B nuc experienced by the hole spin [27]. A magnetic field Bext applied in the plane
of the quantum well yields qubit g-factors of 0.16, 0.22, 0.22, 0.24 for qubits Q1-Q4 respectively
(See section 11.5.1). We allow for an angle θ describing an out-of-plane misalignment in the mag-
netic field, resulting in two important consequences. First, the effective quantization axis of the
qubit is no longer purely in plane. Secondly, due to the anisotropies of the hyperfine coupling
0
and the g-tensor [14], small values of θ result in enhanced ¢ hyperfine interaction strengths. We
define an effective misalignment angle θ 0 = arctan g ⊥ /g || tan(θ) resulting in an effective quanti-
¡
0 0
zation axis Ω̂ = (sin θ , 0, cos θ ). The spin orbit interaction also facilitates a coupling of the g -factor
of the hole spin and electrical noise, which can induce Zeeman noise by displacing or distorting
the wavefunction of the qubit (Fig. 11.1e) [28]. The effect of these noise sources on the quantum
coherence of hole spin qubits is studied in the following sections.

11.2.1. Q UANTUM C OHERENCES

We characterize the magnetic field dependence of the dephasing rates of the four hole spin qubits
in our device. To extract T∗ 2 , we apply a X π/2 -pulse, wait for a time τ, and finally apply a second
X −π/2 -pulse, resulting in a decay characterized by a time T∗ 2 . Figure 11.2a-d shows the extracted
values of T∗ 2 as a function of magnetic field strength Bext between 0.1 T and 1 T. We observe a
clear T∗ ∗
2 ∝ 1/Bext dependence of the dephasing time. The longest T2 = 1.9(1) µs is extracted at
Bext = 0.1 T for qubit Q3. This behaviour is consistent with the expectation that at larger magnetic
fields, the dephasing is dominated by charge noise [12, 18]. Since the g-tensor of hole qubits de-
pends strongly on electric fields [5, 29], and that charge noise primarily couples to the qubit via
it’s electric field gradient, we deduce that charge noise is the dominant decohering mechanism for
Bext > 0.1 T.
The 1/f dependence of the charge noise spectral density permits the use of CPMG pulse sequences
11
to extend the quantum coherence of the qubits. In such a sequence, a qubit is prepared in the spin
down state, then brought to the equator of the Bloch sphere by applying a X π/2 pulse. In this su-
perposition state, the qubit will undergo random phase jumps as a result of environmental noise,
and gradually lose phase coherence. This can be observed by waiting a time t , before projecting
 188 11. QUANTUM C OHERENCE IN G ERMANIUM H OLE S PIN QUBITS

(a) (b)
2000 2000

1500 1500
T *2 (ns)

1000 1000

500 500

0 0
(c) 0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1
(d)
2000 2000

1500 1500
T *2 (ns)

1000 1000

500 500

0 0
0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1
(e) B (T) B (T)

500 B = 0.1 T
( s)

200
100
T CPMG

50
2

20
100 200 500 1000
N

Figure 11.2: Quantum Coherence of hole spin qubits in a 2x2 array as a function of magnetic field. (a-d)
Ramsey pulse sequences are applied to each qubit Q1 (red), Q2 (yellow) Q3 (purple) and Q4 (green) at mag-
netic fields between Bext = 0.1 T and Bext = 1 T. Each qubit shows a T∗ 2 ∝ 1/B dependence. Long dephasing
times of T∗2 = 1.9(1) µs are observed for qubit Q3 at a field of Bext = 0.135 T. (e) Extending quantum coherence
using a CPMG pulse sequence at external magnetic field strength Bext = 0.1 T. Visibility after Nπ = 1000 refo-
cusing pulses vanishes, however coherence can be extended to TCPMG 2 = 504(10) µs for qubit Q3. Error bars
correspond to the 95% confidence intervals of the individual fits.
11
11.2. R ESULTS 189

Xπ/2 Yπ Nπ X−π/2 (c)

(a)
Init.
n=1 n=2 n=3
0.8
τ/2 τ/2

1-P Blocked
0.7
τ
1

0.6
S(t)

AMW
τπ t
-1 CPMG-10
0.5
τ 0 2 4 6 8 10 12
(d) (e)
(b) 100 8
n=1 n=2 n=3 Hahn 1.5 MHz/T
CPMG-2 0.45
6
CPMG-10

f dip (MHz)
|F| 2 /(N ) 2

CPMG-50

t dip ( s)
0.4
4

0.35
2

0.3
CPMG-10 CPMG-10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 0.2 0.25 0.3 0.2 0.25 0.3
B (T) B (T)

Figure 11.3: Probing the 73 Ge nuclear spin precession using CPMG-Nπ filter functions (a) CPMG-Nπ pulse
sequence. A qubit is initialized in the spin down state, then brought to the cardinal +y axis of the Bloch sphere
by applying a resonant X π/2 pulse. Spin and charge Noise cause the qubit to gradually lose phase coherence.
For quasi-static noise, the application of a Yπ pulse will invert the polarity s(t ) of the noise acting on the qubit,
resulting in a recovery of phase coherence at a time τ/2. Subsequent applications of a Yπ pulse will further
increase the coherence of the qubit. A final X −π/2 brings the qubit back to the spin down state. (b) CPMG-Nπ
transmission functions for various Nπ . |F |2 /(Nπ τ)2 for a single (Hahn echo), 2, 10 and 50 applied refocusing
pulses is shown. Transmission maxima appear at ωτ = (2n − 1)/2 for harmonic n. The first three harmonic
positions are shown by dashed blue (n=1), orange (n=2) and yellow (n=3) lines. (c) An exemplary trace, showing
the decay of qubit coherence for Q3 undergoing a CPMG-10 refocussing pulse sequence at Bext = 0.23 T. Clear
dips in coherence are seen in the measured probability of a blocked state 1-PBlocked , as marked by blue, red
and yellow dashed lines. The frequencies corresponding to these dips are the first three harmonics of the 73 Ge
nuclear spin procession frequency respectively. The temporal axis is defined t̃ = τ + τπ , where the Rabi period
of the qubit is taken into account. (d) Magnetic field dependence on the temporal occurrence of each peak. (e)
Corresponding frequency of each peak in (d), converted via the relation f dip = (2n − 1)/2t dip where n = 1,2,3
correspond to the principle, first and second harmonics respectively. The dotted line is the expected Lamor
precission frequency of 73 Ge, γ73Ge = 1.5 MHz/T.

it’s state back onto the z-axis of the Bloch sphere by applying a −π/2 X pulse. If the noise is quasi-
static over a time period t > τ, the effect can be effectively reversed by inserting a Yπ pulse between
two time intervals of length τ/2. This technique reverses the parity s(t ) ∈ {−1, 1} of the noise act-
ing on the qubit, resulting in a phase coherent echo at time t = τ. By increasing the number of
Yπ pulses inserted into the total time window, the frequency spectrum of noise that is corrected
for extends to higher frequencies, improving coherence. Figure 11.2e shows the dependence of
quantum coherence on the number of refocusing pulses Nπ at Bext = 0.1 T. We achieve a record
quantum coherence of TCPMG 2 = 504(10) at Nπ = 1000. Decaying visibility due to pulse imperfec-
tions, sample heating and decoherence during finite pulse times prevent us from measuring pulse
sequences at larger Nπ . Note that our data does not show a clear saturation of TCPMG 2 , indicating
that longer coherence times are possible.
11
11.2.2. P ROBING THE 73 G E L ARMOR P ROCESSION
Unlike electrons, the hyperfine interaction experienced by hole spins is highly anisotropic. The
dipole-dipole interaction has a strong out-of-plane contribution A ⊥ , and couples weakly in-plane
A || [23]. A combination of density functional and k · p theory proves an Ising-type (A || /A ⊥ =
 190 11. QUANTUM C OHERENCE IN G ERMANIUM H OLE S PIN QUBITS

−0.018) hyperfine interaction between the heavy hole qubit states and 73 Ge nuclear spins in a pla-
nar Ge/SiGe heterostructure [30]. Therefore, application of an in-plane magnetic field of modest
strength should almost entirely suppress the hyperfine interaction. In order to investigate whether
hole spin qubits are sensitive to the 7.73% 73 Ge population in the quantum well, we employ the
CPMG-Nπ sequence for frequency selective sensing. A CPMG-Nπ pulse sequence (Fig. 11.3a) has
a characteristic filter function F (Nπ , ωτ) given by [31]
8 sin2 (Nπ ωτ/2)
F (Nπ , ωτ) = sin4 (ωτ/4) (11.1)
πω2 cos2 (ωτ/2)
where Nπ is the number of refocusing pulses, and τ is the wait time between each pulse. Trans-
mission maxima occur at frequencies
2n − 1
ω= (11.2)
2τ
for integer n, making it possible to enhance sensitivity to an arbitrary frequency of noise. Fig-
ure 11.3b shows traces of the filter functions of several CPMG pulse sequences. Rapid dephasing
of qubit coherence is expected when τ = (2n − 1)/2ω in the presence of a noise source with fre-
quency ω. Figure 11.3c shows the coherence of qubit Q3 at Bext = 0.23 T, using a CPMG-10 pulse
sequence. Clear dips are visible in the data, corresponding to coherence loss at specific frequen-
cies. By sweeping the magnetic field between Bext = 0.2 and Bext = 0.3 T, we track the shifting
temporal position of these dips t dip (Fig. 11.3d) and convert them to a frequency f dip using Eq.
11.2. We fit for n = 1,2,3, depicted by blue triangle, red circle and yellow diamond markers, corre-
sponding to the respective principle, first and second harmonics of the noise source (Fig. 11.3e).
These data follow the Lamor procession frequency of the 73 Ge nuclear spin γ73Ge = 1.5 MHz/T, as
represented by the black dashed line. We also confirm using a different qubit, and at a magnetic
field strengths of up to 1 T, these harmonics still follow the same 1.5 MHz/T dependence (See Fig.
11.6). This suggests that the hole spin qubit coherence is affected by interactions with the nuclear
spin bath, and motivates further investigation into the extent of these interactions.

11.2.3. S PIN D EPHASING DUE TO H YPERFINE N OISE

Isotopic purification has led to significant improvements in qubit metrics for silicon based elec-
tron spin qubits. For hole spin qubits in Ge/SiGe, it is unclear whether the dephasing effects of
the nuclear spin bath are large enough to warrant isotopic purification. We therefore aim to quan-
tify the expected dephasing time of hole spin qubits due to hyperfine noise in natural germanium
quantum wells. The dephasing of a hole spin due to a noise spectral density S(ω), while undergo-
ing a CPMG-Nπ refocussing pulse sequence, is given by
µ Z ∞
S(ω)
¶
C (Nπ τ) = exp − d ω 2 F (Nπ , ωτ) (11.3)
−∞ ω
where S(ω) is the spectral density of the noise and F (Nπ , ωτ) is the filter function of a CPMG-Nπ
sequence described in Eq. 11.1. The spectral density can be divided up into two contributions
S(ω) = S δΩ (ω) + S h f (ω). The first term describes spectral diffusion of the hole spin, due to fluctu-
ations in the Overhauser field as well as spin-orbit induced effects due to charge noise. We model
this using an Ornstein-Uhlenbeck type process with characteristic fluctuation time constant τc ,
and interaction strength ∆Ω. We obtain a spectral density by evaluating
dt
Z
11 S δΩ (ω) =
2π
〈〈δΩ(t )δΩ〉〉 . (11.4)

S h f (ω) represents the spectral contribution to S(ω) of hyperfine effects between the hole spin and
the nuclear spin bath, defined to be
d t −i ωt ©
Z
ª
S hf (ω) = e 〈 ĥ nuc (t ), ĥ nuc 〉 . (11.5)
2π
11.2. R ESULTS 191

Nπ
2.6

2.4
100
50
2.2
30
20
2 15
14
1.8 13
1-P Blocked

12
1.6
11
10
9
1.4
8
7
1.2 6
5
4
1
3
2
0.8 1

0 0.5 1 1.5 2 2.5 3 3.5

( s)

Figure 11.4: Studying the hyperfine interaction induced dephasing time of a hole spin qubit. (a) CPMG-Nπ
traces at Bext = 0.25 T for qubit Q3 where Nπ ranges from 1-15, 20, 30, 50 and 100 refocussing pulses. The data
are offset for clarity.

11
 192 11. QUANTUM C OHERENCE IN G ERMANIUM H OLE S PIN QUBITS

Here, ĥ nuc (t ) = exp{i Hnuc t }ĥ nuc exp{−i Hnuc t } defines the time evolution of the transverse
nuclear operator
ĥ nuc = ω⊥
X
k · Ik (11.6)
k
according to a nuclear spin hamiltonian
ωQ X Z 2
ωI I kZ +
X
Hnuc = − (I ) (11.7)
k 2 k k

where ωI = γ73Ge Bext , I kz is the projection of the spin of the k’th nucleus on the axis of Bext , and ωQ
0 0
³ ´
|| ||
is the quadrupolar interaction strength. The term ω⊥ k
= A⊥ k
sin δθ cos δθ − A k cos δθ sin δθ, A k ,
defines the hyperfine coupling strength of the k’th nuclear spin perpendicular to the applied mag-
netic field, allowing for an out of plane misalignment angle θ of Bext with respect to the plane of the
quantum well. Evaluating Eq. 11.5 gives a spectral density contribution of the nuclear spin bath,
parameterized by a hyperfine interaction strength σ⊥ and a quadrupolar interaction strength ωQ .
Figure 11.4 depicts CPMG-Nπ pulse sequences at magnetic field Bext = 0.25 T performed on qubit
Q3, for varying Nπ . The coherence dip due to interactions with the nuclear spin bath becomes
more pronounced as F (Nπ τ) becomes sharper around the 73 Ge procession frequency and the to-
tal time N τ spent sensitive to that frequency increases. Simultaneously fitting each trace (see sec-
tion 11.4.2) assuming zero nuclear polarization, we extract τc = 292 ns, ∆Ω/2π = 54.4 kHz, ωQ /2π
= 4.17 kHz and σ⊥ /2π = 34.4 kHz.
The generality of our model prevents us from drawing strong conclusions about the quasi-static
noise contributions to ∆Ω/2π, since it encompasses the combined effect of two decohering pro-
cesses. These are Overhauser field fluctuations and g-tensor variations due to the presence of
charge noise. Overhauser fields are expected to be weak in Ge/SiGe due to nuclear freezing effects
that suppress nuclear flip-flop interactions [32], the low isotopic concentration of 73 Ge, and as dis-
cussed, is not believed to be the dominant source of noise down to magnetic field strengths of 0.1
T. Therefore it is likely this term primarily represents a charge noise dominated decohering effect
on our system. The extracted quadrupole strength ωQ /2π is higher than the measured bulk value
of ∼100 Hz [33], but significant increases in quadrupolar splitting of up to 26.9 kHz have been ob-
served in strained 75 As nuclei in GaAs [34]. For magnetic fields in the limit of zero motional
p averag-
ing γ73Ge B ext T2∗ < 1, the off-diagonal term σ⊥ /2π will directly determine T2,hf
∗ ≈ 2/σ = 6.54 µs,
⊥
which is very close to the measured value of T2∗ = 1.03 µs. We also note that the Zeeman noise
strength ∆Ω/2π is only marginally stronger. Additionally, by modelling the quantum dot as a cylin-
drical potential of radius 40 nm and depth 7 nm, we can estimate the out-of-plane misalignment
angle to be θ = 0.09. This relatively small value is a plausible misalignment within our experimen-
tal setup, and suggests that achieving the expected hyperfine suppression of an in-plane magnetic
field is challenging in practice.

11.3. C ONCLUSION
We have investigated the coherence of hole spin qubits defined in planar germanium. Dephas-
ing is likely dominated by charge noise at magnetic field strengths 0.1 T<Bext < 1 T. At low field
strength, dephasing timescales up to 1.9(1) µs are measured, extendable to 504(10) µs using dy-
namical decoupling techniques. We observed dephasing of hole spin qubits due to hyperfine cou-
11 pling with the residual non-zero nuclear spin isotope 73 Ge. We developed a theoretical model of
dephasing due to the effects of the nuclear spin bath on hole spin coherence, allowing for a small
out-of-plane misalignment angle of the magnetic field to the quantum well. By simultaneously fit-
ting coherence decays for hole spin qubits undergoing CPMG pulse sequences, we extract a pure
∗ = 6.54 µs comparable to the charge noise limited value at the same
hyperfine dephasing time T2,hf
11.4. M ETHODS 193

field. These findings constitute a novel and powerful tool for analysing nuclear spin noise mech-
anisms in quantum dot spin qubits, while also representing state-of-the-art coherence times for
hole spin qubits. We envision that given the current rate of material improvement in planar ger-
manium heterostructures, isotopic purification of the quantum well will further enhance quantum
coherence metrics, positioning hole spins in planar germanium as a strong contender for future
quantum technologies.

11.4. M ETHODS
11.4.1. D EVICE FABRICATION
The Ge/SiGe wafer is fabricated on a silicon substrate. We use reduced-pressure chemical vapour
deposition to grow a 1.6 µm strain-relaxed Ge layer, a 1 µm reverse graded Si1−x Gex (x varies
from 1 to 0.8), a 500 nm constant composition Si0.2 Ge0.8 , and a 16 nm compressively strained
Ge quantum well. Finally on the quantum well we grow a 55 nm Si0.2 Ge0.8 barrier followed by
an oxidised Si cap layer (<2 nm) [7]. An ohmic contact layer is created by first defining it using
electron beam lithography, etching away the oxidised cap, then depositing 30 nm of Al. This layer
is then covered in 7 nm of Al2 O3 via atomic layer deposition at 300 o C. The gate stack is in two
overlapping layers of Ti:Pd (3/37 nm), separated by 7 nm of Al2 O3 [8].

11.4.2. F ITTING M ETHODS

Ramsey sequences ©are fit to a sinusoid modulated by a decay envelope, of the form PBlocked =
A sin (2πnf τ + φ) exp −(τ)oα /T 2∗ . CPMG-Nπ decays are fit to an exponential decay function PBlocked
ª

= A exp −(τ)α /T 2,CPMG . Coherence dips in figure 11.3c are fit to a function of the form f =
³ ´
a exp −(τ̃ − t dip )2 /2c + d τ̃ + e in the vicinity of each dip, yielding dip position t dip and width
2c used as the error bars. Some data is post processed to aid fitting. Traces in figure 11.4 are fit to
a general coherence function
· Z ∞
S(ω)
¸
P (τ) = β + (P 0 − β) exp − d ω 2 F (Nπ , ωτ) (11.8)
∞ ω

where β is the measurement bias, P 0 = P (0), and S(ω) is paramterized by ∆Ω, τc , ωQ and σ⊥
as discussed in the main text. For each, β and P 0 are fit individually since these parameters can
vary between experiment, while the parameters characterizing the noise spectral density are fit
simultaneously to the set of traces.

11.4.3. R EADOUT
One ohmic lead of each charge sensor is connected in-line to a custom-fabricated niobium-titanium-
nitride (NbTiN) kinetic inductor with expected inductance of L = 2 µH. The other lead is bonded to
a DC-line that is grounded at room temperature. The parasitic capacitance of each charge sensor
results in a tank circuit with resonant frequencies of f = 147.3 MHz and f = 139.9 MHz. Deviations
in the reflected probe tone at these frequencies due to variations in sensor resistance provide a fast
method for reading out the charge state of the capacitively coupled quantum dots.
Spin-to-charge conversion is achieved using a latched Pauli spin blockade method described in [4,
11].
11
 194 11. QUANTUM C OHERENCE IN G ERMANIUM H OLE S PIN QUBITS

11.5. E XTENDED D ATA

11.5.1. H OLE S PIN Q UBIT G -FACTORS
Figure 11.5a shows the measured values of qubit resonance frequency f as a function of applied
magnetic field B between 0.1 T and 1 T. Figure 11.5b shows the extracted g -factors of each qubit
from a linear fit.

(a) (b)
4 Q g||
Q1 0.16
3 Q2 0.23
Q3 0.23
f (GHz)

Q4 0.26
2
(c)
1 1 2
0
0 0.5 1 4 3
B (T)

Figure 11.5: Qubit resonance frequencies. (a) Qubit resonance frequencies as a function of a magnetic field
sweep between 0.1 T and 1 T. (b) Table of extracted in-plane qubit g-factors from a linear fit to (a). (c) Qubit
schematic indicating placement and colour labelling.

11.5.2. O BSERVATION OF 73 G E L AMOR P RECESSION AT H IGH F IELD

Additional experiments were performed on qubit Q1 at magnetic fields between 0.9 T and 1 T. Fig-
ure 11.6a shows the fitted values of t peak at which coherence drops were measured using CPMG-
50 pulse sequences. Figure 11.6b shows the converted frequency values, lying on the expected
Lamor precession frequency of 73 Ge.

(a) (b)
0.38

1.5 1.5 MHz/T

0.37

1.45
f peak (MHz)
t peak ( s)

0.36

0.35 1.4

0.34 1.35
CPMG-50 CPMG-50
0.33 1.3
0.9 0.95 1 0.9 0.95 1
B (T) B (T)

Figure 11.6: Observation of Lamor precession at higher field strength. (a) CPMG-50 pulse sequences applied
to qubit Q1 result in coherence dips around at the wait time t peak as a function of magnetic field strength
between 0.9 T and 1.0 T. (b) Conversion to frequency indicates these coherence dips occur at the Lamor pre-
cession frequency of 73 Ge nuclear spins.
11
R EFERENCES 195

R EFERENCES
[1] DiVincenzo, D. P. The physical implementation of quantum computation. Fortschritte der
Physik 48, 771–783 (2000).

[2] Scappucci, G. et al. The germanium quantum information route. Nature Reviews Materials
(2020).

[3] Lawrie, W. I. et al. Spin Relaxation Benchmarks and Individual Qubit Addressability for Holes
in Quantum Dots. Nano Letters 20, 7237–7242 (2020).

[4] Hendrickx, N. W. et al. A four-qubit germanium quantum processor. Nature 591, 580–585
(2020).

[5] Hendrickx, N. W. et al. A single-hole spin qubit. Nature Communications 11, 1–6 (2020).

[6] Jirovec, D. et al. A singlet triplet hole spin qubit in planar Ge. arXiv 2011.13755 (2020).

[7] Lodari, M. et al. Low percolation density and charge noise with holes in germanium. Materi-
als for Quantum Technology 1, 011002 (2020).

[8] Lawrie, W. I. L. et al. Quantum Dot Arrays in Silicon and Germanium. Applied Physics Letters
116, 080501 (2020).

[9] Hendrickx, N. W. et al. Gate-controlled quantum dots and superconductivity in planar ger-
manium. Nature Communications 9, 1–7 (2018).

[10] Van Riggelen, F. et al. A two-dimensional array of single-hole quantum dots. Applied Physics
Letters 118, 44002 (2021).

[11] Lawrie, W. I. L. et al. Simultaneous driving of semiconductor spin qubits at the fault-tolerant
threshold. arXiv:2109.07837 [cond-mat.mes-hall] (2021).

[12] Bulaev, D. V. & Loss, D. Electric dipole spin resonance for heavy holes in quantum dots. Phys-
ical Review Letters 98, 097202 (2007).

[13] Hendrickx, N. W., Franke, D. P., Sammak, A., Scappucci, G. & Veldhorst, M. Fast two-qubit
logic with holes in germanium. Nature 577, 487–491 (2020).

[14] Watzinger, H. et al. Heavy-Hole States in Germanium Hut Wires. Nano Letters 16, 6879–6885
(2016).

[15] Vandersypen, L. M. K. et al. Interfacing spin qubits in quantum dots and donors—hot, dense,
and coherent. npj Quantum Information 3, 1–10 (2017).

[16] Lodari, M. et al. Light effective hole mass in undoped Ge/SiGe quantum wells. Physical
Review B 100, 041304 (2019).

[17] Sammak, A. et al. Shallow and Undoped Germanium Quantum Wells: A Playground for Spin
and Hybrid Quantum Technology. Advanced Functional Materials 29, 1807613 (2019).

[18] Wang, X. J., Chesi, S. & Coish, W. A. Spin-echo dynamics of a heavy hole in a quantum dot.
11
Phys. Rev. Lett. 109 (2012).

[19] Stockill, R. et al. Quantum dot spin coherence governed by a strained nuclear environment.
Nature Communications 7 (2016).
 196 R EFERENCES

[20] Huthmacher, L. et al. Coherence of a dynamically decoupled quantum-dot hole spin. Physi-
cal Review B 97, 241413 (2018).

[21] Petta, J. R. et al. Applied physics: Coherent manipulation of coupled electron spins in semi-
conductor quantum dots. Science 309, 2180–2184 (2005).

[22] Chekhovich, E. A. et al. Nuclear spin effects in semiconductor quantum dots. Nature Materi-
als 12, 494–504 (2013).

[23] Philippopoulos, P., Chesi, S. & Coish, W. A. First-principles hyperfine tensors for electrons
and holes in GaAs and silicon. Physical Review B 101, 115302 (2020).

[24] Fischer, J., Coish, W. A., Bulaev, D. V. & Loss, D. Spin decoherence of a heavy hole coupled to
nuclear spins in a quantum dot. Physical Review B 78, 155329 (2008).

[25] Malinowski, F. K. et al. Notch filtering the nuclear environment of a spin qubit. Nature Nan-
otechnology 12, 16–20 (2017).

[26] Kerckhoff, J. et al. Magnetic Gradient Fluctuations from Quadrupolar 73 Ge in Si / Si Ge

Exchange-Only Qubits. PRX Quantum 2, 10347 (2021).

[27] Coish, W. A. & Baugh, J. Nuclear spins in nanostructures. Physica Status Solidi (B) Basic
Research 246, 2203–2215 (2009).

[28] Crippa, A. et al. Electrical Spin Driving by g -Matrix Modulation in Spin-Orbit Qubits. Physical
Review Letters 120, 137702 (2018).

[29] Liles, S. D. et al. Electrical control of the g -tensor of the first hole in a silicon MOS quantum
dot. arXiv 2012.04985 (2020).

[30] Philippopoulos, P. Hyperfine and spin-orbit interactions in semiconductor nanostructures.

Ph.D. thesis, MgGill University (2020).

[31] Cywiński, Ł., Lutchyn, R. M., Nave, C. P. & Das Sarma, S. How to enhance dephasing time in
superconducting qubits. Physical Review B 77, 174509 (2008).

[32] Madzik, M. T. et al. Controllable freezing of the nuclear spin bath in a single-atom spin qubit.
Science Advances 6 (2020).

[33] Verkhovskii, S. V. et al. Quadrupole Effects on73 Ge NMR Spectra in Isotopically Controlled
Ge Single Crystals. Zeitschrift fur Naturforschung - Section A Journal of Physical Sciences 55,
105–110 (2000).

[34] Yusa, G., Muraki, K., Takashina, K., Hashimoto, K. & Hirayama, Y. Controlled multiple quan-
tum coherence of nuclear spins in a nanometre-scale device. Nature 434, 1001–1005 (2005).

11
 12
C ONCLUSION & O UTLOOK

197
 198 12. C ONCLUSION & O UTLOOK

12.1. C ONCLUSION
The work in this thesis focused on two different qubit implementations: electron and hole spins
in quantum dots. For the former, the main focus was on the Silicon MOS material platform, while
the latter utilized a SiGe/Ge/SiGe heterostructure, where the quantum dot was defined in a ger-
manium quantum well. The progress in these fields can be summarised as:
• Development of a unified fabrication protocol for defining quantum dots in SiMOS, Si/SiGe
and Ge/SiGe
Chapter 3 presented a fabrication scheme for reliably creating quantum dot devices in three
different group-IV based platforms. Specific fabrication steps are tailored to each platform
based on fundamental challenges or advantages, however for the most part, the protocols
were almost identical. A device fabricated on each platform using the relevant scheme was
presented, showing that single charge occupancy was possible in all cases, for a three quan-
tum dot linear array defined in SiMOS, a five quantum dot linear array defined in Si/SiGe,
and a two-by-two quantum dot array defined in Ge/SiGe. The fabrication methods pre-
sented in this chapter were used to define every device studied in this thesis.
• Demonstration of tuneable tunnel coupling in a SiMOS quantum dot array
Chapter 4 showed that it was possible to tune interdot tunnel couplings in Silicon MOS.
This is a major milestone in the platform, as the high effective mass of electrons, as well as
poor material uniformity at the Si-SiO2 interface has hindered the control of qubit-qubit
interactions in SiMOS. However, by utilizing methods from the previous chapter in device
fabrication, as well as improved material quality, tunnel couple control was possible for two
adjacent quantum dots each occupied by a single electron spin. It was also demonstrated
that control over the tunnel rates was possible, allowing for isolation of the quantum dots
from their reservoirs, and retention of a single electron charge well outside the equilibrium
occupation potentials.
• Demonstration of universal two-qubit logic at elevated temperatures of electron spin
qubits in SiMOS quantum dots
Chapter 5 constituted a milestone in scalability for semiconductor spin qubits, by showing
that a complete set of single and two-qubit gates could be coherently performed at temper-
atures of 1.1 K, more than an order of magnitude higher than conventional dilution refriger-
ator operating temperatures. Single qubit fidelities on the order of 99% were reported, and
dephasing times did not respond to operation temperature between 0.4 K and 1.25 K.
• Demonstration of CROT, CPHASE and SWAP gates in the same device using electron spin
qubits in SiMOS quantum dots at 1.1 K.
We further demonstrated a full suite of exchange interaction mediated two-qubit gates op-
erated at elevated temperatures, and proposed composite two-qubit gates capable of over-
coming limitations in control of the exchange interaction between electron spin qubits in
quantum dots, defined in SiMOS. By applying charge noise models that reflect dephasing
times in the present device, we predicted very high two-qubit gate fidelities possible in this
platform when using composite gate sequences.
• The first demonstration of a single hole spin qubit in a quantum dot
Here, we switched gears to another platform, focus sing instead on exploring planar ger-
manium quantum wells as a host for hole spin qubits. As a first step towards this goal, we
demonstrate coherent control of a single hole spin qubit in a quantum dot, characterizing
spin relaxation times, dephasing times, manipulation speed using EDSR, and g -factor tune-
ability. We find in particular that the resonance frequency of the qubit can be tuned over
a broad range by varying the electrostatic potential of the gates capacitively coupled to the
12 qubit.
12.2. W HAT NEXT ? 199

• Demonstration of long spin lifetimes in a single hole spin.

Extending on previous work, we establish that spin relaxation times are limited by the cou-
pling to nearby reservoirs. As a result, by making the tunnel barriers more opaque, we can
extend spin lifetimes as high as 32 ms. We also show in this work that appreciable spin life-
times above 1 ms can be achieved for hole spin qubits of higher charge occupation num-
bers, specifically for 5 hole spins in a quantum dot. We also investigate the dependence
of hole spin qubit resonance frequency on the electrostatic potential environment, find-
ing once again that the g -factor is highly sensitive to the electrostatic environment of the
quantum dot.
• Universal control of a four qubit quantum processor, defined in a 2x2 array in planar ger-
manium
Significant advances in Ge/SiGe material quality and stack design led to a lower-noise en-
vironment in which to define hole spin qubits. We fabricated a 2x2 quantum dot array with
2 charge sensors, and demonstrated coherent single-, two-, three- and four-qubit quantum
logic, showcased in the demonstration of a 4-qubit GHZ state. This work set a new stan-
dard for spin qubit platforms, as it demonstrated the first two-dimensional array of semi-
conductor spin qubits, opening pathways towards quantum algorithms and simulations of
unprecedented complexity.
• Demonstration of high single qubit fidelities in germanium hole spin qubits
On the same device, we performed single qubit randomized benchmarking, showing that
with careful tuning of the driving power, native gate fidelities of up to 99.990(1)% are achieved.
We also develop a novel randomized benchmarking protocol called n-copy benchmarking,
that allows us to investigate single qubit fidelities while driving multiple qubits at once.
Furthermore, the new protocol allows us to benchmark in such a way that we eliminate
sensitivity to quantum crosstalk, allowing us to selectively quantify the effects of classical
cross-talk errors. We find that single qubit fidelities can be as high as 99.967 % for two-copy
benchmarking, and 99.34(4) % for the four copy case.
• Characterization of the hyperfine interaction strength for holes in natural germanium
quantum wells
The final chapter of the thesis investigated the quantum coherence of hole spin qubits. We
found a clear T 2∗ ∝ 1/B proportionality, suggesting that charge noise is the limiting mech-
anism for magnetic fields above 0.1 T. We also found that refocussing techniques could ex-
tend quantum coherence from T 2∗ = 1.9 µs to T2,CPMG = 504 µs. We also investigated the
strength of the hyperfine interaction between the hole spins and 73 Ge nuclei in the natural
germanium quantum wells. We find that this strength is comparable to the magnitude of
the charge noise acting on the qubits. Given the present rate of material improvements, we
forsee the hyperfine interaction becoming the limiting noise process for qubit coherence in
the near future, necessitating efforts into isotopic purification of the germanium quantum
well.

These efforts listed above constitute stepping stones for the field of quantum computing with spin
qubits defined in quantum dots. They do however, raise some significant challenges for the field.

12.2. W HAT NEXT ?

To repeat the mantra that shows up in just about every article in the field, quantum computing
will require many qubits, controllable with a high degree of accuracy, in order to become a useful
tool. Looking forward, we therefore examine here what is limiting a scale-up of semiconductor
spin qubit based quantum computers. 12
 200 12. C ONCLUSION & O UTLOOK

12.2.1. I N G ERMANIUM
Quantum computers comprising supercoducting qubits have reached qubit numbers exceeding
50 qubits [1]. Trapped ion approaches have been demonstrated with tens of qubits [2]. Spin qubits,
on the other hand, while very promising quantum computing candidates, have been limited to
only a handful of qubits [3–6], performing simple quantum algorithms. This seems at odds with
the common claims of scalability for semiconductor spin qubits, in particular spin states defined
in quantum dots. One major hurdle, is the problem of interconnect scaling, known classically as
Rent’s rule.

We have discussed Rent’s rule in the introduction chapter. The crux of the problem is that per ad-
ditional qubit, current experimental demonstrations require multiple DC and AC lines per qubit,
requiring routing between multiple interconnect layers within a dilution refrigerator. As a result,
the realization of quantum computer comprising millions of qubits using contemporary device
design is untenable. One strategy to mitigate this interconnect problem, is the on-chip integra-
tion of classical electronics. This approach, while elegant, introduces a new problem in that large
amounts of heat will be dissipated at the mK stage, exceeding the cooling power of any modern
dilution refrigerator. As we showed in chapter 5, operation of qubits at higher temperatures that
greatly relieve the required cooling power, is indeed possible, for electron spins in SiMOS quan-
tum dots. Given the universal advantage of hot qubit operation, similar efforts in Ge/SiGe would
strengthen the platform as a scalable contender.

Several challenges present themselves in this effort. With an increase in operational temperature,
the strength of charge noise will increase. For Ge/SiGe devices, charge noise already limits spin
coherence as we saw in the previous chapter. This is further complicated by the fact that the low
in-plane g-factor of holes in Ge/SiGe requires larger magnetic fields to achieve appreciable spin
splittings, increasing the sensitivity of spin states to charge noise. One strategy to mitigate this
could be to switch to an out-of-plane external magnetic field configuration, where the g-factor
is much larger, requiring lower magnetic fields [7]. Such a configuration would require isotopic
purification of the germanium quantum well, since the hyperfine interaction in the out-of-plane
direction is enhanced. A second strategy could be to create ’squeezed’ dots, where charge insen-
sitive points are predicted to occur for quantum dots with a squeezed axis perpendicular to the
applied magnetic field [8]. Such a configuration would allow in-plane configurations, and permit
extremely fast Rabi oscillations at low power cost.

A second important avenue would be exploring shared control of qubits in quantum dots. Cross-
bar style architectures that drastically reduce the required Rent exponent have been proposed [9].
While applicable to both platforms, hole spins, or indeed any qubit candidate exhibiting a strong
spin orbit interaction, are favorably positioned for cross-bar architectures for two reasons. Firstly,
striplines or micromagnets are no longer necessary for qubit manipulation, simplifying fabrication
and design significantly. Secondly, the valence band pinning of the Fermi level in Ge facilitates p-
type Ohmic contact for virtually any metal. This opens an avenue for charge sensor integration
within quantum dot arrays (typical length scales of implantation straggle and thermal diffusion
from annealing are typically larger than quantum dot size), by defining metallic islands flanking
quantum dots throughout an array.

Improvements to the qubits themselves will be necessary to actually perform useful quantum op-
erations on a larger quantum computer. Despite the impressive single qubit fidelities reported
in chapter 10, two-qubit fidelities exceeding 99% have not been demonstrated for either electron
12 spin qubits in SiMOS quantum dots, or hole spin qubits in Ge/SiGe. In both cases, spin dephas-
12.2. W HAT NEXT ? 201

ing times at the exchange-on operational points are too short compared to gate operation times,
in order to perform good two-qubit gates. This is likely due to the fact that at these operational
points, the qubits are highly sensitive to detuning noise. These values could be improved by re-
ducing charge noise values via material and fabrication improvements, or by operating in regimes
where the qubits are insensitive to electric field fluctuations. A challenge in Ge/SiGe is the fact
that symmetry point operation does not fully de-sensitise qubits to charge noise, indicating that
a paradigm shift may be in order. This may come in the form of ‘squeezed dot’ geometries for
charge insensitive operation [8], exploring different quantum well growth directions [10], or by
other means.

We observed in the case of Ge/SiGe that control of the exchange interaction and cross-talk is vital
for good qubit operational quality. A simple strategy to improve these issues could be to design
arrays with larger interdot pitch, relying on the high mobility of the hole wavefunction to couple
adjacent spins. Additionally, isotopic purification and an out-of-plane magnetic field would allow
for larger frequency spacing between qubit resonance frequencies, tuneable via the electrostatic
potentials on local gates. Another limiting factor for hole spin qubits in particular are the initial-
ization and readout fidelities. The spin orbit interaction complicates these processes, and further
efforts to understand and potentially tune the spin-orbit susceptibility of hole spin qubits may be
key to increasing overall operational fidelity in the platform.

12.2.2. I N S ILICON
The achievements in SiMOS included in this thesis have been important milestones for the plat-
form, however difficulties still remain. The Si-SiO2 interface disorder remains a major hurdle for
the platform, as it hampers the control of qubit-qubit interactions, and results in poor unifor-
mity between qubits. For SiMOS to continue as a viable platform, material improvements must
be made on this front. Silicon as a host material however remains highly promising. Si/SiGe het-
erostructures such as those presented in chapter 3 have yielded extremely promising results, such
as long coherence times, high single qubit fidelities, and recently, two-qubit fidelities exceeding
99%. At the time of writing, a six-qubit linear array of electron spin qubits in quantum dots has
been fabricated using the methods developed in chapter 3, marking the largest array of quantum
dot spin qubits to date. However, such linear arrays require micromagnets to create qubit address-
ibility and artificial field gradients to permit qubit manipulation. This makes fabrication of spin
qubit arrays in more than one dimension a major design challenge for the field, as two dimensional
arrays of qubits are a requirement for many error correction codes. Additionally, crossbar archi-
tectures for silicon exist utilizing global ESR fields, however sufficient qubit addressibility places
challenging requirements on the fabrication of superconducting striplines [9].

Reliable valley splittings for electron spin qubits in silicon remains a challenge, particularly for
Si/SiGe platform. A recent technique demonstrated that spiking the silicon quantum well with a
small germanium fraction results in reliable valley splittings higher than 100 µeV, however, it is
unclear how this will affect qubit properties. Therefore, characterization of qubits defined in such
a heterostructure will be an important next step.

12.2.3. I N G ENERAL
At the end of the day, improvements in host material quality and fabrication consistency deter-
mine how good a given qubit is. Additionally, improvements in one of these aspects alone cannot
carry the burden. A wafer with high uniformity means nothing if the material is full of defects due
to electron beam irradiation, while fabricating the perfect device won’t work if the material is full
of impurities or grain boundaries. For this reason, I believe that quantum wells in heterostruc- 12
 202 R EFERENCES

tures are the most likely to succeed in advancing the field. It is also unlikely that devices fabri-
cated in academic cleanrooms using electron beam lithography will ever achieve a standard of
consistency in fabrication required to produce a truly scalable unit cell. However there seem to be
some trends worth pursuing. Deeper quantum wells seem to improve charge noise and mobility in
qubits [11]. This suggests that the majority of charge noise occurs at the surface of the heterostruc-
ture, likely due to a poor quality capping oxide, or atomic layer deposition oxide. Improvements of
heterostructure engineering at these points may be the key to reducing charge noise. Additionally,
positioning quantum wells even deeper may help.

Efforts to optimise the ratio of silicon and germanium in the various layers of heterostructure
stacks are also underway. In Silicon quantum wells, the inclusion of germanium results in a valley
splitting consistently greater than about 100 µeV, while in Ge/SiGe, the reduction of the stoichio-
metric ratio of silicon in the spacer layers results in very high mobilities [12]. It might just be me,
but it seems like as a rule of thumb, the more germanium the better...

R EFERENCES
[1] Arute, F. et al. Quantum supremacy using a programmable superconducting processor. Na-
ture 574, 505–510 (2019).

[2] Egan, L. et al. Fault-tolerant control of an error-corrected qubit. Nature 598, 281–286 (2021).

[3] Bradley, C. E. et al. A Ten-Qubit Solid-State Spin Register with Quantum Memory up to One
Minute. Phys. Rev. X 9, 031045 (2019).

[4] Petit, L. et al. Universal quantum logic in hot silicon qubits. Nature 580, 355–359 (2020).

[5] Philips, S. G. J. et al. Universal control of a six-qubit quantum processor in silicon. arXiv
2202.09252 (2022).

[6] Hendrickx, N. W. et al. A four-qubit germanium quantum processor. Nature 591, 580–585
(2020).

[7] Miller, A. J. et al. Effective out-of-plane g-factor in strained-Ge/SiGe quantum dots. arXiv
2102.01758 (2021).

[8] Bosco, S., Benito, M., Adelsberger, C. & Loss, D. Squeezed hole spin qubits in Ge quantum
dots with ultrafast gates at low power. Phys. Rev. B 104, 115425 (2021).

[9] Li, R. et al. A crossbar network for silicon quantum dot qubits. Sci. Adv. 4, eaar3960 (2018).

[10] Adelsberger, C., Benito, M., Bosco, S., Klinovaja, J. & Loss, D. Hole-spin qubits in Ge nanowire
quantum dots: Interplay of orbital magnetic field, strain, and growth direction. Phys. Rev. B
105, 075308 (2022).

[11] Lodari, M. et al. Low percolation density and charge noise with holes in germanium. Mater.
Quantum Technol. 1, 011002 (2021).

[12] Lodari, M. et al. Lightly-strained germanium quantum wells with hole mobility exceeding
one million. arXiv 2112.11860 (2021).

12
 A CKNOWLEDGEMENTS
Choosing a place to commit four plus years of ones life is an objectively tough call. But I believe I
made the right one with Delft. My time here has really cemented the fact that QuTech is a unique
place to do science, and so I extend the most sincere gratitude to my promotor Menno for giving
me the opportunity to work in such a fantastic place. Your ability to tackle the non-trivial and find
pathways forward where I have seen dead-ends has been inspiring, and I am very grateful for your
guidance. I am also deeply thankful for your support during periods of time where I was struggling
with mental health complications - you showed sincere concern, never gave me a hard time for it,
and pointed me in the right direction for external, professional help. Academia, unfortunately,
seems to have a tendency to impart melancholy on it’s disciples regardless of their capabilities. It
is comforting to know that as the group grows in size and diversity, those who choose to work on
the genuinely exciting science that takes place in Veldhorst lab, will be in good hands.

Thank you also to my copromotor Giordano. Whether it was a question about science or surf-
ing, you have always been an approachable and helpful presence for me at QuTech. Congratula-
tions on all your successes with germanium, which has become an exciting material to work with.
Thanks also to Lieven for providing a collaborative and friendly environment within the spin-qubit
community here in Delft. QuTech is lucky to have you in the drivers seat as scientific director.

Uprooting and moving country is a procedure with which nearly all who choose to pursue
a PhD are painstakingly familiar. From a practical standpoint, it promotes the dissemination of
knowledge and diversity - a process undeniably advantageous to the scientific process. However
from a personal standpoint, it can take it’s toll. As it turns out, in my case the place itself doesn’t
become home, rather the interactions within it. It was a slow process, but I was enormously lucky
to have a beautiful group of people with which I felt a strong sense of belonging.

On that note, thank you to the "OG4", aka. Plant movers aka. Luca, GJ and Nico. Luca, you
were one of my first and lasting connections in the Netherlands. You introduced me to the world
of lifting mass - a world that, admittedly, I rapidly dissociated with. However your support and
reliability during coffees, beers, dinners, lunches, 4PM Red-bulls, trips to Italy, Iceland, Spain and
France has been a cornerstone of my life in Delft. GJ, you taught me most of what I now know
about nano-fabrication, which was a critical skill for me during my PhD. Thank you for that, and
also for all of the Movie nights, gaming nights with BJ, drinks, dinners, arriving at 9 O’clock on the
toll of the Nieuwe Kerk bell to parties. Also, thanks for translating my propositions and summaries!
I wish I could claim a better handle on Dutch after four and a half years but alas, here we are. Nico,
I was hugely fortunate to be able to work with you these last few years. Your immense knowledge
on seemingly everything lab-related was such a strong driving force in our group, and I hope that
I’ve managed to acquire at least a small portion of it via osmosis. Thanks for all the beers, whiskys,
vents, help, and best of luck in Zurich.

There are two important Matteo’s to acknowledge here, however it turns out to be highly non-
trivial to assign qualities that distinguish one from another, as they have identical initials, are both
Italian, both work with diamond, and both play guitar. Even Roma and Romanga sound kind of

203
204 R EFERENCES

similar... Apparently I have a type. So, defaulting to order of appearance, Matteo snr., we met on
day one of my PhD in Delft. You were my office companion, my introduction to the QuTech band,
and one of my primary external vents when things were going wrong. It was also a great com-
fort, to share experiences with you about the challenges of long-distance relationships. I am very
happy that both of us have successfully outlasted the "long distance" component. Best of luck in
the states, and I look forward to our paths crossing again.

Matteo jnr., you joined qutech a little later than I, so our overlap wasn’t as long as I wish it
could have been. But for sure we’ve made the most of the year-and-a-bit we’ve known each other.
It’s a rare thing to meet someone on such a similar musical page. Since bumping into you at one of
the bebop jazz nights, we’ve spent a lot of time jamming, recording, and gigging, together or with
the qutech band. Playing music with you has been a genuine highlight of my life in Delft. Best of
luck with the rest of the PhD.

Floor, I’m really grateful to have learnt the ropes alongside you. A pessimist might describe
our approaches to experiment as conflicting, however I like to think they were complementary. Re-
gardless, working with you has led to a really nice portfolio of work that I’m very proud of. Thanks
also for your support outside the lab, and being there for me when I needed to get something off
my chest, or just grab a coffee.

I’d like to thank everyone in Veldhorst lab. I believe the mix of expertise and backgrounds re-
ally help lay the foundation for a very strong research group. Good luck to the next generation:
Chien, Marcel, Valentin, Hanifa, Francesco, Corentin, Timo, Sayr, as well as everyone else in the
group I’ve had the pleasure of working with: Max R., Sander, Ruoy, Marco, David, Max K. to name
a few, but the sentiment extends to all alumni!

Zooming out, I’d also like to thank people in the extended Vandersypen and Scappucci groups
with whom i’ve worked with or in close proximity to, for the last four years. Cheers to Jelmer,
Anne-Marije, Xiao, Stephan, Sjaak, Florian, Pablo, Tzu-kan, Sergey, Delphine, Tobias, Kostas,
Udit, Tom, Gouji, Patrick, Alice, Andrea, Matteusz, Christian, Job, Alberto, Brian, Diego, Amir,
Luka, Mario, and everyone else in the local spin qubit community! A great deal of my time was
spent in the cleanroom, fine-tuning fabrication recipes and creating devices. In Delft I always felt
spoiled with the amazing fabrication tools and expertise available to me, and am immensely grate-
ful to a huge number of people that made it all possible. First up, I’d like to thank my forerunners
in the clearnoom, who did a huge amount of work towards developing reliable fabrication proto-
cols. Thanks Jelmer, Gabriel, Kanwal and Nima, for really getting silicon fab off the ground. Next,
I am greatly indebted to Sergey, Gertjan and Nico for investing so much time training me on the
various tools and in the various techniques. Finally, thanks to the cleanroom staff who organise
and maintain the cleanroom - cheers to Marc, Marco, Hozanna, Charles, Anja, Arnold, Ewan,
Eugene, and Pauline. Thank you also to Jason, Matt, Olaf, Jelle, Remco, Siebe, Mark and Nico,
who facilitated the smooth operation of all equipment outside the cleanroom, and provided rapid
support whenever something wasn’t sticking, cooling, conducting or beeping the way it should. I
would also like to thank Bill Coish, Pericles Philippopoulos and David Lintau, for our collabora-
tion on hole spin coherence in chapter 11. Additionally, thank you to Jonas Helsen for your help
with all things randomized benchmarking, and to Alex Hamilton and Matt Rendell for hosting me
in Sydney during my visit in 2019.

The common thread through much of my social life in Delft was music. I started joining an
open jam night organized by Chris (thanks) at his house in the center of Delft, playing on possi-
R EFERENCES 205

bly the most out of tune piano in the known universe, sight reading chords to pop songs I’d heard
maybe twice in my life. It was great fun, and I’d like to say thanks to those jam regulars Jonas,
James, Jochain, Holger, Filip, Albert, Alessio, Rob, Florian, Rama, Romy, Gurol, Gabi, Steven,
Matteo, Maud, Vlad, Ege, and many more, for making wednesday nights into an absolute blast.
The same house that hosted the jam became my home for the remainder of my stay in Delft, where
I had the privilege of living with Deniz, Su and Raj. You have all been wonderful companions, and
I will always look back fondly on our house parties (and 5 am post-party debriefs), jam nights,
summer afternoons in the hammock on the terrace, learning to hoola-hoop, BBQs, D&D nights,
and loads more. OL24A was the cornerstone of my social life in Delft and I am so lucky to have
shared it with you.

A huge shout out to the so called ”Bebop crew”, the real core of my social life in Delft. The local
constituents of this community have been in constant flux since inception, but to Almira, Deniz,
GJ, Deepika, Milan, Sona, Luca, Stephan, Stefan, Nina, Milan, Tina, Theo, Kostas, Joana, Kristie,
Kate, Matt, Quentin, Nicole, and the extended family, you made Delft feel like home for me.

I was also fortunate to have been a part of several tabletop RPG campaigns that provided some
much needed escapism, with the added bonus of improving my public speaking confidence. I also
never had to DM - a mind-bendingly time-consuming job, particularly for people who, like me,
struggle with improvisation. Therefore, a massive thanks goes to Ata, Su, Tammy and Jonas, for
taking the time to prepare and deliver some absolutely fantastic campaigns and one-shots. Thanks
also to the fellow players Su, Deniz, Derin, Almira, Tammy, Boris, Hendrick, Edward, Irene, Jan
and Wo for all the stories and world building.

Qutech is a great place to work. That statement is further evidenced by the presence of a totally
institute funded, inclusive and open band which I was so happy to have been a part of. Thanks to
Steven, Matteo, Matteo, Loek, Joe, Hans, Remco, Connor, Julia, Chris, Nico, Floor, Anne-Marije,
Marina, Maia, and Gustavo for all the rehearsals, Il pepperoncino dinners and gigs. And if some-
one could please show Sayr where the jazz scat setting is on the Roland, that would be swell.

Keeping sane throughout lockdowns, curfews and general pandemic related anxiety is abso-
lutely not something one can take for granted. A huge thanks to Su, Matteo, Luiz and Hans for
making those curfewed friday nights into a genuinely wonderful time, where we could learn new
songs together, cook nice dinners, and find some comfort in an otherwise uncertain time.

Keeping in touch with people on the other side of the world in an awkward time zone can be
quite tough, but somehow with Declan, Jacob, Farley and Ellery it has been possible. It was always
a great comfort to be able to jump on discord and watch someones stream, or hang out. Special
thanks to Declan for your countless hours of sacrificed sleep so that we could push through to the
end of our many concurrent gaming projects. Also cheers to Scott, Sonya, James D., Oliver, Sam,
James B., Martin, Shanette and Ryan who have been there in GMT+10, happy to be on the other
side of a (albeit sporadic) catch-up call.

One of the perks of living in a small town is how naturally you get to know the people in your
routines - every morning I bought a large flat white on the way to work. Morning chats with Fleur,
Emily and Leonie always primed me for a good day. Also huge ups to the folks on the other side
of the bar - miming along with instruments during the jazz nights with Mike and Ruben, chats on
break with Arnoud and Xavi, mead tasting with Edward and Baart, D&D with Edward and Su, and
general good vibes from all in the Delft hospitality community.
206 R EFERENCES

When I left Australia four-ish years ago, the move to Delft was an exciting, but lonely adven-
ture that required a substantial adjustment time. At the time of writing, I have now moved once
again, to Copenhagen, though this time not on my own. Kaja you have been a huge support to me
over the last couple years, and I am so excited for our new chapter here in Denmark, and for all the
chapters to come. Jeg elsker dig meget højt.

Finally, thanks to Gran, Mum, Dad, Angus and Douglas. It’s been more than ten years now
since I left WA, but the love and support I have felt throughout has always been my favorite funda-
mental constant - which I promise is a meaningful sentiment coming from a physicist.
 C URRICULUM V ITÆ
William Iain Leonard L AWRIE

21-09-1993 Born in Perth, Australia.

E DUCATION
2005–2010 High School
John Curtin College of the Arts, Fremantle

2012–2015 Bachelor of Science (Physics)

The University of Melbourne, Victoria, Australia

2012–2015 Diploma of Mathematical Sciences (Applied Mathematics)

The University of Melbourne, Victoria, Australia

2015–2017 Master of Science (Experimental Condensed Matter Physics)

The University of Melbourne, Victoria, Australia

2017- PhD. Candidate

Technical University of Delft, The Netherlands
Thesis: Spin Qubits in Silicon and Germanium
Promotor: Dr. M. Veldhorst
Co-promotor: Dr. G. Scappucci

207
 L IST OF P UBLICATIONS
16. W. I. L. Lawrie, F. van Riggelen, N. W. Hendrickx, M. Russ, S. L. de Snoo, A. Sammak, G.
Scappucci, M. Veldhorst Quantum Coherence of Hole Spins in Planar Germanium, In Prepa-
ration.
15. F. van Riggelen, W. I. L. Lawrie, G. Scappucci, M. Veldhorst M. Russ, N. W. Hendrickx, A.
Sammak, M. Rispler, B. M. Terhal, G. Scappucci, M. Veldhorst Phase flip code with semicon-
ductor spin qubits, arXiv preprint, arXiv:2202.11530 (2022).
14. S. G. J. Philips, M. T. Madzik, S. V. Amitonov, S. L. de Snoo, M. Russ, C. Volk, W. I. L. Lawrie,
D. Brousse, L. Tryputen, B. Paquelet-Wutz, A. Sammak, M. Veldhorst, G. Scappucci, L. M. K.
Vandersypen Universal control of a six-qubit quantum processor in silicon, arXiv preprint,
arXiv:2202.09252 (2022).
13. W. I. L. Lawrie, M. Russ, F. van Riggelen, N. W. Hendrickx, S. L. de Snoo, A. Sammak, G.
Scappucci, M. Veldhorst Simultaneous driving of semiconductor spin qubits at the fault-
tolerant threshold, arXiv preprint, arXiv:2109.07837 [cond-mat.mes-hall] (2021).
12. N. W. Hendrickx, W. I. L. Lawrie, M. Russ, F. van Riggelen, S. L. de Snoo, R. N. Schouten, A.
Sammak, G. Scappucci, M. Veldhorst A Four-Qubit Germanium Quantum Processor, Nature
591, 580–585 (2021).
11. M. Lodari, N. W. Hendrickx, W. I. L. Lawrie, T. K. Hsiao, L. M. K. Vandersypen, A. Sammak, M.
Veldhorst, G. Scappucci Low percolation density and charge noise with holes in germanium,
Materials for Quantum Technology 1, 001002 (2021).
10. F. van Riggelen, N. W. Hendrickx, W. I. L. Lawrie, M. Russ, A. Sammak, G. Scappucci, M.
Veldhorst A two-dimensional array of single-hole quantum dots Applied Physics Letters, 118
(4) 044002 (2021).
9. W. I. L. Lawrie, N. W. Hendrickx, F. van Riggelen, M. Russ, L. Petit, A. Sammak, G. Scappucci,
M. Veldhorst Spin relaxation benchmarks and individual qubit addressability for holes in
quantum dots Nano Letters, 20 (10) 7237-7242 (2020).
8. L. Petit, M. Russ, H. G. J. Eenink, W. I. L. Lawrie, J. S. Clarke, L. M. K. Vandersypen, M. Veld-
horst High-fidelity two-qubit gates in silicon above one Kelvin arXiv preprint, arXiv:2007.09034
(2020).
7. N. W. Hendrickx, W. I. L. Lawrie, L. Petit, A. Sammak, G. Scappucci, M. Veldhorst A single-
hole spin qubit, Nature Communications 11 (1), 1–6 (2020).
6. L. Petit, H. G. J. Eenink, M. Russ, W. I. L. Lawrie, J. S. Clarke, L. M. K. Vandersypen, M.
Veldhorst Universal quantum logic in hot silicon qubits Nature, 580 (7803), 355–359 (2020).
5. W. I. L. Lawrie, H. G. J. Eenink, N. W. Hendrickx, J. M. Boter, L. Petit, S. V. Amitonov, M.
Lodari, B. Paquelet Wuetz, C. Volk, S. G. J. Philips, G. Droulers, N. Kalhor, F. van Riggelen, D.
Brousse, A. Sammak, L. M. K. Vandersypen, G. Scappucci, M. Veldhorst Quantum dot arrays
in silicon and germanium, Applied Physics Letters 116 (8) 080501 (2020).
4. L. Petit, H. G. J. Eenink, W. I. L. Lawrie, J. S. Clarke, L. M. K. Vandersypen, M. Veldhorst Tun-
able coupling and isolation of single electrons in silicon metal-oxide-semiconductor quan-
tum dots Nano Letters, 19 (12), 8653-8657 (2019).

209
210 L IST OF P UBLICATIONS

3. D. Holmes, W. I. L. Lawrie, B. C. Johnson, A. Asadpoordarvish, J. C. McCallum, D. R. Mc-

Camey, D. N. Jamieson Activation and electron spin resonance of near-surface implanted
bismuth donors in silicon Physical Review Materials 3 (8) 083403 (2019).
2. D. N. Jamieson, W. I. L. Lawrie, S. G. Robson, A. M. Jakob, B. C. Johnson, J. C. McCallum
Deterministic Doping Materials Science in Semiconductor Processing 62 23-30, (2017).
1. D. N. Jamieson, W. I. L. Lawrie, F. E. Hudson, A. S. Dzurak, A. Morello, S. G. Robson, A.
M. Jakob, B. C. Johnson, J. C. McCallum Deterministic atom placement by ion implantation:
Few and single atom devices for quantum computer technology 2016 21st International Con-
ference on Ion Implantation Technology (IIT), 1-6 (2016).