Circuit Quantum Electrodynamics

Alexandre Blais∗
Institut quantique and Département de Physique,
Université de Sherbrooke,
Sherbrooke J1K 2R1 QC,
Canada
Canadian Institute for Advanced Research, Toronto, ON,
Canada

Arne L. Grimsmo
Centre for Engineered Quantum Systems,
School of Physics,
arXiv:2005.12667v1 [quant-ph] 26 May 2020

The University of Sydney,

Sydney, NSW 2006,
Australia

S. M. Girvin
Yale Quantum Institute,
PO Box 208 334, New Haven,
CT 06520-8263 USA

Andreas Wallraff
Department of Physics,
ETH Zurich, CH-8093, Zurich,
Switzerland.

Quantum mechanical effects at the macroscopic level were first explored in Josephson
junction-based superconducting circuits in the 1980’s. In the last twenty years, the emer-
gence of quantum information science has intensified research toward using these circuits
as qubits in quantum information processors. The realization that superconducting
qubits can be made to strongly and controllably interact with microwave photons, the
quantized electromagnetic fields stored in superconducting circuits, led to the creation
of the field of circuit quantum electrodynamics (QED), the topic of this review. While
atomic cavity QED inspired many of the early developments of circuit QED, the latter
has now become an independent and thriving field of research in its own right. Circuit
QED allows the study and control of light-matter interaction at the quantum level in
unprecedented detail. It also plays an essential role in all current approaches to quantum
information processing with superconducting circuits. In addition, circuit QED enables
the study of hybrid quantum systems, such as quantum dots, magnons, Rydberg atoms,
surface acoustic waves, and mechanical systems, interacting with microwave photons.
Here, we review the coherent coupling of superconducting qubits to microwave photons
in high-quality oscillators focussing on the physics of the Jaynes-Cummings model, its
dispersive limit, and the different regimes of light-matter interaction in this system. We
discuss coupling of superconducting circuits to their environment, which is necessary for
coherent control and measurements in circuit QED, but which also invariably leads to
decoherence. Dispersive qubit readout, a central ingredient in almost all circuit QED
experiments, is also described. Following an introduction to these fundamental concepts
that are at the heart of circuit QED, we discuss important use cases of these ideas in
quantum information processing and in quantum optics. Circuit QED realizes a broad set
of concepts that open up new possibilities for the study of quantum physics at the macro
scale with superconducting circuits and applications to quantum information science in
the widest sense.

May 27, 2020

∗ Send comments and feedback to circuitqed.review@gmail.com

 2

CONTENTS E. Waveguide QED 59

F. Single microwave photon detection 60
I. Introduction 2
IX. Outlook 61
II. Superconducting quantum circuits 4
A. The quantum LC resonator 4 Acknowledgments 62
B. 2D resonators 5
1. Quantized modes of the transmission line A. Hamiltonian of a voltage biased transmon 62
resonator 7 1. Classical gate voltage 62
C. 3D resonators 8 2. Coupling to an LC oscillator 63
D. The transmon artificial atom 9
E. Flux-tunable transmons 12 B. Unitary transformations 63
F. Other superconducting qubits 12 1. Schrieffer-Wolff perturbation theory 64
2. Schrieffer-Wolff for a multilevel system coupled to an
III. Light-matter interaction in circuit QED 13
oscillator in the dispersive regime 64
A. Exchange interaction between a transmon and an
a. The transmon 65
oscillator 13
b. The Jaynes-Cummings model 65
B. The Jaynes-Cummings spectrum 14
C. Dispersive regime 15 3. Bogoliubov approach to the dispersive regime 66
1. Schrieffer-Wolff approach 15 4. Off-resonantly driven transmon 67
2. Bogoliubov approach 16
D. Josephson junctions embedded in multimode C. Input-output theory 67
electromagnetic environments 17
E. Beyond the transmon: multilevel artificial atom 19 References 69
F. Alternative coupling schemes 19

IV. Coupling to the outside world: The role of the

environment 19 I. INTRODUCTION
A. Wiring up quantum systems with transmission lines 20
B. Input-output theory in electrical networks 21 Circuit quantum electrodynamics (QED) is the study
C. Qubit relaxation and dephasing 22
D. Dissipation in the dispersive regime 23
of the interaction of nonlinear superconducting circuits,
E. Multi-mode Purcell effect and Purcell filters 24 acting as artificial atoms or as qubits for quantum infor-
F. Controlling quantum systems with microwave drives 24 mation processing, with quantized electromagnetic fields
in the microwave frequency domain. Inspired by cav-
V. Measurements in circuit QED 25
A. Microwave field detection 25 ity QED (Haroche and Raimond, 2006; Kimble, 1998),
B. Phase-space representations and their relation to field a field of research originating from atomic physics and
detection 28 quantum optics, circuit QED has led to advances in the
C. Dispersive qubit readout 29 fundamental study of light-matter interaction, in the de-
1. Steady-state intra-cavity field 29
2. Signal-to-noise ratio and measurement fidelity 31 velopment of quantum information processing technology
3. Other approaches 33 (Blais et al., 2020; Clarke and Wilhelm, 2008; Kjaergaard
et al., 2019; Krantz et al., 2019; Wendin, 2017), and in
VI. Qubit-resonator coupling regimes 34
A. Resonant regime 35
the exploration of novel hybrid quantum systems (Clerk
1. Bad-cavity limit 35 et al., 2020; Xiang et al., 2013).
2. Bad-qubit limit 35 First steps toward exploring the quantum physics of
3. Strong coupling regime 36 superconducting circuits were made in the mid-1980’s. At
B. Dispersive regime 38
1. Qubit Spectroscopy 38 that time, the question arose whether quantum phenom-
2. AC-Stark shift and measurement-induced ena, such as quantum tunneling or energy level quantiza-
broadening 39 tion, could be observed in macroscopic systems of any kind
C. Beyond strong coupling: ultrastrong coupling regime 41
(Leggett, 1980, 1984b). One example of such a macro-
VII. Quantum computing with circuit QED 42 scopic system is the Josehphson tunnel junction (Joseph-
A. Single-qubit control 42 son, 1962; Tinkham, 2004) formed by a thin insulating
B. Two-qubit gates 44 barrier at the interface between two superconductors and
1. Qubit-qubit exchange interaction 44
2. Flux-tuned 11-02 phase gate 47
in which macroscopic quantities such as the current flow-
3. All-microwave gates 47 ing through the junction or the voltage developed across
4. Parametric gates 49 it are governed by the dynamics of a macroscopic or-
C. Encoding a qubit in an oscillator 51 der parameter (Eckern, 1984). This macroscopic order
VIII. Quantum optics on a chip 53 parameter relates to the density and common phase of
A. Intra-cavity fields 54 Bose-condensed Cooper pairs of electrons. The first ex-
B. Quantum-limited amplification 54 perimental evidences for quantum effects in these circuits
C. Propagating fields and their characterization 56
1. Itinerant single and multi-photon states 56
(Clarke et al., 1988) were the observation of quantum
2. Squeezed microwave fields 56 tunneling of the phase degree of freedom of a Joseph-
D. Remote Entanglement Generation 58 son junction (Devoret et al., 1985), rapidly followed by
 3

the measurement of quantized energy levels of the same of engineered quantum devices with technological applica-
degree of freedom (Martinis et al., 1985). tions. Indeed, after 15 years of continuous development,
While the possibility of observation of coherent quan- circuit QED is now a leading architecture for quantum
tum phenomena in Josephson junction-based circuits, such computation. Simple quantum algorithms have been im-
as coherent oscillations between two quantum states of the plemented (DiCarlo et al., 2009), cloud-based devices are
junction and the preparation of quantum superpositions accessible, demonstrations of quantum-error correction
was already envisaged in the 1980’s (Tesche, 1987), the have approached or reached the so-called break-even point
prospect of realizing superconducting qubits for quantum (Hu et al., 2019; Ofek et al., 2016), and devices with sev-
computation revived interest in the pursuit of this goal eral tens of qubits have been operated with claims of
(Bocko et al., 1997; Bouchiat et al., 1998; Makhlin et al., quantum supremacy (Arute et al., 2019).
2001, 1999; Shnirman et al., 1997). In a groundbreaking More generally, circuit QED is opening new research
experiment, time-resolved coherent oscillations with a directions. These include the development of quantum-
superconducting qubit were observed in 1999 (Nakamura limited amplifiers and single-microwave photon detectors
et al., 1999). Further progress resulted in the observation with applications ranging from quantum information pro-
of coherent oscillations in coupled superconducting qubits cessing to the search for dark matter axions, to hybrid
(Pashkin et al., 2003; Yamamoto et al., 2003) and in signif- quantum systems (Clerk et al., 2020) where different
icant improvements of the coherence times of these devices physical systems such as NV centers (Kubo et al., 2010),
by exploiting symmetries in the Hamiltonian underlying mechanical oscillators (Aspelmeyer et al., 2014), semicon-
the description of the circuits (Vion et al., 2002a). ducting quantum dots (Burkard et al., 2020), or collec-
In parallel to these advances, in atomic physics and tive spin excitations in ferromagnetic crystals (Lachance-
quantum optics, cavity QED developed into an excel- Quirion et al., 2019) are interfaced with superconducting
lent setting for the study of the coherent interactions quantum circuits.
between individual atoms and quantum radiation fields
(Brune et al., 1996; Haroche and Kleppner, 1989; Rempe In this review, we start in Sec. II by introducing the two
et al., 1987; Thompson et al., 1992), and its application main actors of circuit QED: high-quality superconduct-
to quantum communication (Kimble, 2008) and quantum ing oscillators and superconducting artificial atoms. The
computation (Haroche and Raimond, 2006; Kimble, 1998). latter are also known as superconducting qubits in the con-
In the early 2000’s, the concept of realizing the physics of text of quantum information processing. There are many
cavity QED with superconducting circuits emerged with types of superconducting qubits and we choose to focus on
proposals to coherently couple superconducting qubits the transmon (Koch et al., 2007). This choice is made be-
to microwave photons in open 3D cavities (Al-Saidi and cause the transmon is not only the most widely used qubit
Stroud, 2001; Yang et al., 2003; You and Nori, 2003), but also because this allows us to present the main ideas
in discrete LC oscillators (Buisson and Hekking, 2001; of circuit QED without having to delve into the very rich
Makhlin et al., 2001), and in large Josephson junctions physics of the different types of superconducting qubits.
(Blais et al., 2003; Marquardt and Bruder, 2001; Plas- Most of the material presented in this review applies to
tina and Falci, 2003). The prospect of realizing strong other qubits without much modification. Section III is
coupling of superconducting qubits to photons stored in devoted to light-matter coupling in circuit QED including
high-quality coplanar waveguide resonators, together with a discussion of the Jaynes-Cummings model and its dis-
suggestions to use this approach to protect qubits from persive limit. Different methods to obtain approximate
decoherence, to read out their state, and to couple them effective Hamiltonians valid in the dispersive regime are
to each other in a quantum computer architecture ad- presented. Section IV addresses the coupling of supercon-
vanced the study of cavity QED with superconducting ducting quantum circuits to their electromagnetic environ-
circuits (Blais et al., 2004). This possibility of exploring ment, considering both dissipation and coherent control.
both the foundations of light-matter interaction and ad- In Sec. V, we turn to measurements in circuit QED with
vancing quantum information processing technology with an emphasis on dispersive qubit readout. Building on
superconducting circuits motivated the rapid advance in this discussion, Sec. VI presents the different regimes of
experimental research, culminating in the first experi- light-matter coupling which are reached in circuit QED
mental realization of a circuit QED system achieving the and their experimental signatures. In the last sections,
strong coupling regime of light-matter interaction where we turn to two applications of circuit QED: quantum
the coupling overwhelms damping (Chiorescu et al., 2004; computing in Sec. VII and quantum optics in Sec. VIII.
Wallraff et al., 2004). Our objective with this review is to give the reader
Circuit QED combines the theoretical and experimental a solid background on the foundations of circuit QED
tools of atomic physics, quantum optics and the physics rather than showcasing the very latest developments of
of mesoscopic superconducting circuits not only to further the field. We hope that this introductory text will allow
explore the physics of cavity QED and quantum optics in the reader to understand the recent advances of the field
novel parameter regimes, but also to allow the realization and to become an active participant in its development.
 4

II. SUPERCONDUCTING QUANTUM CIRCUITS

Response (arb. units)

V (Φ) (arb. units)
Circuit components with spatial dimensions that are
L C
small compared to the relevant wavelength can be treated |2
as lumped elements (Devoret, 1997), and we start this
section with a particularly simple lumped-element cir- κ
|1
cuit: the quantum LC oscillator. We subsequently discuss ωr
the closely related two- and three-dimensional microwave |0
resonators that play a central role in circuit QED experi-
ments and which can be thought of as distributed versions
of the LC oscillator with a set of harmonic frequencies. 0 −5 0 5
Finally, we move on to nonlinear quantum circuits with Φ/Φ0 δ/κ
Josephson junctions as the source of nonlinearity, and
FIG. 1 (Left) Harmonic potential versus flux of the LC circuit
discuss how such circuits can behave as artificial atoms.
with Φ0 = h/2e the flux quantum. (Right) Response of the
We put special emphasis on the transmon qubit (Koch oscillator to an external perturbation as a function of the
et al., 2007), which is the most widely used artificial atom detuning δ of the perturbation from the oscillator frequency.
design in current circuit QED experiments. Here κ = ωr /Q, with Q the oscillator’s quality factor, is the
full width at half maximum (FWHM) of the oscillator response.
Equivalently, 1/κ is the average lifetime of the single-photon
A. The quantum LC resonator state |1i before it decays to |0i. (Inset) Lumped-element LC
oscillator of inductance L and capacitace C.
An LC oscillator is characterized by its inductance
L and capacitance √C or, equivalently, by its angular p p
frequency with Φzpf =p ~/2ωr C = ~Zr /2 and Qzpf =
p ωr = 1/ LC and characteristic impedance p
Zr = L/C. The total energy of this oscillator is given ~ωr C/2 = ~/2Zr the characteristic magnitude of
by the sum of its charging and inductive energy the zero-point fluctuations of the flux and the charge, re-
spectively. With these definitions, the above Hamiltonian
Q2 Φ2 takes the usual form
HLC = + , (1)
2C 2L
ĤLC = ~ωr (↠â + 1/2), (5)
where Q is the charge on the capacitor and Φ the flux
threading the inductor, see Fig. 1. Charge is related
R t to cur- with eigenstates that satisfy ↠â|ni = n|ni for n =
rent, I, from charge conservation by Q(t) = t0 dt0 I(t0 ), 0, 1, 2, . . . In the rest of this review, we follow the con-
and fluxR to voltage from Faraday’s induction law by vention of dropping from the Hamiltonian the factor of
t
Φ(t) = t0 dt0 V (t0 ), where we have assumed that the 1/2 corresponding to zero-point energy. The action of
p
voltage and current are zero at an initial time t0 , often ↠= 1/2~Zr (Φ̂ − iZr Q̂) is to create a quantized exci-
taken to be in the distant past (Vool and Devoret, 2017). tation of the flux and charge degrees of freedom of the
It is instructive to rewrite HLC as oscillator or, equivalently of the magnetic and electric
fields. In other words, ↠creates a photon of frequency
Q2 1 ωr stored in the circuit.
HLC = + Cωr2 Φ2 . (2)
2C 2 While formally correct, one can wonder if this quanti-
This form emphasizes the analogy of the LC oscillator zation procedure is relevant in practice. In other words,
with a mechanical oscillator of coordinate Φ, conjugate is it possible to operate LC oscillators in a regime where
momentum Q, and mass C. With this analogy in mind, quantum effects are important? For this to be the case, at
quantization proceeds in a manner that should be well least two conditions must be satisfied. First, the oscillator
known to the reader: The charge and flux variables are should be sufficiently well decoupled from uncontrolled
promoted to non-commuting observables satisfying the degrees of freedom such that its energy levels are consid-
commutation relation erably less broad than their separation. In other words,
we require the oscillator’s quality factor Q = ωr /κ, with
[Φ̂, Q̂] = i~. (3) κ the oscillator linewidth or equivalently the photon loss
rate, to be large. An approach to treat the environment
It is further useful to introduce the standard annihilation of a quantum system is described in Sec. IV. Because
â and creation ↠operators of the harmonic oscillator. losses are at the origin of level broadening, superconduc-
With the above mechanical analogy in mind, we choose tors are ideal to reach the quantum regime. In practice,
these operators as most circuit QED devices are made of thin aluminum
films evaporated on low-loss dielectric substrates such as
Φ̂ = Φzpf (↠+ â), Q̂ = iQzpf (↠− â), (4) sapphire or high-resistivity silicon wafers. Mainly for its
 5

larger superconducting gap, niobium is sometimes used (a)

in place of aluminum. In addition to internal losses in
the metals forming the LC circuit, care must also be d
taken to minimize the effect of coupling to the external
circuitry that is essential for operating the oscillator. As
will be discussed below, large quality factors ranging from
Q ∼ 103 to 108 can be obtained in the laboratory (Bruno
et al., 2015; Frunzio et al., 2005; Reagor et al., 2016).
Given that your microwave oven has a quality factor
approaching 104 (Vollmer, 2004), it should not come as w s
a surprise that large Q-factor oscillators can be realized

Transmission, |S21 |
1.0
in state-of-the-art laboratories. Relation to kitchen ap- (b) B E (c)
pliances, however, stops here with the second condition 0.5
requiring that the energy separation ~ωr between adja- t
w s
cent eigenstates be larger than thermal energy kB T . Since 0.0
0 10 20 30
1 GHz × h/kB ∼ 50 mK, the condition ~ωr  kB T can be Frequency (GHz)
satisfied easily with microwave frequency circuits operated
at ∼ 10 mK in a dilution refrigerator. These circuits are FIG. 2 (a) Schematic layout of a λ/2 coplanar waveguide
therefore operated at temperatures far below the critical resonator of length d, center conductor width w, ground plane
temperature (∼ 1 − 10 K) of the superconducting films separation s, together with its capacitively coupled input
from which they are made. and output ports. The cosine shape of the second mode
function (m = 1) is illustrated with pink arrows. Also shown
With these two requirements satisfied, an oscillator is the equivalent lumped element circuit model. Adapted
with a frequency in the microwave range can be operated from Blais et al. (2004). (b) Cross-section cut of the coplanar
in the quantum regime. This means that the circuit waveguide resonator showing the substrate (dark blue), the
can be prepared in its quantum-mechanical ground state two ground planes and the center conductor (light blue) as well
|n = 0i simply by waiting for a time of the order of a as schematic representations of the E and B field distributions.
few photon lifetimes Tκ = 1/κ. It is also crucial to note (c) Transmission versus frequency for an overcoupled resonator.
The first three resonances of frequencies fm = (m + 1)f0
that the vacuum fluctuations in such an oscillator can
are illustrated with f0 = v0 /2d ∼ 10 GHz and linewidth
be made large. For example, taking reasonable values κm /2π = fm /Q.
L ∼ 0.8 nH and C ∼ 0.4 pF, corresponding to ωr /2π ∼ 8
GHz and Zr ∼ 50 Ω, the ground state is characterized by
vacuum fluctuations of the voltagep of variance as large
as ∆V0 = [hV̂ 2 i − hV̂ i2 ]1/2 = ~ωr /2C ∼ 1 µV, with resonators in experiments for radiation detectors (Day
V̂ = Q̂/C. As will be made clear later, this leads to large et al., 2003) and by the understanding of the importance
electric field fluctuations and therefore to large electric- of presenting a clean electromagnetic environment to the
dipole interactions when coupling to an artificial atom. qubits. Early circuit QED experiments were performed
with these 2D coplanar waveguide resonators (Wallraff
et al., 2004), which remains one of the most commonly
B. 2D resonators used architectures today.
A coplanar waveguide resonator consist of a coplanar
Quantum harmonic oscillators come in many shapes and waveguide of finite length formed by a center conductor
sizes, the LC oscillator being just one example. Other of width w and thickness t, separated on both sides by a
types of harmonic oscillators that feature centrally in distance s from a ground plane of the same thickness, see
circuit QED are microwave resonators where the elec- Fig. 2(a) (Pozar, 2012; Simons, 2001). Both conductors
tromagnetic field is confined either in a planar, essen- are typically deposited on a low-loss dielectric substrate of
tially two-dimensional structure (2D resonators) or in a permittivity ε and thickness much larger than the dimen-
three-dimensional volume (3D resonators). The boundary sions w, s, t. This planar structure acts as a transmission
conditions imposed by the geometry of these different line along which signals are transmitted in a way analo-
resonators lead to a discretization of the electromagnetic gous to a conventional coaxial cable. As in a coaxial cable,
field into a set of modes with distinct frequencies, where the coplanar waveguide confines the electromagnetic field
each mode can be thought of as an independent harmonic to a small volume between its center conductor and the
oscillator. Conversely (especially for the 2D case) one can ground, see Fig. 2(b). The dimensions of the center con-
think of these modes as nearly dissipationless acoustic ductor, the gaps, and the thickness of the dielectric are
plasma modes of superconductors. chosen such that the field is concentrated between the
Early experiments in circuit QED were motivated by the center conductor and ground, and radiation in other direc-
observation of large quality factors in coplanar waveguide tions is minimized. This structure supports a quasi-TEM
 6

mode (Wen, 1969), with the electromagnetic field partly resonator was realized with a length of 0.68 m and a fun-
in the dielectric substrate and in the vacuum (or other damental frequency of f0 = 92 MHz. With this frequency
dielectric) above the substrate, and with the largest con- corresponding to a temperature of 4.4 mK, the low fre-
centration in the gaps between the center conductor and quency modes of such long resonators are, however, not in
the ground planes. In practice, the coplanar waveguide the vacuum state. Indeed, according to the Bose-Einstein
can be treated as an essentially dispersion-free, linear distribution, the thermal occupation of the fundamental
dielectric medium. Ideally, the loss of coplanar waveg- mode frequency at 10 mK is n̄κ = 1/(ehf0 /kB T − 1) ∼ 1.8.
uides is only limited by the conductivity of the center Typical circuit QED experiments rather work with res-
conductor and the ground plane, and by the loss tangent onators in the range of 5–15 GHz where, conveniently,
of the dielectric. To minimize losses, superconducting microwave electronics is well developed.
metals such as aluminum, niobium or niobium titanium As already mentioned, entering the quantum regime for
nitride (NbTiN), are used in combination with dielectrics a given mode m requires more than ~ωm  kB T . It is
of low loss tangent, such as sapphire or high-resistivity also important that the linewidth κm be small compared
silicon. to the mode frequency ωm . As for the LC oscillator, the
Similarly to the lumped LC oscillator, the electromag- linewidth can be expressed in terms of the quality factor
netic properties of a coplanar waveguide resonatorp are de- Qm of the resonator mode as κm = ωm /Qm . An expres-
scribed by its characteristic impedance Zr =√ l0 /c0 and sion for the linewidth in terms of circuit parameters is
the speed of light in the waveguide v0 = 1/ l0 c0 , where given in Sec. IV. There are multiple sources of losses and
we have introduced the capacitance to ground c0 and in- it is common to distinguish between internal losses due
ductance l0 per unit length (Simons, 2001). Typical values to coupling to uncontrolled degrees of freedom (dielec-
of these parameters are Zr ∼ 50 Ω and v0 ∼ 1.3 × 108 m/s, tric and conductor losses at the surfaces and interfaces,
or about a third of the speed of light in vacuum (Göppl substrate dielectric losses, non-equilibrium quasiparticles,
et al., 2008). For a given substrate, metal thickness and vortices, two-level fluctuators. . . ) and external losses due
center conductor width, the characteristic impedance can to coupling to the input and output ports used to couple
be adjusted by varying the parameters w, s and t of the signals in and out of the resonator (Göppl et al., 2008).
waveguide (Simons, 2001). In the coplanar waveguide In terms of these two contributions, the total dissipation
geometry, transmission lines of constant impedance Zr rate of mode m is κm = κext,m + κint,m and the total,
can therefore be realized for varying center conductor or loaded, quality factor of the resonator is therefore
width w by keeping the ratio of w/s close to a constant QL,m = (Q−1 −1
ext,m + Qint,m )
−1
. It is always advantageous
(Simons, 2001). This allows the experimenter to fabricate to maximize the internal quality factor and much effort
a device with large w at the edges for convenient inter- have been invested in improving resonator fabrication
facing, and small w away from the edges to minimize the such that values of Qint ∼ 105 are routinely achieved. A
mode volume or simply for miniaturization. dominant source of internal losses in superconducting res-
A resonator is formed from a coplanar waveguide by onators at low power are believed to be two-level systems
imposing boundary conditions of either zero current or (TLSs) that reside in the bulk dielectric, in the metal sub-
zero voltage at the two endpoints separated by a distance strate, and in the metal-vacuum and substrate-vacuum
d. Zero current is achieved by micro-fabricating a gap in interfaces where the electric field is large (Sage et al.,
the center conductor (open boundary), while zero volt- 2011; Wang et al., 2015). Internal quality factors over
age can be achieved by grounding an end point (shorted 106 can be achieved by careful fabrication minimizing the
boundary). A resonator with open boundary conditions occurrence of TLSs and by etching techniques to avoid
at both ends, as illustrated in Fig. 2(a), has a fundamental substrate-vacuum interfaces in regions of high electric
frequency f0 = v0 /2d with harmonics at fm = (m + 1)f0 , fields (Bruno et al., 2015; Calusine et al., 2018; Megrant
and is known as a λ/2 resonator. On the other hand, λ/4 et al., 2012; Vissers et al., 2010).
resonators with fundamental frequency f0 = v0 /4d are On the other hand, the external quality factor can
obtained with one open end and one grounded end. be adjusted by designing the capacitive coupling at the
A typical example is a λ/2 resonator of length 1.0 cm open ends of the resonator to input/output transmission
and speed of light 1.3 × 108 m/s corresponding to a fun- lines. In coplanar waveguide resonators, these input and
damental frequency of 6.5 GHz. This coplanar waveguide output coupling capacitors are frequently chosen either as
geometry is very flexible and a large range of frequen- a simple gap of a defined width in the center conductor,
cies can be achieved. In practice, however, the useful as illustrated in Fig. 2(a), or formed by interdigitated
frequency range is restricted from above by the supercon- capacitors (Göppl et al., 2008). The choice Qext  Qint
ducting gap of the metal from which the resonator is made corresponding to an ‘overcoupled’ resonator is ideal for
(82 GHz for aluminum). Above this energy, losses due to fast qubit measurement, which is discussed in more detail
quasiparticles increase dramatically. Low frequency res- in Sec. V. On the other hand, undercoupled resonators,
onators can be made by using long, meandering, coplanar Qext  Qint , where dissipation is only limited by internal
waveguides. For example, in Sundaresan et al. (2015), a losses which are kept as small as possible, can serve as
 7

quantum memories to store microwave photons for long Cκ L0 Φn Φn+1 Cκ

times. Using different modes of the same resonator (Leek
et al., 2010), or combinations of resonators (Johnson et al., C0
2010; Kirchmair et al., 2013), both regimes of high and x=0 δx x=d
low external losses can also be combined in the same
FIG. 3 Telegrapher model of an open-ended transmission
circuit QED device. A general approach to describe losses
line resonator of length d. L0 and C0 are, respectively, the
in quantum systems is described in Sec. IV. inductance and capacitance associated to each node n of flux
Finally, the magnitude of the vacuum fluctuations of the Φn . The resonator is coupled to external transmission lines
electric field in coplanar waveguide resonators is related (not shown) at its input and output ports via the capacitors
to the mode volume. While the longitudinal dimension Cκ .
of the mode is limited by the length of the resonator,
which also sets the fundamental frequency d ∼ λ/2, the
the classical Hamiltonian corresponding to Figure 3 as
transverse dimension can be adjusted over a broad range.
Commonly chosen transverse dimensions are on the order N
X −1  
1 2 1 2
of w ∼ 10 µm and s ∼ 5 µm (Wallraff et al., 2004). If H= Q + (Φn+1 − Φn ) . (6)
n=0
2C0 n 2L0
desired, the transverse dimension of the center conductor
may be reduced to the sub-micron scale, up to a limit It is useful to consider a continuum limit of this Hamilto-
set by the penetration depth of the superconducting thin nian where the size of a unit cell δx is taken to zero. For
films which is typically of the order of 100 to 200 nm. this purpose, we write C0 = δx c0 and L0 = δx l0 , with
When combining the typical separation s ∼ 5 µm with the c0 and l0 the capacitance and inductance per unit length,
magnitude of the voltage fluctuations ∆V0 ∼ 1 µV already respectively. Moreover, we define a continuum flux field
expected from the discussion of the LC circuit, we find via Φ(xn ) = Φn and charge density field Q(xn ) = Qn /δx.
that the zero-point electric field in coplanar resonator can We can subsequently take the continuum limit δx → 0
be as large as ∆E0 = ∆V0 /s ∼ 0.2 V/m. This is at least while keeping d = N ∆x constant to find
two orders of magnitude larger than the typical amplitude Z d  
of ∆E0 in the 3D cavities used in cavity QED (Haroche 1 2 1 2
H= dx Q(x) + [∂x Φ(x)] , (7)
and Raimond, 2006). As will become clear later, together 0 2c0 2l0
with the large size of superconducting artificial atoms, this
where we have used that ∂x Φ(xn ) = limδx→0 (Φn+1 −
will lead to the very large light-matter coupling strengths
Φn )/δx. In this expression, the charge Q(x, t) =
which are characteristic of circuit QED.
c0 ∂t Φ(x, t) is the canonical momentum to the generalized
Rt
flux Φ(x, t) = −∞ dt0 V (x, t0 ), with V (x, t) the voltage
to ground on the center conductor.
1. Quantized modes of the transmission line resonator
Using Hamilton’s equations together with Eq. (7), we
find that the propagation along the transmission line is
While only a single mode of the transmission line res-
described by the wave equation
onator is often considered, there are many circuit QED
experiments where the multimode structure of the device ∂ 2 Φ(x, t) ∂ 2 Φ(x, t)
plays an important role. In this section, we present the v02 − = 0, (8)
∂x2 ∂t2
standard approach to finding the normal modes of a dis- √
with v0 = 1/ l0 c0 the speed of light in the medium. The
tributed resonator, first using a classical description of
solution to Eq. (8) can be expressed in terms of normal
the circuit.
modes
The electromagnetic properties along the x-direction of
∞
X
a coplanar waveguide resonator of length d can be modeled
Φ(x, t) = um (x)Φm (t), (9)
using a linear, dispersion-free one-dimensional medium.
m=0
Figure 3 shows the telegrapher model for such a system
where the distributed inductance of the resonator’s center 2
with Φ̈m = −ωm Φm a function of time oscillating at the
conductor is represented by the series of lumped elements mode frequency ωm and
inductances and the capacitance to ground by a parallel
combination of capacitances (Pozar, 2011). Using the flux um (x) = Am cos [km x + ϕm ] , (10)
and charge variables introduced in the description of the being the spatial profile of the mode with amplitude
LC oscillator, the energy associated to each capacitance Am . The wavevector km = ωm /v0 and the phase ϕm
is Q2n /2C0 while the energy associated to each inductance are set by the boundary conditions. For an open-ended
is (Φn+1 − Φn )2 /2L0 . In these expressions, Φn is the λ/2-resonator these are
flux variable associated with the nth node and Qn the
conjugate variable which is the charge on that node. Using 1 ∂Φ(x, t)
I(x = 0, d) = − = 0, (11)
the standard approach (Devoret, 1997), we can thus write l0 ∂x x=0,d
 8

corresponding to the fact that the current vanishes at the bifurcation amplifiers (Metcalfe et al., 2007), and to study
two extremities. A λ/4-resonator is modeled by requir- phenomena such as quantum heating (Ong et al., 2013).
ing that the voltage V (x, t) = ∂t Φ(x, t) vanishes at the We discuss the importance of quantum-limited amplifica-
grounded boundary. Asking for Eq. (11) to be satisfied for tion for qubit measurement in Sec. V and applications of
every mode implies that ϕm = 0 and that the wavevector the Kerr nonlinearity in the context of quantum optics in
is discrete with km = mπ/d. Finally, it is useful to choose Sec. VIII. A theoretical treatment of the resonator mode
the normalization constant Am such that functions, frequencies and Kerr nonlinearity in the pres-
Z ence of resonator inhomogeneities, including embedded
1 d junctions, can be found in Bourassa et al. (2012) and is
dx um (x)um0 (x) = δmm0 , (12)
d 0 discussed in Sec. III.D.
√
resulting in Am = 2. This normalization implies that
the amplitude of the modes in a 1D resonator goes down
C. 3D resonators
with the square root of the length d.
Using this normal mode decomposition in Eq. (7), the
Although their physical origin is not yet fully under-
Hamiltonian can now be expressed in the simpler form
stood, dielectric losses at interfaces and surfaces are im-
X∞  2  portant limiting factors to the internal quality factor of
Qm 1 2 2
H= + Cr ωm Φm , (13) coplanar transmission line resonators and lumped element
2Cr 2
m=0 LC oscillators, see Oliver and Welander (2013) for a re-
where Cr = dc0 is the total capacitance of the resonator view. An approach to mitigate the effect of these losses is
and Qm = Cr Φ̇m the charge conjugate to Φm . We im- to lower the ratio of the field energy stored at interfaces
mediately recognize this Hamiltonian to be a sum over and surfaces to the energy stored in vacuum. Indeed, it
independent harmonic oscillators, cf. Eq. (1). has been observed that planar resonators with larger fea-
Following once more the quantization procedure of ture sizes (s and w), and hence weaker electric fields near
Sec. II.A, the two conjugate variables Φm and Qm are the interfaces and surfaces, typically have larger internal
promoted to non-commuting operators quality factors (Sage et al., 2011).
r This approach can be pushed further by using three-
~Zm † dimensional microwave cavities rather than planar cir-
Φ̂m = (âm + âm ), (14) cuits (Paik et al., 2011). In 3D resonators formed by
2
r a metallic cavity, a larger fraction of the field energy is
~
Q̂m = i (↠− âm ), (15) stored in the vacuum inside the cavity rather than close to
2Zm m
the surface. As a result, the surface participation ratio can
p
with Zm = Lm /Cr the characteristic impedance of be as small as 10−7 in 3D cavities, in comparison to 10−5
mode m and L−1 2
m ≡ Cr ωm . Using these expressions in
for typical planar geometries (Reagor, 2015). Another
Eq. (13) immediately leads to the final result potential advantage is the absence of dielectric substrate.
In practice, however, this does not lead to a major gain in
∞
X quality factor since, while coplanar resonators can have
Ĥ = ~ωm â†m âm , (16)
air-substrate participation ratio as large as 0.9, the bulk
m=0
loss tangent of sapphire and silicon substrate is signifi-
with ωm = (m + 1)ω0 the mode frequency and ω0 /2π = cantly smaller than that of the interface oxides and does
v0 /2d the fundamental frequency of the λ/2 transmission- not appear to be the limiting factor (Wang et al., 2015).
line resonator. In practice, three-dimensional resonators come in many
To simplify the discussion, we have assumed here that different shapes and sizes, and can reach higher qual-
the medium forming the resonator is homogenous. In ity factors than lumped-element oscillators and 1D res-
particular, we have ignored the presence of the input onators. Quality factors has high as 4.2 × 1010 have
and output port capacitors in the boundary condition been reported at 51 GHz and 0.8 K with Fabry-Pérot
of Eq. (11). In addition to lowering the external quality cavities formed by two highly-polished cooper mirrors
factor Qext , these capacitances modify the amplitude and coated with niobium (Kuhr et al., 2007). Corresponding
phase of the mode functions, as well as shift the mode to single microwave photon lifetimes of 130 ms, these
frequencies. In some contexts, it can also be useful to cavities have been used in landmark cavity QED exper-
introduce one or several Josephson junctions directly in iments (Haroche and Raimond, 2006). Similar quality
the center conductor of the resonator. As discussed in factors have also been achieved with niobium micromaser
Sec. II.D, this leads to a Kerr nonlinearity, −Kâ†2 â2 /2, cavities at 22 GHz and 0.15 K (Varcoe et al., 2000). In
of the oscillator. Kerr nonlinear coplanar waveguide res- the context of circuit QED, commonly used geometries
onators have been used, for example, as near quantum- include rectangular (Paik et al., 2011; Rigetti et al., 2012)
limited linear amplifiers (Castellanos-Beltran et al., 2008), and coaxial λ/4 cavities (Reagor et al., 2016). The latter
 9

(a) As already mentioned, a major advantage of 3D cav-

ities compared to their 1D or lumped-element analogs
(f) is their high quality factor or, equivalently, long photon
lifetime. A typical internal Q factor for rectangular alu-
minum cavities is 5 × 106 , corresponding to a photon
∼ 1 cm
lifetime above 50 µs (Paik et al., 2011). These numbers
TE110 TE210 are even higher for coaxial cavities where Qint = 7 × 107 ,
Position y (mm) Position y (mm)

(b) (c)
5 5
or above a millisecond of photon storage time, has been
0 0 E reported (Reagor et al., 2016). Moreover, this latter type
-5 -5
10.68 GHz 17.14 GHz of cavity is more robust against imperfections which arise
TE120 TE220 λ/4 when integrating 3D resonators with Josephson-junction-
(d) (e)
5 5
based circuits. Lifetimes up to 2 s have also been reported
0 0
in niobium cavities that were initially developed for accel-
-5 -5
17.47 GHz 21.88 GHz erators (Romanenko et al., 2020). At such long photon
-10 0 10 -10 0 10 lifetimes, microwave cavities are longer-lived quantum
Position x (mm) Position x (mm)
memories than the transmon qubit which we will intro-
FIG. 4 (a) Photograph of a 3D rectangular superconduct- duce in the next section. This has led to a new paradigm
ing cavity showing interior volume of the waveguide enclosure for quantum information processing where information
housing a sapphire chip and transmon qubit, with two symmet- is stored in a cavity with the role of the qubit limited
ric coaxial connectors for coupling signals in and out. Credit: to providing the essential nonlinearity (Mirrahimi et al.,
IBM. (b-e) First four TEmnl modes of a 3D rectangular super- 2014). We come back to these ideas in Sec. VII.C.
conducting cavity obtained from COMSOL™. Credit: Dany
Lachance-Quirion. (f) Schematic representation of a coaxial
λ/4 cavity with electric field (full line) pointing from the inner
conductor to the sidewalls and evanescent field (dashed line) D. The transmon artificial atom
rapidly decaying from the top of the inner conductor. Adapted
from Reagor et al. (2016). Although the oscillators discussed in the previous sec-
tion can be prepared in their quantum mechanical ground
state, it is challenging to observe clear quantum behavior
have important practical advantages in that no current with such linear systems. Indeed, harmonic oscillators are
flows near any seams created in the assembly of the device. always in the correspondence limit and some degree of non-
As illustrated in Fig. 4(a) and in close analogy with linearity is therefore essential to encode and manipulate
the coplanar waveguide resonator, rectangular cavities quantum information in these systems (Leggett, 1984a).
are formed by a finite section of a rectangular waveguide Fortunately, superconductivity allows to introduce non-
terminated by two metal walls acting as shorts. This three- linearity in quantum electrical circuits while avoiding
dimensional resonator is thus simply vacuum surrounded losses. Indeed, the Josephson junction is a nonlinear cir-
on all sides by metal, typically aluminum to maximize cuit element that is compatible with the requirements
the internal quality factor or copper if magnetic field for very high quality factors and operation at millikelvin
tuning of components placed inside the cavity is required. temperatures. The physics of these junctions was first un-
The metallic walls impose boundary conditions on the derstood in 1962 by Brian Josephson, then a 22-year-old
electromagnetic field in the cavity, leading to a discrete set PhD candidate (Josephson, 1962; McDonald, 2001).
of TE and TM cavity modes of frequency (Pozar, 2012) Contrary to expectations (Bardeen, 1962), Josephson
s  2 showed that a dissipationless current, i.e. a supercurrent,
 mπ 2  nπ 2 lπ could flow between two superconducting electrodes sep-
ωmnl = c + + , (17)
a b d arated by a thin insulating barrier. More precisely, he
showed that this supercurrent is given by I = Ic sin ϕ,
labelled by the three integers (l, m, n) and where c is the where Ic is the junction’s critical current and ϕ the phase
speed of light, and a, b and d are the cavity dimensions. difference between the superconducting condensates on
Dimensions of the order of a centimeter lead to resonance either sides of the junction (Tinkham, 1996). The critical
frequencies in the GHz range for the first modes. The current, whose magnitude is determined by the junction
TE modes, to which superconducting artificial atoms cou- size and material parameters, is the maximum current
ple, are illustrated in Fig. 4(b-e). Because these modes that can be supported before Cooper pairs are broken.
are independent, once quantized, the cavity Hamiltonian Once this happens, dissipation kicks in and a finite voltage
again takes the form of Eq. (16) corresponding to a sum of develops across the junction accompanied by a resistive
independent harmonic oscillators. We return to the ques- current. Clearly, operation in the quantum regime re-
tion of quantizing the electromagnetic field in arbitrary quires currents well below this critical current. Josephson
geometries in Sec. III.D. also showed that the time dependence of the phase dif-
 10

ference ϕ is related to the voltage across the junction

according to dϕ/dt = 2πV /Φ0 , with Φ0 = h/2e the flux (b)

V (ϕ) (arb. units)

quantum. It is useful to write
R this expression as ϕ(t) = EJ CS
2πΦ(t)/Φ0 (mod 2π) = 2π dt0 V (t0 )/Φ0 (mod 2π), with
Φ(t) the flux variable already introduced in Sec. II.A.
The mod 2π in the previous equalities takes into account EC |f (c)
the fact that the superconducting phase ϕ is a compact |e
variable on the unit circle, ϕ = ϕ + 2π, while Φ can take ωq Φx
arbitrary real values. |g
(a)
Taken together, the two Josephson relations make it
clear that a Josephson junction relates current I to flux Φ. −2 0 2
This is akin to a geometric inductance whose constitutive ϕ
relation Φ = LI also links these two quantities. For this
FIG. 5 (a) Cosine potential well of the transmon qubit (full
reason, it is useful to define the Josephson inductance line) compared to the quadratic potential of the LC oscillator
 −1 (dashed lines). The spectrum of the former as eigenstates
∂I Φ0 1
LJ (Φ) = = . (18) labelled {|gi, |ei, |f i, |hi . . .} and is characterized by an anhar-
∂Φ 2πIc cos(2πΦ/Φ0 ) monicity −EC . (b) Circuit for the fixed-frequency transmon
In contrast to geometric inductances, LJ depends on the qubit. The square with a cross represents a Josephson junction
with Josephson energy EJ and junction capacitance CJ . (c)
flux. As a result, when operated below the critical current,
By using a SQUID rather than a single junction, the frequency
the Josephson junction can be thought of as a nonlinear of the transmon qubit becomes flux tunable.
inductor.
Replacing the geometric inductance L of the LC os-
cillator discussed in Sec. II.A by a Josephson junction,
as in Fig. 5(b), therefore renders the circuit nonlinear. capacitance CS . In the second line, we have defined the
In this situation, the energy levels of the circuit are no charge number operator n̂ = Q̂/2e, the phase operator
longer equidistant. If the nonlinearity and the quality ϕ̂ = (2π/Φ0 )Φ̂ (mod 2π) and the charging energy EC =
factor of the junction are large enough, the energy spec- e2 /2CΣ . We have also included a possible offset charge
trum resembles that of an atom, with well-resolved and ng = Qg /2e due to capacitive coupling of the transmon
nonuniformly spread spectral lines. We therefore often to external charges. The offset charge can arise from
refer to this circuit as an artificial atom (Clarke et al., spurious unwanted degrees of freedom in the transmon’s
1988). In many situations, and as is the focus of much environment or from an external gate voltage Vg = Qg /Cg .
of this review, we can furthermore restrict our attention As we show below, the choice of EJ and EC is crucial in
to only two energy levels, typically the ground and first determining the system’s sensitivity to the offset charge.
excited states, forming a qubit. The spectrum of ĤT is controlled by the ratio EJ /EC ,
To make this discussion more precise, it is useful to see with different values of this ratio corresponding to dif-
how the Hamiltonian of the circuit of Fig. 5(b) is modified ferent types of superconducting qubits; see for example
by the presence of the Josephson junction taking the place the reviews (Clarke and Wilhelm, 2008; Kjaergaard et al.,
of the linear inductor.
R While the energy
R stored in a linear 2019; Makhlin et al., 2001; Zagoskin and Blais, 2007). Re-
inductor is E = dt V (t)I(t) = dt (dΦ/dt)I = Φ2 /2L, gardless of the parameter regime, one can always P express
where we have used Φ = LI in the last equality, the the Hamiltonian in the diagonal form Ĥ = j ~ωj |jihj|
energy of the nonlinear inductance rather takes the form in terms of its eigenfrequencies ωj and eigenstates |ji.
Z       In the literature, two notations are commonly used to
dΦ 2π 2π
E = Ic dt sin Φ = −EJ cos Φ , label these eigenstates: {|gi, |ei, |f i, |hi . . .} and, when
dt Φ0 Φ0
there is not risk of confusion with resonator Fock states,
(19)
{|0i, |1i, |2i . . .}. Depending on the context, we will use
with EJ = Φ0 Ic /2π the Josephson energy. This quantity
both notation in this review. Figure 6 shows the energy
is proportional the rate of tunnelling of Cooper pairs
difference ωj − ω0 for the three lowest energy levels for
across the junction. Taking into account this contribution,
different ratios EJ /EC as obtained from numerical diago-
the quantized Hamiltonian of the capacitively shunted
nalization of Eq. (20). If the charging energy dominates,
Josephson junction therefore reads (see Appendix A)
EJ /EC < 1, the eigenstates of the Hamiltonian are ap-
 
(Q̂ − Qg )2 2π proximately given by eigenstates of the charge operator,
ĤT = − EJ cos Φ̂ |ji ' |ni, with n̂|ni = n|ni. In this situation, a change
2CΣ Φ0 (20)
in gate charge ng has a large impact on the transition
= 4EC (n̂ − ng )2 − EJ cos ϕ̂.
frequency of the device. As a result, unavoidable charge
In this expression, CΣ = CJ + CS is the total capacitance, fluctuations in the circuit’s environment lead to corre-
including the junction’s capacitance CJ and the shunt sponding fluctuations in the qubit transition frequency
 11

EJ /EC = 2 EJ /EC = 10 EJ /EC = 50 is insensitive to this parameter. Importantly, although

these energies are not sensitive to variations in ng , it is
10.0
Energy (GHz)

still possible to use an external oscillating voltage source

7.5 to cause transition between the transmon states. We
5.0 come back to this later. The last term of Eq. (21) is the
nonlinear correction to this harmonic potential which, for
2.5
EJ /EC  1 and therefore ∆ϕ̂  1 can be truncated to
0.0 its first nonlinear correction
−1 0 1 −1 0 1 −1 0 1
ng ng ng 1 1
Ĥq = 4EC n̂2 + EJ ϕ̂2 − EJ ϕ̂4 . (22)
2 4!
FIG. 6 Frequency difference ωj − ω0 for the first three energy
levels of the transmon Hamiltonian obtained from numeri- As expected from the above discussion, the transmon is
cal diagonalization of Eq. (20) expressed in the charge basis thus a weakly anharmonic oscillator. Note that in this
{|ni} for different EJ /EC ratios and a fixed plasma frequency approximation, the phase ϕ̂ is treated as a continuous
ωp /2π = 5 GHz. For large values of EJ /EC the energy levels variable with eigenvalues extending to arbitrary real val-
become insensitive to the offset charge ng . ues, rather than enforcing 2π-periodicity. This is allowed
as long as the device is in a regime where ϕ̂ is sufficiently
localized, which holds for low-lying energy eigenstates in
and consequently to dephasing. the transmon regime with EJ /EC  1.
To mitigate this problem, a solution is to work in the Following the previous section, it is then useful to
transmon regime where the ratio EJ /EC is large (Koch introduce creation and annihilation operators chosen to
et al., 2007; Schreier et al., 2008). Typical values are diagonalize the first two terms of Eq. (22). Denoting these
EJ /EC ∼ 20 − 80. In this situation, the charge degree of operators b̂† and b̂, in analogy to Eq. (4) we have
freedom is highly delocalized due to the large Josephson
energy. For this reason, and as clearly visible in Fig. 6(c),  1/4
2EC
the first energy levels of the device become essentially ϕ̂ = (b̂† + b̂), (23)
EJ
independent of the the gate charge. It can in fact be  1/4
shown that the charge dispersion, which describes the i EJ
n̂ = (b̂† − b̂). (24)
variation of the energy levels with gate charge, decreases 2 2EC
exponentially with EJ /EC in the transmon regime (Koch
This form makes it quite clear that fluctuations of the
et al., 2007). The net result is that the coherence time
phase ϕ̂ decrease with EJ /EC , while the reverse is true
of the device is much larger than at small EJ /EC . How-
for the conjugate charge n̂. Using these expressions in
ever, as is also clear from Fig. 6, the price to pay for
Eq. (22) finally leads to1
this increased coherence is the reduced anharmonicity of
the transmon, anharmonicity that is required to control p EC †
Ĥq = 8EC EJ b̂† b̂ − (b̂ + b̂)4
the qubit without causing unwanted transitions to higher 12 (25)
excited states. Fortunately, while charge dispersion is EC † †
exponentially small with EJ /EC , the loss of anharmonic- ≈ ~ωq b̂† b̂ − b̂ b̂ b̂b̂,
2
ity has a much weaker dependence on this ratio given by √
∼ (EJ /EC )−1/2 . As will be discussed in more detail in where ~ωq = 8EC EJ − EC . In the second line, we have
Sec. VII, because of the gain in coherence, the reduction kept only terms that have the same number of creation
in anharmonicity is not an impediment to controlling the and annihilation operators. This is reasonable because,
transmon state with high fidelity. in a frame rotating at ωq , any terms with an unequal
While the variance of the charge degree of freedom number of b̂ and b̂† will be oscillating. If the frequency
is large when EJ /EC  1, the variance of its conju- of these oscillations is larger than the prefactor of the
gate oscillating term, then this term rapidly averages out and
p variable ϕ̂ is correspondingly small, with ∆ϕ̂ =
hϕ̂2 i − hϕ̂i2  1. In this situation, it is instructive to can be neglected (Cohen-Tannoudji et al., 1977). This
rewrite Eq. (20) as rotating-wave approximation (RWA) is valid here if ~ωq 
  EC /4, an inequality that is easily satisfied in the transmon
1 1 regime.
ĤT = 4EC n̂2 + EJ ϕ̂2 − EJ cos ϕ̂ + ϕ̂2 , (21)
2 2

the first two terms corresponding to an LC circuit of 1 The approximate Hamiltonian Eq. (25) is not bounded from below
capacitance CΣ and inductance EJ−1 (Φ0 /2π)2 , the linear – an artefact of the truncation of the cosine operator. Care should
part of the Josephson inductance Eq. (18). We have therefore be taken when using this form, and it should strictly
dropped the offset charge ng in Eq. (21) on the basis speaking only be used in a truncated subspace of the original
that the frequency of the relevant low-lying energy levels Hilbert space.
 12
√
The particular combination ωp = 8EC EJ /~ is known nian then reads
as the Josephson plasma frequency and corresponds to
the frequency of small oscillations of the effective particle ĤT = 4EC n̂2 − EJ1 cos ϕ̂1 − EJ2 cos ϕ̂2 , (26)
of mass C at the bottom of a well of the cosine potential
of the Josephson junction. In the transmon regime, this where EJi is the Josephson energy of junction i, and
frequency is renormalized by a ‘Lamb shift’ equal to the ϕ̂i the phase difference across that junction. In the
charging energy EC such that ωq = ωp − EC /~ is the tran- presence of an external flux Φx threading the SQUID
sition frequency between ground and first excited state. loop and in the limit of negligible geometric inductance
Finally, the last term of Eq. (25) is a Kerr nonlinearity, of the loop, flux quantization requires that ϕ̂1 − ϕ̂2 =
with EC /~ playing the role of Kerr frequency shift per 2πΦx /Φ0 (mod 2π) (Tinkham, 1996). Defining the aver-
excitation of the nonlinear oscillator (Walls and Milburn, age phase difference ϕ̂ = (ϕ̂1 + ϕ̂2 )/2, the Hamiltonian
2008). To see this even more clearly, it can be useful can then be rewritten as (Koch et al., 2007; Tinkham,
to rewrite Eq. (25) as Hq = ~ω̃q (b̂† b̂)b̂† b̂, where the fre- 1996)
quency ω̃q (b̂† b̂) = ωq −EC (b̂† b̂−1)/2~ of the oscillator is a
ĤT = 4EC n̂2 − EJ (Φx ) cos(ϕ̂ − ϕ0 ), (27)
decreasing function of the excitation number b̂† b̂. Consid-
ering only the first few levels of the transmon, this simply where
means that the e–f transition frequency is smaller by EC  s  
than the g–e transition frequency, see Fig. 5(a). In other πΦx πΦx
EJ (Φx ) = EJΣ cos 1 + d2 tan2 , (28)
words, in the regime of validity of the approximation made Φ0 Φ0
to obtain Eq. (22), the anharmonicity of the transmon is
−EC with a typical value of EC /h ∼ 100–400 MHz (Koch with EJΣ = EJ2 + EJ1 and d = (EJ2 − EJ1 )/EJΣ the
et al., 2007). Corrections to the anharmonicity from −EC junction asymmetry. The phase ϕ0 = d tan(πΦx /Φ0 ) can
can be obtained numerically or by keeping higher-order be ignored for a time-independent flux (Koch et al., 2007).
terms in the expansion of Eq. (22). According to Eq. (27), replacing the single junction with
While the nonlinearity EC /~ is small with respect to the a SQUID loop yields an effective flux-tunable Josephson
oscillator frequency ωq , it is in practice much larger than energy EJ (Φx ). In turn, thispresults in a flux tunable
the spectral linewidth that can routinely be obtained for transmon frequency ωq (Φx ) = 8EC |EJ (Φx )| − EC /~. 2
these artificial atoms and can therefore easily be spectrally In practice, the transmon frequency can be tuned by as
resolved. As a result, and in contrast to more traditional much as one GHz in as little as 10–20 ns (Rol et al.,
realizations of Kerr nonlinearities in quantum optics, it is 2019; Rol et al., 2020). Dynamic range can also be traded
possible with superconducting quantum circuits to have for faster flux excursions by increasing the bandwidth
a large Kerr nonlinearity even at the single-photon level. of the flux bias lines. This possibility is exploited in
Some of the many implications of this observation will be several applications, including for quantum logical gates
discussed further in this review. For quantum information as discussed in more detail in Sec. VII.
processing, the presence of this nonlinearity is necessary As will become clear later, this additional control knob
to address only the ground and first excited state without can lead to dephasing due to noise in the flux threading
unwanted transition to other states. In this case, the the SQUID loop. With this in mind, it is worth notic-
transmon acts as a two-level system, or qubit. However, ing that a larger asymmetry d leads to a smaller range
it is important to keep in mind that the transmon is a of tunability and thus also to less susceptibility to flux
multilevel system and that it is often necessary to include noise (Hutchings et al., 2017). Finally, first steps towards
higher levels in the description of the device to quantita- realizing voltage tunable transmons where a semiconduct-
tively explain experimental observations. These higher ing nanowire takes the place of the SQUID loop have been
levels can also be used to considerable advantage in some demonstrated (Casparis et al., 2018; Luthi et al., 2018).
cases (Elder et al., 2020; Ma et al., 2019; Reinhold et al.,
2019; Rosenblum et al., 2018).
F. Other superconducting qubits

While the transmon is currently the most extensively

used and studied superconducting qubit, many other types
E. Flux-tunable transmons

A useful variant of the transmon artificial atom is the 2 The absolute value arises because when expanding the Hamilto-
flux-tunable transmon, where the single Josephson junc- nian in powers of ϕ̂ in Eq. (22), the potential energy term must
tion is replaced with two parallel junctions forming a always be expanded around a minimum. This discussion also
superconducting quantum interference device (SQUID), assumes that the ratio |EJ (Φx )|/EC is in the transmon range for
see Fig. 5(c) (Koch et al., 2007). The transmon Hamilto- all relevant Φx .
 13

of superconducting artificial atoms are used in the con-

text of circuit QED. In addition to working with different (a)
ratios of EJ /EC , these other qubits vary in the number 1c
m
of Josephson junctions and the topology of the circuit d∼
in which these junctions are embedded. This includes
charge qubits (Bouchiat et al., 1998; Nakamura et al.,
1999; Shnirman et al., 1997), flux qubits (Mooij et al., µm
00
1999; Orlando et al., 1999) including variations with a ∼3
large shunting capacitance (Yan et al., 2016; You et al.,
2007), phase qubits (Martinis et al., 2002), the quantro-
nium (Vion et al., 2002b), the fluxonium (Manucharyan
(b) (c)
et al., 2009) and the 0 − π qubit (Brooks et al., 2013; Gye- Cg
nis et al., 2019), amongst others. For more details about
these different qubits, the reader is referred to reviews on
the topic (Clarke and Wilhelm, 2008; Kjaergaard et al.,
2019; Krantz et al., 2019; Makhlin et al., 2001; Zagoskin E
and Blais, 2007).
λ/4

III. LIGHT-MATTER INTERACTION IN CIRCUIT QED

A. Exchange interaction between a transmon and an FIG. 7 Schematic representation of a transmon qubit (green)
oscillator coupled to (a) a 1D transmission-line resonator, (b) a lumped-
element LC circuit and (c) a 3D coaxial cavity. Panel (a) is
Having introduced the two main characters of this re- adapted from Blais et al. (2004) and panel (c) from Reagor
et al. (2016).
view, the quantum harmonic oscillator and the transmon
artificial atom, we are now ready to consider their in-
teraction. Because of their large size coming from the sum over m in Eq. (29) to a single term. In this single-
requirement of having a low charging energy (large capaci- mode approximation, the Hamiltonian reduces to a single
tance), transmon qubits can very naturally be capacitively oscillator of frequency denoted ωr coupled to a transmon.
coupled to microwave resonators, see Fig. 7 for schematic It is interesting to note that, regardless of the physical
representations. With the resonator taking the place of nature of the oscillator—for example a single mode of a
the classical voltage source Vg , capacitive coupling to a 2D or 3D resonator—it is possible to represent this Hamil-
resonator can be introduced in the transmon Hamilto- tonian with an equivalent circuit where the transmon is
nian Eq. (20) with a quantized gate voltage ng → −n̂r , capacitively coupled to an LC oscillator as illustrated in
representing the charge bias of the transmon due to the Fig. 7(b). This type of formal representation of complex
resonator (the choice of sign is simply a common con- geometries in terms of equivalent lumped element circuits
vention in the literature that we will adopt here, see is generally known as “black-box quantization” (Nigg
Appendix A). The Hamiltonian of the combined system et al., 2012), and is explored in more detail in Sec. III.D.
is therefore (Blais et al., 2004) As will be discussed in Sec. IV.E, there are many situations
X of experimental relevance where ignoring the multi-mode
Ĥ = 4EC (n̂ + n̂r )2 − EJ cos ϕ̂ − ~ωm â†m âm , (29) nature of the resonator leads to inaccurate predictions.
m Using the creation and annihilation operators intro-
P duced in the previous section, in the single-mode approxi-
where n̂r = m n̂m with n̂m = (Cg /Cm )Q̂m /2e the con- mation Eq. (29) reduces to 3
tribution to the charge bias due to the mth resonator
mode. Here, Cg is the gate capacitance and Cm the as- EC † †
Ĥ ≈ ~ωr ↠â + ~ωq b̂† b̂ − b̂ b̂ b̂b̂
sociated resonator mode capacitance. To simplify these 2 (30)
expressions, we have assumed here that Cg  CΣ , Cm . − ~g(b̂† − b̂)(↠− â),
A derivation of the Hamiltonian of Eq. (29) that goes
beyond the simple replacement of ng by −n̂r and without
the above assumption can be found in Appendix A for
3 One might worry about the term n̂2r arising from Eq. (29). How-
the case of a single LC oscillator coupled to the transmon.
ever, this term can be absorbed in the charging energy term of
Assuming that the transmon frequency is much closer the resonator mode, see Eq. (1), and therefore leads to a renor-
to one of the resonator modes than all the other modes, malization of the resonator frequency which we omit here for
say |ω0 − ωq |  |ωm − ωq | for m ≥ 1, we truncate the simplicity. See Eqs. (A9) and (A10) for further details.
 14

where ωr is the frequency of the mode of interest and significantly larger leading to an important reduction in
 1/4 r the vacuum fluctuations of the electric field. As first
Cg EJ πZr demonstrated in Paik et al. (2011), this can be made
g = ωr , (31)
CΣ 2EC RK without change in the magnitude of g simply by making
the transmon larger thereby increasing its dipole moment.
the oscillator-transmon, or light-matter, coupling con- As illustrated in Fig. 7(c), the transmon then essentially
stant. Here, Zr is the characteristic impedance of the becomes an antenna that is optimally placed within the
resonator mode and RK = h/e2 ∼ 25.8 kΩ the resis- 3D resonator to strongly couple to one of the resonator
tance quantum. The above Hamiltonian can be simplified modes.
further in the experimentally relevant situation where To strengthen the analogy with cavity QED even fur-
the coupling constant is much smaller than the system ther, it is useful to restrict the description of the transmon
frequencies, |g|  ωr , ωq . Invoking the rotating-wave to its first two levels. This corresponds to making the
approximation, it simplifies to replacements b̂† → σ̂+ = |eihg| and b̂ → σ̂− = |gihe|
EC † † in Eq. (30) to obtain the well-known Jaynes-Cummings
Ĥ ≈ ~ωr ↠â + ~ωq b̂† b̂ − b̂ b̂ b̂b̂ Hamiltonian (Blais et al., 2004; Haroche and Raimond,
2 (32)
† † 2006)
+ ~g(b̂ â + b̂â ).
~ωq
As can be seen from Eq. (24), the prefactor ĤJC = ~ωr ↠â + σ̂z + ~g(↠σ̂− + âσ̂+ ), (34)
2
(EJ /2EC )1/4 in Eq. (31) is linked to the size of charge
fluctuations in the transmon. By introducing a length where we use the convention σ̂z = |eihe| − |gihg|. The last
scale l corresponding to the distance a Cooper pair trav- term of this Hamiltonian describes the coherent exchange
els when tunneling across the transmon’s junction, it of a single quantum between light and matter, here real-
is tempting to interpret Eq. (31) as ~g = d0 E0 with ized as a photon in the oscillator or an excitation of the
d0 = 2el(EJ /32EC )1/4 the dipole transmon.
p moment of the trans-
mon and E0 = (ωr /l)(Cg /CΣ ) ~Zr /2 the resonator’s
zero-point electric field as seen by the transmon. Since
B. The Jaynes-Cummings spectrum
these two factors can be made large, especially so in the
transmon regime where d0  2el, the electric-dipole in-
The Jaynes-Cummings Hamiltonian is an exactly solv-
teraction strength g can be made very large, much more
able model which very accurately describes many situ-
so than with natural atoms in cavity QED. It is also
ations in which an atom, artificial or natural, can be
instructive to express Eq. (31) as
considered as a two-level system in interaction with a
 1/4 r single mode of the electromagnetic field. This model
Cg EJ Zr √
g = ωr 2πα, (33) can yield qualitative agreement with experiments in sit-
CΣ 2EC Zvac
uations where only the first two levels of the transmon,
where αp= Zvac /2RK is the fine-structure constant and |σ = {g, e}i, play an important role. It is often the case,
Zvac = µ0 /0 ∼ 377 Ω the impedance of vacuum with however, that quantitative agreement between theoreti-
0 the vacuum permittivity and µ0 the vacuum perme- cal predictions and experiments is obtained only when
ability (Devoret et al., 2007). To find α here should not accounting for higher transmon energy levels and the mul-
be surprising because this quantity characterizes the in- timode nature of the field. Nevertheless, since a great
teraction between the electromagnetic field and charged deal of insight can be gained from this, in this section we
particles. Here, this interaction is reduced by the fact that consider the Jaynes-Cummings model more closely.
both Zr /Zvac and Cg /CΣ are smaller than unity. Very In the absence of coupling g, the bare states of the
large couplings can nevertheless be achieved by working qubit-field system are labelled |σ, ni with σ defined above
with large values of EJ /EC or, in other words, in the and n the photon number. The dressed eigenstates of the
transmon regime. Large g is therefore obtained at the Jaynes-Cummings Hamiltonian, |σ, ni = Û † |σ, ni, can be
expense of reducing obtained from these bare states using the Bogoliubov-
p the transmon’s relative anharmonic-
ity −EC /~ωq ' − EC /8EJ . We note that the coupling like unitary transformation (Boissonneault et al., 2009;
can be increased by boosting the resonator’s impedance, Carbonaro et al., 1979)
something that can be realized, for example, by replacing h i
the resonator’s center conductor with a junction array Û = exp Λ(N̂T )(↠σ̂− − âσ̂+ ) , (35)
(Andersen and Blais, 2017; Stockklauser et al., 2017).
Apart from a change in the details of the expression of where we have defined
the coupling g, the above discussion holds for transmons  p 
arctan 2λ N̂T
coupled to lumped, 2D or 3D resonators. Importantly, by Λ(N̂T ) = p . (36)
going from 2D to 3D, the resonator mode volume is made 2 N̂T
 15

Here, N̂T = ↠â + σ̂+ σ̂− is the operator associated with

the total number of excitations, which commutes with
ĤJC , and λ = g/∆ with ∆ = ωq − ωr the qubit-resonator
detuning. Under this transformation, ĤJC takes the diag- √
|2 2g 2 |1
onal form |0

ĤD = Û † ĤJC Û |1 2g |0
 q  2ωq−EC
= ~ωr ↠â +
~ωq
σ̂z −
~∆
1− 1+ 4λ2 N̂T σ̂z . ωr ωq
2 2
|0
(37) |g |e |f
The dressed-state energies can be read directly
FIG. 8 Box: Energy spectrum of the uncoupled (gray lines)
from this expression and, as illustrated in Fig. 8,
and dressed (blue lines) states of the Jaynes-Cummings Hamil-
the Jaynes-Cummings spectrum consists of doublets tonian at zero detuning, ∆ = ωq − ωr = 0. Transmon states
{|g, ni, |e, n − 1i} of fixed excitation number4 are labelled {|gi, |ei} while photon number in the cavity are
labelled |n = 0, 1, 2 . . .i and are plotted vertically. The de-
~p 2 generacy of the two-dimensional manifolds of states with n
Eg,n = ~nωr − ∆ + 4g 2 n, √
2 (38) quanta is lifted by 2g n by the electric-dipole coupling. The
~p 2 light blue line outside of the main box represents the third
Ee,n−1 = ~nωr + ∆ + 4g 2 n, excited state of the transmon, labelled |f i. Although this is
2
not illustrated here, the presence of this level shifts the dressed
and of the ground state |g, 0i = |g, 0i of energy Eg,0 = states in the manifolds with n ≥ 2 quanta (Fink et al., 2008).
−~ωq /2. The excited dressed states are

|g, ni = cos(θn /2)|g, ni − sin(θn /2)|e, n − 1i, processes. Qubit and resonator are therefore only weakly
(39) entangled and a simplified model obtained below from
|e, n − 1i = sin(θn /2)|g, ni + cos(θn /2)|e, n − 1i,
second-order perturbation theory is often an excellent ap-
√ proximation. As the virtual processes can involve higher
with θn = arctan(2g n/∆).
A crucial feature of the energy spectrum of Eq. (38) is energy levels of the transmon, it is however crucial to
the scaling with the photon number n. In particular, for account for its multi-level nature. For this reason, our
zero detuning, ∆ = 0, the energy levels Eg,n and Ee,n−1 starting point will be the Hamiltonian of Eq. (32) and
√ not its two-level approximation Eq. (34).
are split by 2g n, in contrast to two coupled harmonic
oscillators where the energy splitting is independent of n.
Experimentally probing this spectrum thus constitutes
1. Schrieffer-Wolff approach
a way to assess the quantum nature of the coupled sys-
tem (Carmichael et al., 1996; Fink et al., 2008). We return
To find an approximation to Eq. (32) valid in the disper-
to this and related features of the spectrum in Sec. VI.A.
sive regime, we perform a Schrieffer-Wolff transformation
to second order (Koch et al., 2007). As shown in Ap-
C. Dispersive regime pendix B, as long as the interaction term in Eq. (32) is
sufficiently small, the resulting effective Hamiltonian is
On resonance, ∆ = 0, the dressed-states Eq. (39) are well approximated by
maximally entangled qubit-resonator states implying that
EC † †
the qubit is, by itself, never in a well-defined state, i.e. the Ĥdisp ' ~ωr ↠â + ~ωq b̂† b̂ − b̂ b̂ b̂b̂
reduced state of the qubit found by tracing over the res- 2
∞
X
(40)
onator is not pure. For quantum information processing,
+ ~ Λj + χj ↠â |jihj|,
it is therefore more practical to work in the dispersive j=0
regime where the qubit-resonator detuning is large with
respect to the coupling strength, |λ| = |g/∆|  1. In this where |ji label the eigenstates of the transmon which,
case, the coherent exchange of a quanta between the two under the approximation used to obtain Eq. (25), are just
systems described by the last term of ĤJC is not reso- the eigenstates of the number operator b̂† b̂. Moreover, we
nant, and interactions take place only via virtual photon have defined

Λj = χj−1,j , χj = χj−1,j − χj,j+1 , (41a)

2
4 To arrive at these expressions, we have added ~ωr /2 to ĤD . This jg
χj−1,j = . (41b)
global energy shift is without consequences. ∆ − (j − 1)EC /~
 16

for j > 0 and Λ0 = 0, χ0 = −g 2 /∆. Here χj are known

as dispersive shifts, while Λj are Lamb shifts and are
signatures of vacuum fluctuations (Bethe, 1947; Fragner |1
et al., 2008; Lamb and Retherford, 1947). |2 ωr + χ
Truncating Eq. (40) to the first two levels of the trans-
|0
mon leads to the more standard form of the dispersive ∆
|1 Lorem
Hamiltonian (Blais et al., 2004) ωq + χ ipsum
ωr − χ
~ωq0 |0
Ĥdisp ≈ ~ωr0 ↠â + σ̂z + ~χ↠âσ̂z , (42)
2 |g |e
where χ is the qubit state-dependent dispersive cavity- FIG. 9 Energy spectrum of the uncoupled (gray lines) and
shift with (Koch et al., 2007) dressed states in the dispersive regime (blue lines). The two
lowest transmon states are labelled {|gi, |ei} while photon
g2 g2 numbers in the cavity are labelled |n = 0, 1, 2 . . .i and are
ωr0 = ωr − , ωq0 = ωq + , plotted vertically. In the dispersive regime, the g − e transition
∆ − EC /~ ∆
2
(43) frequency of the qubit in the absence of resonator photons is
g EC /~ Lamb shifted and takes value ωq + χ. Moreover, the cavity
χ= − .
∆(∆ − EC /~) frequency is pulled by its interaction with the qubit and takes
the qubit-state dependent value ωr ± χ.
These dressed frequencies are what are measured experi-
mentally in the dispersive regime. It is important to em-
phasize that the frequencies entering the right-hand-sides (see Appendix B.2.b, Blais et al. (2004) and Boissonneault
of Eq. (43) are the bare qubit and resonator frequencies. et al. (2010)). This agrees with the above result in the
The spectrum of this two-level dispersive Hamiltonian limit of very large EC but, since EC /~ is typically rather
is illustrated in Fig. 9. Much of this review is devoted small compared to ∆ in most transmon experiments, the
to the consequences of this dispersive Hamiltonian for two-level system Jaynes-Cummings model gives a poor
qubit readout and quantum information processing. We prediction for the dispersive shift χ in practice. The
note that the Scrieffer-Wolff transformation also gives rise intuition here is that EC determines the anharmonicity of
to resonator and qubit self-Kerr nonlinearities at fourth the transmon. Two coupled harmonic oscillators can shift
order (Zhu et al., 2013). each other’s frequencies, but only in a state-independent
As already mentioned, these perturbative results are manner. Thus the dispersive shift must vanish in the
valid when the interaction term in Eq. (32) is suffi- limit of EC going to zero.
ciently small compared to the energy splitting of the
bare transmon-oscillator energy levels, |λ| = |g/∆|  1.
Because the matrix elements of the operators involved in 2. Bogoliubov approach
the interaction term scale with the number of photons in
the resonator and the number of qubit excitations, a more We now present an approach to arrive at Eq. (42)
precise bound on the validity of Eq. (40) needs to take into that can be simpler than performing a Schrieffer-Wolff
account these quantities. As discussed in Appendix B.2.a, transformation and which is often used in the circuit QED
we find that for the above second order perturbative re- literature.
sults to be a good approximation, the oscillator photon To proceed, it is convenient to write Eq. (32) as the
number n̄ should be much smaller than a critical photon sum of a linear and a nonlinear part, Ĥ = ĤL + ĤNL ,
number ncrit where
 
1 |∆ − jEC /~|2 ĤL = ~ωr ↠â + ~ωq b̂† b̂ + ~g(b̂† â + b̂↠), (45)
n̄  ncrit ≡ − j , (44)
2j + 1 4g 2 EC † †
ĤNL = − b̂ b̂ b̂b̂. (46)
where j = 0, 1, . . . refers to the qubit state as before. For 2
j = 0, this yields the familiar value ncrit = (∆/2g)2 for
The linear part ĤL can be diagonalized exactly using the
the critical photon number expected from the Jaynes-
Bogoliubov transformation
Cummings model (Blais et al., 2004), while setting j = 1
gives a more conservative bound. In either case, this gives h i
only a rough estimate for when to expect higher-order Ûdisp = exp Λ(↠b̂ − âb̂† ) . (47)
effects to become important.
It is worth contrasting Eq. (43) to the results expected Under this unitary, the annihilation operators transform
† †
from performing a dispersive approximation to the Jaynes- as Ûdisp âÛdisp = cos(Λ)â + sin(Λ)b̂ and Ûdisp b̂Ûdisp =
1
Cummings model Eq. (34), which leads to χ = g 2 /∆ cos(Λ)b̂ − sin(Λ)â. With the choice Λ = 2 arctan(2λ),
 17

this results in the diagonal form (a) (b)

† † †
Ûdisp ĤL Ûdisp = ~ω̃r â â + ~ω̃q b̂ b̂, (48)
Z(ω) Z(ω)
with the dressed frequencies
1 p 
ω̃r = ωr + ωq − ∆2 + 4g 2 , (49a)
2
1 p 
ω̃q = ωr + ωq + ∆2 + 4g 2 . (49b) (c)
2
Applying the same transformation to ĤNL and, in the
dispersive regime, expanding the result in orders of λ
leads to the dispersive Hamiltonian (see Appendix B.3)
† FIG. 10 (a) Transmon qubit coupled to an arbitrary
Ĥdisp = Ûdisp Ĥ Ûdisp
impedance, such as that realized by a 3D microwave cavity.
' ~ω̃r ↠â + ~ω̃q b̂† b̂ (50) (b) The Josephson junction can be represented as a capacitive
~Ka † † ~Kb † † element CJ and a linear inductive element LJ in parallel with
+ â â ââ + b̂ b̂ b̂b̂ + ~χab ↠âb̂† b̂, a purely nonlinear element indicated here by the spiderlike
2 2 symbol. Here, CΣ = CS + CJ is the parallel combinaison of
where we have introduced the Josephson capacitance and the shunting capacitance of
E C  g 4 the transmon. (c) Normal mode decomposition of the parallel
Ka ' − , Kb ' −EC /~, combination of the impedance Z(ω) together with LJ and CΣ
2~ ∆ represented by effective LC circuits.
(51)
g 2 EC /~
χab ' −2 .
∆(∆ − EC /~)
The first two of these quantities are self-Kerr nonlineari- the resonator (Bourassa et al., 2012; Ong et al., 2013), an
ties, while the third is a cross-Kerr interaction. All are approach which is used for example in quantum-limited
negative in the dispersive regime. As discussed in Ap- parametric amplifiers (Castellanos-Beltran et al., 2008) or
pendix B.3, the above expression for χab is obtained after for the preparation of quantum states of the microwave
performing a Schrieffer-Wolff transformation to eliminate electromagnetic field (Holland et al., 2015; Kirchmair
a term of the form b̂† b̂ ↠b̂+H.c. that results from applying et al., 2013; Puri et al., 2017).
Udisp on HNL . Higher-order terms in λ and other terms
rotating at frequency ∆ or faster have been dropped to
arrive at Eq. (50). These terms are given in Eq. (B32). D. Josephson junctions embedded in multimode
Truncating Eq. (50) to the first two levels of the trans- electromagnetic environments
mon correctly leads to Eqs. (42) and (43). Importantly,
these expressions are not valid if the excitation num- So far, we have focussed on the capacitive coupling
ber of the resonator or the transmon is too large or if of a transmon to a single mode of an oscillator. For
|∆| ∼ EC /~. The regime 0 < ∆ < EC , known as the many situations of experimental relevance it is, however,
straddling regime, is qualitatively different from the usual necessary to consider the transmon, or even multiple
dispersive regime. It is characterized by positive self- transmons, embedded in an electromagnetic environment
Kerr and cross-Kerr nonlinearities, Ka , χab > 0, and with a possibly complex geometry, such as a 3D cavity.
is better addressed by exact numerical diagonalization Consider the situation depicted in Fig. 10(a) where
of Eq. (29) (Koch et al., 2007). a capacitively shunted Josephson junction is embedded
A remarkable feature of circuit QED is the large nonlin- in some electromagnetic environment represented by the
earities that are achievable in the dispersive regime. Dis- impedance Z(ω). To keep the discussion simple, we con-
persive shifts larger than the resonator or qubit linewidth, sider here a single junction but the procedure can eas-
χ > κ, γ, are readily realized in experiments, a regime ily be extended to multiple junctions. As discussed in
referred to as strong dispersive coupling (Gambetta et al., Sec. II.D, the Hamiltonian of the shunted junction Eq. (21)
2006; Schuster et al., 2007). Some of the consequences of can be decomposed into a linear term of capacitance
this regime are discussed in Sec. VI.B. It is also possible CΣ = CS + CJ and linear inductance LJ = EJ−1 (Φ0 /2π)2 ,
to achieve large self-Kerr nonlinearities for the resonator, and a purely nonlinear element. This decomposition
Ka > κ.5 These nonlinearities can be enhanced by em- is illustrated in Fig. 10(b), where the spider symbol
bedding Josephson junctions in the center conductor of represents the nonlinear element (Bourassa et al., 2012;
Manucharyan et al., 2007).
We assume that the electromagnetic environment is
5 Of course, the transmon is itself an oscillator with a very large linear, nonmagnetic and has no free charges and currents.
self-Kerr given by ~Kb = −EC . Since CΣ and LJ are themselves linear elements, we might
 18

as well consider them part of the electromagnetic envi- et al., 2016). Consequently, the phase variable can be
ronment too, something that is illustrated by the box in expressed as
Fig. 10(b). Combining all linear contributions, we write a X
Hamiltonian for the entire system, junction plus the sur- ϕ̂ = [ϕm âm + H.c.] , (55)
rounding electromagnetic environment, as Ĥ = ĤL + ĤNL m
with Rx
  where ϕm = i(2π/Φ0 ) x0J dl · Dm (x)/(ωm ε) is the dimen-
J
1 sionless magnitude of the zero-point fluctuations of the
ĤNL = −EJ cos ϕ̂ + ϕ̂2 (52)
2 mth mode as seen by the junction and the boundary
conditions defined as in Fig. 10(a).
the nonlinear part of the transmon Hamiltonian already
Using Eq. (55) in ĤNL we expand the cosine to fourth
introduced in Eq. (21). A good strategy is to first di-
order in analogy with Eq. (22). This means that we
agonalize the linear part, ĤL , which can in principle be
are assuming that the capacitively shunted junction is
done exactly much like was done in Sec. III.C. Subse-
well in the transmon regime, with a small anharmonicity
quently, the phase difference ϕ̂ across the junction can be
relative to Josephson energy. Focusing on the dispersive
expressed as a linear combination of the eigenmodes of
regime where all eigenfrequencies ωm are sufficiently well
ĤL , a decomposition which is then used in ĤNL .
separated, and neglecting fast-rotating terms in analogy
A convenient choice of canonical fields for the elec-
with Sec. III.C, leads to
tromagnetic environment are the electric displacement
field D̂(x) and the magnetic field B̂(x), which can be X 1X
expressed in terms of bosonic creation and annihilation ĤNL ' ~∆m â†m âm + ~Km (â†m )2 â2m ,
m
2 m
operators (Bhat and Sipe, 2006) X (56)
X + ~χm,n â†m âm â†n ân ,
D̂(x) = [Dm (x)âm + H.c.] , (53a) m>n
m P
X where ∆m = 1
χm,n , Km =
χm,m
and
B̂(x) = [Bm (x)âm + H.c.] , (53b) 2 n 2
m
~χm,n = −EJ ϕ2m ϕ2n . (57)
where [âm , â†n ]
= δmn . The more commonly used elec-
tric field is related to the displacement field through It is also useful to introduce the energy participation
D̂(x) = ε0 Ê(x) + P̂(x), where P̂(x) is the polarization ratio pm , defined to be the fraction of the total inductive
of the medium. Moreover, the mode functions Dm (x) energy of mode m that is stored in the junction pm =
and Bm (x) can be chosen to satisfy orthogonality and (2EJ /~ωm )ϕ2m (Minev, 2019) such that we can write
normalization conditions such that
~ωm ωn
X χm,n = − pm pn . (58)
ĤL = ~ωm â†m âm . (54) 4EJ
m
As is clear from the above discussion, finding the nonlin-
In Eqs. (53) and (54), we have implicitly assumed that ear Hamiltonian can be reduced to finding the eigenmodes
the eigenmodes form a discrete set. If some part of the of the system and the zero-point fluctuations of each mode
spectrum is continuous, which is the case for infinite sys- across the junction. Of course, finding the mode frequen-
tems such as open waveguides, the sums must be replaced cies ωm and zero-point fluctuations ϕm , or alternatively
by integrals over the relevant frequency ranges. The re- the energy participation ratios pm , can be complicated for
sult is very general, holds for arbitrary geometries, and a complex geometry. As already mentioned this is, how-
can include inhomogeneities such as partially reflecting ever, an entirely classical electromagnetism problem (Bhat
mirrors, and materials with dispersion (Bhat and Sipe, and Sipe, 2006; Minev, 2019).
2006). We will, however, restrict ourselves to discrete An alternative approach is to represent the linear elec-
spectra in the following. tromagnetic environment seen by the purely nonlinear
Diagonalizing ĤL amounts to determining the mode element as an impedance Z(ω), as illustrated in Fig. 10(c).
functions {D̂m (x), B̂m (x)}, which is essentially a clas- Neglecting loss, any such impedance can be represented
sical electromagnetism problem that can, e.g., be ap- by an equivalent circuit of possibly infinitely many LC
proached using numerical software such as finite element oscillators√connected in series. The eigenfrequencies
solvers (Minev, 2019). Assuming that the mode func- ~ωm = 1/ Lm Cm , can be determined by the real parts
tions have been found, we now turn to ĤNL for which of the zeros of the admittance Y (ω) = Z −1 (ω), and the
we relate ϕ̂ to the bosonic operatorsR âm . This can be effective impedance of the m’th mode as seen by the junc-
done by noting again that ϕ̂(t) = 2π dt0 V̂ (t0 )/Φ0 , where tion can be found from Zm eff
= 2/[ωm Im Y 0 (ωm )] (Nigg
the voltage
R is simplyR the line integral of the electric field et al., 2012; Solgun et al., 2014). The effective impedance
V̂ (t) = dl·Ê(x) = dl·D̂(x)/ε across the junction (Vool is related to the zero-point fluctuations used above as
 19

eff
Zm = 2(Φ0 /2π)2 ϕ2m /~ = RK ϕ2m /(4π). From this point This result is, as already stated, very general, and can
of view, the quantization procedure thus reduces to the be used with a variety of artificial atoms coupled to a
task of determining the impedance Z(ω) as a function of resonator in the dispersive limit. Higher order expressions
frequency. can be found in (Boissonneault et al., 2010; Zhu et al.,
2013).

E. Beyond the transmon: multilevel artificial atom

F. Alternative coupling schemes
In the preceding sections, we have relied on a per-
turbative expansion of the cosine potential of the trans- Coupling the electric dipole moment of a qubit to the
mon under the assumption EJ /EC  1. To go beyond zero-point electric field of an oscillator is the most com-
this regime one can instead resort to exact diagonaliza- mon approach to light-matter coupling in a circuit but
tion of the transmon Hamiltonian. Returning to the full it is not the only possibility. Another approach is to
transmon-resonator Hamiltonian Eq. (29), we write (Koch take advantage of the mutual inductance between a flux
et al., 2007) qubit and the center conductor of a resonator to couple
the qubit’s magnetic dipole to the resonator’s magnetic
Ĥ = 4EC n̂ − EJ cos ϕ̂ + ~ωr ↠â + 8EC n̂n̂r field. Stronger interaction can be obtained by galvanically
X X connecting the flux qubit to the center conductor of a
= ~ωj |jihj| + ~ωr ↠â + i ~gij |iihj|(↠− â), transmission-line resonator (Bourassa et al., 2009). In
j ij
such a situation, the coupling can be engineered to be as
(59)
large, or even larger, than the system frequencies allowing
where |ji are now the eigenstates of the bare transmon
to reach what is known at the ultrastrong coupling regime,
Hamiltonian ĤT = 4EC n̂ − EJ cos ϕ̂ obtained from nu-
see Sec. VI.C.
merical diagonalization and we have defined
Yet another approach is to couple the qubit to the oscil-
Cg lator in such a way that the resonator field does not result
~gij = 2e Qzpf hi|n̂|ji. (60) in qubit transitions but only shifts the qubit’s frequency.
CCΣ
This is known as longitudinal coupling and is represented
The eigenfrequencies ωj and the matrix elements hi|n̂|ji by the Hamiltonian (Billangeon et al., 2015a,b; Didier
can be computed numerically in the charge basis. Alterna- et al., 2015a; Kerman, 2013; Richer and DiVincenzo, 2016;
tively, they can be determined by taking advantage of the Richer et al., 2017)
fact that, in the phase basis, Eq. (20) takes the form of a
Mathieu equation whose exact solution is known (Cottet, ~ωq
Ĥz = ~ωr ↠â + σ̂z + ~gz (↠+ â)σ̂z . (63)
2002; Koch et al., 2007). 2
The second form of Eq. (59) written in terms of energy
Because light-matter interaction in Ĥz is proportional to
eigenstates |ji is a very general Hamiltonian that can
σ̂z rather than σ̂x , the longitudinal interaction does not
describe an arbitrary multilevel artificial atom capacitively
lead to dressing of the qubit by the resonator field of the
coupled to a resonator. Similarly to the discussion
√ of
form discussed in Sec. III.B. Some of the consequences
Sec. III.C, in the dispersive regime where |gij | n + 1 
of this observation, in particular for qubit readout, are
|ωi − ωj − ωr | for all relevant atomic transitions i ↔ j
discussed in Sec. V.C.3.
and with n the oscillator photon number, it is possible
to use a Schrieffer-Wolff transformation to approximately
diagonalize Eq. (59). As discussed in Appendix B.2, to IV. COUPLING TO THE OUTSIDE WORLD: THE ROLE
second order one finds (Zhu et al., 2013) OF THE ENVIRONMENT
X
Ĥ ' ~(ωj + Λj )|jihj| + ~ωr ↠â So far we have dealt with isolated quantum systems. A
j complete description of quantum electrical circuits, how-
X (61)
+ ~χj ↠â|jihj|, ever, must also take into account a description of how
j these systems couple to their environment, including any
measurement apparatus and control circuitry. In fact,
where the environment plays a dual role in quantum technol-
ogy: Not only is a description of quantum systems as
X |gij |2
Λj = , (62a) perfectly isolated unrealistic, as coupling to unwanted
i
ωj − ωi − ωr environmental degrees of freedom is unavoidable, but a
X |gij |2 |gij |2
 perfectly isolated system would also not be very useful
χj = − . (62b) since we would have no means of controlling or observing
ωj − ωi − ωr ωi − ωj − ωr
i it. For these reasons, in this section we consider quantum
 20

systems coupled to external transmission lines. We also Cκ b̂in (t)

introduce the input-output theory which is of central im- Ĥtml
portance in understanding qubit readout in circuit QED
in the next section. b̂out (t)
ωr

A. Wiring up quantum systems with transmission lines ĤS

We start the discussion by considering transmission FIG. 11 LC circuit capacitively coupled to a semi-infinite
lines coupled to individual quantum systems, which are transmission line used to model both damping and driving of
a model for losses and can be used to apply and receive the system. Here, b̂in (t) and b̂out (t) are the oscillators input
and output fields, respectively.
quantum signals for control and measurement. To be
specific, we consider a semi-infinite coplanar waveguide
transmission line capacitively coupled at one end to an In the following, λ(ω) is assumed sufficiently small com-
oscillator, see Fig. 11. The semi-infinite transmission line pared to ωr , such that the interaction can be treated
can be considered as a limit of the coplanar waveguide as a perturbation. In this situation, the system’s Q fac-
resonator of finite length already discussed in Sec. II.B.1 tor is large and the oscillator only responds in a small
where one of the boundaries is now pushed to infinity, bandwidth around ωr . It is therefore reasonable to take
d → ∞. Increasing the length of the transmission line λ(ω) ' λ(ωr ) in Eq. (66). Dropping rapidly oscillating
leads to a densely packed frequency spectrum, which in its terms finally leads to (Gardiner and Zoller, 1999)
infinite limit must be treated as a continuum. In analogy Z ∞
with Eq. (16), the Hamiltonian of the transmission line is
Ĥ ' ĤS + Ĥtml + ~ dω λ(ωr )(âb̂†ω − ↠b̂ω ). (67)
consequently 0
Z ∞
Under the well-established Born-Markov approximations,
Ĥtml = dω ~ω b̂†ω b̂ω , (64) Eq. (67) leads to a Lindblad-form Markovian master equa-
0
tion for the system’s density matrix ρ (Breuer and Petruc-
where the mode operators now satisfy [b̂ω , b̂†ω0 ] = δ(ω − cione, 2002; Carmichael, 2002; Gardiner and Zoller, 1999)
ω 0 ). Similarly, the position-dependent flux and charge
operators of the transmission line are, in analogy with ρ̇ = −i[ĤS , ρ] + κ(n̄κ + 1)D[â]ρ + κn̄κ D[↠]ρ, (68)
Eqs. (9), (10), (14) and (15), given in the continuum limit where κ = 2πλ(ωr )2 = Ztml ωr2 Cκ2 /Cr is the photon de-
by (Yurke, 2004) cay rate, or linewidth, of the oscillator introduced earlier.
Z ∞ r  ωx  Moreover, n̄κ = n̄κ (ωr ) is the number of thermal photons
~ of the transmission line as given by the Bose-Einstein
Φ̂tml (x) = dω cos (b̂†ω + b̂ω ), (65a)
0 πωcv v distribution, hb̂†ω b̂ω0 i = n̄κ (ω)δ(ω − ω 0 ), at the system fre-
Z ∞ r  ωx 
~ωc quency ωr and environment temperature T . The symbol
Q̂tml (x) = i dω cos (b̂†ω − b̂ω ). (65b) D[Ô]• represents the dissipator
0 πv v
1n † o
These are the canonical fields of the transmission line D[Ô]• = Ô • Ô† − Ô Ô, • , (69)
and in the Heisenberg picture under Eq. (64) are related 2
˙ with {·, ·} the anticommutator. Focussing on the second
through√ Q̂tml (x, t) = cΦ̂tml (x, t). In these expressions,
v = 1/ lc is the speed of light in the transmission line, term of Eq. (68), the role of this superoperator can be
with c and l the capacitance and inductance per unit understood intuitively by noting that the term ÔρÔ†
length, respectively. with Ô = â in Eq. (69) acts on the Fock state |ni as
Considering capacitive coupling of the line to the oscil- â|nihn|↠= n|n − 1ihn − 1|. The second term of Eq. (68)
lator at x = 0, the total Hamiltonian takes the form therefore corresponds to photon loss at rate κ. Finite
Z ∞ temperature stimulates photon emission, boosting the loss
rate to κ(n̄κ + 1). On the other hand, the last term of
Ĥ = ĤS + Ĥtml − ~ dω λ(ω)(b̂†ω − b̂ω )(↠− â), (66)
0 Eq. (68) corresponds to absorption of thermal photons by
the system. Because ~ωr  kB T at dilution refrigerator
where ĤS = ~ωr ↠â √
is the p
oscillator Hamiltonian. More- temperatures, it is often assumed that n̄κ → 0. Devia-
over, λ(ω) = (Cκ / cCr ) ωr ω/2πv is the frequency- tions from this expected behavior are, however, common
dependent coupling strength, with Cκ the coupling capac- in practice due to residual thermal radiation propagating
itance and Cr the resonator capacitance. These expres- along control lines connecting to room temperature equip-
sions neglect small renormalizations of the capacitances ment and to uncontrolled sources of heating. Approaches
due to Cκ , as discussed in Appendix A. to mitigate this problem using absorptive components
 21

are being developed (Córcoles et al., 2011; Wang et al., in Eq. (71) leads to the standard input-output relation
2019b). (see Appendix C for details)
√
b̂out (t) − b̂in (t) = κâ(t), (72)
B. Input-output theory in electrical networks
where the input and output fields are defined as
While the master equation describes the system’s Z ∞
damped dynamics, it provides no information on the fields −i
b̂in (t) = √ dω b̂Lω e−i(ω−ωr )t , (73)
radiated by the system. Since radiated signals are what 2π −∞
Z ∞
are measured experimentally, it is of practical importance −i
to include those in our model. b̂out (t) = √ dω b̂Rω e−i(ω−ωr )t (74)
2π −∞
This is known as the input-output theory for which two
standard approaches exist. The first is to work directly and satisfy the commutation relations [b̂in (t), b̂†in (t0 )] =
with Eq. (67) and consider Heisenberg picture equations [b̂out (t), b̂†out (t0 )] = δ(t − t0 ). To arrive at Eq. (72), terms
of motion for the system and field annihilation opera- rotating at ω + ωr have been dropped based on the al-
tors â and b̂ω . This is the route taken by Gardiner and ready mentioned assumption that the system only re-
Collett, which is widely used in the quantum optics liter- sponds to frequencies ω ' ωr such that these terms are
ature (Collett and Gardiner, 1984; Gardiner and Collett, fast rotating (Yurke, 2004). In turn, this approximation
1985). allows to extend the range of integration from (0, ∞) to
An alternative approach is to introduce a decomposition (−∞, ∞) in Eqs. (73) and (74). We have also approxi-
of the transmission line modes in terms of left- and right- mated λ(ω) ' λ(ωr ) over the relevant frequency range.
moving fields, linked by a boundary condition at the These approximations are compatible with those used to
position of the oscillator which we take to be x = 0 with arrive at the Lindblad-form Markovian master equation
the transmission line at x ≥ 0 (Yurke and Denker, 1984). of Eq. (68).
The advantage of this approch is that the oscillator’s The same expressions and approximations can be used
input and output fields are then defined in terms of easily to obtain the equation of motion for the resonator field â(t)
identifiable left- (b̂Lω ) and right-moving (b̂Rω ) radiation in the Heisenberg picture, which takes the form (see Ap-
field components propagating along the transmission line. pendix C for details)
To achieve this, we replace the modes cos(ωx/v)b̂ω in
Eqs. (65a) and (65b) by (b̂Rω eiωx/v +b̂Lω e−iωx/v )/2. Since κ √
˙
â(t) = i[ĤS , â(t)] − â(t) + κb̂in (t). (75)
the number of degrees of freedom of the transmission 2
line has seemingly doubled, the modes b̂L/Rω cannot be This expression shows that the resonator dynamics is
independent. Indeed, the dynamics of one set of modes determined by the input field (in practice, noise or drive),
is fully determined by the other set through a boundary while Eq. (72) shows how the output can, in turn, be found
condition linking the left- and right-movers at x = 0. from the input and the system dynamics. The output field
To see this, it is useful to first decompose the voltage thus holds information about the system’s response to the
˙
V̂ (x, t) = Φ̂tml (x, t) at x = 0 into left-moving (input) and input and which can be measured to, indirectly, give us
right-moving (output) contributions as V̂ (t) = V̂ (x = access to information about the dynamics of the system.
0, t) = V̂in (t) + V̂out (t), where As will be discussed in more detail in Sec. V, this can
Z ∞ r be done, for example, by measuring the voltage at some
~ω iωt †
V̂in/out (t) = i dω e b̂L/Rω + H.c. (70) x > 0 away from the oscillator. Under the approximations
0 4πcv used above, this voltage can be expressed as
The boundary condition at x = 0 follows from Kirchhoff’s r 
current law ~ωr Ztml iωr x/v−iωr t
V̂ (x, t) ' e b̂out (t)
2
 (76)
ˆ = V̂out (t) − V̂in (t) ,
I(t) (71)
Ztml + e−iωr x/v−iωr t b̂in (t) + H.c. .
ˆ ˙
where the left-hand side I(t) = (Cκ /Cr )Q̂r (t) is the
Note that this approximate expression assumes that all
current injected by the sample, with Q̂r the oscillator
relevant frequencies are near ωr and furthermore neglects
charge (see Appendix C for a derivation), while the right-
all non-Markovian time-delay effects.
hand side is the transmission line voltage difference at
x = 0.6 A mode expansions of the operators involved

at x = 0, it would follow that V̂in (t) = V̂out (t), i.e. the endpoint
6 Note that if instead we have a boundary condition of zero current simply serves as a mirror reflecting the input signal.
 22

rameters controlling their transition frequency and by

dispersive coupling to other degrees of freedom in their
environment. For a transmon, a phenomenological model
for dephasing can be introduced by adding the following
term to the master equation (Carmichael, 2002)
Φx
ĤS 2γϕ D[b̂† b̂]ρ, (78)

with γϕ the pure dephasing rate. Because of its insensi-

FIG. 12 Transmon qubit capacitively coupled to a semi- tivity to charge noise (see Fig. 6), γϕ is often very small
infinite transmission line and inductively to a flux line. These for the 0-1 transition of transmon qubits (Koch et al.,
ports are used to control the qubit state and to change its
2007). Given that charge dispersion increases exponen-
transition frequency. They also lead to qubit decay into the
transmission line and to dephasing due to flux noise. tially with level number, dephasing due to charge noise
can, however, be apparent on higher transmon levels, see
for example Egger et al. (2019). Another source of de-
In this section we have considered a particularly simple phasing for the transmon is the residual thermal photon
setup: A single quantum system connected to the endpoint population of a resonator to which the transmon is disper-
of a semi-infinite transmission line. More generally, quan- sively coupled. This can be understood from the form of
tum systems can be made to interact by coupling them to the interaction in the dispersive regime, χab ↠âb̂† b̂, where
a common transmission line, and multiple transmission fluctuations of the photon number lead to a fluctuations
lines can be used to form quantum networks. These more in the qubit frequency and therefore to dephasing (Bertet
complex setups can be treated using the SLH formalism, et al., 2005; Gambetta et al., 2006; Rigetti et al., 2012;
which generalizes the results in this section (Combes et al., Schuster et al., 2005). We note that a term of the form
2016; Gough and James, 2009). of Eq. (78) but with b̂† b̂ replaced by ↠â can be added
to the master equation of the oscillator to model this
aspect. Oscillator dephasing rates are, however, typically
C. Qubit relaxation and dephasing small and this contribution is often neglected (Reagor
et al., 2016). Other sources of relaxation and dephasing
The master equation Eq. (68) was derived for an os- include two-level systems within the materials and inter-
cillator coupled to a transmission line, but this form of faces of the devices (Müller et al., 2019), quasiparticles
the master equation is quite general. In fact, Eq. (66) is (Glazman and Catelani, 2020) generated by a number of
itself a very generic system-bath Hamiltonian that can phenomenon including infrared radiation (Barends et al.,
be used to model dissipation due to a variety of different 2011; Córcoles et al., 2011) and even ionizing radiation
noise sources (Caldeira and Leggett, 1981). To model (Vepsäläinen et al., 2020).
damping of an arbitrary quantum system, for example a Combining the above results, the master equation for a
transmon qubit or a coupled resonator-transmon system, transmon subject to relaxation and dephasing assuming
the operator â in Eq. (66) is simply replaced with the n̄γ → 0 is
relevant system operator that couples to the transmission
line (or, more generally, the bath). ρ̇ = − i[Ĥq , ρ] + γD[b̂]ρ + 2γϕ D[b̂† b̂]ρ. (79)
For the case of a transmon, see Fig. 12, ĤS in Eq. (68)
is replaced with the Hamiltonian Ĥq of Eq. (25) to- It is common to express this master equation in the two-
gether with the additional replacements D[â]• → D[b̂]•, level approximation of the transmon, something that
D[↠]• → D[b̂† ]•, and κ → γ. Here, γ = 2πλ(ωq )2 is the is obtained simply by taking Ĥq → ~ωa σ̂z /2, b̂† b̂ →
relaxation rate of the artificial atom which is related to (σ̂z + 1) /2, b̂ → σ̂− and b̂† → σ̂+ .
the qubit-environment coupling strength evaluated at the Note that the rates γ and γϕ appearing in the above
qubit frequency. This immediately leads to the master expressions are related to the characteristic T1 relaxation
equation time and T2 coherence time of the artificial atom which
are defined as (Schoelkopf et al., 2003)
ρ̇ = −i[Ĥq , ρ] + γ(n̄γ + 1)D[b̂]ρ + γ n̄γ D[b̂† ]ρ, (77)
1 1 1
T1 = = ' , (80a)
where ρ now refers to the transmon state and n̄γ is the γ1 γ↓ + γ↑ γ
thermal photon number of the transmon’s environment. 1 γ −1
1
It is often assumed that n̄γ → 0 but, just like for the T2 = = + γϕ , (80b)
γ2 2
oscillator, a residual thermal population is often observed
in practice (Córcoles et al., 2011; Wang et al., 2019b). where γ↓ = (n̄γ + 1)γ, γ↑ = n̄γ γ. The approximation
Superconducting quantum circuits can also suffer from in Eq. (80a) is for n̄γ → 0. At zero temperature, T1 is
dephasing caused, for example, by fluctuations of pa- the characteristic time for the artificial atom to relax
 23

from its first excited state to the ground state. On the

other hand, T2 is the dephasing time, which quantifies
the characteristic lifetime of coherent superpositions, and
includes both a contribution from pure dephasing (γϕ )
and relaxation (γ1 ). Current best values for the T1 and
T2 time of transmon qubits is in the 50 to 120 µs range
for aluminum-based transmons (Devoret and Schoelkopf, ωr
2013; Kjaergaard et al., 2019; Nersisyan et al., 2019; Wei
et al., 2019). Relaxation times above 300 µs have been
reported in transmon qubits where the transmon pads Φx ωr
have been made with tantalum rather than aluminum,
but the Josephson junction still made from aluminum and
aluminum oxide (Place et al., 2020). Other superconduct-
ing qubits also show large relaxation and coherence times. FIG. 13 Because the dressed states in the dispersive regime
Examples are T1 , T2 ∼ 300 µs for heavy-fluxonium qubits are entangled qubit-cavity states, cavity damping at the rate
(Zhang et al., 2020), and T1 ∼ 1.6 ms and T2 ∼ 25 µs for κ leads to qubit relaxation at the Purcell rate γκ . Conversely,
qubit relaxation leads to cavity decay at the rate inverse
the 0 − π qubit (Gyenis et al., 2019).
Purcell rate κγ . Adding a Purcell filter (not shown) reduces
Qubit relaxation and incoherent excitation occur due the cavity density of states at the qubit frequency and therefore
to uncontrolled exchange of GHz frequency photons be- suppresses Purcell decay.
tween the qubit and its environment. These processes are
observed to be well described by the Markovian master
equation of Eq. (79). In contrast, the dynamics leading dissipative processes influence the transmon and the res-
to dephasing are typically non-Markovian, happening at onator in completely independent manners. However,
low-frequencies (i.e. slow time scales set by the phase because Ĥ entangles the two systems, the loss, for exam-
coherence time itself). As a result, these processes can- ple, of a resonator photon can lead to qubit relaxation.
not very accurately be described by a Markovian master Moving to the dispersive regime, a more complete picture
equation such as Eq. (79). This equation thus repre- of dissipation emerges after applying the unitary trans-
sents a somewhat crude approximation to dephasing in formation Eq. (47) not only on the Hamiltonian but also
superconducting qubits. That being said, in practice, the on the above master equation. Neglecting fast rotating
Markovian theory is still useful in particular because it terms and considering corrections to second order in λ
correctly predicts the results of experiments probing the (which is consistent if κ, γ, γϕ = O(EC g 2 /∆2 )), leads
steady-state response of the system. to the dispersive master equation (Boissonneault et al.,
2009)

D. Dissipation in the dispersive regime ρ̇disp = − i[Ĥdisp , ρdisp ]

We now turn to a description of dissipation for the + (κ + κγ )D[â]ρdisp + (γ + γκ )D[b̂]ρdisp

(82)
coupled transmon-resonator system of Sec. III. Assum- + 2γϕ D[b̂† b̂]ρdisp
ing that the transmon and the resonator are coupled to
+ γ∆ D[↠b̂]ρdisp + γ∆ D[b̂† â]ρdisp ,
independent baths as illustrated in Fig. 13, the master
equation for this composite system is (taking n̄κ,γ → 0 where
for simplicity)  g 2  g 2  g 2
ρ̇ = − i[Ĥ, ρ] + κD[â]ρ + γD[b̂]ρ γκ = κ, κγ = γ, γ∆ = 2 γϕ , (83)
∆ ∆ ∆
(81)
+ 2γϕ D[b̂† b̂]ρ, †
and ρdisp = Ûdisp ρÛdisp is the density matrix in the disper-
where ρ is now a density matrix for the total system, and sive frame. This expression has three new rates, the first
Ĥ describes the coupled system as in Eq. (32). Impor- of which is known as the Purcell decay rate γκ (Purcell,
tantly, the above expression is only valid at small values 1946). This rate captures the fact that the qubit can relax
of g/(ωr , ωq ). This is because energy decay occurs via by emission of a resonator photon. It can be understood
transitions between system eigenstates while the above simply following Eq. (39) from the form of the dressed
expression describes transitions between the uncoupled eigenstate |e, 0i ∼ |e, 0i + (g/∆)|g, 1i which is closest to
bare states. A derivation of the master equation valid at a qubit excitation |ei. This state is the superposition of
arbitrary g can be found, for example, in Beaudoin et al. the qubit first excited state with no photon and, with
(2011). probability (g/∆)2 , the qubit ground state with a photon
More important to the present discussion is the fact in the resonator. The latter component can decay at the
that, at first glance, Eq. (81) gives the impression that rate κ taking the dressed excited qubit to the ground state
 24

|g, 0i with a rate γκ . A similar intuition also applies to E. Multi-mode Purcell effect and Purcell filters
κγ , now associated with a resonator photon loss through
a qubit decay event. Up to now we have considered dissipation for a qubit
dispersively coupled to a single-mode oscillator. Replacing
the latter with a multi-mode resonator leads to dressing
The situation is more subtle for the last line of Eq. (82).
of the qubit by all of the resonator modes and there-
Following Boissonneault et al. (2008, 2009), an effective
fore to a modification of the Purcell decay rate. Fol-
master equation for the transmon only can be obtained
lowing the above discussion, one may then expect the
from Eq. (82) by approximately eliminating the resonator
contributions
P∞ to add up, leading to the modified rate
degrees of freedom. This results in transmon relaxation
m=0 (gm /∆ m )2 κm , with m the mode index. However,
and excitation rates given approximately by n̄γ∆ , with
when accounting for the frequency dependence of κm , gm
n̄ the average photon number in the resonator. Com-
and ∆m , this expression diverges (Houck et al., 2008). It
monly known as dressed-dephasing, this leads to spurious
is possible to cure this problem using a more refined model
transitions during qubit measurement and can be inter-
including the finite size of the transmon and the frequency
preted as originating from dephasing noise at the detuning
dependence of the impedance of the resonator’s input and
frequency ∆ that is up- or down-converted by readout
output capacitors (Bourassa, 2012; Malekakhlagh et al.,
photons to cause spurious qubit state transitions.
2017).
Given that damping rates in quantum electrical circuits
Because we have taken the shortcut of applying the are set by classical system parameters (Leggett, 1984b),
dispersive transformation on the master equation, the a simpler approach to compute the Purcell rate exists.
above discussion neglects the frequency dependence of It can indeed be shown that γκ = Re[Y (ωq )]/CΣ , with
the various decay rates. In a more careful derivation, the Y (ω) = 1/Z(ω) the admittance of the electromagnetic
dispersive transformation is applied on the system plus environment seen by the transmon (Esteve et al., 1986;
bath Hamiltonian, and only then is the master equation Houck et al., 2008). This expression again makes it clear
derived (Boissonneault et al., 2009). The result has the that relaxation probes the environment (here represented
same form as Eq. (82), but with different expressions for by the admittance) at the system frequency. It also sug-
the rates. Indeed, it is useful to write κ = κ(ωr ) and gests that engineering the admittance Y (ω) such that it is
γ = γ(ωq ) to recognize that, while photon relaxation purely reactive at ωq can cancel Purcell decay (see the in-
is probing the environment at the resonator frequency set of Fig. 13). This can be done, for example, by adding
ωr , qubit relaxation is probing the environment at ωq . a transmission-line stub of appropriate length and termi-
With this notation, the first two rates of Eq. (83) become nated in an open circuit at the output of the resonator,
in the more careful derivation γκ = (g/∆)2 κ(ωq ) and something which is known as a Purcell filter (Reed et al.,
κγ = (g/∆)2 γ(ωr ). In other words, Purcell decay occurs 2010b). Because of the increased freedom in optimizing
by emitting a photon at the qubit frequency and not at the system parameters (essentially decoupling the choice
the resonator frequency as suggested by the completely of κ from the qubit relaxation rate), various types of
white noise model used to derive Eq. (83). In the same Purcell filters are commonly used experimentally (Bronn
way, it is useful to write the dephasing rate as γϕ = et al., 2015; Jeffrey et al., 2014; Walter et al., 2017).
γϕ (ω → 0) to recognize the importance of low-frequency
noise. Using this notation, the rates in the last two terms
of Eq. (82) become, respectively, γ∆ = 2(g/∆)2 γϕ (∆) and F. Controlling quantum systems with microwave drives
γ−∆ = 2(g/∆)2 γϕ (−∆) (Boissonneault et al., 2009). In
short, dressed dephasing probes the noise responsible for While connecting a quantum system to external trans-
dephasing at the transmon-resonator detuning frequency mission lines leads to losses, such connections are nev-
∆. This observation was used to probe this noise at GHz ertheless necessary to control and measure the system.
frequencies by Slichter et al. (2012). Consider a continuous microwave tone of frequency ωd
and phase φd applied to the input port of the resonator.
A simple approach to model this drive is based on the
It is important to note that the observations in this
input-output approach of Sec. IV.B. Indeed, the drive
section result from the qubit-oscillator dressing that oc-
can be taken into account by replacing the input field
curs under the Jaynes-Cummings Hamiltonian. For this
b̂in (t) in Eq. (75) with b̂in (t) → b̂in (t) + β(t), where
reason, the situation is very different if the electric-dipole
β(t) = A(t) exp(−iωd t−iφd ) is the coherent classical part
interaction leading to the Jaynes-Cummings Hamiltonian
of
√ the input field of amplitude A(t). The resulting term
is replaced by a longitudinal interaction of the form of
κβ(t) in the Langevin equation can be absorbed in the
Eq. (63). In this case, there is no light-matter dressing
system Hamiltonian with the replacement ĤS → ĤS + Ĥd
and consequently no Purcell decay or dressed-dephasing
where
(Billangeon et al., 2015a; Kerman, 2013). This is one of  
the advantages of this alternative light-matter coupling. Ĥd = ~ ε(t)↠e−iωd t−iφd + ε∗ (t)âeiωd t+iφd ), (84)
 25
√
with ε(t) = i κA(t) the possibly time-dependent ampli- 300 K 4K 10 mK
tude of the drive as seen by the resonator mode. Gener- b̂out
alizing to multiple drives on the resonator and/or drives
on the transmon is straightforward.
RF Att. Att.
b̂in
Moreover, the Hamiltonian Ĥd is the generator of dis-
placement in phase space of the resonator. As a result, IF âamp
by choosing appropriate parameters for the drive, evolu- FPGA ADC
tion under Ĥd will bring the intra-resonator state from b̂in
vacuum to an arbitrary coherent state (Carmichael, 2002;
LO
Gardiner and Zoller, 1999)
X∞ Signal gen.
2 αn Circulator HEMT amp.
|αi = D̂(α)|0i = e−|α| /2
√ |ni, (85)
n=0 n! IQ Mixer Att. Attenuator Q-limited amp.

where D̂(α) is known as the displacement operator and FIG. 14 Schematic representation of the microwave mea-
takes the form surement chain for field detection in circuit QED, with the
resonator depicted as a Fabry-Perot cavity. The signal (RF)
†
−α∗ â
D̂(α) = eαâ . (86) from a microwave source is applied to the input port of the
resonator first passing through attenuators to reduce the level
As discussed in the next section, coherent states play an of thermal radiation. After passing through a circulator, the
important role in qubit readout in circuit QED. resonator’s output field is first amplified by a quantum-limited
amplifier, such as a JPA or a JTWPA, and then by an HEMT
It is important to note that Ĥd derives from Eq. (75)
amplifier. The signal is then mixed with a local oscillator
which is itself the result of a rotating-wave approxi- (LO). The signal at the output of the mixer is digitized with
mation. As can be understood from Eq. (20), before an analog-to-digital converter (ADC) and can be further pro-
this approximation, the drive rather takes the form cessed by a field-programmable gate array (FPGA). The two
i~ε(t) cos(ωd t + φd )(↠− â). Although Ĥd is sufficient lines at the output of the mixer correspond to the two quadra-
in most cases of practical interest, departures from the tures of the field. The temperature at which the different
predictions of Eq. (84) can been seen at large drive am- components are operated is indicated.
plitudes (Pietikäinen et al., 2017; Verney et al., 2019).
moreover inhibited when working in the dispersive regime
V. MEASUREMENTS IN CIRCUIT QED where the effective qubit-resonator interaction Eq. (42) is
such that even the probe-tone photons are not absorbed
Before the development of circuit QED, the quantum by the qubit. As a result, the backaction on the qubit is
state of superconducting qubits was measured by fab- to a large extent limited to the essential dephasing that
ricating and operating a measurement device, such as quantum measurements must impart on the measured
a single-electron transistor, in close proximity to the system leading, in principle, to a quantum non-demolition
qubit (Makhlin et al., 2001). A challenge with such an (QND) qubit readout.
approach is that the readout circuitry must be strongly Because of the small energy of microwave photons with
coupled to the qubit during measurement so as to ex- respect to optical photons, single-photon detectors in the
tract information on a time scale much smaller than T1 , microwave frequency regime are still being developed, see
while being well decoupled from the qubit when the mea- Sec. VIII.F. Therefore, measurements in circuit QED rely
surement is turned off to avoid unwanted back-action. on amplification of weak microwave signals followed by
Especially given that measurement necessarily involves detection of field quadratures using heterodyne detection.
dissipation (Landauer, 1991), simultaneously satisfying Before discussing qubit readout, the objective of the next
these two requirements is challenging. Circuit QED, how- subsection is to explain these terms and go over the main
ever, has several advantages to offer over the previous challenges related to such measurements in the quantum
approaches. Indeed, as discussed in further detail in this regime.
section, qubit readout in this architecture is realized by
measuring scattering of a probe tone off an oscillator cou-
pled to the qubit. This approach first leads to an excellent A. Microwave field detection
measurement on/off ratio since qubit readout only occurs
in the presence of the probe tone. A second advantage Figure 14 illustrates a typical measurement chain in
is that the necessary dissipation now occurs away from circuit QED. The signal of a microwave source is directed
the qubit, essentially at a voltage meter located at room to the input port of the resonator first going through
temperature, rather than in a device fabricated in close a series of attenuators thermally anchored at different
proximity to the qubit. Unwanted energy exchange is stages of the dilution refrigerator. The role of these atten-
 26

uators is to absorb the room-temperature thermal noise violate the Heisenberg uncertainty relation between the
propagating towards the sample. The field transmitted two quadratures.
by the resonator is first amplified, then mixed with a ref- In a standard parametric amplifier, âf in Eq. (88) rep-
erence signal, converted from analog to digital, and finally resents the amplitude of the signal mode and h represents
processed with an FPGA or recorded. Circulators are the amplitude of a second mode called the idler. The
inserted before the amplification stage to prevent noise physical interpretation of Eq. (88) is that an ideal ampli-
generated by the amplifier from reaching the resonator. fier performs a Bogoljubov transformation on the signal
Circulators are directional devices that transmit signals and idler modes. The signal mode is amplified, but the
in the forward direction while strongly attenuating signals requirement that the transformation be canonical im-
propagating in the reverse direction (here coming from the plies that the (phase conjugated and amplified) quantum
amplifier) (Pozar, 2011). In practice, circulators are bulky noise from the idler port must appear in the signal out-
off-chip devices relying on permanent magnets that are put port. Ideally, the input to the idler is vacuum with
not compatible with the requirement for integration with hĥ† ĥi = 0 and hĥĥ† i = 1, so the amplifier only adds quan-
superconducting quantum circuits. They also introduce tum noise. Near quantum-limited amplifiers with ∼ 20 dB
additional losses, for example due to insertion losses and power gain approaching this ideal behavior are now rou-
off-chip cable losses. Significant effort is currently being tinely used in circuit QED experiments. These Josephson
devoted to developing compact, on-chip, superconducting junction-based devices, as well as the distinction between
circuit-based circulators (Abdo et al., 2019; Chapman phase-preserving and phase-sensitive amplification, are
et al., 2017; Kamal et al., 2011; Lecocq et al., 2017). discussed further in Sec. VIII.B.
In practice, the different components and cables of the To measure the very weak signals that are typical in
measurement chain have a finite bandwidth which we will circuit QED, the output of the first near-quantum limited
assume to be larger than the bandwidth of the signal of amplifier is further amplified by a low-noise high-electron-
interest b̂out (t) at the output of the resonator. To account mobility transistor (HEMT) amplifier. The latter acts
for the finite bandwidth of the measurement chain and to on the signal again following Eq. (88), now with a larger
simplify the following discussion, it is useful to consider power gain ∼ 30 − 40 dB but also larger added noise
the filtered output field photon number. The very best cryogenic HEMT ampli-
fiers in the 4-8 GHz band have noise figures as low as
âf (t) = (f ? b̂out )(t) hĥ† ĥi ∼ 5 − 10. However, the effect of attenuation due to
Z ∞
cabling up to the previous element of the amplification
= dτ f (t − τ )b̂out (τ ) chain, i.e. a quantum-limited amplifier or the sample of in-
−∞ (87)
Z ∞ h√ i terest itself, can degrade this figure significantly. A more
= dτ f (t − τ ) κâ(τ ) + b̂in (τ ) , complete understanding of the added noise in this situa-
−∞
tion can be derived from Fig. 15(a). There, beam splitters
which is linked to the intra-cavity field â via the input- of transmissivity η1,2 model the attenuation leading to
ouput boundary condition Eq. (72) which we have used the two amplifiers of gain labelled G1 and G2 . Taking
in the last line. In this Rexpression, the filter func- into account vacuum noise v̂1,2 at the beam splitters, the
∞ input-output expression of this chain can be cast under
tion f (t) is normalized to −∞ dt|f (t)|2 = 1 such that
the form of Eq. (88) with a total gain GT = η1 η2 G1 G2
[âf (t), â†f (t)] = 1. As will be discussed later in the con-
text of qubit readout, in addition to representing the and noise mode ĥ†T corresponding to the total added noise
measurement bandwidth, filter functions are used to opti- number
mize the distinguishability between the qubit states. 1 h
NT = η1 (G1 − 1)G2 (N1 + 1)
Ignoring the presence of the circulator and assuming GT − 1
i
that a phase-preserving amplifier (i.e. an amplifier that + (G2 − 1)(N2 + 1) − 1 (89)
amplifies both signal quadratures equally) is used, in  
the first stage of the measurement chain the signal is 1 N2
≈ 1 + N1 + − 1,
transformed according to (Caves, 1982; Clerk et al., 2010) η1 η2 G1
√ √
âamp = Gâf + G − 1ĥ† , (88) with Ni = hĥ†i ĥi i with i = 1, 2, T . The last expression
corresponds to the large gain limit. If the gain G1 of the
where G is the power gain and ĥ† accounts for noise first amplifier is large, the noise of the chain is dominated
added by the amplifier. The presence of this added noise by the noise N1 of the first amplifier. This emphasizes the
is required for the amplified signal to obey the bosonic importance of using near quantum-limited amplifiers with
commutation relation, [âamp , â†amp ] = 1. Equivalently, the low noise in the first stage of the chain. In the literature,
noise must be present because the two quadratures of the quantum efficiency η = 1/(NT + 1) is often used to
the signal are canonically conjugate. Amplification of characterize the measurement chain, with η = 1 in the
both quadratures without added noise would allow us to ideal case NT = 0.
 27
t1 t2
(a) η1 η2
Q
G1 G2
v̂1 ĥ1 v̂2 ĥ2 π/2

(b) η̄
LO
G/η̄ âamp
I
v̂
t v̂
FIG. 15 (a) Amplification chain with amplifiers of gain G1,2
and noise mode ĥ1,2 with attenuation modeled by beam split-
ters of transmitivity η1,2 . The beam splitters each have a vac- FIG. 16 Schematic representation of an IQ mixer. The RF
†
uum port with vacuum mode v̂1,2 such that hv̂1,2 v̂1,2 i = 0. The signal âamp is split into two parts at a power divider, here
quantum efficiency derived from this model is η = 1/(NT +1) ≤ illustrated as a beam splitter to account for added noise due
to internal modes. Ideally, only vacuum noise v̂ is introduced
1, with NT = hĥ†T ĥT i the total added noise number given in
at that stage. The two outputs are combined with a local
Eq. (89). (b) Alternative model where a noisy amplifier is
oscillator (LO) at mixers. By phase shifting the LO by π/2 in
modeled by a noiseless amplifier preceded by a beam splitter
one of the two arms, it is possible to simultaneously measure
of transmitivity η̄. The quantum efficiency derived from this
the two quadratures of the field.
model is η̄ = 1/(2A + 1) ≤ 1/2 with A the added noise given
in Eq. (92).

be written as η̄ = 1/(2A + 1) ≤ 1/2 and is found to be

It is worthwhile to note that another definition of the bounded by 1/2 in the ideal case. Importantly, the concept
quantum efficiency can often be found in the literature. of quantum efficiency is not limited to amplification, and
This alternative definition is based on Fig. 15(b) where a can be applied to the whole measurement chain illustrated
noisy amplifier of gain G is replaced by a noiseless ampli- in Fig. 14.
fier of gain G/η̄ preceded by a fictitious beam splitter of Using Eqs. (76) and (88), the voltage after amplification
transmittivity η̄ adding vacuum noise to the amplifier’s in- can be expressed as
put (Leonhardt and Paul, 1993). The quantum efficiency r
corresponds, here, to the transmittivity η̄ of the fictitious ~ωRF Ztml  −iωRF t 
V̂amp (t) ' e âamp + H.c. , (93)
beam splitter. The input-ouput relation of the network 2
of Fig. 15(b) with
p its √ noiseless√phase-preserving amplifier
reads âamp = G/η̄( η̄âf + 1 − η̄v̂), something which where ωRF is the signal frequency. To simplify the expres-
can be expressed as sions, we have dropped the phase associated to the finite
  cable length. We have also dropped the contribution from
2 G 1 2 the input field b̂in (t) moving towards the amplifier in the
h|âamp | i = (1 − η̄) + η̄h|âf | i , (90)
η̄ 2 opposite direction at this point, cf. Fig. 14, because this
field is not amplified and therefore gives a very small
with h|Ô|2 i = h{Ô† , Ô}i/2 the symmetrized fluctuations. contribution compared to the amplified output field. Re-
The first term of the above expression corresponds to the call, however, the contribution of this field to the filtered
noise added by the amplifier, here represented by vacuum signal Eq. (87).
noise added to the signal before amplification, while the Different strategies can be used to extract information
second term corresponds to noise in the signal at the input from the amplified signal, and here we take the next
of the amplifier. On the other hand, Equation (88) for a stage of the chain to be an IQ-mixer. As schematically
noisy amplifier can also be cast in the form of Eq. (90) illustrated in Fig. 16, in this microwave device the signal
with first encounters a power divider, illustrated here as a beam
splitter to account for added noise due to internal modes,
h|âamp |2 i = G(A + h|âf |2 i), (91) followed in each branch by mixers with local oscillators
where we have introduced the added noise (LO) that are offset in phase by π/2. The LO consists in a
  reference signal of well-defined amplitude ALO , frequency
(G − 1) † 1 ωLO and phase φLO :
A= hĥ ĥi + . (92)
G 2
VLO (t) = ALO cos(ωLO t − φLO ). (94)
In the limit of low amplifier noise hĥ† ĥi → 0 and large
gain, the added noise is found to be bounded by A ≥ Mixers use nonlinearity to down-convert the input sig-
(1 − G−1 )/2 ' 1/2 corresponding to half a photon of noise nal to a lower frequency referred to as the intermediate
(Caves, 1982). Using Eqs. (90) and (91), the quantum frequency (IF) signal. Describing first the signal as a clas-
efficiency of a phase-preserving amplifier can therefore sical voltage VRF (t) = ARF cos(ωRF t + φRF ), the output
 28

at one of these mixers is (Pozar, 2011) to extract X̂f (t) and P̂f (t).
We note that the case ωIF = 0 is generally known as
Vmixer (t) = KVRF (t)VLO (t) homodyne detection (Gardiner and Zoller, 1999; Leon-
1 hardt, 1997; Pozar, 2011; Wiseman and Milburn, 2010).
= KALO ARF {cos[(ωLO − ωRF )t − φLO ]
2 In this situation, the IF signal in one of the arms of the
+ cos[(ωLO + ωRF )t − φLO ]} , IQ-mixer is proportional to
(95)
â†f eiφLO + âf e−iφLO
where K accounts for voltage conversion losses. Accord- X̂f,φLO =
2 (99)
ing to the above expression, mixing with the LO results
in two sidebands at frequencies ωLO ± ωRF . The high = X̂f cos φLO + P̂f sin φLO .
frequency component is filtered out with a low-pass filter
While this is in appereance simpler and therefore advan-
(not shown) leaving only the lower sideband of frequency
tageous, this approach is succeptible to 1/f noise and
ωIF = ωLO − ωRF . The choice ωIF 6= 0 is known as hetero-
drift because the homodyne signal is at DC. It is also
dyne detection. Taking the LO frequency such that ωIF
worthwhile to note that homodyne detection as realized
is in the range of few tens to a few hundreds of MHz, the
with the approach described here differs from optical ho-
signal can be digitized using an analog to digital converter
modyne detection which can be performed in a noiseless
(ADC) with a sampling rate chosen in accordance with
fashion (in the present case, noise is added at the very
the bandwidth requirements of the signal to be recorded.
least by the phase-preserving amplifiers and the noise
This bandwidth is set by the choice of IF frequency and
port of the IQ mixer). The reader is referred to Schuster
the signal bandwidth, typically a few MHz to a few tens of
et al. (2005) and Krantz et al. (2019) for more detailed
MHz if set by the bandwidth κ/2π of the cavity chosen for
discussions of the different field measurement techniques
the specific circuit QED application such as qubit readout.
and their applications in the context of circuit QED.
The recorded signal can then be averaged, or analyzed in
more complex ways, using real-time field-programmable
gate array (FPGA) electronics or processed offline. A de-
B. Phase-space representations and their relation to field
tailed discussion of digital signal processing in the context
detection
of circuit QED can be found in Salathé et al. (2018).
Going back to a quantum mechanical description of the
In the context of field detection, it is particularly useful
signal by combining Eqs. (93) and (95), the IF signals at
to represent the quantum state of the electromagnetic field
the I and Q ports of the IQ-mixer read
using phase-space representations. There exists several
h i
such representations and here we focus on the Wigner
V̂I (t) = VIF X̂f (t) cos(ωIF t) − P̂f (t) sin(ωIF t)
(96a) function and the Husimi-Q distribution (Carmichael, 2002;
+ V̂noise,I (t), Haroche and Raimond, 2006). This discussion applies
h i equality well to the intra-cavity field â as to the filtered
V̂Q (t) = −VIF P̂f (t) cos(ωIF t) + X̂f (t) sin(ωIF t) output field âf .
(96b)
+ V̂noise,Q (t), The Wigner function is a quasiprobability distribution
given by the Fourier transform
where we have taken φLO = 0 in the I arm of the IQ- ZZ ∞
mixer, and φp LO = π/2 in the Q arm. We have defined
1 0 0
Wρ (x, p) = 2 dx0 dp0 Cρ (x0 , p0 )e2i(px −xp ) (100)
VIF = KALO κGZtml ~ωRF /2, and V̂noise,I/Q as the con- π −∞
tributions from the amplifier noise and any other added
noise. We have also introduced the quadratures of the characteristic function
n o
â†f + âf i(â†f − âf ) Cρ (x, p) = Tr ρ e2i(pX̂−xP̂ ) . (101)
X̂f = , P̂f = , (97)
2 2
With ρ the state of the electromagnetic field, Cρ (x, p) can
the dimensionless position and momentum operators of
be understood as the expectation value of the displace-
the simple harmonic oscillator, here defined such that
ment operator
[X̂f , P̂f ] = i/2. Taken together, V̂I (t) and V̂Q (t) trace a
circle in the xf − pf plane and contain information about †
−α∗ â
the two quadratures at all times. It is therefore possible D̂(α) = e2i(pX̂−xP̂ ) = eαâ , (102)
to digitally transform the signals by going to a frame
with α = x + ip, see Eq. (86).
where they are stationary using the rotation matrix
  Coherent states, already introduced in Eq. (85), have
cos(ωIF t) − sin(ωIF t) particularly simple Wigner functions. Indeed, as illus-
R(t) = (98)
sin(ωIF t) cos(ωIF t) trated schematically in Fig. 17, the Wigner function
 29

terms of the Wigner function, the Q-function takes the

Xφ form (Carmichael, 2002)
Z
2 ∞ 2 2

P Qρ (α) = d β Wρ (β)e−2|α−β| = Wρ (α) ∗ W|0i (α).

π −∞
P (Xφ ) Xφ+π/2 (106)
Xφ The Husimi-Q distribution Qρ (α) is thus obtained by con-
volution of the Wigner function with a Gaussian, and is
φ
therefore smoother than Wρ (α). As made clear by the sec-
X ond equality, this Gaussian is in fact the Wigner function
of the vacuum state, W|0i (α), obtained from Eq. (103)
with β = 0. In other words, the Q-function for ρ is ob-
FIG. 17 Pictorial phase-space distribution of a coherent state
tained from the Wigner function of the same state after
and its marginal along an axis Xφ rotated by φ from X.
adding vacuum noise. As already illustrated in Fig. 16,
heterodyne detection with an IQ mixer adds (ideally)
W|βi (α) of the coherent state |βi is simply a Gaussian vaccum noise to the signal before detection. This leads
centered at β in phase space: to the conclusion that the probability distributions for
the simultaneous measurement of two orthogonal quadra-
2 −2|α−β|2 tures in heterodyne detection is given by the marginals
W|βi (α) = e . (103)
π of the Husimi-Q distribution rather than of the Wigner
√ function (Caves et al., 2012).
The width 1/ 2 of the Gaussian is a signature of quantum
noise and implies that coherent states saturate the Heisen-
berg inequality, ∆X∆P = 1/2 with ∆O2 = hÔ2 i − hÔi2 . C. Dispersive qubit readout
We note that, in contrast to Eq. (103), Wigner functions
take negative values for non-classical states of the field. 1. Steady-state intra-cavity field
In the context of dispersive qubit measurements, the
Wigner function is particularly useful because it is related As discussed in Sec. III.C, in the dispersive regime the
to the probability distribution for the outcome of measure- transmon-resonator Hamiltonian is well approximated by
ments of the quadratures X̂ and P̂ . Indeed, the marginals ~ωq
P (x) and P (p), obtained by integrating Wρ (x, p) along Ĥdisp ≈ ~ (ωr + χσ̂z ) ↠â + σ̂z . (107)
2
the orthogonal quadrature, are simply given by
Z ∞ To simplify the discussion, here we have truncated the
P (x) = dp Wρ (x, p) = hx|ρ|xi, (104a) transmon Hamiltonian to its first two levels, absorbed
−∞ Lamb shifts in the system frequencies, and neglected a
Z ∞
transmon-induced nonlinearity of the cavity [Ka appear-
P (p) = dx Wρ (x, p) = hp|ρ|pi, (104b) ing in Eq. (50)]. As made clear by the first term of the
−∞
above expression, in the dispersive regime, the resonator
where |xi and |pi are the eigenstate of X̂ and P̂ , re- frequency becomes qubit-state dependent: If the qubit
spectively. This immediately implies that the probabil- is in |gi then hσ̂z i = −1 and the resonator frequency is
ity distribution of the outcomes of an ideal homodyne ωr − χ. On the other hand, if the qubit is in |ei, hσ̂z i = 1
measurement of the quadrature X̂φ is given by P (xφ ) and ωr is pulled to ωr + χ. In this situation, driving the
obtained by integrating the Wigner function Wρ (α) along cavity results in a qubit-state dependent coherent state,
the orthogonal quadrature X̂φ+π/2 . This is schematically |αg,e i. Thus, if the qubit is initialized in the superpo-
illustrated for a coherent state in Fig. 17. sition cg |gi + ce |ei, the system evolves to an entangled
Another useful phase-space function is the Husimi-Q qubit-resonator state of the form
distribution which, for a state ρ, takes the simple form
cg |g, αg i + ce |e, αe i. (108)
1
Qρ (α) = hα|ρ|αi. (105) To interpret this expression, let us recall the paradigm
π
of the Stern-Gerlach experiment. There, an atom passes
This function represents the probability distribution of through a magnet and the field gradient applies a spin-
finding ρ in the coherent state |αi and, in contrast to dependent force to the atom that entangles the spin state
Wρ (α), it is therefore always positive. of the atom with the momentum state of the atom (which
Since Qρ (α) and Wρ (α) are both complete descrip- in turn determines where the atom lands on the detector).
tions of the state ρ, it is not surprising that one can The experiment is usually described as measuring the
be expressed in terms of the other. For example, in spin of the atom, but in fact it only measures the final
 30

(a) p time. To see how this goal can be reached, it is useful to

-2 -1 1 2
x first evaluate more precisely the evolution of the intra-
cavity field under such a measurement. The intra-cavity
-0.5 αe αg field is obtained from the Langevin equation Eq. (75)
-1.0 with ĤS = Ĥdisp and by taking into account the cavity
drive as discussed in Sec. IV.F. Doing so, we find that the
P(x) Signal complex field amplitude hâiσ = ασ given that the qubit
0.4 is in state σ = {g, e} satisfies
Vacuum
noise
0.2 α̇e (t) = −iε(t) − (δr + χ)αe (t) − καe (t)/2, (109a)
Error
-2 -1 1 2 x α̇g (t) = −iε(t) − (δr − χ)αg (t) − καg (t)/2, (109b)
SNR/2 τm /(κ/2)

(b ) 0.4 1.0 a (d)

p 0.8
with δr = ωr − ωd the detuning of the measurement
-0.4 drive to the bare cavity frequency. The time evolution
0.6
(c) 0 of these two cavity fields in phase space are illustrated


0.4 for three different values of 2χ/κ by dashed gray lines in

p -1 0.2 Fig. 18(a-c).
c b Focusing for the moment on the steady-state (α̇σ = 0)
-2 0
-1 0 1 0 5 10 15 20 response
x 2χ/κ
s −ε
FIG. 18 (a) Path in phase space leading up to steady-state αe/g = , (110)
(δr ± χ) − iκ/2
of the intra-cavity pointer states αg and αe for 2χ/κ = 1, a
measurement drive at the bare cavity frequency with an ampli- with + for e and − for g, results in the steady-state
tude leading to one measurement photon at steady-state, and intra-cavity quadratures
assuming infinite qubit relaxation time. (top). Corresponding
marginals along x with the signal, noise and √ error defined ε(δr ± χ)
in the text (bottom). The circles of radius 1/ 2 represent hX̂i2e/g = , (111a)
vacuum noise. (b) Path in phase space for 2χ/κ = 10 and (c) (δr ± χ)2 + (κ/2)2
2χ/κ = 0.2. (d) Signal-to-noise ratio as a function of 2χ/κ for εκ/2
an integration time τm /κ = 200 (dark blue) and τm /κ = 10 hP̂ i2e/g = . (111b)
(δr ± χ)2 + (κ/2)2
(light blue). The maximum of the SNR at short integration
time is shifted away from 2χ/κ. The letters correspond to the When driving the cavity at its bare frequency, δr = 0,
ratio 2χ/κ of the three previous panels. information about the qubit is only contained in the X
quadrature, see Fig. 18(a-c).
It is also useful to define the steady-state amplitude
position of that atom on the detector. However, since the
q 2ε
spin and position are entangled, we can uniquely infer
Ase/g = hX̂i2e/g + hP̂ i2e/g = p
the spin from the position, provided there is no overlap (κ/2) + (δr ± χ)2
2
in the final position distributions for the two spin states. (112)
In this case we have effectively performed a projective and phase
measurement of the spin. !  
By analogy, if the spin-dependent coherent states of s hX̂ie/g δr ± χ
φe/g = arctan = arctan . (113)
the microwave field, αe,g , can be resolved by heterodyne hP̂ ie/g κ/2
detection, then they act as pointer states in the qubit
measurement. Moreover, since Ĥdisp commutes with the These two quantities are plotted in Fig. 19. As could
observable that is measured, σ̂z , this is a QND (quantum already have been expected from the form of Ĥdisp , a
non-demolition) measurement (Braginsky et al., 1980) coherent tone of frequency ωr ± χ on the resonator is
(in contrast to the Stern-Gerlach measurement which is largely transmitted if the qubit is in the ground (excited)
destructive). Note that for a system initially in a superpo- state, and mostly reflected if the qubit is in the excited
sition of eigenstates of the measurement operator, a QND (ground) state. Alternatively, driving the resonator at its
measurement does in fact change the state by randomly bare frequency ωr leads to a different phase accumulation
collapsing it onto one of the measurement eigenstates. for the transmitted signal depending on the state of the
The true test of ‘QNDness’ is that subsequent measure- qubit. In particular, on resonance with the bare resonator,
ment results are not random but simply reproduce the δr = 0, the phase shift of the signal associated to the two
first measurement result. qubit states is simply ± arctan(2χ/κ). As a result, in
The objective in a qubit measurement is to maximize the dispersive regime, measuring the amplitude and/or
the readout fidelity in the shortest possible measurement the phase of the transmitted or reflected signal from the
 31

Transmission (arb. units) 2. Signal-to-noise ratio and measurement fidelity

90

Phase shift (degrees)

|g |e Except for the last paragraph, the above discussion
45
concerned the steady-state intra-cavity field from which
∆ we can infer the steady-state heterodyne signal. It is,
0
however, also crucial to consider the temporal response
−45 of the resonator’s output field to the measurement drive
since, in the context of quantum computing, qubit readout
−90 should be as fast as possible. Moreover, the probability
ωr − χ ωr + χ ωq of assigning the correct outcome to a qubit measurement,
Frequency or more simply put the measurement fidelity, must also
be large. As the following discussion hopes to illustrate,
FIG. 19 Resonator transmission (dashed lines) and corre- simultaneously optimizing these two important quantities
sponding phase shifts (full lines) for the two qubit states
requires some care.
(blue: ground; red: excited). When driving the resonator close
to its pulled frequencies, the resonator response strongly de- As discussed in Sec. V.A, the quadratures X̂f (t) and
pends on the state of the qubit. Adapted from (Blais et al., P̂f (t) are extracted from heterodyne measurement of the
2007). resonator output field. Combining these signals and inte-
grating for a time τm , the operator corresponding to this
measurement takes the form
Z τm n h i
M̂ (τm ) = dt wX (t) VIF X̂f (t) + V̂noise,Xf (t)
0
resonator reveals information about the qubit state (Blais h io
+wP (t) VIF P̂f (t) + V̂noise,Pf (t) ,
et al., 2004). On the other hand, when driving the res-
onator at the qubit frequency, for example, to realize a (114)
logical gate discussed further in Sec. VII.A, the phase
shift of the resonator field only negligibly depends on where V̂noise,Xf /Pf (t) is the noise in the Xf /Pf quadrature.
the qubit state. This results in negligible entanglement The weighting functions wX (t) = |hX̂f ie − hX̂f ig | and
between the resonator, and consequently on negligible wP (t) = |hP̂f ie − hP̂f ig | are chosen such as to increase
measurement-induced dephasing on the qubit. the discrimination of the two qubit states (Bultink et al.,
2018; Magesan et al., 2015; Ryan et al., 2015; Walter et al.,
It is very important to note that for purposes of sim- 2017). Quite intuitively, because of qubit relaxation, these
plification, all of the above discussion has been couched functions give less weight to the cavity response at long
in terms of the amplitude and phase of the oscillating times since it will always reveal the qubit to be in its
electric field internal to the microwave resonator. In prac- ground state irrespective of the prepared state (Gambetta
tice, we can typically only measure the field externally et al., 2007). Moreover, for the situation illustrated in
in the transmission line(s) coupled to the resonator. The Fig. 18, there is no information on the qubit state in the
relation between the two is the subject of input-output P quadrature. Reflecting this, wP (t) = 0 which prevents
theory discussed in Sec. IV and Appendix C. The main the noise in that quadrature from being integrated.
ideas can be summarized rather simply. Consider an Following Secs. V.A and V.B, the probability distribu-
asymmetric cavity with one port strongly coupled to the tion for the outcome of multiple shots of the measurement
environment and one port weakly coupled. If driven from of M̂ (τm ) is expected to be Gaussian and characterized
the weak port, nearly all of the information about the by the marginal of the Q-function of the intra-cavity
state of the qubit is in the field radiated by the cavity field. Using the above expression, the signal-to-noise ratio
into the strongly coupled port. The same is true if the (SNR) of this measurement can be defined as illustrated
cavity is driven from the strongly coupled side, but now in Fig. 18(a) for the intra-cavity field: it is the separation
the output field is a superposition of the directly reflected of the average combined heterodyne signals corresponding
drive plus the field radiated by the cavity. If the drive to the two qubit states divided by the standard deviation
frequency is swept across the cavity resonance, the signal of the signal, an expression which takes the form
undergoes a phase shift of π in the former case and 2π
in the latter. This affects the sensitivity of the output |hM̂ (t)ie − hM̂ (t)ig |2
SNR2 (t) ≡ 2 (t)i + hM̂ 2 (t)i
. (115)
field to the dispersive shift induced by the qubit. If the hM̂N e N g
cavity is symmetric, then half the information about the
state of the qubit appears at each output port so this Here, hM̂ iσ is the average integrated heterodyne signal
configuration is inefficient. Further details can be found given that the qubit is in state σ, and M̂N = M̂ − hM̂ i
in (Clerk et al., 2010). the noise operator which takes into account the added
 32

noise but also the intrinsic vacuum noise of the quantum the steady-state average intra-cavity measurement photon
states. number. Another approach to improve the SNR is there-
In addition to the SNR, another important quantity is fore to work at large measurement photon number n̄. This
the measurement fidelity (Gambetta et al., 2007; Walter idea, however, cannot be pushed too far since increasing
et al., 2017) 7 the measurement photon number leads to a breakdown
of the approximations that have been used to derive the
Fm = 1 − [P (e|g) + P (g|e)] ≡ 1 − Em , (116) dispersive Hamiltonian Eq. (107). Indeed, as discussed in
Sec. III.C the small parameter in the perturbation theory
where P (σ|σ 0 ) is the probability that a qubit in that leads to the dispersive approximation is not g/∆
state σ is measured to be in state σ 0 . In the sec- but rather n̄/ncrit , with ncrit the critical photon number
ond equality, we have defined the measurement er- introduced in Eq. (44). Well before reaching n̄/ncrit ∼ 1,
ror Em which, as illustrated in Fig. 18(a), is simply higher-order terms in the dispersive approximation start
the overlap of the marginals Pσ (x) of the Q-functions to play a role and lead to departures from the expected
for the two R qubit states. This can be expressed behavior. For example, it is commonly experimentally
as Em = dxφLO +π/2 min[P0 (xφLO +π/2 ), P1 (xφLO +π/2 )], observed that the dispersive measurement loses its QND
where the LO phase is chosen to minimize Em . Using this character well before n̄ ∼ ncrit and often at measurement
expression, the measurement fidelity is found to be related photon populations as small as n̄ ∼ 1 − 10 (Johnson et al.,
to the SNR by Fm = 1 − erfc(SNR/2), where erfc is the 2011; Minev et al., 2019). Because of these spurious qubit
complementary error function (Gambetta et al., 2007). It flips, measurement photon numbers are typically chosen
is important to note that this last result is valid only if the to be well below ncrit (Walter et al., 2017). While this
marginals are Gaussian. In practice, qubit relaxation and non-QNDness at n̄ < ncrit is expected from the discussion
higher-order effects omitted in the dispersive Hamiltonian of dressed-dephasing found in Sec. IV.D, the predicted
Eq. (107) can lead to distortion of the coherent states and measurement-induced qubit flip rates are smaller than
therefore to non-Gaussian marginals (Gambetta et al., often experimentally observed. We note that qubit transi-
2007; Hatridge et al., 2013). Kerr-type nonlinearities that tions at n̄ > ncrit caused by accidental resonances within
are common in circuit QED tend to create a banana- the qubit-resonator system have been studied in Sank
shaped distortion of the coherent states in phase space, et al. (2016).
something that is sometimes referred to as bananization
To reach high fidelities, it is also important for the
(Boutin et al., 2017; Malnou et al., 2018; Sivak et al.,
measurement to be fast compared to the qubit relaxation
2019).
time T1 . A strategy to speed-up the measurement is to
Although we are interested in short measurement times,
use a low-Q oscillator which leads to a faster readout rate
it is useful to consider the simpler expression for the long-
simply because the measurement photons leak out more
time behavior of the SNR which suggests different strate-
rapidly from the resonator to be detected. However, this
gies to maximize the measurement fidelity. Assuming
should not be done at the price of increasing the Purcell
δr = 0 and ignoring the prefactors related to gain and
rate γκ to the point where this mechanism dominates qubit
mixing, we find (Gambetta et al., 2008)
decay (Houck et al., 2008). As discussed in Sec. IV.E, it is
√ possible to avoid this situation to a large extent by adding
SNR(τm → ∞) ' (2ε/κ) 2κτm |sin 2φ| , (117)
a Purcell filter at the output of the resonator (Bronn et al.,
where φ is given by Eq. (113); see (Didier et al., 2015a) 2015; Jeffrey et al., 2014; Reed et al., 2010b).
for a detailled derivation of this expression. The reader Fixing κ so as to avoid Purcell decay and working at the
can easily verify that the choice χ/κ = 1/2 maximizes optimal χ/κ ratio, it can be shown that the steady-state
Eq. (117), see Fig. 18(d) (Gambetta et al., 2008). This response is reached in a time ∝ 1/χ (Walter et al., 2017).
ratio is consequently often chosen experimentally (Walter Large dispersive shifts can therefore help to speed up the
et al., 2017). While leading to a smaller steady-state SNR, measurement. As can be seen from Eq. (43), χ can be
other choices of the ratio χ/κ can be more advantageous increased by working at larger qubit anharmonicity or,
at finite measurement times. in other words, larger charging energy EC . Once more,
In the small χ limit, the factor 2/κ in SNR(τm → ∞) this cannot be pushed too far since the transmon charge
can be interpreted using Eq. (109b) as the square-root of dispersion and therefore its dephasing rate increase with
EC .
The above discussion shows that QND qubit measure-
ment in circuit QED is a highly constrained problem.
7 An alternative definition known as the assignment fidelity is The state-of-the-art for such measurements is currently
1 − 12 [P (e|g) − P (g|e)] (Magesan et al., 2015). This quantity
takes values in [0, 1] while formally Fm ∈ [−1, 1]. Negative values
of Fm ∼ 98.25% in τm = 48 ns, when minimizing readout
are, however, not relevant in practice. Indeed, because Fm = −1 time, and 99.2% in 88 ns, when maximizing the fidelity, in
corresponds to systematically reporting the incorrect value, a both cases using n̄ ∼ 2.5 intra-cavity measurement pho-
fidelity of 1 is recovered after flipping the measurement outcomes. tons (Walter et al., 2017). These results were obtained by
 33

detailed optimization of the system parameters following high photon-number state (Dykman and Krivoglaz, 1980;
the concepts introduced above but also given an under- Manucharyan et al., 2007). By dispersively coupling a
standing of the full time response of the measurement qubit to the nonlinear resonator, this bifurcation can be
signal |hM̂ (t)i1 − hM̂ (t)i0 |. The main limitation in these made qubit-state dependent (Vijay et al., 2009). It is
reported fidelity was the relatively short qubit relaxation possible to exploit the fact that the low- and high-photon-
time of 7.6 µs. Joint simultanteous dispersive readout number states can be easily distinguished to realize high-
of two transmon qubits capacitively coupled to the same fildelity single-shot qubit readout (Mallet et al., 2009).
resonator has also been realized (Filipp et al., 2009).
The very small photon number used in these experi-
ments underscores the importance of quantum-limited b. High-power readout and qubit ’punch out’ Coming back
amplifiers in the first stage of the measurement chain, to linear resonators, while the non-QNDness at moderate
see Fig. 14. Before the development of these amplifiers, measurement photon number mentioned above leads to
which opened the possibility to perform strong single-shot small measurement fidelity, it was observed by a fearless
(i.e. projective) measurements, the SNR in dispersive mea- graduate student that, in the limit of very large measure-
surements was well below unity, forcing the results of these ment power, a fast and high-fidelity single-shot readout
weak measurements to be averaged over tens of thousands is recovered (Reed et al., 2010a). An intuitive under-
of repetitions of the experiment to extract information standing of this observation can be obtained from the
about the qubit state (Wallraff et al., 2005). The advent Jaynes-Cummings Hamiltonian Eq. (34) (Bishop √ et al.,
of near quantum-limited amplifiers (see Sec. VIII.B) has 2010; Boissonneault et al., 2010). Indeed, for n  n, the
made it possible to resolve the qubit state in a single-shot first term of this Hamiltonian dominates over the qubit-
something which has led, for example, to the observa- oscillator interaction ∝ g such that the cavity responds
tion of quantum jumps of a transmon qubit (Vijay et al., at its bare frequency ωr despite the presence of the trans-
2011). mon. This is sometimes referred to as ‘punching out’ the
Finally, we point out that the quantum efficiency, η, of qubit and can be understood as a quantum-to-classical
the whole measurement chain can be extracted from the transition where, in the correspondence limit, the system
SNR using (Bultink et al., 2018) behaves classically and therefore responds at the bare cav-
ity frequency ωr . Interestingly, with a multi-level system
SNR2 such as the transmon, the power at which this transition
η= , (118)
4βm occurs depends on the state of the transmon, leading to
Rτ a high-fidelity measurement. This high-power readout
where βm = 2χ 0 m dt Im[αg (t)αe (t)∗ ] is related to the
measurement-induced dephasing discussed further in is, however, obtained at the expense of completely losing
Sec. VI.B.2. This connection between quantum efficiency, the QND nature of the dispersive readout (Boissonneault
SNR, and measurement-induced dephasing results from et al., 2010).
the fundamental link between the rate at which infor-
mation is gained in a quantum measurement and the √
unavoidable backaction on the measured system (Clerk c. Squeezing Finally, the n scaling of SNR(τm → ∞)
et al., 2010; Korotkov, 2001). mentioned above can be interpreted as resulting from
populating the cavity with a coherent state and is known
as the standard quantum limit. It is natural to ask if
3. Other approaches replacing the coherent measurement tone with squeezed
input radiation (see Sec. VIII.C.2) can lead to Heisenberg-
a. Josephson Bifurcation Amplifier While the vast major- limited scaling for which the SNR scales linearly with the
ity of circuit QED experiments rely on the approach measurement photons number (Giovannetti et al., 2004).
described above, several other qubit-readout methods To achieve this, one might imagine squeezing a quadra-
have been theoretically explored or experimentally imple- ture of the field to reduce the overlap between the two
mented. One such alternative is known as the Josephson pointer states. In Fig. 18, this corresponds to squeezing
Bifurcation Amplifier (JBA) and relies on using, for ex- along X. The situation is not so simple since the large
ample, a transmission-line resonator that is made non- dispersive coupling required for high-fidelity qubit readout
linear by incorporating a Josephson junction in its cen- leads to a significant rotation of the squeezing angle as
ter conductor (Boaknin et al., 2007). This circuit can the pointer states evolve from the center of phase space
be seen as a distributed version of the transmon qubit to their steady-state. This rotation results in increased
and is well described by the Kerr-nonlinear Hamiltonian measurement noise due to contributions from the anti-
of Eq. (25) (Bourassa et al., 2012). With a relatively squeezed quadrature (Barzanjeh et al., 2014). Borrowing
weak Kerr nonlinearity (∼ −500 kHz) and under a co- the idea of quantum-mechanics-free subsystems (Tsang
herent drive of well chosen amplitude and frequency, this and Caves, 2012), it has been shown that Heisenberg-
system bifurcates from a low photon-number state to a limited scaling can be reached with two-mode squeezing
 34

by dispersively coupling the qubit to two rather than one 2015a,b; Didier et al., 2015a; Kerman, 2013; Richer and
resonator (Didier et al., 2015b). DiVincenzo, 2016; Richer et al., 2017). Another approach
to realize these ideas is to strongly drive a resonator
dispersively coupled to a qubit (Blais et al., 2007; Das-
d. Longitudinal readout An alternative approach to qubit sonneville et al., 2020). Indeed, the strong drive leads to
readout is based on the Hamiltonian Ĥz of Eq. (63) with a large displacement of the cavity field â → â + α which
its longitudinal qubit-oscillator coupling gz (↠+ â)σ̂z . In on the dispersive Hamiltonian leads to
contrast to the dispersive Hamiltonian which leads to
a rotation in phase space, longitudinal coupling gener- χ↠âσ̂z → χ↠âσ̂z + αχ(↠+ â)σ̂z + χα2 σ̂z , (120)
ates a linear displacement of the resonator field that is
conditional on the qubit state. As a result, while under where we have assumed α to be real for simplicity. For χ
the dispersive evolution there is little information gain small and α large, the second term dominates therefore
about the qubit state at short times [see the poor pointer realizing a synthetic longitudinal interaction af amplitude
state separation at short times in Fig. 18(a)], Ĥz rather gz = αχ. In other words, longitudinal readout can be
generates the ideal dynamics for a measurement with a realized as a limit of the dispersive readout where χ
180◦ out-of-phase displacements of the pointer states αg approaches zero while α grows such that χα is constant.
and αe . It is therefore expected that this approach can A simple interpretation of this observation is that, for
lead to much shorter measurement times than is possible strong drives, the circle on which the pointer states rotate
with the dispersive readout (Didier et al., 2015a). due to the dispersive interaction has a very large radius α
Another advantage is that Ĥz commutes with the mea- such that, for all practical purposes, the motion appears
sured observable, [Ĥz , σ̂z ] = 0, corresponding to a QND linear.
measurement. While the dispersive Hamiltonian Ĥdisp A variation of this approach which allows for larger
also commutes with σ̂z , it is not the case for the full longitudinal coupling strengh was experimentally realized
Hamiltonian Eq. (32) from which Ĥdisp is perturbatively by Touzard et al. (2019) and Ikonen et al. (2019) and
derived. As already discussed, this non-QNDness leads relies on driving the qubit at the frequency of the res-
to Purcell decay and to a breakdown of the dispersive onator. This is akin to the cross-resonance gate discussed
approximation when the photon populations is not signif- further in Sec. VII.B.3 and which leads to the desired
icantly smaller than the critical photon number ncrit . On longitudinal interaction, see the last term of Eq. (144). A
the other hand, because Ĥz is genuinely QND it does not more subtle approach to realize a synthetic longitudinal
suffer from these problems and the measurement photon interaction is to drive a qubit with a Rabi frequency ΩR
number can, in principle, be made larger under longitudi- while driving the resonator at the sideband frequencies
nal than under dispersive coupling. Moreover, given that ωr ± ΩR . This idea was implemented by Eddins et al.
Ĥz leads to displacement of the pointer states rather than (2018) who also showed improvement of qubit readout
to rotation in phase space, single-mode squeezing can also with single-mode squeezing. Importantly, because these
be used to increase the measurement SNR (Didier et al., realizations are based on the dispersive Hamiltonian, they
2015a). suffer from Purcell decay and non-QNDness. Circuits
Because the longitudinal coupling can be thought of realizing dispersive-like interactions that are not derived
as a cavity drive of amplitude ±gz with the sign being from a Jaynes-Cummings interaction have been studied
conditional on the qubit state, Ĥz leads in steady-state (Dassonneville et al., 2020; Didier et al., 2015a).
to a pointer state displacement ±gz /(ωr + iκ/2), see
Eq. (110). With ωr  gz , κ in practice, this displacement
is negligible and cannot realistically be used for qubit VI. QUBIT-RESONATOR COUPLING REGIMES
readout. One approach to increase the pointer state
separation is to activate the longitudinal coupling by We now turn to a discussion of the different coupling
modulating gz at the resonator frequency (Didier et al., regimes that are accessible in circuit QED and how these
2015a). Taking gz (t) = g̃z cos(ωr t) leads, in a rotating regimes are probed experimentally. We first consider the
frame and after dropping rapidly oscillating terms, to the resonant regime where the qubit is tuned in resonance
Hamiltonian with the resonator, before moving on to the dispersive
g̃z † regime characterized by large qubit-resonator detuning.
H̃z = (â + a)σ̂z . (119) While the situation of most experimental interest is the
2
strong coupling regime where the coupling strength g
Under this modulation, the steady-state displacement now overwhelms the decay rates, we also touch upon the so-
becomes ±g̃z /κ and can be significant even for moderate called bad-cavity and bad-qubit limits because of their
modulation amplitudes g̃z . historical importance and their current relevance to hy-
Circuits realizing the longitudinal coupling with trans- brid quantum systems. Finally, we briefly consider the
mon or flux qubits have been studied (Billangeon et al., ultrastrong coupling regime where g becomes comparable
 35

or is even larger than the system’s frequencies. To sim- 1. Bad-cavity limit

plify the discussion, we will treat the artificial atom as a
simple two-level system throughout this section. We first consider the bad-cavity limit realized when
the cavity decay rate overwhelms the coupling g which
is itself larger than the qubit linewidth: κ > g  γ2 .
A. Resonant regime
This situation corresponds to an overdamped oscillator
and, at qubit-oscillator resonance, leads to rapid decay
of the qubit. A simple model for this process is obtained
The low-energy physics of the Jaynes-Cummings model
using the truncated Hilbert space discussed above where
is well described by the ground state |g, 0i = |g, 0i and
we now drop the cavity drive for simplicity. Because of
first two excited states
the very large decay rate κ, we can asume the oscillator
√ to rapidly reach its steady-state ∂t hâi = 0. Using the
|g, 1i = (|g, 1i − |e, 0i)/ 2,
√ (121) resulting expression for hâi in Eq. (123) immediately leads
|e, 0i = (|g, 1i + |e, 0i)/ 2, to
 
which, as already illustrated in Fig. 8, are split in fre- γ1 + γk0
hσ̂˙ − i = − + γϕ hσ̂− i, (125)
quency by 2g. As discussed in Sec. V.C in the context 2
of the dispersive readout, the coupled qubit-resonator
system can be probed by applying a coherent microwave where we have defined the Purcell decay rate γκ0 = 4g 2 /κ.
tone to the input of the resonator and measuring the The expression for this rate has a rather different form
transmitted or reflected signal. than the Purcell rate γκ = (g/∆)2 κ given in Eq. (83).
To arrive at an expression for the expected signal in such These two results are, however, not incompatible but have
an experiment, we consider the equations of motion for been obtained in very different regimes. An expression
the field and qubit annihilation operators in the presence for the Purcell rate that interpolates between the two
of a coherent drive of amplitude ε and frequency ωd on above expressions can be obtained and takes the form
the resonator’s input port. In a frame rotating at the κg 2 /[(κ/2)2 + ∆2 ] (Sete et al., 2014).
drive frequency, these equations take the form The situation described here is illustrated for κ/g = 10
and γ1 = 0 in Fig. 20(a) which shows the probability
κ  for the qubit to be in its excited state versus time after
˙ =−
hâi + iδr hâi − ighσ̂− i − iε, (122)
2 initializing the qubit in its excited state and the resonator
hσ̂˙ − i = − (γ2 + iδq ) hσ̂− i + ighâσ̂z i, (123) in vacuum. Even in the absence of qubit T1 , the qubit
is seen to quickly relax to its ground state something
with δr = ωr − ωd and δq = ωq − ωd , and where γ2 is which, as discussed in Sec. IV.D, is due to qubit-oscillator
defined in Eq. (80b). These expressions are obtained us- hybridization. Figure 20(b) shows the transmitted power
ing ∂t hÔi = Trρ̇Ô and the master equations of Eqs. (68) versus drive frequency in the presence of a very weak
and (79) at zero temperature and in the two-level approx- coherent tone populating the cavity with n̄  1 photons.
imation for the transmon. Alternatively, the expression The response shows a broad Lorentzian peak of width
for ∂t hâi is simply the average of Eq. (75) with ĤS the κ together with a narrow electromagnetically induced
Jayne-Cummings Hamiltonian. transparency (EIT)-like window of width γκ0 (Mlynek
et al., 2014; Rice and Brecha, 1996). This effect which is
At very low excitation amplitude ε, it is reasonable
due to interference between the intra-cavity field and the
to truncate the Hilbert space to the first three levels
probe tone vanishes in the presence of qubit dephasing.
defined above. In this subspace, hâσ̂z i = −hâi since
Although not the main regime of interest in circuit QED,
â acts nontrivially only if the qubit is in the ground
the bad-cavity limit offers an opportunity to engineer the
state (Kimble, 1994). It is then simple to compute the
dissipation seen by the qubit. For example, this regime
steady-state transmitted homodyne power by solving the
has been used to control the lifetime of long-lived donor
above expressions with ∂t hâi = ∂t hσ− i = 0 and using
spins in silicon in a hybrid quantum system (Bienfait
Eq. (112) to find
et al., 2016).
 2 2
εVIF δq − iγ2
|A|2 = . (124)
2 (δq − iγ2 )(δr − iκ/2) − g 2 2. Bad-qubit limit

This expression is exact in the low excitation power limit. The bad qubit limit corresponds to the situation where
Taking the qubit and the oscillator to be on resonance, a high-Q cavity with large qubit-oscillator coupling is real-
∆ = ωq − ωr = 0, we now consider the result of cavity ized, while the qubit dephasing and/or energy relaxation
transmission measurements in three different regimes of rates is large: γ2 > g  κ. Although this situation is
qubit-cavity interaction. not typical of circuit QED with transmon qubits, it is
 36
κ
reader should also be aware that qubit-resonator detuning-
1.0 4
dependent dispersive shifts of the cavity resonance can be
|0, e observed in this bad-qubit limit. The observation of such
γκ dispersive shifts on its own should not be mistaken for an
Pe

0.5 |1, g 2
observation of strong coupling (Wallraff et al., 2013).

0.0
0 1000 2000 −20 0 20 3. Strong coupling regime
1.0
2
We now turn to the case where the coupling strength
overwhelms the qubit and cavity decay rates, g > κ, γ2 .
↠â

0.5 κγ 1
In this regime, light-matter interaction is strong enough
for a single quantum to be coherently exchanged between
the electromagnetic field and the qubit before it is ir-
0.0 0
reversibly lost to the environment. In other words, at
0 1000 2000 −2 0 2 resonance ∆ = 0 the splitting 2g between the two dressed
1.0 1.0 eigenstates {|g, 1i, |e, 0i} of Eq. (121) is larger than their
2g
linewidth κ/2 + γ2 and can be resolved spectroscopically.
We note that, with the eigenstates being half-photon and
Pe

0.5 0.5
half-qubit8 , the above expression for the dressed-state
linewidth is simply the average of the cavity and of the
0.0 0.0 qubit linewitdh (Haroche, 1992). Figure 20(f) shows cav-
0 10 20 −200 0 200 ity transmission for (κ, γ1 , γϕ )/g = (0.1, 0.1, 0) and at low
t (ns) δr /2π (MHz) excitation power such that, on average, there is signifi-
cantly less than one photon in the cavity. The resulting
FIG. 20 Numerical simulations of the qubit-oscillator mas- doublet of peaks located at ωr ± g is the direct signature
ter equation for (a,c,e) the time evolution starting from of the dressed-states {|g, 1i, |e, 0i} and is known as the
the bare state |0, ei (light blue) or |1, gi (blue), and (b,d,f) vacuum Rabi splitting. The observation of this doublet is
steady-state response A2 = |hâi|2 as a function of cav- the hallmark of the strong coupling regime.
ity drive frequency (dark blue) for the three coupling
The first observation of this feature in cavity QED
regimes. Pe : qubit excited state population. (a,b) Bad-
cavity limit: (κ, γ1 , g)/2π = (10, 0, 1) MHz. (c,d) Bad-qubit with a single atom and a single photon was reported
limit: (κ, γ1 , g)/2π = (0, 10, 1) MHz. (e,f) Strong coupling: by (Thompson et al., 1992). In this experiment, the num-
(κ, γ1 , g)/2π = (0.1, 0.1, 100) MHz (dashed lines in panel (e) ber of atoms in the cavity was not well controlled and it
and solid in (f) and (κ, γ1 , g)/2π = (1, 1, 100) (full lines in e). could only be determined that there was on average one
The light blue line in panel (f) is computed with a thermal atom in interaction with the cavity field. This distinction
photon number of n̄κ = 0.35 rather than n̄κ = 0 for all the is important because, in the presence
other results. √ of N atoms, the
collective interaction strength is g N and the observed
splitting correspondingly larger (Fink et al., 2009; Tavis
relevant for some hybrid systems that suffer from signif- and Cummings, 1968). Atom number fluctuation is obvi-
icant dephasing. This is the case, for example, in early ously not a problem in circuit QED and, with the very
experiments with charge qubits based on semiconduc- strong coupling and relatively small linewidths that can
tor quantum dots coupled to superconducting resonators routinely be experimentally achieved, reaching the strong
(Frey et al., 2012; Petersson et al., 2012; Viennot et al., coupling regime is not particularly challenging in this
2014). system. In fact, the very first circuit QED experiment
of Wallraff et al. (2004) reported the observation of a
In analogy to the bad-cavity case, the strong damping of
clear vacuum Rabi splitting with 2g/(κ/2 + γ2 ) ∼ 10, see
the qubit together with the qubit-resonator coupling leads
Fig. 21(a). This first demonstration used a charge qubit
to the photon decay rate κ0γ = 4g 2 /γ1 which is sometimes
which, by construction, has a much smaller coupling g
known as the ‘inverse’ Purcell rate. This is illustrated in
than typical transmon qubits. As a result, more recent
Fig. 20(c) which shows the time-evolution of the coupled
experiments with transmon qubits can display ratios of
system starting with a single photon in the resonator and
peak separation to linewidth in the several hundred, see
the qubit in the ground state. In this situation, the cavity
Fig. 21(b) (Schoelkopf and Girvin, 2008).
response is a simple Lorentzian broadened by the inverse
Purcell rate, see Fig. 20(d). If the qubit were to be probed
directly rather than indirectly via the cavity, the atomic
response would show the EIT-like feature of Fig. 20(b), 8 According to some authors, these dressed states should therefore
now with a dip of width κ0γ (Rice and Brecha, 1996). The be referred to as quton and phobit (Schuster, 2007).
 37

Probe frequency (MHz)

0.1 (a) |â|2

Transmission
0.08 0.030
5100 ωq 0.025
0.06
0.020
0.04 5000
0.02
ωr 0.015

0.010
0 4900 0.005
6.02 6.03 6.04 6.05 6.06 6.07
Probe frequency (GHz) 4600 4800 5000 5200 5400
0.12 (b) Qubit frequency (MHz)
Transmission

FIG. 22 Vacuum Rabi splitting revealed in numerical simula-

0.08
tions of the cavity transmission A2 = |hâi|2 as a function of
probe frequency and qubit transition frequency for the same
0.04 parameters as in Fig. 20(f) The bare cavity and qubit fre-
quencies are indicated by the horizontal and diagonal dashed
0.00 lines, respectively. The vacuum Rabi splitting of Fig. 20(f)
is obtained at resonance (ωr = ωq ) along the vertical dotted
6.75 6.85 6.95 7.05
line.
Probe frequency (GHz)

FIG. 21 (a) Transmission-line resonator transmission versus

probe frequency in the first observation of vacuum Rabi split- matter coupling can also be displayed in time-resolved
ting in circuit QED (full blue line). The qubit is a Cooper measurements (Brune et al., 1996). Starting from the
pair box qubit with EJ /h ≈ 8 GHz and EC /h ≈ 5.2 GHz. The qubit-oscillator ground state, this can be done, for exam-
full red line is a calculated spectra with 2g/2π ≈ 11.6 MHz, ple, by first pulsing the qubit to its first excited state and
κ/2π ≈ 0.8 MHz and γ2 /2π ≈ 0.7 MHz. As a reference, the then bringing it on resonance with the cavity. As illus-
dashed light blue line is the measured transmission with the trated in Fig. 20(e), this results in oscillations in the qubit
qubit strongly detuned from the resonator. Adapted from
and cavity populations at the vacuum Rabi frequency 2g.
Wallraff et al. (2004). (b) Resonator transmission with a trans-
mon qubit. The vacuum Rabi splitting is even more resolved Time-resolved vacuum Rabi oscillations in circuit QED
with 2g/2π = 350 MHz, κ/2π ∼ 800 kHz and γ2 /2π ∼ 200 were first performed with a flux qubit coupled to a discrete
kHz. Notice the change in probe frequency range from panel LC oscillator realized in the bias circuitry of the device
(a). Adapted from Schoelkopf and Girvin (2008). (Johansson et al., 2006). This experiment was followed
by a similar observation with a phase qubit coupled to a
coplanar waveguide resonator (Hofheinz et al., 2008).
Figure 22 shows the qubit-oscillator spectrum as a In the limit of weak excitation power which we have
function of probe frequency, as above, but now also as a considered so far, the coupled qubit-oscillator system is
function of the qubit frequency allowing to see the full indistinguishable from two coupled classical linear oscil-
qubit-resonator avoided crossing. The horizontal dashed lators. As a result, while the dressed-states that are
line corresponds to the bare cavity frequency while the probed in these experiments are entangled, the observa-
diagonal dashed line is the bare qubit frequency. The tion of an avoided crossing cannot be taken as a conclusive
vacuum Rabi splitting of Fig. 20(f) is obtained from a demonstration that the oscillator field is quantized or of
linecut (dotted vertical line) at resonance between the qubit-oscillator entanglement. Indeed, a vacuum Rabi
bare qubit frequency ωq and the bare cavity frequency ωr . splitting can be interpreted as the familiar normal mode
Because it is the cavity that is probed here, the response splitting of two coupled classical oscillators.
is larger when the dressed-states are mostly cavity-like A clear signature of the quantum nature
and disappears away from the cavity frequency where the √ of the system
can, however, be obtained by probing the n dependence
cavity no longer responds to the probe (Haroche, 1992). of the spacing of the higher excited states of the Jaynes-
It is interesting to note that the splitting predicted by Cummings ladder already discussed in Sec. III.B. This
Eq. (124) for the transmitted homodyne signal is in fact dependence results from the matrix element of the opera-
smaller than 2g in the presence of finite relaxation and tor â and is consequently linked to the quantum nature of
dephasing. Although not significant in circuit QED with the field (Carmichael et al., 1996). Experimentally, these
transmon qubits, this correction can become important in transitions can be accessed in several ways including by
systems such as charge qubits in quantum dots that are two-tone spectroscopy (Fink et al., 2008), by increasing
not very deep in the strong coupling regime. We also note the probe tone power (Bishop et al., 2009), or by increas-
that the observed splitting can be smaller when measured ing the system temperature (Fink et al., 2010). The light
in reflection rather than in transmission. blue line in Fig. 20(f) shows cavity transmission with a
Rather than spectroscopic measurements, strong light- thermal photon number of n̄κ = 0.35 rather than n̄κ = 0
 38

ments used the ability to create entanglement between

a qubit and a photon through the resonant interaction
with a resonator, e.g. in the context of quantum computa-
|e, 1 √ tion (Mariantoni et al., 2011b), for entangling resonator
2g 2 modes (Mariantoni et al., 2011a), and transferring quan-
|g, 2 tum states (Sillanpää et al., 2007).

|e, 0
2g B. Dispersive regime
|g, 1
For most quantum computing experiments, it is com-
mon to work in the dispersive regime where, as already
discussed in Sec. III.C, the qubit is strongly detuned from
= |g, 0 the oscillator with |∆|  g. There, the dressed eigenstates
FIG. 23 Ground state and first two doublets of the Jaynes- are only weakly entangled qubit-oscillator states. This
Cummings ladder. The dark blue arrows correspond to the is to be contrasted to the resonant regime where these
transitions that are probed in a vacuum Rabi experiment. The eigenstates are highly entangled resulting in the qubit and
transitions illustrated with light blue arrows lead to additional the oscillator to completely lose their individual character.
peaks at transition frequencies lying inside of the vacuum Rabi In the two-level system approximation, the dispersive
doublet at elevated temperature or increased probe power. On
regime is well described by the Hamiltonian Ĥdisp of
the other hand, the matrix element associated with the red
transitions would lead to response at transition frequencies Eq. (107). There, we had interpreted the dispersive cou-
outside of the vacuum Rabi doublet. Those transition are, pling as a qubit-state dependent shift of the oscillator
however, suppressed and are not observed (Rau et al., 2004). frequency. This shift can be clearly seen in Fig. 22 as the
deviation of the oscillator response from the bare oscillator
frequency away from resonance (horizontal dashed line).
(dark blue line). At this more elevated temperature, ad- This figure also makes it clear that the qubit frequency,
ditional pairs of peaks with smaller separation are now whose bare value is given by the diagonal dashed line, is
observed in addition to the original peaks separated by also modified by the dispersive coupling to the oscilla-
2g. As illustrated in Fig. 23, these additional √ structures tor. To better understand this qubit-frequency shift, it is
are due to multi-photon transitions and their n scaling instructive to rewrite Ĥdisp as
reveal the anharmonicity of the Jaynes-Cummings ladder.   
Interestingly, the matrix elements of transitions that lie † ~ † 1
Ĥdisp ≈ ~ωr â â + ωq + 2χ â â + σ̂z , (126)
outside of the original vacuum Rabi splitting peaks are 2 2
suppressed and these transitions are therefore not ob-
served, see the red arrow in Fig. 23 (Rau et al., 2004). We where it is now clear that the dispersive interaction of
also note that, at much larger power or at elevated tem- amplitude χ not only leads to a qubit-state dependent
perature, the system undergoes a quantum-to-classical frequency pull of the oscillator, but also to a photon-
transition and a single peak at the resonator frequency number dependent frequency shift of the qubit given
ωr is observed (Fink et al., 2010). In short, the impact by 2χ↠â. This is known as the ac-Stark shift (or the
of the qubit on the system is washed away in the corre- quantized light shift) and is here accompanied by a Lamb
spondence limit. This is to be expected from the form of shift corresponding to the factor of 1/2 in the last term of
the Jaynes-Cummings Hamiltonian Eq. (34) √ where the Eq. (126) and which we had dropped in Eq. (107). In this
qubit-cavity coupling ~g(↠σ̂− + âσ̂+ ) with its n scaling section, we explore some consequences of this new point
is overwhelmed by the free cavity Hamiltonian ~ωr ↠â of view on the dispersive interaction, starting by first
which scales as n. This is the same mechanism that leads reviewing some of the basic aspects of qubit spectroscopic
to the high-power readout discussed in Sec. V.C.3. measurements.
Beyond this spectroscopic evidence, the quantum na-
ture of the field and qubit-oscillator entanglement was
also demonstrated in a number of experiments directly 1. Qubit Spectroscopy
measuring the joint density matrix of the dressed states.
For example, Eichler et al. (2012b) have achieved this To simplify the discussion, we first consider spectro-
by creating one of the entangled states {|g, 1i, |e, 0i} in scopically probing the qubit assuming that the oscillator
a time-resolved vacuum Rabi oscillation experiment and, remains in its vacuum state. This is done by applying a
subsequently, measuring the qubit state in a dispersive coherent field of amplitude αd and frequency ωd to the
measurement and the photon state using a linear detec- qubit, either via a dedicated voltage gate on the qubit or
tion method (Eichler et al., 2012a). A range of experi- to the input port of the resonator. Ignoring the resonator
 39

(a) (b) 0 experiment (Vion et al., 2002a).9

phase, φ (deg)
0.4 −10 In typical optical spectroscopy of atoms in a gas, one

γq /2π (MHz)
directly measures the absorption of photons by the gas
2 −20 as a function of the frequency of the photons. In circuit
Pe

0.2 QED, one typically performs quantum jump spectroscopy

−30
0
(c) by measuring the probability that an applied microwave
0 1 −40 drive places the qubit into its excited state. The variation
0.0 (ΩR /2π)2 (MHz2 ) in qubit population with qubit drive can be measured
−10 0 10 0.0 0.5 1.0 by monitoring the change in response of the cavity to
δq /2π (MHz) (ΩR /2π)2 (MHz2 ) the spectroscopy drive. As discussed in Sec. V.C, this is
realized using two-tone spectroscopy by measuring the
FIG. 24 Power broadening of the qubit line. (a) Excited qubit cavity transmission, or reflection, of an additional drive
population (left vertical axis) and phase (right vertical axis) of frequency close to ωr . In the literature, this second
as a function of the drive detuning δq for the Rabi amplitudes drive is often referred to as the probe or measurement
ΩR /2π = 0.1 MHz (light blue), 0.5 MHz (blue) and 1 MHz tone, while the spectroscopy drive is also known as the
(dark blue). The phase is obtained from φ = arctan(2χhσ̂z i/2),
pump tone. As shown by Eq. (113), the phase of the
with 2χ/κ = 1. (b) Excited qubit population and phase at
δq = 0 and as a function of Ω2R . The horizontal dashed gray transmitted probe tone is related to the qubit population.
line corresponds to qubit saturation, Pe = 1/2. (c) Qubit In particular, with the probe tone at the bare cavity
linewidth as a function of Ω2R . All three panels have been frequency and in the weak dispersive limit χ  κ, this
obtained from numerical simulations of the dispersive qubit phase is simply proportional to the qubit population,
master equation with γ1 /2π = 0.1 MHz and γϕ /2π = 0.1 MHz, φs = arctan(2χhσ̂z is /2) ≈ 2χhσ̂z is /κ. Monitoring φs as
to the exception of the dashed blue lines in panels (b) and a function of the spectroscopy tone frequency therefore
(c) that correspond to the analytical expressions found in the
directly reveals the Lorentzian qubit lineshape (Schuster
text.
et al., 2005).

2. AC-Stark shift and measurement-induced broadening

for the moment, this situation is described by the Hamil-
tonian δq σ̂z /2 + ΩR σ̂x /2, where δq = (ωq + χ) − ωd is
In the above discussion, we have implicitly assumed that
the detuning between the Lamb-shifted qubit transition
the amplitude of the measure tone is such that the intra-
frequency and the drive frequency, and ΩR ∝ αd is the
cavity photon population is vanishingly small, h↠âi →
Rabi frequency. Under this Hamiltonian and using the
0. As is made clear by Eq. (126), increase in photon
master equation Eq. (79) projected on two levels of the
population leads to a qubit frequency shift by an average
qubit, the steady-state probability Pe = (hσ̂z is + 1)/2 for
value of 2χh↠âi. Figure 25(a) shows this ac-Stark shift
the qubit to be in its excited state (or, equivalently, the
in the steady-state qubit population as a function of
probability to be in the ground state, Pg ) is found to be
spectroscopy frequency for three different probe drive
(Abragam, 1961)
powers. Taking advantage of the dependence of the qubit
frequency on measurement power, prior knowledge of the
value of χ allows one to infer the intra-cavity photon
1 Ω2R
Pe = 1 − Pg = . (127) number as a function of input pump power from such
2 γ1 γ2 + δq2 γ1 /γ2 + Ω2R

The Lorantzian lineshape of Pe as a function of the drive 9 Different quantities associated with the dephasing time are used
frequency is illustrated in Fig. 24(a). In the limit of in the literature, the three most common being T2 , T2∗ and T2echo .
strong qubit drive, i.e. large Rabi frequency ΩR , the While T2 corresponds to the intrinsic or “natural” dephasing time
of the qubit, T2∗ ≤ T2 accounts for inhomogeneous broadening.
steady-state qubit population reaches saturation with For example, for a flux-tunable transmon, this broadening can be
Pe = Pg = 1/2, see Fig. 24(b). Moreover, as the power due to random fluctuations of the flux treading the qubit’s SQUID
increases, the full width at half maximum (FWHM) of loop. A change of the flux over the time of the experiment needed
to extract T2 results in a qubit frequency shifts, something that
the qubit lineshape evolves
p from the bare qubit linewidth
is measured as a broadening of the qubit’s intrinsic linewidth.
given by γq = 2γ2 to 2 1/T22 + Ω2R T1 /T2 , something that Notably, the slow frequency fluctuations can be cancelled by
is known as power broadening and which is illustrated in applying a π-pulse midway through a Ramsey fringe experiment.
Fig. 24(c). In practice, the unbroadenend dephasing rate The measured dephasing time is then known as T2echo and is
usually longer than T2∗ with its exact value depending on the
γ2 can be determined from spectroscopic measurements spectrum of the low-frequency noise affecting the qubit (Martinis
by extrapolating to zero spectroscopy tone power the et al., 2003). The method of dynamical decoupling which relies
2
linear dependence of νHWHM . This quantity can also be on more complex pulse sequences can be used to cancel higher-
determined in the time domain from a Ramsey fringe frequency components of the noise (Bylander et al., 2011).
 40

0.3
herent nature of the cavity field. For a thermal cavity
field, a n̄(n̄ + 1) dependence is rather expected and ob-
0.2 2χ n̂ served (Bertet et al., 2005; Kono et al., 2017).
Pe

This change in qubit linewidth due to photon shot

0.1 noise in the coherent measurement tone populating the
cavity can be interpreted as the unavoidable dephas-
0.0 ing that a quantum system undergoes during measure-
−1 0 1 2 3 4 5 ment. Using a polaron-type transformation familiar
from condensed-matter theory, the cavity can be inte-
0.02 |3 grated out of the qubit-cavity master equation and, in
2χ
|2 |4
this way, the associated measurement-induced dephas-
|5 ing rate can be expressed in the dispersive regime as
Pe

|1
0.01
|6 γm (t) = 2χIm[αg (t)αe∗ (t)], where αg/e (t) are the time-
|n = 0 |7 dependent coherent state amplitudes associated with the
0.00 two qubit states (Gambetta et al., 2008). In the long time
limit, the above rate can be expressed in the more intuitive
0 20 40 60 80
form γm = κ|αe − αg |2 /2, where αe − αg is the distance
δq /2π (MHz)
between the two steady-state pointer states (Gambetta
FIG. 25 Excited state population as a function of the qubit et al., 2008). Unsurprisingly, measurement-induced de-
drive frequency. (a) Dispersive regime with χ/2π = 0.1 MHz phasing is faster when the pointer states are more easily
and (b) strong dispersive limit with χ/2π = 5 MHz. The distinguishable and the measurement thus more efficient.
resolved peaks correspond to different cavity photon numbers. This last expression can also be directly obtained from
The spectroscopy drive amplitude is fixed to ΩR /2π = 0.1 the entangled qubit-pointer state Eq. (108) whose coher-
MHz and the damping rates to γ1 /2π = κ/2π = 0.1 MHz. In ence decay, at short times, at the rate γm under photon
panel (a) the measurement drive is on resonance with the bare
cavity frequency, with amplitude /2π = (0, 0.2, 0.4) MHz for
loss (Haroche and Raimond, 2006).
the light blue, blue and dark blue line, respectively. In panel Using the expressions Eqs. (109a) and (109b) for the
(b) the measurement drive is at the pulled cavity frequency pointer states amplitude, γm can be expressed as
ωr − χ with amplitude /2π = 0.1 MHz.
κχ2 (n̄g + n̄e )
γm = , (128)
δr2 + χ2 + (κ/2)2
measurements (Schuster et al., 2005). However, care with n̄σ = |ασ |2 the average cavity photon number given
must be taken since the linear dependence of the qubit that the qubit is state σ. The distinction between n̄g
frequency on power predicted in Eq. (107) is only valid and n̄e is important if the measurement drive is not
well inside the dispersive regime or, more precisely, at symmetrically placed between the two pulled cavity fre-
small n̄/ncrit . We come back to this shortly. quencies corresponding to the two qubit states. Taking
As is apparent from Fig. 25(a), in addition to causing a δr = ωr − ωd = 0 and thus n̄g = n̄e ≡ n̄ for a two-level
frequency shift of the qubit, the cavity photon population system, the measurement-induced dephasing rate takes,
also causes a broadening of the qubit linewidth. This can in the small χ/κ limit, the simple form γm ∼ 8χ2 n̄/κ.
be understood simply by considering again the form of Thus as announced above, the qubit linewitdth scales
Ĥdisp in Eq. (126). Indeed, while in the above discus- with n̄. With the cautionary remarks that will come be-
sion we considered only the average qubit frequency shift, low, measuring this linewidth versus the drive power is
2χh↠âi, the actual shift is rather given by 2χ↠â such thus another way to infer n̄ experimentally.
that the full photon-number distribution is important. So far, we have been concerned with the small χ/κ limit.
As a result, when the cavity is prepared in a coherent However, given the strong coupling and high-quality fac-
state by the measurement tone, each Fock state |ni of the tor that can be experimentally realized in circuit QED,
coherent field leads to its own qubit frequency shift 2χn. it is also interesting to consider the opposite limit where
In the weak dispersive limit corresponding to χ/κ small, χ/κ is large. A first consequence of this strong disper-
the observed qubit lineshape is thus the result of the sive regime, illustrated in Fig. 25(b), is that the qubit
inhomogeneous broadening due to the Poisson statistics frequency shift per photon can then be large enough
of the coherent state populating the cavity. This effect to be resolved spectroscopically (Gambetta et al., 2006;
becomes more apparent as the average measurement pho- Schuster et al., 2007). More precisely, this occurs if 2χ is
ton number n̄ increases and results in a crossover from a larger than γ2 + (n̄ + n)κ/2, the width of the nth photon
Lorentzian qubit lineshape whose linewidth scales with peak (Gambetta et al., 2006). Moreover, the amplitude
n̄ to√a Gaussian lineshape whose linewidth rather scales of each spectroscopic line is a measure of the probabil-
as n̄ (Gambetta et al., 2006; Schuster et al., 2005). ity of finding the corresponding photon number in the
This square-root dependence can be traced to the co- cavity. Using this idea, it is possible, for example, to
 41

ωri/2π (GHz)
experimentally distinguish between coherent and thermal (a) (b) (c)
population of the cavity (Schuster et al., 2007). This 7.01
strong dependence of the qubit frequency on the exact 7.00
photon number also allows for conditional qubit-cavity 6.99
logical operations where, for example, a microwave pulse 2 levels 3 levels 6 levels
is applied such that qubit state is flipped if and only if 0 20 40 60 0 20 40 60 30 40 50
there are n photons in the cavity (Johnson et al., 2010). 20log( /1 MHz)
Although challenging, this strong dispersive limit has also
been acheived in some cavity QED experiments (Gleyzes FIG. 26 Change of effective resonator frequency ωrσ with
et al., 2007; Guerlin et al., 2007). This regime has also increasing measurement drive power for the different states σ
been acheived in hybrid quantum systems, for example of the transmon qubit (dotted blue line: ground state, full red
in phonon-number resolving measurements of nanome- line: first excited state and dashed gray line: second excited
state). The horizontal green dashed line is the bare resonator
chanical oscillators (Arrangoiz-Arriola et al., 2019; Sletten
frequency. (a) Two-level artificial atom, (b) taking into account
et al., 2019) and magnon-number resolving measurements three levels of the transmon and (c) 6 levels of the transmon.
(Lachance-Quirion et al., 2017). The system parameters are chosen such that (ω01 , ω12 , g)/2π =
We now come back to the question of inferring the (6, 5.75, 0.1) GHz. Adapted from Boissonneault et al. (2010).
intra-cavity photon number from ac-Stark shift or qubit
linewidth broadening measurements. As mentioned previ-
ously, the linear dependence of the ac-Stark shift on the pendence of the qubit ac-Stark with power (Gambetta
measurement drive power predicted from the dispersive et al., 2006). We can only repeat that care must be taken
Hamiltonian Eq. (107) is only valid at small n̄/ncrit . In- when extracting the intra-cavity photon number in the
deed, because of higher-order corrections, the cavity pull dispersive regime.
itself is not constant with n̄ but rather decreases with in-
creasing n̄ (Gambetta et al., 2006). This change in cavity
pull is illustrated in Fig. 26(a) which shows the effective C. Beyond strong coupling: ultrastrong coupling regime
resonator frequency given that the qubit is in state σ as a
function of drive amplitude, ωrσ (n) = Eσ,n+1 −Eσ,n , with We have discussed consequences of the strong coupling,
Eσ,n+1 the dressed state energies defined in Eq. (38) (Bois- g > κ, γ2 , and strong dispersive, χ > κ, γ2 , regimes which
sonneault et al., 2010). At very low drive amplitude, the can both easily be realized in circuit QED. Although
cavity frequency is pulled to the expected value ωr ± χ de- the effect of light-matter interaction has important conse-
pending on the state of the qubit. As the drive amplitude quences, in both these regimes g is small with respect to
increases, and with it the intra-cavity photon number, the system frequencies, ωr , ωq  g, a fact that allowed
the pulled cavity frequency goes back to its bare value ωr . us to safely drop counter-rotating terms from Eq. (30).
Panels (b) and (c) show the pulled frequencies taking into In the case of a two-level system this allowed us to work
account three and six transmon levels, respectively. In with the Jaynes-Cummings Hamiltonian Eq. (34). The
contrast to the two-level approximation and as expected situation where these terms can no longer be neglected is
from Eq. (40), in this many-level situation the symmetry known as the ultrastrong coupling regime.
that was present in the two-level case is broken and the As discussed in Sec. III.A, the relative smallness of g
pulled frequencies are not symmetrically placed around with respect to the system frequencies √ can be traced to
ωr . We note that this change in effective cavity frequency Eq. (33) where we see that g/ωr ∝ α, with α ∼ 1/137
is at the heart of the high-power readout already discussed the fine-structure constant. This is, however, not a fun-
in Sec. V.C.2. damental limit and it is possible to take advantage of the
Because of this change in cavity pull, which can be flexibility of superconducting quantum circuits to engi-
interpreted as χ itself changing with photon numbers, the neer situations
√ where light-matter coupling rather scales
ac-Stark shift and the measurement-induced dephasing as ∝ 1/ α. In this case, the smallness of α now helps
do not necessarily follow the simple linear dependence boost the coupling rather than constraining it. A circuit
expected from Ĥdisp . For this reason, it is only possible to realizing this idea was first proposed in (Devoret et al.,
safely infer the intra-cavity photon number from measure- 2007) and is commonly known as the in-line transmon.
ment of the ac-Stark shift or qubit linewidth broadening It simply consists of a transmission line resonator whose
at small photon number. It is worth nothing that, in some center conductor is interrupted by a Josephson junction.
cases, the reduction in cavity pull can move the cavity Coupling strengths has large as g/ωr ∼ 0.15 can in prin-
frequency closer to the drive frequency, thereby leading ciple be obtained in this way but increasing this ratio
to an increase in cavity population. For some system further can be challenging because it is done at the ex-
parameters, these two nonlinear effects – reduction in pense of reducing the transmon anharmonicity (Bourassa
cavity pull and increase in cavity population – can partly et al., 2012).
compensate each other, leading to an apparent linear de- An alternative approach relies on galvanically coupling
 42

a flux qubit to the center conductor of a transmission-line commonly used gates to illustrate the basic principles.
resonator. In this configuration, light-matter coupling Unless otherwise noted, in this section we will assume the
can be made very large by increasing the impedance of qubits to be dispersively coupled to the resonator.
the center conductor of the resonator in the vicinity of the
qubit, something that can be realized by interrupting the
center conductor of the resonator by a Josephson junction A. Single-qubit control
or a junction array (Bourassa et al., 2009). In this way,
coupling strengths of g/ωq ∼ 1 or larger can be achieved. Arbitrary single-qubit rotations can be realized in an
These ideas were first realized in (Forn-Dı́az et al., 2010; NMR-like fashion with voltage drives at the qubit fre-
Niemczyk et al., 2010) with g/ωq ∼ 0.1 and more recently quency (Blais et al., 2007, 2004). One approach is to
with coupling strengths as large as g/ωq ∼ 1.34 (Yoshihara drive the qubit via one of the resonator ports (Wallraff
et al., 2016). Similar results have also been obtained in et al., 2005). Because of the large qubit-resonator de-
the context of waveguide QED where the qubit is coupled tuning, a large fraction of the input power is reflected
to an open transmission line rather than to a localized at the resonator, something that can be compensated by
cavity mode (Forn-Diaz et al., 2016). increasing the power emitted by the source. The reader
A first consequence of reaching this ultrastrong coupling will recognize that this approach is very similar to a qubit
regime is that, in addition to a Lamb shift g 2 /∆, the qubit measurement but, because of the very large detuning, with
transition frequency is further modified by the so-called δr  χ such that αe − αg ∼ 0 according to Eq. (110). As
Bloch-Siegert shift of magnitude g 2 /(ωq + ωr ) (Bloch and illustrated in Fig. 19, this far off-resonance drive therefore
Siegert, 1940). Another consequence is that the ground causes negligible measurement-induced dephasing (Blais
state of the combined system is no longer the factorizable et al., 2007). We also note that in the presence of multiple
state |g0i but is rather an entangled qubit-resonator state. qubits coupled to the same resonator, it is important that
An immediate implication of this observation is that the the qubits be sufficiently detuned in frequency from each
master equation Eq. (81), whose steady-state is |g0i, is not other to avoid the control drive intended for one qubit to
an appropriate description of damping in the ultrastrong inadvertently affect the other qubits.
coupling regime (Beaudoin et al., 2011). The reader Given this last constraint, an often more convenient
interested in learning more about this regime of light- approach, already illustrated in Fig. 13, is to capacitively
matter interaction can consult the reviews by Forn-Dı́az couple the qubit to an additional transmission line from
et al. (2019) and Frisk Kockum et al. (2019). which the control drives are applied. Of course, the
coupling to this additional control port must be small
enough to avoid any impact on the qubit relaxation time.
Following Sec. IV.F, the amplitude of the drive as seen
VII. QUANTUM COMPUTING WITH CIRCUIT QED √
by the qubit is given by ε = −i γβ, where β is the
amplitude of the drive at the input port, and γ is set by the
One of the reasons for the rapid growth of circuit QED
capacitance between the qubit and the transmission line.
as a field of research is its prominent role in quantum
A small γ, corresponding to a long relaxation time, can be
computing. The transmon is today the most widely used
compensated by increasing the drive amplitude |β|, while
superconducting qubit, and the dispersive measurement
making sure that any heating due to power dissipation
described in Sec. V is the standard approach to qubit
close to the qubit does not affect qubit coherence. Design
readout. Moreover, the capacitive coupling between trans-
guidelines for wiring, an overview of the power dissipation
mons that are fabricated in close proximity can be used to
induced by drive fields in qubit drive lines, and their effect
implement two-qubit gates. Alternatively, the transmon-
on qubit coherence is discussed, for example, in Krinner
resonator electric dipole interaction can also be used to
et al. (2019).
implement such gates between qubits that are separated
Similarly to Eq. (84), a coherent drive of time-
by distances as large as a centimeter, the resonator acting
dependent amplitude ε(t), frequency ωd and phase φd
as a quantum bus to mediate qubit-qubit interactions. As
on a transmon is then modeled by
illustrated in Fig. 27, realizing a quantum computer ar-  
chitecure, even of modest size, requires bringing together Ĥ(t) = Ĥq + ~ε(t) b̂† e−iωd t−iφd + b̂eiωd t+iφd , (129)
in a single working package essentially all of the elements
discussed in this review. where Ĥq = ~ωq b̂† b̂ − E2C (b̂† )2 b̂2 is the transmon Hamil-
In this section, we describe the basic principles behind tonian. Going to a frame rotating at ωd , Ĥ(t) takes the
one- and two-qubit gates in circuit QED. Our objective simpler form
is not to give a complete overview of the many different  
gates and gate-optimization techniques that have been Ĥ 0 = Ĥq0 + ~ε(t) b̂† e−iφd + b̂eiφd , (130)
developed. We rather focus on the key aspects of how light-
matter interaction facilitates coherent quantum operations where Ĥq0 = ~δq b̂† b̂ − E2C (b̂† )2 b̂2 with δq = ωq − ωd the
for superconducting qubits, and describe some of the more detuning between the qubit and the drive frequencies.
 43

1 mm
Qubits Coupling Flux lines Charge Feedline Readout Purcell
resonators lines resonators filters

FIG. 27 False colored optical microscope image of a four-transmon device. The transmon qubits are shown in yellow, the
coupling resonators in cyan, the flux lines for single qubit tuning in green, the charge lines for single-qubit manipulation in
pink, and a common feedline for multiplexed readout in purple, with transmission line resonators for dispersive readout (red)
employing Purcell filters (blue). Adapted from Andersen et al. (2019).

Truncating to two levels of the transmon as in Eq. (34), two quadratures of the qubit with the envelope of the
Ĥ 0 takes the form second quadrature chosen to be the time-derivative of
~δq ~ΩR (t) the envelope of the first quadrature (Gambetta et al.,
Ĥ 0 = σ̂z + [cos(φd )σ̂x + sin(φd )σ̂y ] , (131) 2011b; Motzoi et al., 2009). More generally, one can cast
2 2
the problem of finding an optimal drive as a numerical
where we have introduced the standard notation ΩR = optimization problem which can be tackled with optimal
2ε for the Rabi frequency. This form of Ĥ 0 makes it control approaches such as the GRadient Ascent Pulse
clear how the phase of the drive, φd , controls the axis of Engineering (GRAPE) (Khaneja et al., 2005).
rotation on the qubit Bloch sphere. Indeed, for δq = 0,
the choice φd = 0 leads to rotations around the X-axis Experimental results from Chow et al. (2010) compar-
while φd = π/2 to rotations around the Y -axis. Since ing the error in single-qubit gates with and without DRAG
any rotation on the Bloch sphere can be decomposed are shown in Fig. 28. At long gate times, decoherence
into X and Y rotations, arbitrary single-qubit control is is the dominant source of error such that both Gaussian
therefore possible using sequences of on-resonant drives and DRAG pulses initially improve as the gate time is
with appropriate phases. reduced. However, as the pulses get shorter and their
Implementing a desired gate requires turning on and off frequency bandwidth become comparable to the trans-
the drive amplitude. To realize as many logical operations mon anharmonicity, leakage leads to large errors for the
as possible within the qubit coherence time, the gate time Gaussian pulses. In contrast, the DRAG results continue
should be as short as possible and square pulses are opti- to improve as gates are made shorter and are consistent
mal from that point of view. In practice, however, such with a two-level system model of the transmon. These
pulses suffer from important deformation as they propa- observations show that small anharmonicity is not a fun-
gate down the finite-bandwidth transmission line from the damental obstacle to fast and high-fidelity single-qubit
source to the qubit. Moreover, for a weakly anharmonic gates. Indeed, thanks to pulse shaping techniques and
multi-level system such as a transmon, high-frequency long coherence times, state of the art single-qubit gate
components of the square pulse can cause unwanted tran- errors are below 10−3 , well below the threshold for topo-
sitions to levels outside the two-level computational sub- logical error correcting codes (Barends et al., 2014; Chen
space. This leakage can be avoided by using smooth et al., 2016).
(e.g. Gaussian) pulses, but this leads to longer gate times. While rotations about the Z axis can be realized by
Another solution is to shape the pulse so as to remove concatenating the X and Y rotations described above,
the unwanted frequency components. A widely used ap- several other approaches are used experimentally. Work-
proach that achieves this is known as Derivative Removal ing in a rotating frame as in Eq. (131) with δq = 0, one
by Adiabatic Gate (DRAG). It is based on driving the alternative method relies on changing the qubit transition
 44

Pulse width σ (ns) B. Two-qubit gates

1 2 3 4 6

0.1 Two-qubit gates are generally more challenging to re-

alize than single-qubit gates. Error rates for current
two-qubit gates are typically around one to a few per-
Error per gate

cent, which is an order of magnitude higher than those

of single-qubit gates. Recent experiments are, however,
closing this gap (Foxen et al., 2020). Improving two-qubit
0.01 gate fidelities at short gate times is a very active area
of research, and a wide variety of approaches have been
developed. A key challenge in realizing two-qubit gates is
Benchmarking with Gaussian pulses the ability to rapidly turn interactions on and off. While
Three-level simulation for single-qubit gates this is done by simply turning on
Benchmarking with DRAG pulses
and off a microwave drive, two-qubit gates require turn-
Two-level simulation
0.001 ing on a coherent qubit-qubit interaction for a fixed time.
10 15 20 25 Achieving large on/off ratios is far more challenging in
Total gate length (ns) this situation.
FIG. 28 Single-qubit gate errors extracted from randomized Broadly speaking, one can divide two-qubit gates into
benchmarking for Gaussian and DRAG pulses as a fonction different categories depending on how the qubit-qubit in-
of total gate time and pulse width σ for Gaussian pulses. teraction is activated. The main approaches discussed in
The experimental results (symbols) are compared to numerical the following are illustrated schematically in Fig. 29. An
simulations (lines) with two or three transmon levels. Adapted
important distinction between these different schemes is
from Chow et al. (2010).
whether they rely on frequency-tunable qubits or not. Fre-
quency tunability is convenient because it can be used to
frequency such that δq = 6 0 for a determined duration. In controllably tune qubits into resonance with one another
the absence of drive, ΩR = 0, this leads to phase accu- qubit or with a resonator. Using flux-tunable transmons
mulation by the qubit state and therefore to a rotation has led to some of the fastest and highest fidelity two-
about the Z axis. As discussed in Sec. II.E, fast changes qubit gates to date, see Fig. 29(a,b) (Arute et al., 2019;
of the qubit transition frequency are possible by, for exam- Barends et al., 2014; Chen et al., 2014). However, as men-
ple, applying a magnetic field to a flux-tunable transmon. tioned previously this leads to additional qubit dephasing
However, working with flux-tunable transmons is done at due to flux noise. An alternative are all-microwave gates
cost of making the qubit susceptible to dephasing due to which only use microwave drives, either on the qubits
flux noise. To avoid this, the qubit transition frequency or on a coupler bus such as a resonator, to activate an
can also be tuned without relying on a flux-tunable device effective qubit-qubit interaction, see Fig. 29(c). Finally,
by applying a strongly detuned microwave tone on the yet another category of gates are parametric gates where
qubit. For ΩR /δq  1, this drive does not lead to Rabi a system parameter is modulated in time at a frequency
oscillations but induces an ac-Stark shift of the qubit fre- which bridges an energy gap between the states of two
quency due to virtual transition caused by the drive (Blais qubits. Parametric gates can be all-microwave but, in
et al., 2007). Indeed, as shown in Appendix B.4, to second some instances, involve modulating system frequencies
order in ΩR /δq and assuming for simplicity a constant using external magnetic flux, see Fig. 29(d).
drive amplitude, this situation is described by the effective
Hamiltonian
 
00 1 EC Ω2R 1. Qubit-qubit exchange interaction
Ĥ ' ~ωq − σ̂z . (132)
2 2 δq2
a. Direct capacitive couplingOne of the conceptually sim-
The last term can be turned on and off with the amplitude plest ways to realize two-qubit gates is through direct
of the detuned microwave drive and can therefore be used capacitive coupling between the qubits, see Fig. 29(a). In
to realize Z rotations. analogy with Eq. (32), the Hamiltonian describing this
Finally, since the X and Y axis in Eq. (131) are de- situation reads
fined by the phase φd of the drive, a particularly simple
approach to realize a Z gate is to add the desired phase
Ĥ = Ĥq1 + Ĥq2 + ~J(b̂†1 b̂2 + b̂1 b̂†2 ), (133)
offset to the drive fields of all subsequent X and Y rota-
tions and two-qubit gates. This so-called virtual Z-gate
can be especially useful if the computation is optimized to where Ĥqi = ~ωqi b̂†i b̂i − ECi (b̂†i )2 b̂2i /2 is the Hamiltonian
use a large number of Z-rotations (McKay et al., 2017). of the ith transmon and b̂i the corresponding annihilation
 45

(a) (b)
J g1 g2
Q1 Q2 Q1 C Q2
Tunable Tunable Tunable Coupler Tunable
qubit 1 qubit 2 qubit 1 qubit 2
(c) (d)

g1 g2 g1 g2
Q1 C Q2 Q1 C Q2
Qubit 1 Coupler Qubit 2 Qubit 1 Qubit 2

FIG. 29 Schematic illustration of some of the two-qubit gate schemes discussed in the text. Exchange interaction between two
qubits (a) from direct capacitive coupling or (b) mediated by a coupler such as a bus resonator. The qubits are tuned in and
out of resonance with each other to activate and deactivate the interaction, respectively. (c) All-microwave gates activated by
microwave drives on the qubits and/or a coupler such as a bus resonator. In this scheme, the qubits can have a fixed frequency.
(d) Parametric gates involving modulating a system parameter, such a tunable coupler. Inspired from Yan et al. (2018).

operator. The interaction amplitude J takes the form (J 2 /∆12 )σ̂z1 σ̂z2 remains. This unwanted interaction in
 1/4 the off state of the gate leads to a conditional phase ac-
2EC1 EC2 EJ1 EJ2 cumulation on the qubits. As a result, the on-off ratio
~J = × , (134)
ECc 2EC1 2EC2 of this direct coupling gate is estimated to be ∼ ∆12 /J.
In practice, this ratio cannot be made arbitrarily small
with EJi and ECi the transmon Josephson and charg- because increasing the detuning of one pair of qubits in
ing energies, and ECc = e2 /2Cc the charging energy of a multi-qubit architecture might lead to accidental reso-
the coupling capacitance labelled Cc . This beam-splitter nance with a third qubit. This direct coupling approach
Hamiltonian describes the coherent exchange of an exci- was implemented by Barends et al. (2014) using frequency
tation between the two qubits. In the two-level approxi- tunable transmons with a coupling J/2π = 30 MHz and
mation, assuming tuned to resonance qubits, ωq1 = ωq2 , an on/off ratio of 100. We note that the unwanted phase
and moving to a frame rotating at the qubit frequency, accumulation due to the residual σ̂z1 σ̂z2 can in principle
Eq. (133) takes the familiar form be eliminated using refocusing techniques borrowed from
Ĥ 0 = ~J(σ̂+1 σ̂−2 + σ̂−1 σ̂+2 ). (135) nuclear magnetic resonance (Slichter, 1990).

Evolution under this √Hamiltonian for a time π/(4J) leads

to an entangling iSWAP gate which, up to single- Another approach to turn on and off the swap interac-
qubit rotations, is equivalent to a controlled NOT gate tion is to make the J coupling itself tunable in time. This
(CNOT) (Loss and DiVincenzo, 1998). is conceptually simple, but requires more complex cou-
As already mentioned, to precisely control the evolution pling circuitry typically involving flux-tunable elements
under Ĥ 0 and therefore the gate time, it is essential to be that can open additional decoherence channels for the
able to vary the qubit-qubit interaction with a large on/off qubits. One advantage is, however, that tuning a coupler
ratio. There are essentially two approaches to realizing rather than qubit transition frequencies helps in reducing
this. The most straightforward way is to tune the qubits the frequency crowding problem. This approach is used,
in resonance to perform a two-qubit gate, and to strongly for example, by (Chen et al., 2014) where two transmon
detune them to stop the coherent exchange induced by qubits are coupled via a flux-tunable inductive coupler.
Ĥ 0 (Bialczak et al., 2010; Blais et al., 2003; Dewes et al., In this way, it was possible to realize an on/off ratio of
2012). Indeed, for J/∆12  1 where ∆12 = ωq1 − ωq2 1000, with√a maximum coupling of 100 MHz correspond-
is the detuning between the two qubits, the coherent ex- ing to a iSWAP gate in 2.5 ns. A simpler approach
change J is suppressed and can be dropped from Eq. (133) based on a frequency tunable transmon qubit acting as
under the RWA. A more careful analysis following the coupler, as suggested in Yan et al. (2018), was also used
same arguments and approach used to describe the dis- to tune qubit-qubit coupling from 5 MHz to −40 MHz
persive regime (cf. Sec. III.C) shows that, to second order going through zero coupling with a gate time of ∼ 12 ns
in J/∆12 , a residual qubit-qubit interaction of the form and a gate infidelity of ∼ 0.5% (Arute et al., 2019).
 46

b. Resonator mediated coupling An alternative to the 18

above approach is to use a resonator as a quantum

Frequency (GHz)
bus mediating interactions between two qubits, see ζ
17
Fig. 29(b) (Blais et al., 2007, 2004; Majer et al., 2007). An
advantage compared to direct coupling is that the qubits
|2, 0, 0
do not have to be fabricated in close proximity to each
16
other. With the qubits coupled to the same resonator,
and in the absence of any direct coupling between the |1, 1, 0
qubits, the Hamiltonian describing this situation is |0, 2, 0
15
15 16 17 18
2
X 10
Ĥ = Ĥq1 + Ĥq2 + ~ωr ↠â + ~gi (↠b̂i + âb̂†i ). (136)

Frequency (GHz)
9
i=1
2J
One way to make use of this pairwise interaction is, as- 8 |1, 0, 0
2g
suming the resonator to be in the vacuum state, to first 7
tune one of the two qubits in resonance with the resonator |0, 0, 0
for half a vacuum Rabi oscillation cycle, swapping an ex- 6
citation from the qubit to the resonator, before tuning it |0, 1, 0
back out of resonance. The second qubit is then tuned 5
10 12 14 16 18
in resonance mapping the excitation from the resonator
Second qubit frequency (GHz)
to the second qubit (Sillanpää et al., 2007). While this
sequence of operations can swap the quantum state of the FIG. 30 Spectrum of two transmon qubits coupled to a com-
first qubit to the second, clearly demonstrating the role mon resonator as a function of the frequency of the second
of the resonator as a quantum bus, it does not correspond qubit in the (a) 2-excitation and (b) 1-excitation manifold.
to an entangling two-qubit gate. The full lines are obtained by numerical diagonalization of
Eq. (136) in the charge basis with 5 transmon levels and 5 res-
Alternatively, a two-qubit gate can be performed by
onator levels, and with parameters adapted from DiCarlo et al.
only virtually populating the resonator mode by work- (2009): EJ1(2) /h = 28.48(42.34) GHz, EC1(2) /h = 317(297)
ing in the dispersive regime where both qubits are far MHz and g1(2) /2π = 199(190) MHz. In the one-excitation
detuned from the resonator (Blais et al., 2007, 2004; Ma- manifold, both the 2g anticrossing of the first qubit with the
jer et al., 2007). Building on the results of Sec. III.C, resonator and the 2J anticrossing of the two qubits are vis-
in this situation the effective qubit-qubit interaction is ible. In the two-excitation manifold, the 11-02 anticrossing
revealed by using of magnitude ζ can be seen. Notice the change in horizon-
hP theapproximate i dispersive transforma- tal scale between the two panels. The states are labelled as
gi † †
tion Û = exp i ∆i â b̂i − âb̂i on Eq. (136). Making |1st qubit, 2nd qubit, resonatori. The dashed light blue lines
use of the Baker-Campbell-Hausdorff expansion Eq. (B2) are guides to the eye following the bare frequency of the first
to second order in gi /∆i , we find qubit.

Ĥ 0 = Ĥq1
0 0
+ Ĥq2 + ~J(b̂†1 b̂2 + b̂1 b̂†2 ) coupling in Ĥ 0 takes the form
2
X  
+ ~ω̃r ↠â + ~χabi ↠âb̂†i b̂i J=
g1 g2 1
+
1
(138)
(137)
i=1 2 ∆1 ∆2
X  
+ ~Ξij b̂†i b̂i b̂†j b̂i + b̂†i b̂j , and reveals itself in the frequency domain by an anticross-
i6=j ing of size 2J between the qubit states |01i and |10i. This
is illustrated in Fig. 30(b) which shows the eigenenergies
with Hqi 0
' ~ω̃qi b̂†i b̂i − E2Ci (b̂†i )2 b̂2i the transmon Hamilto- of the Hamiltonian Eq. (136) in the 1-excitation manifold.
nians, χabi ' −2ECi gi2 /∆2i a cross-Kerr coupling between In this figure, the frequency of qubit 1 is swept while that
the resonator and the ith qubit. The frequencies ω̃qi and of qubit 2 is kept constant at ∼ 8 GHz with the resonator
ω̃r include the Lamb shift. The last line can be understood at ∼ 7 GHz. From left to right, we first see the vacuum
as an excitation number dependent exchange interaction Rabi splitting of size 2g at ωq1 = ωr , followed by a smaller
with Ξij = ECi gi gj /(2∆i ∆j ). Since this term is much anticrossing of size 2J at the qubit-qubit resonance. It is
smaller than the J-coupling it can typically be neglected. worth mentioning that the above expression for J is only
Note that we have not included a self-Kerr term of order valid for single-mode oscillators and is renormalized in the
χabi on the resonator. This term is of no practical con- presence of multiple modes (Filipp et al., 2011; Solgun
sequences in the dispersive regime where the resonator et al., 2019).
is only virtually populated. The resonator-induced J To understand the consequence of the J-coupling in
 47

the time domain, it is useful to note that, if the resonator gate since
is initially in the vacuum state, it will remain in that
state under the influence of Ĥ 0 . In other words, the ĈZ (φ) = diag(1, 1, 1, eiφ )
(140)
resonator is only virtually populated by its dispersive 1
= R̂Z 2
(−φ10 )R̂Z (−φ01 )ĈZ (φ01 , φ10 , φ11 ),
interaction with the qubits. For this reason, with the
resonator initialized in the vacuum state, the second line R i
with φ = φ11 − φ01 − φ10 = dt ζ(t) and where R̂Z (θ) =
of Eq. (137) can for all practical purposes be ignored and iθ
diag(1, e ) is a single qubit phase gate acting on qubit
we are back to the form of the direct coupling Hamiltonian i. For φ 6= 0 this is an entangling two-qubit gate and, in
of Eq. (133). Consequently, when both qubits are tuned particular, for φ = π it is a controlled-Z gate (CPHASE).
in resonance with each other, but still dispersive with Rather than adiabatically tuning the flux in and out of
respect to the resonator, the latter acts as a quantum bus the 11−02 resonance, an alternative is to non-adiabatically
mediating interactions between the qubits. An entangling pulse to this anti-crossing (DiCarlo et al., 2010; Strauch
gate can thus be performed in the same way as with direct et al., 2003; Yamamoto et al., 2010). In this sudden
capacitive coupling, either by tuning the qubits in and out approximation, leaving the system there for a time t, the
of resonance with each other (Majer et al., 2007) or by state |11i evolves into cos(ζt/2~)|11i+sin(ζt/2~)|02i. For
making the couplings gi tunable (Gambetta et al., 2011a; t = h/ζ, |11i is mapped back into itself but acquires a
Srinivasan et al., 2011). minus sign in the process. On the other hand, since they
are far from any resonance, the other logical states evolve
trivially. This therefore again results in a CPHASE gate.
In this way, fast controlled-Z gates are possible. For direct
2. Flux-tuned 11-02 phase gate
qubit-qubit coupling in particular, some of the fastest and
highest fidelity two-qubit gates have been achieved this
The 11-02 phase gate is a controlled-phase gate that way with error rates below the percent level and gate
is well suited to weakly anharmonic qubits such as trans- times of a few tens of ns (Barends et al., 2014; Chen et al.,
mons (DiCarlo et al., 2009; Strauch et al., 2003). It is 2014).
obtained from the exchange interaction of Eq. (133) and Despite its advantages, a challenge associated with this
can thus be realized through direct (static or tunable) gate is the distortions in the flux pulses due to the finite
qubit-qubit coupling or indirect coupling via a resonator bandwidth of the control electronics and line. In addition
bus. to modifying the waveform experienced by the qubit, this
√
In contrast to the iSWAP gate, the 11-02 phase gate can lead to long time scale distortions where the flux
is not based on tuning the qubit transition frequencies at the qubit at a given time depends on the previous
between the computational states into resonance with flux excursions. This situation can be partially solved by
each other, but rather exploits the third energy level of pre-distorting the pulses taking into account the known
the transmon. The 11-02 gate relies on tuning the qubits distortion, but also by adapting the applied flux pulses
to a point where the states |11i and |02i are degenerate in to take advantage of the symmetry around the transmon
the absence of J coupling. As illustrated in Fig. 30(a), the sweet-spot to cancel out unwanted contributions (Rol
qubit-qubit coupling lifts this degeneracy by an energy ζ et al., 2019).
whose value can be found perturbatively (DiCarlo et al.,
2009). Because of this repulsion caused by coupling to
the state |02i, the energy E11 of the state |11i is smaller 3. All-microwave gates
than E01 + E10 by ζ. Adiabatically flux tuning the qubits
in and out of the 11 − 02 anticrossing therefore leads to a Because the on/off ratio of the gates discussed above
conditional phase accumulation which is equivalent to a is controlled by the detuning between the qubits, it is
controlled-phase gate. necessary to tune the qubit frequencies over relatively
To see this more clearly, it is useful to write the unitary large frequency ranges or, alternatively, to have tunable
corresponding to this adiabatic time evolution as coupling elements. In both cases, having a handle on
the qubit frequency or qubit-qubit coupling opens the
  system to additional dephasing. Moreover, changing the
1 0 0 0 qubit frequency over large ranges can lead to accidental
 0 eiφ01 0 0 
ĈZ (φ01 , φ10 , φ11 ) = 
 0 0 eiφ10 0  ,
 (139) resonance with other qubits or uncontrolled environmental
modes, resulting in energy loss. For these reasons, it
0 0 0 eiφ11 can be advantageous to control two-qubit gates in very
R much the same way as single-qubit gates: by simply
where φab = dt Eab (t)/~ is the dynamical phase ac- turning on and off a microwave drive. In this section, we
cumulated over total flux excursion. Up to single-qubit describe two so-called all-microwave gates: the resonator-
rotations, this is equivalent to a standard controlled-phase induced phase (RIP) gate and the cross-resonance (CR)
 48

gate. Both are based on fixed-frequency far off-resonance the resonator mediates a σ̂z1 σ̂z2 interaction between the
qubits with an always-on qubit-resonator coupling. The two qubits and therefore leads to a conditional phase
RIP gate is activated by driving a common resonator gate. This expression also makes it clear that the need to
and the CR gate by driving one of the qubits. Other all- avoid measurement-induced dephasing with δ̃r  κ limits
microwave gates which will not be discussed further here the effective interaction strength and therefore leads to
include the sideband-based iSWAP (Leek et al., 2009), the relatively long gate times. This can, however, be mitigated
bSWAP (Poletto et al., 2012), the microwave-activated by taking advantage of pulse shaping techniques (Cross
CPHASE (Chow et al., 2013) and the fg-ge gate (Egger and Gambetta, 2015) or by using squeezed radiation to
et al., 2019; Zeytinoğlu et al., 2015). erase the which-qubit information in the output field of
the resonator (Puri and Blais, 2016). Similarly to the
longitudinal readout protocol discussed in Sec. V.C.3,
a. Resonator-induced phase gate The RIP gate relies on longitudinal coupling also offers a way to overcome many
two strongly detuned qubits that are dispersively coupled of the limitations of the conventional RIP gate (Royer
to a common resonator mode. The starting point is thus et al., 2017).
Eq. (137) where we now neglect the J coupling by taking Some of the advantages of this two-qubit gate are that it
|ωq1 − ωq2 |  J. In the two-level approximation and can couple qubits that are far detuned from each other and
accounting for a drive on the resonator, this situation is that it does not introduce significant leakage errors (Paik
described by the Hamiltonian et al., 2016). This gate was demonstrated by Paik et al.
(2016) with multiple transmons coupled to a 3D resonator,
ω̃q1 ω̃q2
Ĥ 0 /~ = σ̂z1 + σ̂z2 + ω˜r ↠â achieving error rates of a few percent and gate times of
2 2 several hundred nanoseconds.
X 2 (141)
+ χi ↠âσ̂zi + ε(t)(↠e−iωd t + âeiωd t ),
i=1
b. Cross-resonance gate The cross-resonance gate is based
where ε(t) is the time-dependent amplitude of the res- on qubits that are detuned from each other and cou-
onator drive and ωd its frequency. Note that we also pled by an exchange term J of the form of Eq. (133) or
neglect the resonator self-Kerr nonlinearity. Eq. (137) (Chow et al., 2011; Rigetti and Devoret, 2010).
The gate is realized by adiabatically ramping on and off While the RIP gate relies on off-resonant driving of a
the drive ε(t), such that the resonator starts and ends in common oscillator mode, this gate is based on directly
the vacuum state. Crucially, this means that the resonator driving one of the qubits at the frequency of the other.
is unentangled from the qubits at the start and end of the Moreover, since the resonator is not directly used and, in
gate. Moreover, to avoid measurement-induced dephasing, fact, ideally remains in its vacuum throughout the gate,
the drive frequency is chosen to be far from the cavity the J coupling can be mediated by a resonator or by
mode, δ̃r = ω̃r − ωd  κ. Despite this strong detuning, direct capactitive coupling.
the dispersive shift causes the resonator frequency to In the two-level approximation and in the absence of
depend on the state of the two qubits and, as a result, the the drive, this interaction takes the form
resonator field evolves in a closed path in phase space that ~ωq1 ~ωq2
is qubit-state dependent. This leads to a different phase Ĥ = σ̂z1 + σ̂z2 +~J(σ̂+1 σ̂−2 + σ̂−1 σ̂+2 ). (143)
2 2
accumulation for the different qubit states, and therefore
to a controlled-phase gate of the form of Eq. (139). To see how this gate operates, it is useful to diagonalize Ĥ
This conditional phase accumulation can be made using the two-level system version of the transformation
more apparent by moving Eq. (141) to a frame rotat- Eq. (47). The result takes the same general form as
ing at the drive frequency and by applying the po- Eq. (48) and Eq. (49), after projecting to two levels.
laron transformation Û = exp[α̂0 (t)↠− α̂∗0 (t)â] with In this frame, the presence of the J coupling leads to a
0
P renormalization of the qubit frequencies which for strongly
α (t) = α(t) − i χi σ̂zi /δ̃r on the resulting Hamiltonian.
This leads to the approximate effective Hamiltonian (Puri detuned qubits, |∆12 | = |ωq1 − ωq2 |  |J|, take the values
and Blais, 2016) ω̃q1 ≈ ωq1 + J 2 /∆12 and ω̃q2 ≈ ωq2 − J 2 /∆12 to second
" # order in J/∆12 . In the same frame, a drive on the first
X δ̃qi qubit, ~ΩR (t) cos(ωd t)σ̂x1 , takes the form (Chow et al.,
00
Ĥ ' ~ + χ|α(t)| σ̂zi + ~δr ↠â
2
2011)
i
2
2
(142) ~ΩR (t) cos(ωd t) (cos θσ̂x1 + sin θσ̂z1 σ̂x2 )
X 2χ 1 χ 2 |α(t)|2
 
+ ~χi ↠âσ̂zi − ~ σ̂z1 σ̂z2 , J (144)
i=1
δr ≈ ~ΩR (t) cos(ωd t) σ̂x1 + σ̂z1 σ̂x2 ,
∆12
with δ̃x = ω̃x − ωd and where the field amplitude α(t) with θ = arctan(2J/∆12 )/2 and where the second line
satisfies α̇ = −iδ̃r α − i(t). In this frame, it is clear how is valid to first order in J/∆12 . As a result, driving the
 49

first qubit at the frequency of the second qubit, ωd = ω̃q2 , 2013; Sheldon et al., 2016). The σ̂z1 σ̂z2 interaction is
activates the term σ̂z1 σ̂x2 which can be used to realize a detrimental to the gate fidelity as it effectively makes the
CNOT gate. frequency of the second qubit dependent on the logical
More accurate expressions for the amplitude of the CR state of the first qubit. Because of this, the effective
term σ̂z1 σ̂x2 can be obtained by taking into account more dressed frequency of the second qubit cannot be known
levels of the transmons. In this case, the starting point is in general, such that it is not possible to choose the drive
the Hamiltonian Eq. (133) with, as above, a drive term frequency ωd to be on resonance with the second qubit,
on the first qubit irrespective of the state of the first. As a consequence, the
CR term σ̂z1 σ̂x2 in Eq. (146) will rotate at an unknown
Ĥ = Ĥq1 + Ĥq2 + ~J(b̂†1 b̂2 + b̂1 b̂†2 ) qubit-state dependent frequency, leading to a gate error.
(145)
+ ~ε(t)(b̂†1 e−iωd t + b̂1 eiωd t ), The σ̂z1 σ̂z2 term should therefore be made small, which
ultimately limits the gate speed. Interestingly, for a pair
where ωd ∼ ωq2 . Similarly to the previous two-level of qubits with equal and opposite anharmonicity, χ12 = 0
system example, it is useful to eliminate the J-coupling. and this unwanted effect is absent. This cannot be re-
We do this by moving to a rotating frame at the drive alized with two conventional transmons, but is possible
frequency for both qubits, followed by a Schireffer-Wolff with other types of qubits (Ku et al., 2020; Winik, 2020).
transformation to diagonalize the first line of Eq. (145) Since J 0 is small, another caveat of the CR gate is that
to second order in J, see Appendix B.1. The drive term large microwave amplitudes ε are required for fast gates.
is modified under the same transformation by using the For the typical low-anharmonicity of transmon qubits, this
explicit expression for the Schrieffer-Wolff generator Ŝ = can lead to leakages and to effects that are not captured
Ŝ (1) + . . . given in Eq. (B6), and the Baker-Campbell- by the second-order perturbative results of Eqs. (144)
Hausdorff formula Eq. (B2) to first order: eŜ b̂1 e−Ŝ ' and (146). More detailed modeling based on the Hamilto-
b̂1 + [Ŝ (1) , b̂1 ]. The full calculation is fairly involved and nian of Eq. (145) suggests that classical crosstalk induced
here we only quote the final result after truncating to the on the second qubit from driving the first qubit can be
two lowest levels of the transmon qubits (Magesan and important and is a source of discrepancy between the
Gambetta, 2018; Tripathi et al., 2019) simple two-level system model and experiments (Magesan
and Gambetta, 2018; Tripathi et al., 2019; Ware et al.,
~δ̃q1 ~δ̃q2 ~χ12 2019). Because of these spurious effects, CR gate times
Ĥ 0 ' σ̂z1 + σ̂z2 + σ̂z1 σ̂z2
2  2 2  have typically been relatively long, of the order of 300
(146)
0 EC1 J 0 to 400 ns with gate fidelities ∼ 94–96% (Córcoles et al.,
+~ε(t) σ̂x1 − J σ̂x2 − σ̂z1 σ̂x2 .
~ ∆12 2013). However, with careful calibration and modeling
beyond Eq. (146), it has been possible to push gate times
In this expression, the detunings include frequency shifts down to the 100–200 ns range with error per gates at the
due to the J coupling with δ̃q1 = ωq1 + J 2 /∆12 + χ12 − ωd percent level (Sheldon et al., 2016).
and δ̃q2 = ωq2 − J 2 /∆12 + χ12 − ωd . The parameters χ12 Similarlrly to the RIP gate, advantages of the CR gate
and J 0 are given by include the fact that realizing this gate can be realized
using the same drive lines that are used for single-qubit
J2 J2
χ12 = EC2
− EC1
, (147a) gates. Moreover, it works with fixed frequency qubits
∆12 + ~ ∆12 − ~ which often have longer phase coherence times than their
J flux-tunable counterparts. However, both the RIP and the
J0 = . (147b)
∆12 −
EC1 CR gate are slower than what can now be achieved with
~
additional flux control of the qubit frequency or of the
Equations (144) and (146) agree in the limit of large coupler. We also note that, due to the factor EC1 /~∆12 in
anharmonicity EC1,2 and we again find that a drive on the amplitude of the σ̂z1 σ̂x2 term, the detuning of the two
the first qubit at the frequency of the second qubit acti- qubits cannot be too large compared to the anharmonicity,
vates the CR term σ̂z1 σ̂x2 . However, there are important putting further constraints on the choice of the qubit
differences at finite EC1/2 , something which highlights the frequencies. This may lead to frequency crowding issues
importance of taking into account the multilevel nature when working with large numbers of qubits.
of the transmon. Indeed, the amplitude of the CR term
is smaller here than in Eq. (144) with a two-level system.
Moreover, in contrast to the latter case, when taking into 4. Parametric gates
account multiple levels of the transmon qubits we find
a spurious interaction σ̂z1 σ̂z2 of amplitude χ12 between As we have already discussed, a challenge in realizing
the two qubits, as well as a drive on the second qubit two-qubit gates is activating a coherent interaction be-
of amplitude J 0 ε(t). This unwanted drive can be echoed tween two qubits with a large on/off ratio. The gates
away with additional single-qubit gates (Córcoles et al., discussed so far have aimed to achieve this in different
 50
√
ways. The iSWAP and the 11-02 gates are based on flux- Taking ωq1 (t) = ωq1 + ε sin(ωm t), the transition frequency
tuning qubits into a resonance condition or on a tunable of the first qubit develops frequency modulation (FM)
coupling element. The RIP gate is based on activating sidebands. The two qubits can then be effectively brought
an effective qubit-qubit coupling by driving a resonator into resonance by choosing the modulation to align one
and the CR gate by driving one of the qubits. Another of the FM sidebands with ωq2 , thereby rendering the
approach is to activate an off-resonant interaction by J effectively coupling resonant. This can be seen more
modulating a qubit frequency, a resonator frequency, or clearly by moving to a rotating frame defined by the
the coupling parameter itself at an appropriate frequency. unitary
This parametric modulation provides the energy necessary Rt
i
dt0 ωq1 (t0 )σ̂z1 −iωq2 tσ̂z2 /2
to bridge the energy gap between the far detuned qubit Û = e− 2 0 e , (152)
states. Several such schemes, known as parametric gates,
have been theoretically developed and experimentally re- where the Hamiltonian takes the form (Beaudoin et al.,
alized, see for example Beaudoin et al. (2012); Bertet et al. 2012; Strand et al., 2013)
(2006); Caldwell et al. (2018); Didier et al. (2018); Kapit ∞  
X ε
(2015); Liu et al. (2007); McKay et al. (2016); Naik et al. 0
Ĥ = J Jn in ei(∆12 −nωm )t σ̂+1 σ̂−2
(2017); Niskanen et al. (2007, 2006); Reagor et al. (2018); n=−∞
ωm
 (153)
Sirois et al. (2015); and Strand et al. (2013).
n i(ωq1 +ωq2 −nωm t)t
The key idea behind parametric gates is that modula- +i e σ̂+1 σ̂+2 + H.c. .
tion of a system parameter can induce transitions between
energy levels that would otherwise be too far off-resonance To arrive at the above expression,
P∞we have used the Jacobi-
to give any appreciable coupling. We illustrate the idea Anger expansion eiz cos θ = n=−∞ in
Jn (z)e
inθ
, with
first with two directly coupled qubits described by the Jn (z) Bessel functions of the first kind. Choosing the
Hamiltonian modulation frequency such that nωm = ∆12 aligns the
~ωq1 ~ωq2 nth sideband with the resonator frequency such that a
Ĥ = σ̂z1 + σ̂z2 + J(t)σ̂x1 σ̂x2 , (148)
2 2 resonant qubit-resonator interaction is recovered. The
largest contribution comes from the first sideband with
where we assume that the coupling is periodically modu-
J1 which has a maximum around J1 (1.84) ' 0.58, thus
lated at the frequency ωm , J(t) = J0 + J˜ cos(ωm t). Mov-
corresponding to an effective coupling that is a large frac-
ing to a rotating frame at the qubit frequencies, the above
tion of the bare J coupling. Note that the assumption of
Hamiltonian takes the form
 having a simple sinusoidal modulation of the frequency
Ĥ 0 = J(t) ei(ωq1 −ωq2 )t σ̂+1 σ̂−2 neglects the fact that the qubit frequency has a nonlin-
ear dependence on external flux for tunable transmons.
 (149)
This behavior can still be approximated by appropriately
+ ei(ωq1 +ωq2 )t σ̂+1 σ̂+2 + H.c. . varying Φx (t) (Beaudoin et al., 2012).
Parametric gates can also be mediated by modulating
Just as in Sec. VII.B.1.a, if the coupling is constant J(t) = the frequency of a resonator bus to which qubits are dis-
J0 , and J0 /(ωq1 − ωq2 ), J0 /(ωq1 + ωq2 )  1, then all the persively coupled (McKay et al., 2016). Much as with
terms of Ĥ 0 are fast-rotating and can be neglected. In this flux-tunable transmons, the resonator is made tunable by
situation, the gate is in the off state. On the other hand, inserting a SQUID loop in the center conductor of the res-
by appropriately choosing the modulation frequency ωm , onator (Castellanos-Beltran and Lehnert, 2007; Sandberg
it is possible to selectively activate some of these terms. et al., 2008). Changing the flux threading the SQUID
Indeed, for ωm = ωq1 − ωq2 , the terms σ̂+1 σ̂−2 + H.c. are loop changes the SQUID’s inductance and therefore the
no longer rotating and are effectively resonant. Dropping effective length of the resonator. As in a trombone, this
the rapidly rotating terms, this leads to leads to a change of the resonator frequency. An advan-
tage of modulating the resonator bus over modulating
J˜
Ĥ 0 ' (σ̂+1 σ̂−2 + σ̂−1 σ̂+2 ) . (150) the qubit frequency is that the latter can have a fixed
2 frequency, thus reducing its susceptibility to flux noise.
As already discussed, this interaction √ can be used to Finally, it is worth pointing out that while the speed of
generate entangling gates such as the iSWAP. If rather the cross-resonance gate is reduced when the qubit-qubit
ωm = ω1 + ω2 then σ̂+1 σ̂+2 + H.c. is instead selected. detuning is larger than the transmon anharmonicity, para-
In practice, it can sometimes be easier to modulate metric gates do not suffer from this problem. As a result,
a qubit or resonator frequency rather than a coupling there is more freedom in the choice of the qubit frequen-
strength. To see how this leads to a similar result, consider cies with parametric gates, which is advantageous to avoid
the Hamiltonian frequency crowding related issues such as addressability
~ωq1 (t) ~ωq2 errors and crosstalk. We also note that the modulation
Ĥ = σ̂z1 + σ̂z2 + J σ̂x1 σ̂x2 . (151) frequencies required to activate parametric gates can be a
2 2
 51

few hundred MHz, in contrast to the RIP gate or the CR et al., 2016a), and Gottesman-Kitaev-Preskill (GKP)
gate which require microwave drives. Removing the need codes (Campagne-Ibarcq et al., 2019; Flühmann et al.,
for additional microwave generators simplifies the control 2019; Gottesman et al., 2001), as well as a two-mode am-
electronics and may help make the process more scalable. plitude damping code described in (Chuang et al., 1997).
A counterpoint is that fast parametric gates often require To understand the basic idea behind this approach, we
large modulation amplitudes, which can be challenging. first consider the simplest instance of the binomial code
in which a qubit is encoded in the following two states of
a resonator mode (Michael et al., 2016a)
C. Encoding a qubit in an oscillator
1
|0L i = √ (|0i + |4i) , |1L i = |2i, (154)
So far we have discussed encoding of quantum informa- 2
tion into the first two energy levels of an artificial atom, with Fock states |ni. The first aspect to notice is that for
the cavity being used for readout and two-qubit gates. both logical states, the average photon number is n̄ = 2
However, cavity modes often have superior coherence and, as a result, the likelihood of a photon loss event is
properties than superconducting artificial atoms, some- the same for both states. An observer detecting a loss
thing that is especially true for the 3D cavities discussed event will therefore not gain any information allowing
in Sec. II.C (Reagor et al., 2016). This suggests that her to distinguish whether the loss came from |0L i or
encoding quantum information in the oscillator mode can from |1L i. This is a necessary condition for a quantum
be advantageous. Using oscillator modes to store and state encoded using the logical states Eq. (154) to not be
manipulate quantum information can also be favorable ‘deformed’ by a photon loss event. Moreover, under the
for quantum error correction which is an essential aspect action of â, the arbitrary superposition c0 |0L i + c1 |1L i
of scalable quantum computer architectures (Nielsen and becomes c0 |3i + c1 |1i after normalization. The coefficients
Chuang, 2000). c0 and c1 encoding the quantum information are intact
Indeed, in addition to their long coherence time, oscil- and the original state can in principle be recovered with a
lators have a simple and relatively well-understood error unitary transformation. By noting that while the original
model: to a large extent, the dominant error is single- state only has support on even photon numbers, the
photon loss. Taking advantage of this, it is possible to state after a photon loss only has support on odd photon
design quantum error correction codes that specifically numbers, we see that the photon loss event can be detected
correct for this most likely error. This is to be contrasted by measuring photon number parity P̂ = (−1)n̂ . The
to more standard codes, such as the surface code, which parity operator thus plays the role of a stabilizer for this
aim at detecting and correcting both amplitude and phase code (Michael et al., 2016b; Nielsen and Chuang, 2000).
errors (Fowler et al., 2012). Moreover, as will become This simple encoding should be compared to directly
clear below, the infinite dimensional Hilbert space of a using the Fock states {|0i, |1i} to store quantum infor-
single oscillator can be exploited to provide the redun- mation. Clearly, in this case, a single photon loss on
dancy which is necessary for error correction thereby, in c0 |0i + c1 |1i leads to |0i and the quantum information
principle, allowing using less physical resources to protect has been irreversibly lost. Of course, this disadvantage
quantum information than when using two-level systems. should be contrasted to the fact that the rate at which
Finally, qubits encoded in oscillators can be concatenated photons are lost, which scales with n̄, is (averaged over the
with conventional error correcting codes, where the latter code words) four times as large when using the encoding
should be optimized to exploit the noise resilience pro- Eq. (154), compared to using the Fock states {|0i, |1i}.
vided by the oscillator encoding (Grimsmo et al., 2020; This observation reflects the usual conundrum of quantum
Guillaud and Mirrahimi, 2019; Puri et al., 2019; Tuckett error correction: using more resources (here more pho-
et al., 2018, 2019a,b). tons) to protect quantum information actually increases
Of course, as we have already argued, nonlinearity re- the natural error rate. The protocol for detecting and cor-
mains essential to prepare and manipulate quantum states recting errors must be fast enough and accurate enough
of the oscillator. When encoding quantum information in to counteract this increase. The challenge for experimen-
a cavity mode, a dispersively coupled artificial atom (or tal implementations of quantum error correction is thus
other Josephson junction-based circuit element) remains to reach and go beyond the break-even point where the
present but only to provide nonlinearity to the oscillator encoded qubit, here Eq. (154), has a coherence time ex-
ideally without playing much of an active role. ceeding the coherence time of the unencoded constituent
Oscillator encodings of qubits investigated in the con- physical components, here the Fock states {|0i, |1i}. Near
text of quantum optics and circuit QED include cat codes break-even performance with the above binomial code
(Cochrane et al., 1999; Gilchrist et al., 2004; Grimm has been experimentally reported by Hu et al. (2019).
et al., 2019; Lescanne et al., 2019b; Mirrahimi et al., The simplest binomial code introduced above is able to
2014; Ofek et al., 2016; Puri et al., 2017; Ralph et al., correct a single amplitude-damping error (photon loss).
2003), the related binomial codes (Hu et al., 2019; Michael Thus if the correction protocol is applied after a time
 52

4-qubit code Simplest binomial code

1 √1 (|0i
Code word |0L i √
2
(|0000i + |1111i) 2
+ |4i)
1
Code word |1L i √ (|1100i
2
+ |0011i) |2i
Mean excitation number n̄ 2 2
4
Hilbert space dimension 2 = 16 {0, 1, 2, 3, 4} = 5
Number of correctable errors ˆ σ− , σ− , σ− , σ− } = 5
{I, ˆ a} = 2
{I,
1 2 3 4
Stabilizers Ŝ1 = Ẑ1 Ẑ2 , Ŝ2 = Ẑ3 Ẑ4 , Ŝ3 = X̂1 X̂2 X̂3 X̂4 P̂ = (−1)n̂
Number of Stabilizers 3 1
Approximate QEC? Yes, 1st order in γt Yes, 1st order in κt

TABLE I Comparison of qubit and bosonic codes for amplitude damping. γ and κ are respectively the qubit and oscillator
energy relaxation rates.

interval δt, the probability of an uncorrectable error is single mode, there is only a single error, namely photon
reduced from O(κ δt) to O((κ δt)2 ), where κ is the cavity loss (or no loss), and it can be detected by measuring a
energy decay rate. single stabilizer, the photon number parity. It turns out
To better understand the simplicity and efficiency ad- that, unlike in ordinary quantum optics, photon number
vantages of bosonic QEC codes, it is instructive to do a parity is relatively easy to measure in circuit QED with
head-to-head comparison of the simplest binomial code high fidelity and minimal state demolition (Ofek et al.,
to the simplest qubit code for amplitude damping. The 2016; Sun et al., 2014). It is for all these reasons that,
smallest qubit code able to protect logical information unlike the four-qubit code, the bosonic code Eq. (154)
against a general single-qubit error requires five qubits has already been demonstrated experimentally to (very
(Bennett et al., 1996; Knill et al., 2001; Laflamme et al., nearly) reach the break-even point for QEC (Hu et al.,
1996). However, the specific case of the qubit amplitude- 2019; Ma et al., 2020). Generalizations of this code to pro-
damping channel can be corrected to first order against tect against more than a single photon loss event, as well
single-qubit errors using a 4-qubit code (Leung et al., 1997) as photon gain and dephasing, are described in Michael
that, like the binomial code, satisfies the Knill-Laflamme et al. (2016a).
conditions (Knill and Laflamme, 1997) to lowest order Operation slightly exceeding break-even has been re-
and whose two logical codewords are ported by (Ofek et al., 2016) with cat-state bosonic en-
coding which we describe now. In the encoding used in
1 that experiment, each logical code word is a superposition
|0L i = √ (|0000i + |1111i) , (155a)
2 of four coherent states referred to as a four-component
1 cat code (Mirrahimi et al., 2014):
|1L i = √ (|1100i + |0011i) . (155b)
2
|0L i = N0 (|αi + |iαi + | − αi + | − iαi) , (156a)
This four-qubit amplitude damping code and the single- |1L i = N1 (|αi − |iαi + | − αi − | − iαi) , (156b)
mode binomial bosonic code for amplitude damping are
compared in Table VII.C. Note that, just as in the bino- where Ni are normalization constants, with N0 ' N1 for
mial code, both codewords have mean excitation number large |α|. The Wigner functions for the |0L i codeword is
equal to two and so are equally likely to suffer an excita- shown in Fig. 31(a) for α = 4. The relationship between
tion loss. The logical qubit of Eq. (155) lives in a Hilbert this encoding and the simple code in Eq. (154) can be seen
space of dimension 24 = 16 and has four different physical by writing Eq. (156) using the expression Eq. (85) for |αi
sites at which the damping error can occur. Counting in terms of Fock states. One immediately finds that |0L i
the case of no errors, there are a total of five different only has support on Fock states |4ni with n = 0, 1, . . . ,
error states which requires measurement of three distinct while |1L i has support on Fock states |4n + 2i, again
error syndromes Ẑ1 Ẑ2 , Ẑ3 Ẑ4 , and X̂1 X̂2 X̂3 X̂4 to diag- for n = 0, 1, . . . . It follows that the two codewords are
nose (where P̂i refers to Pauli operator P̂ acing on qubit mapped onto orthogonal states under the action of â,
i). The required weight-two and weight-four operators just as the binomial code of Eq. (154). Moreover, the
have to date not been easy to measure in a highly QND average photon number n̄ is approximately equal for the
manner and with high fidelity, but some progress has been two logical states in the limit of large |α|. The protection
made towards this goal (Chow et al., 2014, 2015; Corcoles offered by this encoding is thus similar to that of the
et al., 2015; Ristè et al., 2015; Takita et al., 2016). In binomial code in Eq. (154). In fact, these two encodings
contrast, the simple bosonic code in Eq. (154) requires belong to a larger class of codes characterized by rotation
only the lowest five states out of the (formally infinite) symmetries in phase space (Grimsmo et al., 2020).
oscillator Hilbert space. Moreover, since there is only a We end this section by discussing an encoding that is
 53

5.0 W (β) overhead necessary for fault-tolerant quantum computa-

0.3 tion (Guillaud and Mirrahimi, 2019; Puri et al., 2019).
2.5
0.1 The basic intuition behind this statement is that the qubit
Im β

0.0 defined by Eq. (157) can be used in an error correcting

0.0
code tailored to predominantly correct the most likely
−2.5 -0.1 error (here, phase flips) rather than devoting ressources
−5.0 -0.3 to correcting both amplitude and phase errors (Tuckett
−5 0 5 −5 0 5
et al., 2018, 2019a,b).
Re β Re β Another bosonic encoding that was recently demon-
strated in circuit QED is the Gottesman-Kitaev-Preskill
FIG. 31 Plot of Wigner function W (β) obtained numerically
for four-component (a) and two-component (b) cat states with (GKP) code (Campagne-Ibarcq et al., 2019). This demon-
α = 4. Red is positive and blue is negative. stration is the first QEC experiment able to correct all
logical errors and it came close to reaching the break-
even point. While all the bosonic codes described above
even simpler than Eq. (156), sometimes referred to as a are based on codewords that obey rotation symmetry in
two-component cat code. In this case, the codewords are phase space, the GKP code is instead based on translation
defined simply as |+L i = N0 (|αi + | − αi) and |−L i = symmetry. We will not describe the GKP encoding in
N1 (|αi − | − αi (Cochrane et al., 1999; Gilchrist et al., more detail here, but refer the reader to the review by
2004; Mirrahimi et al., 2014; Puri et al., 2017; Ralph Terhal et al. (2020).
et al., 2003). The Wigner function for |+L i is shown
in Fig. 31(b). The choice to define the above codewords
in the logical X̂L basis instead of the ẐL basis is, of course,
just a convention, but turns out to be convenient for this
particular cat code. In contrast to Eqs. (154) and (156), VIII. QUANTUM OPTICS ON A CHIP
these two states are not mapped to two orthogonal states
under the action of â. To understand this encoding, it is The strong light-matter interaction realized in circuit
useful to consider the logical ẐL basis states in the limit QED together with the flexibility allowed in designing and
of large |α| operating superconducting quantum circuits has opened
1 2
the possibility to explore the rich physics of quantum
|0L i = √ (|+L i + |−L i) = |αi + O(e−2|α| ), (157a) optics at microwave frequencies in circuits. As discussed
2
previously it has, for example, made possible the clear
1 2
|1L i = √ (|+L i − |−L i) = | − αi + O(e−2|α| ). (157b) observation of vacuum Rabi splitting, of photon-number
2 splitting in the strong-dispersive regime, as well as of
As is made clear by the second equality, for large enough signatures of ultrastrong light-matter coupling. The new
|α| these logical states are very close to coherent states of parameter regimes that can be achieved in circuit QED
the same amplitude but opposite phase. The action of â have also made it possible to test some of the theoret-
is thus, to a very good approximation, a phase flip since ical predictions from the early days of quantum optics
â|0L /1L i ∼ ±|0L /1L i. and to explore new research avenues. A first indication
The advantage of this encoding is that, while photon that circuit QED is an ideal playground for these ideas
loss leads to phase flips, the bit-flip rate is exponentially is the strong Kerr nonlinearity relative to the decay rate,
small with |α|. This can be immediately understood from K/κ, that can readily be achieved in circuits. Indeed,
the golden rule whose relevant matrix element for bit flips from the point of view of quantum optics, a transmon
2
is h1L |â|0L i ∼ h−α|â|αi = αe−2|α| . In other words, if is a Kerr nonlinear oscillator that is so nonlinear that
the qubit is encoded in a coherent state with many pho- it exhibits photon blockade. Given the very high Q fac-
tons, losing one simply does not do much. This is akin to tors that can be achieved in 3D superconducting cavities,
the redundancy required for quantum error correction. As such levels of nonlinearity can also readily be obtained
a result, the bit-flip rate (1/T1 ) decreases exponentially in microwave resonators by using transmons or other
with |α|2 while the phase flip rate increases only linearly Josephson junction-based circuits to induce nonlinearity
with |α|2 . The crucial point is that the bias between bit in electromagnetic modes.
and phase flip error rates increases exponentially with Many of the links between circuit QED and quantum
α, which has been verified experimentally (Grimm et al., optics have already been highlighted in this review. In
2019; Lescanne et al., 2019b). While the logical states this section, we continue this discussion by presenting
Eq. (157) do not allow for recovery from photon-loss er- some further examples. The reader interested in learning
rors, the strong asymmetry between different types of more about quantum optics at microwave frequencies can
errors can be exploited to significantly reduce the qubit consult the review article by Gu et al. (2017).
 54

|0 + |3 + |6 |0 + eiπ/8 |3 + |6 |0 + eiπ/4 |3 + |6 |0 + e3iπ/8 |3 + |6 |0 + i |3 + |6

+2/π
2
1
0 +1/π
–1
–2

W(α)
Im(α)
0
2
1
0 –1/π

–1
–2
–2/π
–2 –1 0 1 2 –2 –1 0 1 2 –2 –1 0 1 2 –2 –1 0 1 2 –2 –1 0 1 2
Re(α)

FIG. 32 Wigner function of the intra-cavity field for Fock state superpositions |0i + eiϕ |3i + |6i for five values of the phase ϕ,
see panel titles. Top row: theory, Bottom row: experimental data. Figure adapted from Hofheinz et al. (2009).

A. Intra-cavity fields parable to the Fock state lifetime, limiting the Fock states
which can be reached and the fidelity of the resulting
Because superconducting qubits can rapidly be tuned states.
over a wide frequency range, it is possible to bring them Taking advantage of the very large χ/κ which can be
in and out of resonance with a cavity mode on a time reached in 3D cavities, an alternative to create such states
scale which is fast with respect to 1/g, the inverse of the is to cause qubit transitions conditioned on the Fock state
qubit-cavity coupling strength. For all practical purposes, of the cavity. Together with cavity displacements, these
this is equivalent to the thought experiment of moving photon-number dependent qubit transitions can be used
an atom in and out of the cavity in cavity QED. An to prepare arbitrary cavity states (Heeres et al., 2015;
experiment by Hofheinz et al. (2008) took advantage Krastanov et al., 2015). Combining these ideas with
of this possibility to prepare the cavity in Fock states numerical optimal control has allowed Heeres et al. (2017)
up to |n = 6i. With the qubit and the cavity in their to synthesize cavity states with high fidelity such as Fock
respective ground states and the two systems largely states up to |n = 6i and four-legged cat states.
detuned, their approach is to first π-pulse the qubit to The long photon lifetime that is possible in 3D super-
its excited state. The qubit frequency is then suddenly conducting cavities together with the possibility to realize
brought in resonance with the cavity for a time 1/2g such a single-photon Kerr nonlinearity which overwhelms the
as to swap the qubit excitation to a cavity photon as the cavity decay, K/κ > 1, has enabled a number of similar
system evolves under the Jaynes-Cummings Hamiltonian experiments such as the observation of collapse and re-
Eq. (34). The interaction is then effectively stopped by vival of a coherent state in a Kerr medium (Kirchmair
moving the qubits to its original frequency, after which the et al., 2013) and the preparation of cat states with nearly
cycle is repeated until n excitations have been swapped in 30 photons (Vlastakis et al., 2013). Another striking ex-
this way. Crucially, because the swap frequency between
√ ample is the experimental encoding of qubits in oscillator
the states |e, n − 1i and |g, ni is proportional to n, the states already discussed in Sec. VII.C.
time during which qubit and cavity are kept in resonance
must
√ be adjusted accordingly at each cycle. The same
n dependence is then used to probe the cavity state B. Quantum-limited amplification
using the qubit as a probe (Brune et al., 1996; Hofheinz
et al., 2008). Driven by the need for fast, high-fidelity single-shot
Building on this technique and using a protocol pro- readout of superconducting qubits, superconducting low-
posed by Law and Eberly (1996) for cavity QED, the noise linear microwave amplifiers are a subject of intense
same authors have demonstrated the preparation of ar- research. There are two broad classes of linear amplifiers.
bitrary states of the cavity field and characterized these First, phase-preserving amplifiers that amplify equiva-
states by measuring the cavity Wigner function (Hofheinz lently both quadratures of the signal. Quantum mechanics
et al., 2009). Figure 32 shows the result of this Wigner imposes that these amplifiers add a minimum of half a pho-
tomography for superpositions involving up to six cavity ton of noise to the input signal (Caves, 1982; Caves et al.,
photons (top row: theory, bottom row: data). As noted 2012; Clerk et al., 2010). Second, phase-sensitive ampli-
in Hofheinz et al. (2008), a downside of this sequential fiers which amplify one quadrature of the signal while
method is that the preparation time rapidly becomes com- squeezing the orthogonal quadrature. This type of ampli-
 55

fier can in principle operate without adding noise (Caves, Using appropriate filtering in the frequency domain, single-
1982; Clerk et al., 2010). Amplifiers adding the minimum mode parametric amplifiers can be operated in a phase-
amount of noise allowed by quantum mechanics, phase sensitive mode, when detecting the emitted radiation over
preserving or not, are referred to as quantum-limited the full bandwidth of the physical mode, see e.g. Eichler
amplifiers. We note that, in practice, phase sensitive et al. (2011a). This is also called the degenerate mode
amplifiers are useful if the quadrature containing the rele- of operation. Alternatively, the same single-oscillator-
vant information is known in advance, a condition that mode amplifier can be operated in the phase-preserving
is realized when trying to distinguish between two coher- mode, when separating frequency components above and
ent states in the dispersive qubit readout discussed in below the pump in the experiment, e.g. by using ap-
Sec. V.C. propriately chosen narrow-band filters, see for example
While much of the development of near-quantum- Eichler et al. (2011a). Parametric amplifiers with two or
limited amplifiers has been motivated by the need to multiple physical modes are also frequently put to use
improve qubit readout, Josephson junction based ampli- (Roy and Devoret, 2016) and can be operated both in
fiers have been theoretically investigated (Yurke, 1987) the phase-sensitive and phase-preserving modes, e.g. in
and experimentally demonstrated as early as the late degenerate or non-degenerate mode of operation, as for
80‘s (Yurke et al., 1989, 1988). These amplifiers have now example demonstrated in Eichler et al. (2014).
found applications in a broad range of contexts. In their Important parameters which different approaches for
simplest form, such an amplifier is realized as a driven implementing JPAs aim at optimizing include amplifier
oscillator mode rendered weakly nonlinear by incorporat- gain, bandwidth and dynamic range. The latter refers
ing a Josephson junction and are generically known as a to the range of power over which the amplifier acts lin-
Josephson parametric amplifier (JPA). early, i.e. powers at which the amplifier output is linearly
For weak nonlinearity, the Hamiltonian of a driven related to its input. Above a certain input power level,
nonlinear oscillator is well approximated by the correction terms in Eq. (159) resulting from the junc-
tion nonlinearity can no longer be ignored and lead to
K †2 2 saturation of the gain (Abdo et al., 2013; Boutin et al.,
H = ω0 ↠â + â â + p (↠e−iωp t + âeiωp t ), (158)
2 2017; Kochetov and Fedorov, 2015; Liu et al., 2017; Planat
et al., 2019). For this reason, while transmon qubits are
where ω0 is the system frequency, K the negative Kerr operated in a regime where the single-photon Kerr nonlin-
nonlinearity, and p and ωp are the pump amplitude earity is large and overwhelms the decay rate, JPAs are
and frequency, respectively. The physics of the JPA is operated in a very different regime with |K|/κ ∼ 10−2 or
best revealed by applying a displacement transformation smaller.
D̂† (α)âD̂(α) = a + α to H with α chosen to cancel the
An approach to increase the dynamic range of JPAs
pump term. Doing so leads to the transformed Hamilto-
is to replace the Josephson junction of energy EJ by an
nian
array of M junctions, each of energy M EJ (Castellanos-
1
Beltran et al., 2008; Castellanos-Beltran and Lehnert,
HJPA = δ↠â + 2 â†2 + ∗2 â2 + Hcorr , (159) 2007; Eichler and Wallraff, 2014). Because the voltage
2
drop is now distributed over the array, the bias on any
where δ = ω0 + 2|α|2 K − ωp is the effective detuning, single junction is M times smaller and therefore the ef-
2 = α2 K, and are Hcorr correction terms which can fective Kerr nonlinearity of the device is reduced from
be dropped for weak enough pump amplitude and Kerr K to K/M 2 . As a result, nonlinear effects kick-in only
nonlinearity, i.e. when κ is large in comparison to K and at increased input signal powers leading to an increased
thus the drive does not populate the mode enough for dynamic range. Importantly, this can be done without
higher-order nonlinearity to become important (Boutin degrading the amplifier’s bandwidth (Eichler and Wall-
et al., 2017). The second term, of amplitude 2 , is a raff, 2014). Typical values are ∼ 50 MHz bandwidth with
two-photon pump which is the generator of quadrature ∼ −117 dBm saturation power for ∼ 20 dB gain (Planat
squeezing. Depending on the size of the measurement et al., 2019). Impedance engineering can be used to im-
bandwidth, this leads to phase preserving or sensitive prove these numbers further (Roy et al., 2015).
amplification when operating the pdevice close to but under Because the JPA is based on a localized oscillator mode,
the parametric threshold 2 < δ 2 + (κ/2)2 , with κ the the product of its gain and bandwidth is approximately
device’s single-photon loss rate (Wustmann and Shumeiko, constant. Therefore, increase in one must be done at the
2013). Rather than driving the nonlinear oscillator as in expense of the other (Clerk et al., 2010; Eichler and Wall-
Eq. (158), an alternative approach to arrive at HJPA is raff, 2014). As a result, it has proven difficult to design
to replace the junction by a SQUID and to apply a flux JPAs with enough bandwidth and dynamic range to si-
modulation at 2ω0 (Yamamoto et al., 2008). multaneously measure more than a few transmons (Jeffrey
Equation (159) is the Hamiltonian for a parametric et al., 2014).
amplifier working with a single physical oscillator mode. To avoid the constant gain-bandwidth product which
 56

results from relying on a resonant mode, a drastically noise to quantify and subtract from the data the noise
different strategy, known as the Josephson traveling-wave introduced by the measurement chain. In this way, it is
parametric amplifier (JTWPA), is to use an open nonlin- possible to extract, for example, first- and second-order
ear medium in which the signal co-propagates with the coherence functions of the microwave field. Remarkably,
pump tone. While in a JPA the signal interacts with the with enough averaging, this approach does not require
nonlinearity for a long time due to the finite Q of the cir- quantum-limited amplifiers, although the number of re-
cuit, in the JTWPA the long interaction time is rather a quired measurement runs is drastically reduced when such
result of the long propagation length of the signal through amplifiers are used when compared to HEMT amplifiers.
the nonlinear medium (O’Brien et al., 2014). In practice, This approach was used to measure second-order coher-
JTWPA are realized with a metamaterial transmission ence functions, G2 (t, t + τ ) = h↠(t)↠(t + τ )â(t + τ )â(t)i,
line whose center conductor is made from thousands of in the first demonstration of antibunching of a pulsed
Josephson junctions in series (Macklin et al., 2015). This single microwave-frequency photon source (Bozyigit et al.,
device does not have a fixed gain-bandwidth product and 2011). The same technique also enabled the observation
has been demonstrated to have 20 dB over as much as of resonant photon blockade at microwave frequencies
3 GHz bandwidth while operating close to the quantum (Lang et al., 2011) and, using two single photon sources
limit (Macklin et al., 2015; Planat et al., 2020; White at the input of a microwave beam splitter, the indistin-
et al., 2015). Because every junction in the array can guishability of microwave photons was demonstrated in a
be made only very weakly nonlinear, the JTWPA also Hong-Ou-Mandel correlation function measurement (Lang
offers large enough dynamic range for rapid multiplexed et al., 2013). Moreover, a similar approach was used to
simultaneously readout of multiple qubits (Heinsoo et al., characterize the blackbody radiation emitted by a 50 Ω
2018). load resistor (Mariantoni et al., 2010).
Building on these results, it is also possible to recon-
struct the quantum state of itinerant microwave fields
C. Propagating fields and their characterization from measurement of the fields moments. This technique
relies on interleaving two types of measurements: mea-
1. Itinerant single and multi-photon states surements on the state of interest and ones in which the
field is left in the vacuum as a reference to subtract away
In addition to using qubits to prepare and characterize the measurement chain noise (Eichler et al., 2011b). In
quantum states of intra-cavity fields, it is also possible to this way, the Wigner function of arbitrary superpositions
take advantage of the strong nonlinearity provided by a of vacuum and one-photon Fock states have been recon-
qubit to prepare states of propagating fields at the output structed (Eichler et al., 2011b; Kono et al., 2018). This
of a cavity. This can be done, for example, in a cavity technique was extended to propagating modes containing
with relatively large decay rate κ by tuning a qubit into multiple photons (Eichler et al., 2012a). Similarly, entan-
and out of resonance with the cavity (Bozyigit et al., 2011) glement between a (stationary) qubit and a propagating
or by applying appropriatly chosen drive fields (Houck mode was quantified in this approach with joint state to-
et al., 2007). Alternatively, it is also possible to change mography (Eichler et al., 2012a,b). Quadrature-histogram
the cavity decay rate in time to create single-photon analysis also enabled, for example, the measurement of
states (Sete et al., 2013; Yin et al., 2013). correlations between radiation fields (Flurin et al., 2015),
The first on-chip single-photon source in the microwave and the observation of entanglement of itinerant photon
regime was realized with a dispersively coupled qubit en- pairs in waveguide QED (Kannan et al., 2020).
gineered such that the Purcell decay rate γκ dominates
the qubit’s intrinsic non-radiative decay rate γ1 (Houck
et al., 2007). In this situation, exciting the qubit leads 2. Squeezed microwave fields
to rapid qubit decay by photon emission. In the absence
of single-photon detectors working at microwave frequen- Operated in the phase-sensitive mode, quantum-limited
cies, the presence of a photon was observed by using a amplifiers are sources of squeezed radiation. Indeed, for
nonlinear circuit element (a diode) whose output signal is δ = 0 and ignoring the correction terms, the JPA Hamil-
proportional to the square of the electric field, ∝ (↠+ â)2 , tonian of Eq. (159) is the generator of the squeezing
and therefore indicative of the average photon number, transformation
h↠âi, in repeated measurements. 1
S(ζ) = e 2 ζ
∗ 2
â − 12 ζâ†2
, (160)
Rather than relying on direct power measurements,
techniques have also been developed to reconstruct arbi- which takes vacuum to squeezed vacuum, |ζi = S(ζ)|0i.
trary correlation functions of the cavity output field from In this expression, ζ = reiθ with r the squeezing parame-
the measurement records of the field qudratures (Men- ter and θ the squeezing angle. As illustrated in Fig. 33(a),
zel et al., 2010; da Silva et al., 2010). These approaches the action of S(ζ) on vacuum is to ‘squeeze’ one quadra-
rely on multiple detection channels with uncorrelated ture of the field at the expense of ‘anti-squeezing’ the
 57

(a) 2 (b)
θ/2 10
preparations of the same squeezed state, together with
1 maximum likelihood techniques, was used to reconstruct
e−r

S (dB)
the Wigner function of the propagating microwave field.
Im α

0 0
Moreover, the photon number distribution of a squeezed
−1 field was measured using a qubit in the strong dispersive
er −10
regime (Kono et al., 2017). As is clear from the form of
−2
−2 0 2 0.0 0.5 1.0 the squeezing transformation S(ζ), squeezed vacuum is
Re α φ/2π composed of a superposition of only even photon numbers
(c) (Schleich and Wheeler, 1987), something which Kono et al.
Transmon in (2017) confirmed in experiments.
3D cavity
Thanks to the new parameter regimes that can be
achieved in circuit QED, it is possible to experimentally
Pump
test some long-standing theoretical predictions of quan-
tum optics involving squeezed radiation. For example, in
Squeezed
vacuum
the mid-80’s theorists predicted how dephasing and reso-
Circulator
nance fluorescence of an atom would be modified in the
Josephson parametric presence of squeezed radiation (Carmichael et al., 1987;
amplifier Gardiner, 1986). Experimentally testing these ideas in the
context of traditional quantum optics with atomic systems,
FIG. 33 (a) Wigner function of a squeezed vacuum state however, represents a formidable challenge (Carmichael,
S(ζ)|0i with r = 0.75 and θ = π/2. The white countour
line is an ellipse of semi-minor axis e−r /2 and semi-major
2019; Turchette et al., 1998). The situation is different
axis er /2. (b) Squeezing level versus φ for r = 0.5, 1.0, 1.5 in circuits where squeezed radiation can easily be guided
and θ = π. The horizontal line corresponds to vacuum noise from the source of squeezing to the qubit playing the role
2
∆Xvac = 1/2. (c) Experimental setup used by Murch et al. of artificial atom. Moreover, the reduced dimensionality
(2013b) to prepare a squeezed vacuum state using a Josephson in circuits compared to free-space atomic experiments
parametric amplifier and to send, via a circulator (gray box), limits the number of modes that are involved, such that
this state to a transmon qubit in a 3D cavity (blue box). Panel the artificial atom can be engineered so as to preferentially
c) is adapted from Murch et al. (2013b).
interact with a squeezed radiation field.
Taking advantage of the possibilities offered by circuit
orthogonal quadrature while leaving the total area in QED, Murch et al. (2013a) confirmed the prediction that
phase space unchanged. As a result, squeezed states, like squeezed radiation can inhibit phase decay of an (artifical)
coherent states, saturate the Heisenberg inequality. atom (Gardiner, 1986). In this experiment, the role of the
This can be seen more clearly from the variance of the two-level atom was played by the hybridized cavity-qubit
quadrature operator X̂φ which takes the form state {|g, 0i, |e, 0i}. Moreover, squeezing was produced
by a JPA over a bandwidth much larger than the natural
1  2r 2 
linewitdh of the two-level system, see Fig. 33(c). Accord-
∆Xφ2 = e sin φ̃ + e−2r cos2 φ̃ , (161)
2 ing to theory, quantum noise below the vacuum level along
the squeezed quadrature leads to a reduction of dephas-
where we have defined φ̃ = φ − θ/2. In experiments, the
ing. Conversely, along the anti-squeezed quadrature, the
squeezing level is often reported in dB computed using
enhanced fluctuations lead to increased dephasing. For
the expression
the artificial atom, this results in different time scales for
∆Xφ2 dephasing along orthogonal axis of the Bloch sphere. In
S = 10 log10 2
. (162) the experiment, phase decay inhibition along the squeezed
∆Xvac
quadrature was such that the associated dephasing time
Figure 33(b) shows this quantity as a function of φ. It increased beyond the usual vacuum limit of 2T1 . By mea-
reaches its minimal value e−2r /2 at φ = [θ + (2n + 1)π]/2 suring the dynamics of the two-level atom, it was moreover
where the variance ∆Xφ2 dips below the vacuum noise possible to reconstruct the Wigner distribution of the itin-
2 erant squeezed state produced by the JPA. Using a similar
level ∆Xvac = 1/2 (horizontal line).
Squeezing in Josephson devices was observed already in setup, Toyli et al. (2016) studied resonance fluorescence
the late 80’s (Movshovich et al., 1990; Yurke et al., 1989, in the presence of squeezed vacuum and found excellent
1988), experiments that have been revisited with the devel- agreement with theoretical predictions (Carmichael et al.,
opment of near quantum-limited amplifiers (Castellanos- 1987). In this way, it was possible to infer the level of
Beltran et al., 2008; Zhong et al., 2013). Quantum state squeezing (3.1 dB below vacuum) at the input of the
tomography of an itinerant squeezed state at the output cavity.
of a JPA was reported by Mallet et al. (2011). There, The discussion has so far been limited to squeezing of a
homodyne detection with different LO phases on multiple single mode. It is also possible to squeeze a pair of modes,
 58

which is often referred to as two-mode squeezing. Labeling Crucially, the cavity pull associated with odd-parity states
the modes as â1 and â2 , the corresponding squeezing {|01i, |10i} is zero while it is ±2χ for the even-parity states
transformation reads {|00i, |11i}. As a result, for χ  κ, a tone at the bare
cavity frequency leads to a large cavity field displacement
1 ∗
â1 â2 − 12 ζâ†1 â†2
S12 (ζ) = e 2 ζ . (163) for the even-parity subspace. On the other hand, the
displacement is small or negligible for the odd-parity sub-
Acting on vacuum, S12 generates a two-mode squeezed space. Starting with a uniform unentangled superposition
state which is an entangled state of modes â1 and â2 . As of the states of the qubits, homodyne detection therefore
a result, in isolation, the state of one of the two entangled stochastically collapses the system to one of these sub-
modes appears to be in a thermal state where the role of spaces thereby preparing an entangled state of the two
the Boltzmann factor exp(−β~ωi ), with ωi=1,2 the mode qubits (Lalumière et al., 2010), an idea that was realized
frequency, is played by tanh2 r (Barnett and Radmore, experimentally (Ristè et al., 2013).
2002). In this case, correlations and therefore squeezing is
The same concept was used by Roch et al. (2014) to
revealed when considering joint quadratures of the form
entangle two transmon qubits coupled to two 3D cavities
X̂1 ± X̂2 and P̂1 ± P̂2 , rather than the quadratures of a
separated by more than a meter of coaxial cable. There,
single mode as in Fig. 33(a). In Josephson-based devices,
the measurement tone transmitted through the first cav-
two-mode squeezing can be produced using nondegenerate
ity is sent to the second cavity, only after which it is
parametric amplifiers of different types (Roy and Devoret,
measured by homodyne detection. In this experiment,
2016). Over 12 dB of squeezing below vacuum level be-
losses between the two cavities – mainly due to the pres-
tween modes of different frequencies, often referred to as
ence of a circulator preventing any reflection from the
signal and idler in this context, has been reported (Eich-
second cavity back to the first cavity – as well as finite
ler et al., 2014). Other experiments have demonstrated
detection efficiency was the main limit to the acheivable
two-mode squeezing in two different spatial modes, i.e. en-
concurrence, a measure of entanglement, to 0.35.
tangled signals propagating along different transmission
While the above protocol probabilistically entangles a
lines (Bergeal et al., 2012; Flurin et al., 2012).
pair of qubits, a more powerful but also more experimen-
tally challenging approach allows, in principle, to realize
D. Remote Entanglement Generation this in a fully deterministic fashion (Cirac et al., 1997).
Developed in the context of cavity QED, this scheme
Several approaches to entangle nearby qubits have been relies on mapping the state of an atom strongly coupled
discussed in Sec. VII. In some instances it can, however, to a cavity to a propagating photon. By choosing its wave
be useful to prepare entangled states of qubits separated packet to be time-symmetric, the photon is absorbed with
by larger distances. Together with protocols such as unit probability by a second cavity also containing an
quantum teleportation, entanglement between distant atom. In this way, it is possible to exchange a quantum
quantum nodes can be the basis of a ‘quantum internet’ state between the two cavities. Importantly, this protocol
(Kimble, 2008). Because optical photons can travel for rel- relies on having a unidirectional channel between the cav-
atively long distances in room temperature optical fiber ities such that no signal can propagate from the second to
while preserving their quantum coherence, this vision the first cavity. At microwave frequencies, this is achieved
appears easier to realize at optical than at microwave fre- by inserting a circulator between the cavities. By first
quencies. Nevertheless, given that superconducting cables entangling the emitter qubit to a partner qubit located
at millikelvin temperatures have similar losses per meter in the same cavity, the quantum-state transfer protocol
as optical fibers (Kurpiers et al., 2017), there is no reason can be used to entangle the two nodes.
to believe that complex networks of superconductor-based Variations on this more direct approach to entangle
quantum nodes cannot be realized. One application of remote nodes have been implemented in circuit QED (Ax-
this type of network is a modular quantum computer line et al., 2018; Campagne-Ibarcq et al., 2018; Kurpiers
architecture where the nodes are relatively small-scale et al., 2018). All three experiments rely on producing
error-corrected quantum computers connected by quan- time-symmetric propagating photons by using the interac-
tum links (Chou et al., 2018; Monroe et al., 2014). tion between a transmon qubit and cavity mode. Multiple
One approach to entangle qubits fabricated in distant approaches to shape and catch propagating photons have
cavities relies on entanglement by measurement, which been developed in circuit QED. For example, Wenner et al.
is easy to understand in the case of two qubits coupled (2014) used a transmission-line resonator with a tunable
to the same cavity. Assuming the qubits to have the input port to catch a shaped microwave pulse with over
same dispersive shift χ due to coupling to the cavity, the 99% probability. Time-reversal-symmetric photons have
dispersive Hamiltonian in a doubly rotating frame takes been created by Srinivasan et al. (2014) using 3-island
the form transmon qubits (Gambetta et al., 2011a; Srinivasan et al.,
2011) in which the coupling to a microwave resonator is
H = χ(σ̂z1 + σ̂z2 )↠â. (164) controlled in time so as to shape the mode function of
 59

spontaneously emitted photons. In a similar fashion, as a scatterer. This extinction results from destructive
shaped single photons can be generated by modulating interference of the light beam with the collinearly emitted
the boundary condition of a semi-infinite transmission line radiation from the scatterer. Ideally, this results in 100%
using a SQUID (Forn-Dı́az et al., 2017) which effectively reflection. In practice, because the scatterer emits in all
controls the spontaneous emission rate of a qubit coupled directions, there is poor mode matching with the focused
to the line and emitting the photon. beam and reflection of ∼ 10% is observed with a single
Alternatively, the remote entanglement generation ex- atom (Tey et al., 2008) and ∼ 30% with a single molecule
periment of Kurpiers et al. (2018) rather relies on a (Maser et al., 2016).
microwave-induced amplitude- and phase-tunable cou- Mode matching can, however, be made to be close to
pling between the qubit-resonator |f 0i and |g1i states, ideal with electromagnetic fields in 1D superconducting
akin to the fg-ge gate already mentioned in Sec. VII.B.3 transmission lines and superconducting artificial atoms
(Zeytinoğlu et al., 2015). Exciting the qubit to its |f i where the artificial atoms can be engineered to essentially
state followed by a π-pulse on the f 0 − g1 transition only emit in the forward and backward directions along
transfers the qubit excitation to a single resonator photon the line (Shen and Fan, 2005). In the first realization of
which is emitted as a propagating photon. This single- this idea in superconducting quantum circuits, Astafiev
photon wave packet can be shaped to be time-symmetric et al. (2010) observed extinction of the transmitted signal
by tailoring the envelope of the f 0 − g1 pulse (Pechal by as much as 94% by coupling a single flux qubit to a
et al., 2014). By inducing the reverse process with a time- superconducting transmission line. Experiments with a
reversed pulse on a second resonator also containing a transmon qubit have seen extinction as large as 99.6%
transmon, the itinerant photon is absorbed by this second (Hoi et al., 2011). Pure dephasing and non-radiative de-
transmon. In this way, an arbitrary quantum state can cay into other modes than the transmission line are the
be transferred with a probability of 98.1% between the cause of the small departure from ideal behavior in these
two cavities separated by 0.9 m of coaxial line bisected by experiments. Nevertheless, the large observed extinction
a circulator (Kurpiers et al., 2018).√ By rather preparing is a clear signature that radiative decay in the transmis-
the emitter qubit in a (|ei + |f i)/ 2 superposition, the sion line γr (i.e. Purcell decay) overwhelms non-radiative
same protocol deterministically prepares an entangled decay γnr . In short, in this cavity-free system referred to
state of the two transmons with a fidelity of 78.9% at a as waveguide QED, γr /γnr  1 is the appropriate defi-
rate of 50 kHz (Kurpiers et al., 2018). The experiments nition of strong coupling and is associated with a clear
of Axline et al. (2018) and Campagne-Ibarcq et al. (2018) experimental signature: the extinction of transmission by
reported similar Bell-state fidelities using different ap- a single scatterer.
proaches to prepare time-symmetric propagating photons Despite its apparent simplicity, waveguide QED is a rich
(Pfaff et al., 2017). The fidelity reported by the three toolbox with which a number of physical phenomena have
experiments suffered from the presence of a circulator been investigated (Roy et al., 2017). This includes Autler-
bisecting the nearly one meter-long coaxial cable sepa- Townes splitting (Abdumalikov et al., 2010), single-photon
rating the two nodes. Replacing the lossy commercial routing (Hoi et al., 2011), the generation of propagating
circulator by an on-chip quantum-limited version could nonclassical microwave states (Hoi et al., 2012), as well
improve the fidelity (Chapman et al., 2017; Kamal et al., as large cross-Kerr phase shifts at the single-photon level
2011; Metelmann and Clerk, 2015). By taking advantage (Hoi et al., 2013).
of the multimode nature of a meter long transmission line, In another experiment, Hoi et al. (2015) studied the
it was also possible to deterministically entangle remote radiative decay of an artificial atom placed in front of a
qubits without the need of a circulator. In this way, a mirror, here formed by a short to ground of the waveg-
bidirectional communication channel between the nodes uide’s center conductor. In the presence of a weak drive
is established and deterministic Bell pair production with field applied to the waveguide, the atom relaxes by emit-
79.3% fidelity has been reported (Leung et al., 2019). ting a photon in both directions of the waveguide. The
radiation emitted towards the mirror, assumed here to
be on the left of the atom, is reflected back to interact
E. Waveguide QED again with the atom after having acquired a phase shift
θ = 2×2πl/λ+π, where l is the atom-mirror distance and
The bulk of this review is concerned with the strong λ the wavelength of the emitted radiation. The additional
coupling of artificial atoms to the confined electromagnetic phase factor of π accounts for the hard reflection at the
field of a cavity. Strong light-matter coupling is also mirror. Taking into account the resulting multiple round
possible in free space with an atom or large dipole-moment trips, this modifies the atomic radiative decay rate which
molecule by tightly confining an optical field in the vicinity takes the form γ(θ) = 2γr cos2 (θ/2) (Glaetzle et al., 2010;
of the atom or molecule (Schuller et al., 2010). A signature Hoi et al., 2015; Koshino and Nakamura, 2012).
of strong coupling in this setting is the extinction of the For l/λ = 1/2, the radiative decay rate vanishes corre-
transmitted light by the single atom or molecule acting sponding to destructive interference of the right-moving
 60

field and the left-moving field after multiple reflections between the dark state of the effective cavity and the qubit
on the mirror. In contrast, for l/λ = 1/4, these fields playing the role of atom were observed, confirming that
interfere constructively leading to enhanced radiative re- the strong-coupling regime of cavity QED was achieved.
laxation with γ(θ) = 2γr . The ratio l/λ can be modi-
fied by shorting the waveguide’s center conductor with
a SQUID. In this case, the flux threading the SQUID F. Single microwave photon detection
can be used to change the boundary condition seen by
the qubit, effectively changing the distance l (Sandberg The development of single-photon detectors at infrared,
et al., 2008). The experiment of Hoi et al. (2015) rather optical and ultraviolet frequencies has been crucial to
relied on flux-tuning of the qubit transition frequency, the field of quantum optics and in fundamental tests of
thereby changing λ. In this way, a modulation of the quantum physics (Eisaman et al., 2011; Hadfield, 2009).
qubit decay rate by a factor close to 10 was observed. A High-efficiency photon detectors are, for example, one of
similar experiment has been reported with a trapped ion the elements that allowed the loophole-free violation of
in front of a movable mirror (Eschner et al., 2001). Bell’s inequality (Giustina et al., 2015; Hensen et al., 2015;
Engineering vacuum fluctuations in this system has Shalm et al., 2015). Because microwave photons have
been pushed even further by creating microwave photonic orders of magnitude less energy than infrared, optical or
bandgaps in waveguides to which transmon qubits are ultraviolet photons, the realization of a photon detector
coupled (Liu and Houck, 2016; Mirhosseini et al., 2018). at microwave frequencies is more challenging. Yet, photon
For example, Mirhosseini et al. (2018) have coupled a detectors in that frequency range would find a number of
transmon qubit to a metamaterial formed by periodically applications, including in quantum information processing
loading the waveguide with lumped-element microwave (Kimble, 2008; Narla et al., 2016), for quantum radars
resonators. By tuning the transmon frequency in the band (Barzanjeh et al., 2015; Barzanjeh et al., 2019; Chang
gap where there is zero or only little density of states to et al., 2019), and for the detection of dark matter axions
accept photons emitted by the qubit, an increase by a (Lamoreaux et al., 2013).
factor of 24 of the qubit lifetime was observed. Non-destructive counting of microwave photons local-
An interpretation of the ‘atom in front of a mirror’ ized in a cavity has already been demonstrated experi-
experiments is that the atom interacts with its mirror mentally by using an (artificial) atom as a probe in the
image. Rather than using a boundary condition (i.e. a strong dispersive regime (Gleyzes et al., 2007; Schuster
mirror) to study the resulting constructive and destructing et al., 2007). Similar measurements have also been done
interferences and change in the radiative decay rate, it using a transmon qubit mediating interactions between
is also possible to couple a second atom to the same two cavities, one containing the photons to be measured
waveguide (Lalumière et al., 2013; van Loo et al., 2013). and a second acting as a probe (Johnson et al., 2010).
In this case, photons (real or virtual) emitted by one The detection of itinerant microwave photons remains,
atom can be absorbed by the second atom leading to however, more challenging. A number of theoretical pro-
interactions between the atoms separated by a distance posals have appeared (Fan et al., 2014; Helmer et al.,
2l. Similar to the case of a single atom in front of a 2009; Koshino et al., 2013, 2016; Kyriienko and Sorensen,
mirror, when the separation between the atoms is such 2016; Leppäkangas et al., 2018; Romero et al., 2009; Royer
that 2l/λ = 1/2, correlated decay of the pair of atoms et al., 2018; Sathyamoorthy et al., 2014; Wong and Vav-
at the enhanced rate 2γ1 is expected (Chang et al., 2012; ilov, 2017). One common challenge for these approaches
Lalumière et al., 2013) and experimentally observed (van based on absorbing itinerant photons in a localized mode
Loo et al., 2013). On the other hand, at a separation before detecting them can be linked to the quantum Zeno
of 2l/λ = 3/4, correlated decay is replaced by coherent effect. Indeed, continuous monitoring of the probe mode
energy exchange between the two atoms mediated by will prevent the photon from being absorbed in the first
virtual photons (Chang et al., 2012; Lalumière et al., place. Approaches to mitigate this problem have been
2013; van Loo et al., 2013). We note that the experiments introduced, including using an engineered, impedance
of van Loo et al. (2013) with transmon qubits agree with matched Λ-system used to deterministically capture the
a Markovian model of the interaction of the qubits with incoming photon (Koshino et al., 2016), and using the
the waveguide (Chang et al., 2012; Lalumière et al., 2013; bright and dark states of an ensemble of absorbers (Royer
Lehmberg, 1970). Deviations from these predictions are et al., 2018).
expected as the distance between the atoms increases Despite these challenges, first itinerant microwave pho-
(Zheng and Baranger, 2013). ton detectors have been achieved in the laboratory (Chen
Finally, following a proposal by Chang et al. (2012), et al., 2011; Inomata et al., 2016; Narla et al., 2016; Oel-
an experiment by Mirhosseini et al. (2019) used a pair sner et al., 2017), in some cases achieving photon detection
of transmon qubits to act as an effective cavity for a without destroying the photon in the process (Besse et al.,
third transmon qubit, all qubits being coupled to the 2018; Kono et al., 2018; Lescanne et al., 2019a). Notably,
same waveguide. In this way, vacuum Rabi oscillations a microwave photon counter was used to measure a su-
 61

perconducting qubit with a fidelity of 92% without using in mesoscopic systems (Schmidt and Koch, 2013). Few
a linear amplifier between the source and the detector resonator- and qubit-devices are also promising for analog
(Opremcak et al., 2018). Despite these advances, the real- quantum simulations. Examples are the exploration of
ization of a high-efficiency, large-bandwith, QND single a simple model of the light harvesting process in photo-
microwave photon detector remains a challenge for the synthetic complexes in a circuit QED device under the
field. influence of both coherent and incoherent drives (Potočnik
et al., 2018), and the analog simulation of dissipatively
stabilized strongly correlated quantum matter in a small
IX. OUTLOOK photon BoseHubbard lattice (Ma et al., 2019).
Because it is a versatile platform to interface quantum
Fifteen years after its introduction (Blais et al., 2004; devices with transition frequencies in the microwave do-
Wallraff et al., 2004), circuit QED is a leading architecture main to photons stored in superconducting resonators at
for quantum computing and an exceptional platform to ex- similar frequencies, the ideas of circuit QED are also now
plore the rich physics of quantum optics in new parameter used to couple to a wide variety of physical systems. An ex-
regimes. Circuit QED has, moreover, found applications ample of such hybrid quantum systems are semiconducter-
in numerous other fields of research as discussed in the based double quantum dots coupled to superconducting
body of the review and in the following. In closing this microwave resonators. Here, the position of an electron
review, we turn to some of these recent developments. in a double dot leads to a dipole moment to which the
Although there remain formidable challenges before resonator electric field couples. First experiments with
large-scale quantum computation becomes a reality, the gate-defined double quantum dots in nanotubes (Delbecq
increasing number of qubits that can be wired up, as well et al., 2011), GaAs (Frey et al., 2012; Toida et al., 2013;
as the improvements in coherence time and gate fidelity, Wallraff et al., 2013), and InAs nanowires (Petersson et al.,
suggests that it will eventually be possible to perform 2012) have demonstrated dispersive coupling and its use
computations on circuit QED-based quantum processors for characterizing charge states of quantum dots (Burkard
that are out of reach of current classical computers. Quan- et al., 2020). These first experiments were, however, lim-
tum supremacy on a 53-qubit device has already been ited by the very large dephasing rate of the quantum dot’s
claimed (Arute et al., 2019), albeit on a problem of no charge states, but subsequent experiments have been able
immediate practical interest. There is, however, much ef- to reach the strong coupling regime (Bruhat et al., 2018;
fort deployed in finding useful computational tasks which Mi et al., 2017; Stockklauser et al., 2017). Building on
can be performed on Noisy Intermediate-Scale Quantum these results and by engineering an effective spin-orbit
(NISQ) devices (Preskill, 2018). First steps in this direc- interaction (Beaudoin et al., 2016; Pioro-Ladrière et al.,
tion include the determination of molecular energies with 2008), it has been possible to reach the strong coupling
variational quantum eigensolvers (Colless et al., 2018; regime with single spins (Landig et al., 2018; Mi et al.,
Kandala et al., 2017; O’Malley et al., 2016) or boson 2018; Samkharadze et al., 2018).
sampling approachs (Wang et al., 2019a) and machine When the coupling to a single spin cannot be made
learning with quantum-enhanced features (Havlı́ček et al., large enough to reach the strong coupling regime, it can
2019). be possible to rely on an ensemble of spins to boost the
Engineered circuit QED-based devices also present an effective coupling. Indeed, in the presence of an ensemble
exciting avenue toward performing analog quantum simu- of N emitters,√ the coupling strength to the ensemble is
lations. In contrast to quantum computing architectures, enhanced by N (Fink√et al., 2009; İmamoğlu, 2009), such
quantum simulators are usually tailored to explore a sin- that for large enough g N the strong coupling regime can
gle specific problem. An example are arrays of resonators be reached. First realization of these ideas used ensembles
which are capacitively coupled to allow photons to hop of ∼ 1012 spins to bring the coupling from a few Hz to
from resonator to resonator. Taking advantage of the flex- ∼ 10 MHz with NV centers in diamond (Kubo et al.,
ibility of superconducting quantum circuits, it is possible 2010) and Cr3+ spins in ruby (Schuster et al., 2010). One
to create exotic networks of resonators such as lattices in objective of these explorations is to increase the sensitivity
an effective hyperbolic space with constant negative cur- of electron paramagnetic resonance (EPR) or electron spin
vature (Kollár et al., 2019). Coupling qubits to each res- resonance (ESR) spectroscopy for spin detection with
onator realizes a Jaynes-Cummings lattice which exhibits the ultimate goal of achieving the single-spin limit. A
a quantum phase transition similar to the superfluid-Mott challenge in reaching this goal is the long lifetime of single
insulator transition in BoseHubbard lattices (Houck et al., spins in these systems which limits the repetition rate of
2012). Moreover, the nonlinearity provided by capacti- the experiment. By engineering the coupling between the
tively coupled qubits, or of Josephson junctions embedded spins and an LC oscillator fabricated in close proximity, it
in the center conductor of the resonators, creates photon- has been possible to take advantage of the Purcell effect to
photon interactions. This leads to effects such as photon reduce the relaxation time from 103 s to 1s (Bienfait et al.,
blockade bearing some similarities to Coulomb blockade 2016). This faster time scale allows for faster repetition
 62

rates thereby boosting the sensitivity, which (a) ΦB (b)

√ could lead BC AB
to spin sensitivities on the order of 0.1 spin/ Hz (Haikka
et al., 2017). Cg
Mechanical systems operated in the quantum regime EJ CS L C
ΦC Vg ΦA
also benefited from the ideas of circuit QED (Aspelmeyer
et al., 2014). An example is a suspended aluminium
membrane that plays the role of a vacuum gap capacitor
in a microwave LC oscillator. The frequency of this FIG. 34 (a) Voltage-biased transmon qubit with the three
oscillator depends on the separation between the plates of relevant flux branches. (b) Replacing the classical voltage
the capacitor leading to a coupling between the oscillator source by a LC oscillator. The dashed arrows indicate the sign
and the flexural mode of the membrane. Strong coupling convention.
between mechanical motion and the LC oscillator has
been demonstrated (Teufel et al., 2011b), which allowed
Appendix A: Hamiltonian of a voltage biased transmon
to sideband cool the motion of the mechanical oscillator
to phonon occupation number as small as nphonon ∼ 0.34.
An excellent introduction to the quantization of electro-
(Teufel et al., 2011a). Squeezed radiation generated by
magnetic circuits can be found in Vool and Devoret (2017).
a Josephson parametric amplifier was also used to cool
Here, we only give a brief introduction to this topic by
beyond the quantum backaction limit to nphonon ∼ 0.19
means of two examples that are used throughout this
(Clark et al., 2017). Building on these ideas, entanglement
review: a transmon qubit biased by an external voltage
of the mechanical motion and the microwave fields was
source, and a transmon coupled to an LC oscillator.
demonstrated (Palomaki et al., 2013b) as well as coherent
state transfer between itinerant microwave fields and a
mechanical oscillator (Palomaki et al., 2013a). 1. Classical gate voltage
Hybrid systems are also important in the context of mi-
crowave to optical frequency transduction in the quantum Consider first the circuit shown in Fig. 34(a), illustrat-
regime. This is a very desirable primitive for quantum ing a transmon biased by an external voltage Vg . Follow-
networks, as it would allow quantum processors based on ing Vool and Devoret (2017), we start by associating a
Rt
circuit QED to be linked optically over large distances. A branch flux Φi (t) = −∞ dt0 Vi (t0 ) to each branch of the
variety of hybrid systems are currently being investigated circuit, with Vi the voltage across branch i = A, B, C
for this purpose, including electro-optomechanical, electro- indicated in Fig. 34(a). Because Kirchoff’s laws impose
optic and magneto-optic ones (Higginbotham et al., 2018; constraints between the branch fluxes, these fluxes are not
Lambert et al., 2019; Lauk et al., 2020). Two other hy- independent variables and are therefore not independent
brid quantum systems that have recently emerged are degrees of freedom of the circuit. Indeed, Kirchoff’s volt-
quantum surface acoustic waves interacting with super- age law dictates that VC + VB + VA = Vg + Φ̇B + Φ̇A = 0
conducting qubits (Gustafsson et al., 2014), and quantum where we have used the sign convention dictated by the
magnonics where quanta of excitation of spin-wave modes (arbitrarily chosen) orientation of the arrows in Fig. 34(a).
known as magnon are strongly coupled to the field of a This constraint allows us to eliminate ΦB in favor of ΦA .
3D microwave cavity (Lachance-Quirion et al., 2019). Moreover, following Kirchoff’s current law, the currents
IA and IB flowing into and out of the node indicated
by the black dot in Fig. 34(a) obey IA = IB . This con-
straint can be expressed in terms of the branch fluxes
using the constitutive relations for the capacitances Cg
ACKNOWLEDGMENTS and CΣ = CS + CJ

QA = CΣ Φ̇A , QB = Cg Φ̇B , (A1)

We thank Alexandre Choquette-Poitevin, Agustin Di
Paolo, Christopher Eichler, Jens Koch, Dany Lachance- as well as the Josephson current relation
Quirion, Yasunobu Nakamura, William Oliver, Alexandru
Petrescu, Baptiste Royer, Gerd Schön and Irfan Siddiqi IJ = Ic sin ϕA , (A2)
for discussions and comments on the manuscript. This
work was undertaken thanks to funding from NSERC, where ϕA = (2π/Φ0 )ΦA and Ic is the critical current. We
the Canada First Research Excellence Fund, the U.S. can thus write IA = Q̇A + IJ = CΣ Φ̈A + Ic sin ϕA and
Army Research Office Grant No. W911NF-18-1-0411 and IB = Q̇B = Cg Φ̈B . Combining the above expressions, we
W911NF-18-1-0212, and the Australian Research Coun- arrive at
cil (ARC) via Discovery Early Career Research Award
(DECRA) DE190100380. CΣ Φ̈A + Ic sin ϕA = −Cg (Φ̈A + Φ̈C ). (A3)
 63

Here, Φ̇C = Vg is the applied bias voltage and the only Defining the vector of conjugate momenta Q =
dynamical variable in the above equation is thus ΦA . As (QA , QC )T , the Hamiltonian is then (Goldstein et al.,
can easily be verified, this equation of motion for ΦA can 2001)
equivalently be derived from the Euler-Lagrange equation
for ΦA with the Lagrangian 1 T −1
H= Q C Q+V
2
CΣ 2 Cg (C + Cg ) 2 Cg
LT = Φ̇ + (Φ̇A + Φ̇C )2 + EJ cos ϕA , (A4) = QA + 2 QA QC − EJ cos ϕA (A9)
2 A 2 2C̄ 2 C̄
where EJ = (Φ0 /2π)Ic . (CΣ + Cg ) 2 ΦC
+ QC + ,
The corresponding Hamiltonian can be found by first 2C̄ 2 2L
identifying the canonical momentum associated to the
coordinate ΦA , QA = ∂LT /∂ Φ̇A = (CΣ + Cg )Φ̇A + Cg Φ̇C , where we have defined C̄ 2 = Cg CΣ + Cg C + CΣ C. The
and performing a Legendre transform to obtain (Goldstein limit Cg  CΣ , C results in the simplified expression
et al., 2001)  2
Cg
QA + C QC
(QA − Cg Vg )2 (A10)
H' − EJ cos ϕA + HLC ,
HT = QA Φ̇A − LT = − EJ cos ϕA , (A5) 2CΣ
2(CΣ + Cg )
Q2
where we have made the replacement Φ̇C = Vg and with HLC = 2CC + Φ2LC the Hamiltonian of the LC circuit.
dropped the term Cg Vg2 /2 which only leads to an overall By promoting the flux and charge variables to operators,
shift of the energies. Promoting the conjugate variables and defining n̂ = Q̂A /2e, n̂r = (Cg /C)Q̂C /2e and diago-
to non-commuting operators [Φ̂A , Q̂A ] = i~, we arrive nalizing ĤLC as in Sec. II.A, we arrive at Eq. (29) for a
at Eq. (20) where we have assumed that Cg  CΣ to single mode m = r.
simplify the notation. Equation (A10) can easily be generalized to capacitive
coupling between other types of circuits, such as resonator-
resonator, transmon-transmon or transmon-transmission
2. Coupling to an LC oscillator line coupling by simply replacing the potential energy
terms −EJ cos ϕA and Φ2C /2L to describe the type of
As a model for the simplest realization of circuit QED, circuits in question. This leads, for example, to Eq. (133)
we now replace the voltage source by an LC oscillator, see for two capacitively coupled transmons after introducing
Fig. 34(b). The derivation follows the same steps as before, ladder operators as in Eqs. (23) and (24).
now with Vg + Φ̇B − Φ̇A = 0 and IA + IB = 0 because of
the different choice of orientation for branch A. Moreover,
at the node labelled BC we have IB = IC . Eliminating Appendix B: Unitary transformations
ΦB as before and using the constitutive relations for the
capacitance C and inductance L of the LC oscillator We introduce a number of unitary transformations often
to express the current through the oscillator branch as employed in the field of circuit QED. The starting point
IC = C Φ̈C + ΦC /L, we find is the usual transformation
ΦC ˙
C Φ̈C + = Cg (Φ̈A − Φ̈C ). (A6) ĤU = Û † Ĥ Û − i~Û † Û, (B1)
L
In contrast to the above example, ΦC is now a dynamical of a Hamiltonian under a time-dependent unitary Û with
variable in it’s own right rather than being simply set the corresponding transformation for the states |ψU i =
by a voltage source. Together with Equation (A3) which Û † |ψi. Since the unitary can be written as Û = exp(−Ŝ)
still holds, Eq. (A6) can equivalently be derived using the with Ŝ an anti-Hermitian operator, a very useful result
Euler-Lagrange equations with the Lagrangian in this context is the Baker-Campbell-Hausdorff (BCH)
formula, which holds for any two operators Ŝ and Ĥ
L = LT + LLC , (A7)
1
eŜ Ĥe−Ŝ = Ĥ + [Ŝ, H] + [Ŝ, [Ŝ, Ĥ]] + . . .
where LT is given in Eq. (A4) and LLC = C2 Φ̇2C − 2L
1
ΦC . 2!
It is convenient to write Eq. (A7) as L = T − V with X∞
1 n (B2)
T = 12 ΦT CΦ and V = ΦC /2L−EJ cos ϕA where we have = CŜ [Ĥ],
n!
defined the vector Φ = (ΦA , ΦC )T and the capacitance n=0

matrix where in the last line we have introduced the short-hand

! n times
CΣ + Cg −Cg notation CŜn [Ĥ] =[Ŝ, [Ŝ, [Ŝ, . . . , Ĥ]]] and CŜ0 [Ĥ] = Ĥ (Bois-
C= . (A8)
−Cg C + Cg sonneault et al., 2009).
 64

1. Schrieffer-Wolff perturbation theory For the reader’s convenience, the explicit results up to
k = 2 are for the generator (with ε = 1)
We often seek unitary transformation that diagonalize
the Hamiltonian of an interacting system. Exact diag- hµ, n|V̂ |ν, li
onalization can, however, be impractical, and we must hµ, n|Ŝ (1) |ν, li = , (B6a)
Eµ,n − Eν,l
resort to finding an effective Hamiltonian which describes
X hµ, n|V̂ |ν, ki hν, k|V̂ |ν, li
the physics at low energies using perturbation theory. A hµ, n|Ŝ (2) |ν, li =
general approach to perturbation theory which we follow Eµ,n − Eν,l Eµ,n − Eν,k
k
here is known as a SchriefferWolff transformation (Bravyi ! ,
hµ, n|V̂ |µ, ki hµ, k|V̂ |ν, li
et al., 2011; Schrieffer and Wolff, 1966). The starting −
point is a generic Hamiltonian of the form Eµ,n − Eν,l Eµ,k − Eν,l
(B6b)
Ĥ = Ĥ0 + V̂ , (B3)
6 µ, while block-diagonal matrix element where
for ν =
with typically Ĥ0 some free Hamiltonian and V̂ a pertur- µ = ν are zero, and
bation. We divide the total Hilbert space of our system
into different subspaces such that Ĥ0 does not couple Ĥ (0) = Ĥ0 , (B7a)
states in different subspaces while V̂ does. The goal of X
the Schrieffer-Wolff transformation is to arrive at an ef- Ĥ (1) = P̂µ V̂ P̂µ , (B7b)
µ
fective Hamiltonian for which the different subspaces are X
completely decoupled. hµ, n|Ĥ (2) |µ, mi = hµ, n|V̂ |ν, lihν, l|V̂ |µ, mi
The different subspaces, which we label by an index ν6=µ,l
µ, can conveniently be defined by a set of projection   (B7c)
1 1 1
operators (Cohen-Tannoudji et al., 1998; Zhu et al., 2013) × + ,
2 Eµ,n − Eν,l Eµ,m − Eν,l
X
P̂µ = |µ, nihµ, n|, (B4) for the transformed Hamiltonian (block off-diagonal ma-
n
trix element are zero, i.e., hµ, n|Ĥ (2) |ν, mi = 0 for µ 6= ν).
where |µ, ni, n = 0, 1, . . . , is an orthonormal basis In these expressions, Eµ,n refers to the bare energy of
for the subspace labeled µ. For the Schrieffer-Wolff |µ, ni under the unperturbed Hamiltonian Ĥ0 . An explicit
transformation to be valid, we must assume that V̂ formulae for Ĥ (k) up k = 4 can be found, e.g. in Winkler
is a small perturbation. Formally, the operator norm (2003).
||V̂ || = max|ψi ||Ô|ψi|| should be smaller than half the
energy gap between the subspaces we intend to decouple;
see Eq. (3.1) of Bravyi et al. (2011). While V̂ is often 2. Schrieffer-Wolff for a multilevel system coupled to an
formally unbounded in circuit QED applications, the op- oscillator in the dispersive regime
erator is always bounded when restricting the problem to
physically relevant states. As an application of the general result of Eq. (B7)
The Schrieffer-Wolff transformation is based on finding we consider in this section a situation that is commonly
a unitary transformation Û = e−Ŝ which approximately encoutered in circuit QED: An arbitrary artificial atom
decouples the different subspaces µ by truncating the coupled to a single mode oscillator in the dispersive regime.
Baker-Campbell-Hausdorff formula Eq. (B2) at a desired Both the transmon artificial atom and the two-level sys-
order. We first expand both Ĥ and Ŝ in formal power tem discussed in Sec. III.C are special cases of this more
series general example. The artificial atom, here taken to be a
generic multilevel system, is described
P in its eigenbasis
Ĥ = Ĥ (0) + εĤ (1) + ε2 Ĥ (2) + . . . , (B5a) with the Hamiltonian Ĥatom = j ~ωj |jihj|. The full
Hamiltonian is therefore given by
Ŝ = εŜ (1) + ε2 Ŝ (2) + . . . , (B5b)
X  
where ε is a fiducial parameter introduced to simplify Ĥ = ~ωr ↠â + ~ωj |jihj| + B̂↠+ B̂ † â , (B8)
order counting and which we can ultimately set to ε → 1. j
The Schrieffer-Wolff transformation is found by insert-
ing Eq. (B5) back into Eq. (B2), and collecting terms at where B̂ is an arbitrary operator of the artificial atom
each order εk . We can then iteratively solve for S (k) and which couples to the oscillator. For example, in the case
Ĥ (k) by requiring that the resulting Hamiltonian ĤU is of capacitive coupling, it is proportional to the charge
block-diagonal (i.e. it does not couple different subspaces operator with B̂ ∼ in̂, cf. Eq. (29).
P
µ) at each order, and the additional requirement that Ŝ By inserting resolutions of the identity Iˆ = j |jihj|,
is itself block off-diagonal (Bravyi et al., 2011). the interaction term can be re-expressed in the atomic
 65

eigenbasis as (Koch et al., 2007) j|ji, with j = 0, 1, . . . , ∞. Moreover, the coupling opera-
X tor is B̂ = ~g b̂, and thus
Ĥ = ~ωr ↠â + ~ωj |jihj| p ∗
j gj,j+1 = ghj|b̂|j + 1i = g j + 1 = gj,j+1 , (B16)
X
(B9)
+ ~ gij |iihj|↠+ gij
∗
|jihi|â , while all other matrix elements gij are zero. We conse-
ij quently find
jg 2
where ~gij = hi|B̂|ji, and with gij = gji if B̂ = B̂ † . Λj = χj−1,j = , (B17a)
To use Eq. (B7), we identify the first line of Eq. (B9) as ωq − EC /~(j − 1) − ωr
Ĥ0 and the second line as the perturbation V̂ . The sub- χj = χj−1,j − χj,j+1
spaces labeled by µ are in this situation one-dimensional,  
j j+1 (B17b)
P̂µ = |µihµ|, with |µi = |n, ji = |ni ⊗ |ji, |ni an oscil- = g2 − ,
ωj − ωj−1 − ωr ωj+1 − ωj − ωr
lator number state and |ji an artificial atom eigenstate.
A straightforward calculation yields the second order re- for j > 0, while for j = 0 we have Λ0 = 0 and χ0 =
sult (Zhu et al., 2013) −χ01 = −g 2 /∆ where ∆ ≡ ωq − ωr . In the two-level
X approximation of Eq. (B13), this becomes (Koch et al.,
Ĥdisp = eŜ Ĥe−Ŝ ' ~ωr ↠â + ~(ωj + Λj )|jihj| 2007)
j
X χ12 g2
† ωr0 = ωr − = ωr − , (B18a)
+ ~χj â â|jihj|, 2 ∆ − EC /~
j
g2
(B10) ωq0 = ω1 − ω0 + χ01 = ωq + , (B18b)
∆
where
X X χ12 g 2 EC /~
χ = χ01 − =− , (B18c)
Λj = χij , χj = (χij − χji ) , (B11) 2 ∆ (∆ − EC /~)
i i
which are the results quoted in Eq. (43).
with Recall that this Schrieffer-Wolff perturbation theory
is only valid if the perturbation V̂ is sufficiently small.
|gji |2
χij = . (B12) Following Bravyi et al. (2011), a more precise statement is
ωj − ωi − ωr that we require 2||V̂ || < ∆min , where ∆min is the smallest
Note that we are following here the convention of Koch energy gap between any of the bare energy eigenstates
et al. (2007) rather than of Zhu et al. (2013) for the |ni ⊗ |ji, where |ni is a number state for the oscillator.
definition of χij . Here, V̂ = g(b̂† â+ b̂↠) is formally unbounded but physical
Projecting Eq. (B10) on the first two-atomic levels states have finite excitation numbers. Therefore, replacing
j = 0, 1 with the convention σ̂z = |1ih1| − |0ih0| we obtain the operator norm by hn, j|V̂ † V̂ |n, ji1/2 and using ∆min =
|∆ − jEC /~| corresponding to the minimum energy gap
~ωq0 between neighboring states |n, ji and |n ± 1, j ∓ 1i, we
Ĥdisp ' ~ωr0 ↠â + σ̂z + ~χ↠âσ̂z , (B13)
2 find that a more precise criterion for the validity of the
above perturbative results is
where we have dropped a constant term and defined  
ωr0 = ωr + (χ0 + χ1 )/2, ωq0 = ω1 − ω0 + Λ1 − Λ0 and 1 |∆ − jEC /~|2
n  ncrit ≡ − j . (B19)
χ = (χ1 − χ0 )/2. 2j + 1 4g 2
Setting j = 0, this gives the familiar expression ncrit =
a. The transmon (∆/2g)2 , while setting j = 1 gives a more conservative
estimate. As quoted in the main text, the appropriate
The transmon capacitively coupled to an oscillator is small parameter is therefore n̄/ncrit , with n̄ the average
one example of the above result. From Eq. (32), we oscillator photon number. For the second order effective
identify the free Hamiltonian as Hamiltonian Ĥdisp to be an accurate description of the
system requires n̄/ncrit to be significantly smaller than
EC † † unity (it is difficult to make a precise statement but the
Ĥ0 = ~ωr ↠â + ~ωq b̂† b̂ − b̂ b̂ b̂b̂, (B14)
2 criteria n̄/ncrit . 0.1 is often used).
and the perturbation as

V̂ = ~g(b̂† â + b̂↠). (B15) b. The Jaynes-Cummings model

In this nonlinear oscillator approximation for the trans- It is interesting to contrast the above result in which the
mon, the transmon eigenstates are number states b̂† b̂|ji = transmon is treated as a multilevel system with the result
 66

obtained if the artificial atom is truncated to a two-level cos(Λ)b̂ − sin(Λ)â, leading to

system before performing the Schireffer-Wolff transfor-
mation. That is, we start with the Jaynes-Cummings ĤL0 = Û † ĤL Û = ω˜r ↠â + ω̃q b̂† b̂
 
Hamiltonian ∆ (B25)
+ g cos(2Λ) − sin(2Λ) (↠b̂ + âb̂† ),
~ωq 2
ĤJC = ~ωr ↠â + σ̂z + ~g(↠σ̂− + âσ̂+ ). (B20)
2
where
Identifying the first two terms as the unperturbed Hamil-
tonian Ĥ0 and the last term as the interaction V̂ , we ω̃r = cos2 (Λ)ωr + sin2 (Λ)ωq − g sin(2Λ), (B26)
can again apply Eq. (B7). Alternatively, the result 2 2
ω̃q = cos (Λ)ωq + sin (Λ)ωr + g sin(2Λ). (B27)
can be found more directly from Eq. (B13) by taking
∗
g01 = g01 = g and all other gij = 0. The result is To cancel the last term of ĤL0 , we take Λ = 12 arctan(2λ)
with λ = g/∆ and ∆ = ωq − ωr to obtain the diagonal
g2 g2 form
ωr0 = ωr , ωq0 = ωq + , χ= , (B21)
∆ ∆
ĤL0 = ~ω̃r ↠â + ~ω̃q b̂† b̂, (B28)
with ∆ = ωq − ωr as before. We see that the results
agree with Eq. (B18) only in the limit EC /~  ∆, g. with the mode frequencies
Importantly, since EC is relatively small compared to
the detuning ∆ in most transmon experiments, the value 1 p 
ω̃r = ωr + ωq − ∆2 + 4g 2 , (B29)
for χ predicted from the Jaynes-Cummings model is far 2
from the multi-level case. Moreover, following the same 1 p 
ω̃q = ωr + ωq + ∆2 + 4g 2 . (B30)
argument as above, we find that the Schrieffer-Wolff trans- 2
formation is valid for photon numbers n̄ < ncrit with
ncrit = (∆/2g)2 − j with j = 0, 1 for the ground and The same transformation on ĤNL gives
excited qubit states, respectively. 0
ĤNL = Û † ĤNL Û
It is interesting to note that the transformation used
here to approximately diagonalize the Jaynes-Cummings EC EC
= − cos4 (Λ)(b̂† )2 b̂2 − sin4 (Λ)(↠)2 â2
Hamiltonian can be obtained by Taylor expanding the 2 2
generator Λ(N̂T ) of the unitary transformation Eq. (35) − 2EC cos2 (Λ) sin2 (Λ)↠âb̂† b̂
which exactly diagonalizes ĤJC . This exercise also leads  
to the conclusion that n̄/ncrit , with ncrit = (∆/2g)2 , is + EC cos3 (Λ) sin(Λ) b̂† b̂ ↠b + H.c.
 
the appropriate small parameter. Alternatively, Ĥdisp can
+ EC cos(Λ) sin3 (Λ) ↠â âb̂† + H.c.
also be obtained simply by Taylor expanding the diagonal
form Eq. (37) of ĤJC (Boissonneault et al., 2010). EC
− cos(Λ)2 sin(Λ)2 [(↠)2 b̂2 + H.c.].
2
(B31)
3. Bogoliubov approach to the dispersive regime Note that, at this stage, the transformation is exact.
In the dispersive regime, we expand the mode frequencies
We derive the results presented in Sec. III.C.2. Our 0
and ĤNL in powers of λ. For the nonlinear part of the
starting point is thus the transmon-resonator Hamilto- Hamiltonian, this yields
nian Eq. (32) and our final result the dispersive Hamilto-
nian of Eq. (50). 0 EC † 2 2 EC † 2 2
ĤNL = − (b̂ ) b̂ − λ4 (â ) â
It is first useful to express Eq. (32) as a sum of a linear 2 2
and a nonlinear part, Ĥ = ĤL + ĤNL where − 2λ2 EC ↠âb̂† b̂
+ λEC (b̂† b̂ ↠b̂ + H.c.) (B32)
ĤL = ~ωr ↠â + ~ωq b̂† b̂ + ~g(b̂† â + b̂↠), (B22)
EC † † + λ3 EC (↠â âb̂† + H.c.)
ĤNL = − b̂ b̂ b̂b̂. (B23) EC † 2 2
2 − λ2 [(â ) b̂ + H.c.] + O(λ5 ).
2
The linear Hamiltonian ĤL can be diagonalized exactly
with the Bogoliubov transformation The magnitude λ2 EC of the cross-Kerr term ↠âb̂† b̂ in
h i this expression does not coincide with Eq. (3.12) of Koch
Û = exp Λ(↠b̂ − âb̂† ) . (B24) et al. (2007). To correct this situation, we apply an
additional transformation to eliminate the third line of
Under this unitary transformation, the annihilation op- Eq. (B32). This term is important because it corresponds,
erators transform as Û † âÛ = cos(Λ)â + sin(Λ)b̂, Û † b̂Û = roughly, to an exchange interaction ↠b̂ + b̂† â with an
 67

additional number operator b̂† b̂ which distinghuishes the ε(t) = ε and with φd = 0 for simplicity, our starting point
different transmon levels. To eliminate this term, we is
apply a Schrieffer-Wolff transformation to second order
with the generator S = λ0 (b̂† b̂ ↠b̂ − H.c.) where λ0 = EC † 2 2
Ĥ = ~δq b̂† b̂ − (b̂ ) b̂ + ~ε(b̂† + b̂), (B40)
λEC /[∆ + EC (1 − 2λ2 )]. Neglecting the last two lines 2
of Eq. (B32) and ommiting a correction of order λ2 , we
in a frame rotating at ωd and where δq = ωq − ωd . We
arrive at Eq. (50) which agrees with Koch et al. (2007).
treat the drive as a perturbation and apply the second
order formula Eq. (B7) to obtain
4. Off-resonantly driven transmon
EC † 2 2
ĤU ' ~δq b̂† b̂ − (b̂ ) b̂
We derive Eq. (132) describing the ac-Stark shift re- 2 !
sulting from an off-resonant drive on a transmon qubit. X j j+1
2
+|ε| − |jihj| (B41)
Our starting point is Eq. (129) which takes the form δq − EC (j−1)
δq − E~C j
j ~
EC † 2 2 EC † 2 2 |ε|2 |ε|2
Ĥ(t) = ~ωq b̂† b̂ − (b̂ ) b̂ + ~(t)b̂† + ~∗ (t)b̂, (B33) ' ~δq b̂† b̂ − (b̂ ) b̂ − 2EC 2 b̂† b̂ − ,
2 2 δq δq
where we have defined (t) = ε(t)e−iωd t−iφd . To account
where |ji is used to label transmon states, as before.
for a possible time-dependence of the drive envelope ε(t)
In the last approximation we have kept only terms to
it is useful to apply the time-dependent displacement
O(jEC /δq ). This agrees with Eq. (B38) for |α|2 = |ε/δq |2 .
transformation
More accurate expressions can be obtained by going to
∗
(t)b̂−α(t)b̂† higher order in perturbation theory (Schneider et al.,
Û (t) = eα . (B34)
2018).
Under Û (t), b̂ transforms to Û † b̂Û = b̂ − α(t), while
˙
Û † Û = α̇∗ (t)b̂ − α̇(t)b̂† . (B35) Appendix C: Input-output theory

Using these expressions, the transformed Hamiltonian Following closely Yurke (2004) and Yurke and Denker
becomes (1984), we derive the input-output equations of Sec. IV.B.
˙ As illustrated in Fig. 11, we consider an LC oscillator
Ĥ 0 = Û † H Û − iÛ † Û
located at x = 0 and which is capacitively coupled to
' ~ωq (b̂† b̂ − α∗ b̂ − αb̂† ) a semi-infinite transmission line extending from x = 0
EC † 2 2 (B36) to ∞. In analogy with Eq. (7), the Hamiltonian for the
− [(b̂ ) b̂ + 4|α|2 b̂† b̂) transmission line is
2
+ ~b̂† + ~∗ b̂ − i~(α̇∗ b̂ − α̇b̂† ), h i2
Z ∞  ∂x Φ̂tml (x) 
Q̂tml (x)2
where we have dropped fast-rotating terms and a scalar. Ĥtml = dx θ(x) + , (C1)
−∞ 2c 2l
The choice

α̇(t) = −iωq α(t) + i(t), (B37) where c and l are, respectively, the capacitance and induc-
tance per unit length, and θ(x) the Heaviside step func-
cancels the linear drive term leaving tion. The flux and charge operators satisfy the canonical
commutation relation [Φ̂tml (x), Q̂tml (x0 )] = i~δ(x − x0 ).
EC † 2 2
Ĥ 0 (t) ' [~ωq − 2EC |α(t)|2 ]b̂† b̂ − (b̂ ) b̂ . (B38) On the other hand,√the Hamiltonian of the LC oscillator
2 of frequency ωr = 1/ Lr Cr is ĤS = Q̂2r /(2Cr ) + Φ̂2r /(2Lr )
Taking a constant envelope ε(t) = ε for simplicity such and the interaction Hamiltonian takes the form
that |α(t)|2 = (ε/δq )2 , the above expression takes the Z ∞
Cκ
compact form Ĥint = dx δ(x) Q̂r Q̂tml (x), (C2)
  −∞ cC r
00 1 Ω2R
Ĥ (t) ' ~ωq − EC 2 σ̂z , (B39) where Cκ is the coupling capacitance between the oscilla-
2 2δq
tor and the line. In deriving Eq. (C2), we have neglected
in the two-level approximation which is Eq. (132) of the renormalizations of c and Cr due to Cκ (c.f Appendix A).
main text. The
R ∞ total Hamiltonian is thus Ĥ = ĤS + Ĥtml + Ĥint =
It is instructive to obtain the same result now using the −∞
dx H, where we have introduced the Hamiltonian
Schrieffer-Wolff approach. Assuming a constant envelope density H in the obvious way.
 68

Using these results, Hamilton’s equations for the field finally arrive at the boundary condition of Eq. (71) at
in the transmission line take the form x=0
Cκ ˙
˙ Q̂tml (x) Cκ V̂out (t) − V̂in (t) = Ztml Q̂r , (C10)
Φ̂tml (x) = θ(x) + δ(x) Q̂r , (C3) Cr
c Cc
" # r where we have introduced the standard notation
˙ ∂x Φ̂tml (x) V̂in/out (t) = V̂L/R (x = 0, t).
Q̂tml (x) = ∂x θ(x) . (C4)
l Using the mode expansion of the fields in Eq. (C8)
together with Eq. (4) for the LC oscillator charge operator
These two equations can be combined into a wave equation in terms of the ladder operator â, Eq. (C10) can be
for Φ̂tml which, for x > 0, reads expressed as
Z ∞ r
¨ ω −i(ω−ωr )t  
Φ̂tml (x) = v 2 ∂x2 Φ̂tml (x), (C5) dω e b̂Rω − b̂Lω
0 4πcv
√ r (C11)
and where v = 1/ lc is the speed of light in the line. At Cκ ωr Cr
= − iωr Ztml â,
the location x = 0 of the oscillator, we instead find Cr 2
h i where we have neglected terms rotating at ω + ωr . After
¨
Φ̂tml (x) = θ(x)v 2 δ(x)∂x Φ̂tml (x) + ∂x2 Φ̂tml (x) some re-arrangements this can be written in the form of
Cκ ˙ (C6) the standard input-output boundary condition (Collett
+ δ(x) Q̂r , and Gardiner, 1984; Gardiner and Collett, 1985)
Cr c
√
b̂out (t) − b̂in (t) = κâ(t), (C12)
where we have used ∂x θ(x) = δ(x). We integrate the last
equation over −ε < x < ε and subsequently take ε → 0 with input and output fields defined as
to find the boundary condition Z ∞
−i
b̂in (t) = √ dω b̂Lω e−i(ω−ωr )t , (C13a)
Cκ ˙ 2π −∞
v 2 ∂x Φ̂tml (x = 0) = − Q̂r . (C7) Z ∞
Cr c −i
b̂out (t) = √ dω b̂Rω e−i(ω−ωr )t . (C13b)
2π −∞
From Eq. (C5), we find that the general solution and the photon loss rate κ is given by
for the flux and charge fields, defined as Q̂tml (x, t) =
c∂t Φ̂tml (x, t), can be written for x > 0 as Φ̂tml (x, t) = Ztml Cκ2 ωr2
κ= . (C14)
Φ̂L (x, t) + Φ̂R (x, t) and Q̂tml (x, t) = Q̂L (x, t) + Q̂R (x, t), Cr
with the subscript L/R denoting left- and right-moving There are two further approximations which are made
fields when going from Eq. (C11) to Eq. (C12): We have ex-
Z ∞ r tended the range of integration over frequency√ from [0, ∞)
~ √
Φ̂L/R (x, t) = dω e±iωx/v+iωt b̂†L/Rω to (−∞, ∞), and we have replaced the factor ω by ωr
0 4πωcv (C8a) inside the integrand. Both approximations are made
+ H.c., based on the assumptions that only terms with ω ' ωr
Z ∞ r contribute significantly to the integral in Eq. (C11).
~ωc ±iωx/v+iωt †
Q̂L/R (x, t) = i dω e b̂L/Rω Moreover, we rewrite Eq. (C9) as
0 4πv (C8b) h i
− H.c. ∂x Φ̂tml (x, t) = Ztml Q̂L (x, t) − Q̂R (x, t)
h i (C15)
In this expression, we introduced the operators b̂νω satis- = Ztml 2Q̂L (x, t) − Q̂tml (x, t) ,
fying [b̂νω , b̂µω0 ] = δνµ δ(ω − ω 0 ) for ν =L, R.
where in the last equality we have used Q̂tml (x, t) =
Because of the boundary condition at x = 0, the left-
Q̂L (x, t) + Q̂R (x, t). At x = 0, this gives
and right-moving fields are not independent. To see this,
we first note that, following from the form of Φ̂tml (x, t), 1 Cκ
Q̂tml (x = 0, t) = 2Q̂L (x = 0, t) + Q̂r (t). (C16)
v Cr
∂x Φ̂tml (x, t) ˙ ˙ (C9) Using this result in the Heisenberg representation equa-
Ztml = Φ̂L (x, t) − Φ̂R (x, t),
l tions of motion for the LC oscillator,
p
with Ztml = l/c the characteristic impedance of the ˙ i Q̂r Cκ
Φ̂r = [Ĥ, Φ̂r ] = + Q̂tml (x = 0), (C17)
transmission line. Noting that I(x) ˆ = ∂x Φ̂tml (x) is the ~ Cr Cr c
˙ ˙ i Φ̂r
current and defining voltages V̂L/R (x) = Φ̂L/R (x), we can Q̂r = [Ĥ, Q̂r ] = − , (C18)
recognize Eq. (C9) as Ohm’s law. Using Eq. (C7), we ~ Lr
 69

we arrive at a single equation of motion for the oscillator Bardeen, J., 1962, Phys. Rev. Lett. 9(4), 147.
charge Barends, R., J. Kelly, A. Megrant, A. Veitia, D. Sank, E. Jef-
   frey, T. C. White, J. Mutus, A. G. Fowler, B. Campbell,
¨ Cκ 1 Cκ Y. Chen, Z. Chen, et al., 2014, Nature 508(7497), 500,
Q̂r = −ωr2 Q̂r + Q̂r + 2Q̂in . (C19) ISSN 0028-0836, URL http://www.nature.com/nature/
c v Cr
journal/v508/n7497/full/nature13171.html.
Again writing Q̂r in terms of bosonic creation and an- Barends, R., J. Wenner, M. Lenander, Y. Chen, R. C. Bialczak,
J. Kelly, E. Lucero, P. O’Malley, M. Mariantoni, D. Sank,
nihilation operators, it is possible to express Eq. (C19)
H. Wang, T. C. White, et al., 2011, Appl. Phys. Lett. 99(11),
in the form of the familiar Langevin equation Eq. (75) 113507 (pages 3), URL http://link.aip.org/link/?APL/
for the mode operator â(t). This standard expression is 99/113507/1.
obtained after neglecting fast rotating terms and making Barnett, S., and P. Radmore, 2002, Methods in Theoretical
the following “slowly varying envelope” approximations Quantum Optics (Oxford University Press, New York), 2nd
(Yurke, 2004) edition.
Barzanjeh, S., D. P. DiVincenzo, and B. M. Terhal, 2014,
d2 −iωr t ˙ −iωr t ,
Phys. Rev. B 90, 134515, URL http://link.aps.org/doi/
âe ' − ωr2 âe−iωr t − 2iωr âe (C20a) 10.1103/PhysRevB.90.134515.
dt2
d −iωr t Barzanjeh, S., S. Guha, C. Weedbrook, D. Vitali, J. H.
âe ' − iωr âe−iωr t , (C20b) Shapiro, and S. Pirandola, 2015, Phys. Rev. Lett.
dt 114, 080503, URL http://link.aps.org/doi/10.1103/
d PhysRevLett.114.080503.
b̂−ω e−iωt ' − iωr b̂−ω e−iωt . (C20c)
dt Barzanjeh, S., S. Pirandola, D. Vitali, and J. M. Fink, 2019,
arXiv e-prints , arXiv:1908.03058eprint 1908.03058.
Equation (75) can be viewed as a Heisenberg picture Beaudoin, F., J. M. Gambetta, and A. Blais, 2011, Phys. Rev.
analog to the Markovian master equation Eq. (68). A 84, 043832, URL http://link.aps.org/doi/10.1103/
PhysRevA.84.043832.
Beaudoin, F., D. Lachance-Quirion, W. A. Coish, and M. Pioro-
REFERENCES Ladrière, 2016, Nanotechnology 27(46), 464003.
Beaudoin, F., M. P. da Silva, Z. Dutton, and A. Blais, 2012,
Abdo, B., N. T. Bronn, O. Jinka, S. Olivadese, A. D. Córcoles, Phys. Rev. A 86, 022305, URL https://link.aps.org/
V. P. Adiga, M. Brink, R. E. Lake, X. Wu, D. P. Pappas, doi/10.1103/PhysRevA.86.022305.
and J. M. Chow, 2019, Nature Communications 10(1), 3154. Bennett, C. H., D. P. DiVincenzo, J. A. Smolin, and W. K.
Abdo, B., A. Kamal, and M. Devoret, 2013, Phys. Rev. B Wootters, 1996, Phys. Rev. A 54(5), 3824.
87(1), 014508. Bergeal, N., F. Schackert, L. Frunzio, and M. H. Devoret, 2012,
Abdumalikov, A. A., O. Astafiev, A. M. Zagoskin, Y. A. Phys. Rev. Lett. 108, 123902, URL http://link.aps.org/
Pashkin, Y. Nakamura, and J. S. Tsai, 2010, Phys. Rev. doi/10.1103/PhysRevLett.108.123902.
Lett. 104(19), 193601. Bertet, P., I. Chiorescu, G. Burkard, K. Semba, C. J. P. M. Har-
Abragam, A., 1961, Principles of Nuclear Magnetism (Oxford mans, D. P. DiVincenzo, and J. E. Mooij, 2005, Phys. Rev.
University Press). Lett. 95, 257002, URL http://link.aps.org/abstract/
Al-Saidi, W. A., and D. Stroud, 2001, Phys. Rev. B 65(1), PRL/v95/e257002.
014512. Bertet, P., C. Harmans, and J. Mooij, 2006, Physical Review
Andersen, C. K., and A. Blais, 2017, New Journal of Physics B 73(6), 064512.
19(2), 023022, ISSN 1367-2630, URL http://stacks.iop. Besse, J.-C., S. Gasparinetti, M. C. Collodo, T. Walter,
org/1367-2630/19/i=2/a=023022. P. Kurpiers, M. Pechal, C. Eichler, and A. Wallraff, 2018,
Andersen, C. K., A. Remm, S. Lazar, S. Krinner, J. Heinsoo, Phys. Rev. X 8, 021003, URL https://journals.aps.org/
J.-C. Besse, M. Gabureac, A. Wallraff, and C. Eichler, 2019, prx/abstract/10.1103/PhysRevX.8.021003.
npj Quantum Information 5(1), 69. Bethe, H. A., 1947, Phys. Rev. 72(4), 339.
Arrangoiz-Arriola, P., E. A. Wollack, Z. Wang, M. Pechal, Bhat, N. A. R., and J. E. Sipe, 2006, Phys. Rev. A
W. Jiang, T. P. McKenna, J. D. Witmer, R. Van Laer, and 73, 063808, URL https://link.aps.org/doi/10.1103/
A. H. Safavi-Naeini, 2019, Nature 571(7766), 537. PhysRevA.73.063808.
Arute, F., K. Arya, R. Babbush, D. Bacon, J. C. Bardin, Bialczak, R. C., M. Ansmann, M. Hofheinz, E. Lucero, M. Nee-
R. Barends, R. Biswas, S. Boixo, F. G. Brandao, D. A. ley, A. D. O’Connell, D. Sank, H. Wang, J. Wenner, M. Stef-
Buell, et al., 2019, Nature 574(7779), 505. fen, A. N. Cleland, and J. M. Martinis, 2010, Nat. Phys.
Aspelmeyer, M., T. J. Kippenberg, and F. Marquardt, 2014, 6(6), 409, ISSN 1745-2473, URL http://dx.doi.org/10.
Rev. Mod. Phys. 86, 1391, URL http://link.aps.org/ 1038/nphys1639.
doi/10.1103/RevModPhys.86.1391. Bienfait, A., J. J. Pla, Y. Kubo, X. Zhou, M. Stern, C. C.
Astafiev, O., A. M. Zagoskin, A. A. Abdumalikov Jr., Y. A. Lo, C. D. Weis, T. Schenkel, D. Vion, D. Esteve, J. J. L.
Pashkin, T. Yamamoto, K. Inomata, Y. Nakamura, and Morton, and P. Bertet, 2016, Nature (London) 531(7592),
J. S. Tsai, 2010, Science 327(5967), 840, URL http://www. 74, ISSN 0028-0836, URL http://dx.doi.org/10.1038/
sciencemag.org/cgi/content/abstract/327/5967/840. nature16944.
Axline, C. J., L. D. Burkhart, W. Pfaff, M. Zhang, K. Chou, Billangeon, P.-M., J. S. Tsai, and Y. Nakamura, 2015a, Phys.
P. Campagne-Ibarcq, P. Reinhold, L. Frunzio, S. M. Girvin, Rev. B 91, 094517, URL http://link.aps.org/doi/10.
L. Jiang, M. H. Devoret, and R. J. Schoelkopf, 2018, Nature 1103/PhysRevB.91.094517.
Physics 14(7), 705. Billangeon, P.-M., J. S. Tsai, and Y. Nakamura, 2015b, Phys.
 70

Rev. B 92, 020509, URL http://link.aps.org/doi/10. Córcoles, A. A. Houck, J. M. Gambetta, and J. M.

1103/PhysRevB.92.020509. Chow, 2015, Applied Physics Letters 107(17), 172601,
Bishop, L. S., J. M. Chow, J. Koch, A. A. Houck, M. H. URL http://scitation.aip.org/content/aip/journal/
Devoret, E. Thuneberg, S. M. Girvin, and R. J. Schoelkopf, apl/107/17/10.1063/1.4934867.
2009, Nat. Phys. 5(2), 105, ISSN 1745-2473. Brooks, P., A. Kitaev, and J. Preskill, 2013, Phys. Rev.
Bishop, L. S., E. Ginossar, and S. M. Girvin, 2010, Phys. Rev. A 87(5), URL http://dx.doi.org/10.1103/PhysRevA.87.
Lett. 105(10), 100505. 052306.
Blais, A., A. M. van den Brink, and A. M. Zagoskin, 2003, Bruhat, L. E., T. Cubaynes, J. J. Viennot, M. C. Dartiailh,
Phys. Rev. Lett. 90(12), 127901. M. M. Desjardins, A. Cottet, and T. Kontos, 2018, Phys.
Blais, A., J. Gambetta, A. Wallraff, D. I. Schuster, S. M. Rev. B 98, 155313, URL https://link.aps.org/doi/10.
Girvin, M. H. Devoret, and R. J. Schoelkopf, 2007, 1103/PhysRevB.98.155313.
Phys. Rev. A 75(3), 032329, URL http://pra.aps.org/ Brune, M., F. Schmidt-Kaler, A. Maali, J. Dreyer, E. Hagley,
abstract/PRA/v75/i3/e032329. J. M. Raimond, and S. Haroche, 1996, Phys. Rev. Lett.
Blais, A., S. M. Girvin, and W. D. Oliver, 2020, Na- 76(11), 1800.
ture Physics 16(3), 247, URL https://doi.org/10.1038/ Bruno, A., G. de Lange, S. Asaad, K. L. van der Enden, N. K.
s41567-020-0806-z. Langford, and L. DiCarlo, 2015, Appl. Phys. Lett. 106,
Blais, A., R.-S. Huang, A. Wallraff, S. M. Girvin, and R. J. 182601, URL http://arxiv.org/abs/1502.04082.
Schoelkopf, 2004, Phys. Rev. A 69, 062320, URL https: Buisson, O., and F. Hekking, 2001, in Macroscopic Quantum
//link.aps.org/doi/10.1103/PhysRevA.69.062320. Coherence and Quantum Computing, edited by A. D. V.,
Bloch, F., and A. Siegert, 1940, Phys. Rev. 57, 522. B. Ruggiero, and P. Silvestrini (Kluwer, New York).
Boaknin, E., V. E. Manucharyan, S. Fissette, M. Met- Bultink, C. C., B. Tarasinski, N. Haandbk, S. Poletto,
calfe, L. Frunzio, R. Vijay, I. Siddiqi, A. Wallraff, R. J. N. Haider, D. J. Michalak, A. Bruno, and L. DiCarlo,
Schoelkopf, and M. Devoret, 2007, arXiv:cond-mat/0702445 2018, Applied Physics Letters 112(9), 092601, URL https:
eprint arXiv:cond-mat/0702445, URL http://arxiv.org/ //doi.org/10.1063/1.5015954.
abs/cond-mat/0702445. Burkard, G., M. J. Gullans, X. Mi, and J. R. Petta, 2020,
Bocko, M., A. Herr, and M. Feldman, 1997, IEEE Transactions Nature Reviews Physics 2(3), 129.
on Applied Superconductivity 7(2), 36383641, ISSN 1051- Bylander, J., S. Gustavsson, F. Yan, F. Yoshihara, K. Harrabi,
8223. G. Fitch, D. G. Cory, Y. Nakamura, J.-S. Tsai, and O. W.
Boissonneault, M., J. M. Gambetta, and A. Blais, 2008, Phys. D., 2011, Nat. Phys. 7, 565, URL http://www.nature.com/
Rev. A 77(6), 305. nphys/journal/v7/n7/full/nphys1994.html.
Boissonneault, M., J. M. Gambetta, and A. Blais, 2009, Phys. Caldeira, A. O., and A. J. Leggett, 1981, Phys. Rev.
Rev. A 79(1), 013819 (pages 17), URL http://link.aps. Lett. 46, 211, URL https://link.aps.org/doi/10.1103/
org/abstract/PRA/v79/e013819. PhysRevLett.46.211.
Boissonneault, M., J. M. Gambetta, and A. Blais, 2010, Phys. Caldwell, S. A., N. Didier, C. A. Ryan, E. A. Sete, A. Hud-
Rev. Lett. 105(10), 100504. son, P. Karalekas, R. Manenti, M. P. da Silva, R. Sinclair,
Bouchiat, V., D. Vion, P. Joyez, D. Esteve, and M. H. Devoret, E. Acala, N. Alidoust, J. Angeles, et al., 2018, Phys. Rev.
1998, Phys. Scr. T76, 165, URL http://stacks.iop.org/ Applied 10, 034050, URL https://link.aps.org/doi/10.
1402-4896/T76/165. 1103/PhysRevApplied.10.034050.
Bourassa, J., 2012, Non-linéarité et couplage lumière-matière Calusine, G., A. Melville, W. Woods, R. Das, C. Stull,
en électrodynamique en circuit,, Ph.D. thesis, Université de V. Bolkhovsky, D. Braje, D. Hover, D. K. Kim, X. Miloshi,
Sherbrooke. D. Rosenberg, A. Sevi, et al., 2018, Applied Physics Letters
Bourassa, J., F. Beaudoin, J. M. Gambetta, and A. Blais, 112(6), 062601.
2012, Phys. Rev. A 86, 013814, URL http://link.aps. Campagne-Ibarcq, P., A. Eickbusch, S. Touzard, E. Zalys-
org/doi/10.1103/PhysRevA.86.013814. Geller, N. Frattini, V. Sivak, P. Reinhold, S. Puri,
Bourassa, J., J. M. Gambetta, J. A. A. Abdumalikov, S. Shankar, R. Schoelkopf, et al., 2019, arXiv:1907.12487 .
O. Astafiev, Y. Nakamura, and A. Blais, 2009, Phys. Rev. Campagne-Ibarcq, P., E. Zalys-Geller, A. Narla, S. Shankar,
A 80(3), 032109 (pages 8), URL http://link.aps.org/ P. Reinhold, L. Burkhart, C. Axline, W. Pfaff, L. Frunzio,
abstract/PRA/v80/e032109. R. J. Schoelkopf, and M. H. Devoret, 2018, Phys. Rev. Lett.
Boutin, S., D. M. Toyli, A. V. Venkatramani, A. W. Ed- 120, 200501, URL https://link.aps.org/doi/10.1103/
dins, I. Siddiqi, and A. Blais, 2017, Phys. Rev. Ap- PhysRevLett.120.200501.
plied 8, 054030, URL https://link.aps.org/doi/10. Carbonaro, P., G. Compagno, and F. Persico, 1979, Phys. Lett.
1103/PhysRevApplied.8.054030. A 73, 97.
Bozyigit, D., C. Lang, L. Steffen, J. M. Fink, C. Eichler, Carmichael, H. J., 2002, Statistical Methods in Quantum
M. Baur, R. Bianchetti, P. J. Leek, S. Filipp, M. P. da Silva, Optics 1: Master Equations and Fokker-Planck Equations
A. Blais, and A. Wallraff, 2011, Nat. Phys. 7(2), 154, ISSN (Springer, Berlin), 2 edition.
1745-2473, URL http://dx.doi.org/10.1038/nphys1845. Carmichael, H. J., 2019, Physics 9, 77.
Braginsky, V. B., Y. I. Vorontsov, and K. S. Thorne, 1980, Carmichael, H. J., P. Kochan, and B. C. Sanders, 1996, Phys.
Science 209(4456), 547, ISSN 0036-8075, URL https:// Rev. Lett. 77(4), 631.
science.sciencemag.org/content/209/4456/547. Carmichael, H. J., A. S. Lane, and D. F. Walls, 1987, Phys.
Bravyi, S., D. P. DiVincenzo, and D. Loss, 2011, Annals of Rev. Lett. 58, 2539, URL https://link.aps.org/doi/10.
physics 326(10), 2793. 1103/PhysRevLett.58.2539.
Breuer, H. P., and F. Petruccione, 2002, The theory of open Casparis, L., M. R. Connolly, M. Kjaergaard, N. J. Pear-
quantum systems (Oxford University Press). son, A. Kringhøj, T. W. Larsen, F. Kuemmeth, T. Wang,
Bronn, N. T., Y. Liu, J. B. Hertzberg, A. D. C. Thomas, S. Gronin, G. C. Gardner, M. J. Manfra, et al.,
 71

2018, Nature Nanotechnology . Rev. A 56(2), 1114, URL http://journals.aps.org/pra/

Castellanos-Beltran, M. A., K. D. Irwin, G. C. Hilton, abstract/10.1103/PhysRevA.56.1114.
L. R. Vale, and K. W. Lehnert, 2008, Nat. Phys. Cirac, J. I., P. Zoller, H. J. Kimble, and H. Mabuchi, 1997,
4, 929, URL http://www.nature.com/nphys/journal/v4/ Phys. Rev. Lett. 78(16), 3221.
n12/abs/nphys1090.html. Clark, J. B., F. Lecocq, R. W. Simmonds, J. Aumentado,
Castellanos-Beltran, M. A., and K. W. Lehnert, 2007, Appl. and J. D. Teufel, 2017, Nature 541(7636), 191, URL http:
Phys. Lett. 91(8), 083509, URL http://link.aip.org/ //dx.doi.org/10.1038/nature20604.
link/?APL/91/083509/1. Clarke, J., A. N. Cleland, M. H. Devoret, D. Esteve, and J. M.
Caves, C. M., 1982, Phys. Rev. D 26, 1817. Martinis, 1988, Science 239, 992.
Caves, C. M., J. Combes, Z. Jiang, and S. Pandey, 2012, Phys. Clarke, J., and F. K. Wilhelm, 2008, Nature 453(7198),
Rev. A 86, 063802, URL https://link.aps.org/doi/10. 1031, ISSN 0028-0836, URL http://dx.doi.org/10.1038/
1103/PhysRevA.86.063802. nature07128.
Chang, C. W. S., A. M. Vadiraj, J. Bourassa, B. Balaji, Clerk, A. A., M. H. Devoret, S. M. Girvin, F. Marquardt, and
and C. M. Wilson, 2019, Applied Physics Letters 114(11), R. J. Schoelkopf, 2010, Rev. Mod. Phys. 82(2), 1155.
112601. Clerk, A. A., K. W. Lehnert, P. Bertet, J. R. Petta, and
Chang, D. E., L. Jiang, A. V. Gorshkov, and H. J. Kimble, Y. Nakamura, 2020, Nature Physics 16(3), 257.
2012, New J. Phys. 14(6), 063003, URL http://stacks. Cochrane, P. T., G. J. Milburn, and W. J. Munro, 1999,
iop.org/1367-2630/14/i=6/a=063003. Phys. Rev. A 59, 2631, URL https://dx.doi.org/10.
Chapman, B. J., E. I. Rosenthal, J. Kerckhoff, B. A. 1103/PhysRevA.59.2631.
Moores, L. R. Vale, J. A. B. Mates, G. C. Hilton, K. La- Cohen-Tannoudji, C., B. Diu, and F. Laloe, 1977, Quantum
lumière, A. Blais, and K. W. Lehnert, 2017, Phys. Rev. Mechanics, Volume 2 (John Wiley & Sons, New York).
X 7, 041043, URL https://link.aps.org/doi/10.1103/ Cohen-Tannoudji, C., J. Dupont-Roc, and G. Grynberg, 1998,
PhysRevX.7.041043. Atom-Photon Interactions: Basic Processes and Applica-
Chen, Y., C. Neill, P. Roushan, N. Leung, M. Fang, tions (Wiley Science Paperback Series).
R. Barends, J. Kelly, B. Campbell, Z. Chen, B. Chiaro, Colless, J. I., V. V. Ramasesh, D. Dahlen, M. S. Blok, M. E.
A. Dunsworth, E. Jeffrey, et al., 2014, Phys. Rev. Lett. Kimchi-Schwartz, J. R. McClean, J. Carter, W. A. de Jong,
113, 220502, URL http://link.aps.org/doi/10.1103/ and I. Siddiqi, 2018, Phys. Rev. X 8, 011021, URL https:
PhysRevLett.113.220502. //link.aps.org/doi/10.1103/PhysRevX.8.011021.
Chen, Y.-F., D. Hover, S. Sendelbach, L. Maurer, S. T. Merkel, Collett, M. J., and C. W. Gardiner, 1984, Phys. Rev. A 30(3),
E. J. Pritchett, F. K. Wilhelm, and R. McDermott, 2011, 1386.
Phys. Rev. Lett. 107, 217401, URL http://link.aps.org/ Combes, J., J. Kerckhoff, and M. Sarovar, 2016,
doi/10.1103/PhysRevLett.107.217401. arXiv:1611.00375 URL http://arxiv.org/abs/1611.
Chen, Z., J. Kelly, C. Quintana, R. Barends, B. Camp- 00375.
bell, Y. Chen, B. Chiaro, A. Dunsworth, A. G. Fowler, Corcoles, A., E. Magesan, S. J. Srinivasan, A. W. Cross,
E. Lucero, E. Jeffrey, A. Megrant, et al., 2016, Phys. M. Steffen, J. M. Gambetta, and J. M. Chow, 2015,
Rev. Lett. 116, 020501, URL http://link.aps.org/doi/ Nat Commun 6, , URL http://dx.doi.org/10.1038/
10.1103/PhysRevLett.116.020501. ncomms7979.
Chiorescu, I., P. Bertet, K. Semba, Y. Nakamura, C. J. P. M. Córcoles, A. D., J. M. Chow, J. M. Gambetta, C. Rigetti, J. R.
Harmans, and J. E. Mooij, 2004, Nature 431(7005), 159. Rozen, G. A. Keefe, M. B. Rothwell, M. B. Ketchen, and
Chou, K. S., J. Z. Blumoff, C. S. Wang, P. C. Reinhold, C. J. M. Steffen, 2011, Appl. Phys. Lett. 99(18), 181906 (pages 3),
Axline, Y. Y. Gao, L. Frunzio, M. H. Devoret, L. Jiang, and URL http://link.aip.org/link/?APL/99/181906/1.
R. J. Schoelkopf, 2018, Nature 561(7723), 368. Córcoles, A. D., J. M. Gambetta, J. M. Chow, J. A. Smolin,
Chow, J. M., A. D. Córcoles, J. M. Gambetta, C. Rigetti, M. Ware, J. Strand, B. L. T. Plourde, and M. Steffen, 2013,
B. R. Johnson, J. A. Smolin, J. R. Rozen, G. A. Keefe, Phys. Rev. A 87, 030301, URL http://link.aps.org/doi/
M. B. Rothwell, M. B. Ketchen, and M. Steffen, 2011, Phys. 10.1103/PhysRevA.87.030301.
Rev. Lett. 107, 080502, URL http://link.aps.org/doi/ Cottet, A., 2002, Implementation of a quantum bit in a super-
10.1103/PhysRevLett.107.080502. conducting circuit, Ph.D. thesis, Université Paris 6.
Chow, J. M., L. DiCarlo, J. M. Gambetta, F. Motzoi, L. Frun- Cross, A. W., and J. M. Gambetta, 2015, Phys. Rev.
zio, S. M. Girvin, and R. J. Schoelkopf, 2010, Phys. Rev. A A 91, 032325, URL http://link.aps.org/doi/10.1103/
82(4), 040305. PhysRevA.91.032325.
Chow, J. M., J. M. Gambetta, A. W. Cross, S. T. Merkel, Dassonneville, R., T. Ramos, V. Milchakov, L. Planat, E. Du-
C. Rigetti, and M. Steffen, 2013, New J. Phys. 15(11), mur, F. Foroughi, J. Puertas, S. Leger, K. Bharadwaj,
115012, URL http://stacks.iop.org/1367-2630/15/i= J. Delaforce, C. Naud, W. Hasch-Guichard, et al., 2020,
11/a=115012. Phys. Rev. X 10, 011045, URL https://link.aps.org/
Chow, J. M., J. M. Gambetta, E. Magesan, D. W. Abra- doi/10.1103/PhysRevX.10.011045.
ham, A. W. Cross, B. R. Johnson, N. A. Masluk, C. A. Day, P. K., H. G. LeDuc, B. A. Mazin, A. Vayonakis, and
Ryan, J. A. Smolin, S. J. Srinivasan, and M. Steffen, J. Zmuidzinas, 2003, Nature 425(6960), 817, ISSN 0028-
2014, Nat. Commun. 5, 4015, URL http://dx.doi.org/10. 0836, URL http://dx.doi.org/10.1038/nature02037.
1038/ncomms5015. Delbecq, M. R., V. Schmitt, F. D. Parmentier, N. Roch, J. J.
Chow, J. M., S. J. Srinivasan, E. Magesan, A. D. Cor- Viennot, G. Fève, B. Huard, C. Mora, A. Cottet, and
coles, D. W. Abraham, J. M. Gambetta, and M. Steffen, T. Kontos, 2011, Phys. Rev. Lett. 107, 256804, URL http:
2015, arXiv:1509.02815 URL http://arxiv.org/abs/1509. //link.aps.org/doi/10.1103/PhysRevLett.107.256804.
02815. Devoret, M., and R. J. Schoelkopf, 2013, Science 339, 1169.
Chuang, I. L., D. W. Leung, and Y. Yamamoto, 1997, Phys. Devoret, M. H., 1997, in Quantum Fluctuations: Les Houches
 72

Session LXIII, edited by S. Reynaud, E. Giacobino, and Eisaman, M. D., J. Fan, A. Migdall, and S. V. Polyakov,
J. Zinn-Justin (Elsevier), pp. 351–386. 2011, Review of Scientific Instruments 82(7), 071101,
Devoret, M. H., S. Girvin, and R. Schoelkopf, 2007, Ann. URL http://scitation.aip.org/content/aip/journal/
Phys. 16(10-11), 767, URL http://dx.doi.org/10.1002/ rsi/82/7/10.1063/1.3610677.
andp.200710261. Elder, S. S., C. S. Wang, P. Reinhold, C. T. Hann, K. S. Chou,
Devoret, M. H., J. M. Martinis, and J. Clarke, 1985, Phys. B. J. Lester, S. Rosenblum, L. Frunzio, L. Jiang, and R. J.
Rev. Lett. 55, 1908, URL http://link.aps.org/doi/10. Schoelkopf, 2020, Phys. Rev. X 10, 011001, URL https:
1103/PhysRevLett.55.1908. //link.aps.org/doi/10.1103/PhysRevX.10.011001.
Dewes, A., F. R. Ong, V. Schmitt, R. Lauro, N. Boulant, Eschner, J., C. Raab, F. Schmidt-Kaler, and R. Blatt, 2001,
P. Bertet, D. Vion, and D. Esteve, 2012, Phys. Rev. Lett. Nature 413, 495, URL http://www.nature.com/nature/
108, 057002, URL http://link.aps.org/doi/10.1103/ journal/v413/n6855/full/413495a0.html.
PhysRevLett.108.057002. Esteve, D., M. H. Devoret, and J. M. Martinis, 1986, Phys.
DiCarlo, L., J. M. Chow, J. M. Gambetta, L. S. Bishop, B. R. Rev. B 34, 158, URL http://link.aps.org/doi/10.1103/
Johnson, D. I. Schuster, J. Majer, A. Blais, L. Frunzio, PhysRevB.34.158.
S. M. Girvin, and R. J. Schoelkopf, 2009, Nature 460(7252), Fan, B., G. Johansson, J. Combes, G. J. Milburn, and T. M.
240, ISSN 0028-0836, URL http://dx.doi.org/10.1038/ Stace, 2014, Phys. Rev. B URL http://journals.aps.org/
nature08121. prb/abstract/10.1103/PhysRevB.90.035132.
DiCarlo, L., M. D. Reed, L. Sun, B. R. Johnson, J. M. Chow, Filipp, S., M. Göppl, J. M. Fink, M. Baur, R. Bianchetti,
J. M. Gambetta, L. Frunzio, S. M. Girvin, M. H. Devoret, L. Steffen, and A. Wallraff, 2011, Phys. Rev. A 83(6),
and R. J. Schoelkopf, 2010, Nature 467, 574. 063827, URL http://pra.aps.org/abstract/PRA/v83/
Didier, N., J. Bourassa, and A. Blais, 2015a, Phys. Rev. i6/e063827.
Lett. 115, 203601, URL http://journals.aps.org/prl/ Filipp, S., P. Maurer, P. J. Leek, M. Baur, R. Bianchetti, J. M.
abstract/10.1103/PhysRevLett.115.203601. Fink, M. Göppl, L. Steffen, J. M. Gambetta, A. Blais, and
Didier, N., A. Kamal, W. D. Oliver, A. Blais, and A. Wallraff, 2009, Phys. Rev. Lett. 102(20), 200402, URL
A. A. Clerk, 2015b, Phys. Rev. Lett. 115, 093604, http://link.aps.org/abstract/PRL/v102/e200402.
URL http://journals.aps.org/prl/abstract/10.1103/ Fink, J. M., R. Bianchetti, M. Baur, M. Göppl, L. Steffen,
PhysRevLett.115.093604. S. Filipp, P. J. Leek, A. Blais, and A. Wallraff, 2009, Phys.
Didier, N., E. A. Sete, M. P. da Silva, and C. Rigetti, 2018, Rev. Lett. 103(8), 083601, URL http://link.aps.org/
Phys. Rev. A 97, 022330, URL https://journals.aps. abstract/PRL/v103/e083601.
org/pra/abstract/10.1103/PhysRevA.97.022330. Fink, J. M., M. Göppl, M. Baur, R. Bianchetti, P. J.
Dykman, M., and M. Krivoglaz, 1980, Physica A: Sta- Leek, A. Blais, and A. Wallraff, 2008, Nature 454(7202),
tistical Mechanics and its Applications 104(3), 480 , 315, ISSN 0028-0836, URL http://dx.doi.org/10.1038/
ISSN 0378-4371, URL http://www.sciencedirect.com/ nature07112.
science/article/pii/0378437180900102. Fink, J. M., L. Steffen, P. Studer, L. S. Bishop, M. Baur,
Eckern, U., 1984, Phys. Rev. B; Physical Review B 30(11), R. Bianchetti, D. Bozyigit, C. Lang, S. Filipp, P. J.
6419. Leek, and A. Wallraff, 2010, Phys. Rev. Lett. 105(16),
Eddins, A., S. Schreppler, D. M. Toyli, L. S. Martin, 163601, URL http://prl.aps.org/abstract/PRL/v105/
S. Hacohen-Gourgy, L. C. G. Govia, H. Ribeiro, A. A. i16/e163601.
Clerk, and I. Siddiqi, 2018, Phys. Rev. Lett. 120, 040505, Flühmann, C., T. L. Nguyen, M. Marinelli, V. Negnevit-
URL https://link.aps.org/doi/10.1103/PhysRevLett. sky, K. Mehta, and J. P. Home, 2019, Nature 566,
120.040505. 513, ISSN 1476-4687, URL https://doi.org/10.1038/
Egger, D., M. Ganzhorn, G. Salis, A. Fuhrer, P. Müller, s41586-019-0960-6.
P. Barkoutsos, N. Moll, I. Tavernelli, and S. Filipp, 2019, Flurin, E., N. Roch, F. Mallet, M. H. Devoret, and B. Huard,
Phys. Rev. Applied 11, 014017, URL https://link.aps. 2012, Phys. Rev. Lett. 109, 183901, URL http://link.aps.
org/doi/10.1103/PhysRevApplied.11.014017. org/doi/10.1103/PhysRevLett.109.183901.
Eichler, C., D. Bozyigit, C. Lang, M. Baur, L. Steffen, J. M. Flurin, E., N. Roch, J. Pillet, F. Mallet, and B. Huard, 2015,
Fink, S. Filipp, and A. Wallraff, 2011a, Phys. Rev. Lett. Phys. Rev. Lett. 114, 090503, URL http://link.aps.org/
107, 113601, URL http://link.aps.org/doi/10.1103/ doi/10.1103/PhysRevLett.114.090503.
PhysRevLett.107.113601. Forn-Diaz, P., J. J. Garcia-Ripoll, B. Peropadre, J.-L. Or-
Eichler, C., D. Bozyigit, C. Lang, L. Steffen, J. Fink, and giazzi, M. A. Yurtalan, R. Belyansky, C. M. Wilson,
A. Wallraff, 2011b, Phys. Rev. Lett. 106(22), 220503, URL and A. Lupascu, 2016, Nat Phys advance online pub-
http://prl.aps.org/abstract/PRL/v106/i22/e220503. lication, , ISSN 1745-2481, URL http://dx.doi.org/10.
Eichler, C., D. Bozyigit, and A. Wallraff, 2012a, Phys. Rev. 1038/nphys3905.
A 86, 032106, URL http://pra.aps.org/abstract/PRA/ Forn-Dı́az, P., L. Lamata, E. Rico, J. Kono, and E. Solano,
v86/i3/e032106. 2019, Rev. Mod. Phys. 91, 025005, URL https://link.
Eichler, C., C. Lang, J. M. Fink, J. Govenius, S. Filipp, and aps.org/doi/10.1103/RevModPhys.91.025005.
A. Wallraff, 2012b, Phys. Rev. Lett. 109, 240501, URL http: Forn-Dı́az, P., J. Lisenfeld, D. Marcos, J. J. Garcı́a-Ripoll,
//link.aps.org/doi/10.1103/PhysRevLett.109.240501. E. Solano, C. J. P. M. Harmans, and J. E. Mooij, 2010,
Eichler, C., Y. Salathe, J. Mlynek, S. Schmidt, and A. Wallraff, Phys. Rev. Lett. 105(23), 237001.
2014, Phys. Rev. Lett. 113, 110502, URL http://link.aps. Forn-Dı́az, P., C. W. Warren, C. W. S. Chang, A. M.
org/doi/10.1103/PhysRevLett.113.110502. Vadiraj, and C. M. Wilson, 2017, Phys. Rev. Applied
Eichler, C., and A. Wallraff, 2014, EPJ Quantum Tech- 8, 054015, URL https://link.aps.org/doi/10.1103/
nology 1(1), 2, ISSN 2196-0763, URL http://www. PhysRevApplied.8.054015.
epjquantumtechnology.com/content/1/1/2. Fowler, A. G., M. Mariantoni, J. M. Martinis, and A. N.
 73

Cleland, 2012, Phys. Rev. A 86, 032324, URL http:// org/10.1063/1.3010859.

link.aps.org/doi/10.1103/PhysRevA.86.032324. Gottesman, D., A. Kitaev, and J. Preskill, 2001, Physical
Foxen, B., C. Neill, A. Dunsworth, P. Roushan, B. Chiaro, Review A 64(1), 012310.
A. Megrant, J. Kelly, Z. Chen, K. Satzinger, R. Barends, Gough, J., and M. R. James, 2009, IEEE Transactions on
F. Arute, K. Arya, et al., 2020, arXiv e-prints , Automatic Control 54(11), 2530.
arXiv:2001.08343eprint 2001.08343. Grimm, A., N. E. Frattini, S. Puri, S. O. Mundhada,
Fragner, A., M. Göppl, J. M. Fink, M. Baur, R. Bianchetti, S. Touzard, M. Mirrahimi, S. M. Girvin, S. Shankar, and
P. J. Leek, A. Blais, and A. Wallraff, 2008, Science M. H. Devoret, 2019, arXiv e-prints , arXiv:1907.12131eprint
322(5906), 1357, URL http://www.sciencemag.org/cgi/ 1907.12131.
content/abstract/322/5906/1357. Grimsmo, A. L., J. Combes, and B. Q. Baragiola, 2020, Phys.
Frey, T., P. J. Leek, M. Beck, A. Blais, T. Ihn, K. En- Rev. X 10, 011058, URL https://link.aps.org/doi/10.
sslin, and A. Wallraff, 2012, Phys. Rev. Lett. 108(046807), 1103/PhysRevX.10.011058.
046807, URL http://prl.aps.org/abstract/PRL/v108/ Gu, X., A. F. Kockum, A. Miranowicz, Y. xi Liu, and F. Nori,
i4/e046807. 2017, Physics Reports 718-719, 1 , ISSN 0370-1573, mi-
Frisk Kockum, A., A. Miranowicz, S. De Liberato, S. Savasta, crowave photonics with superconducting quantum circuits,
and F. Nori, 2019, Nature Reviews Physics 1(1), 19, URL URL http://www.sciencedirect.com/science/article/
https://doi.org/10.1038/s42254-018-0006-2. pii/S0370157317303290.
Frunzio, L., A. Wallraff, D. Schuster, J. Majer, and Guerlin, C., J. Bernu, S. Deleglise, C. Sayrin, S. Gleyzes,
R. Schoelkopf, 2005, IEEE T. Appl. Supercon. 15(2), 860, S. Kuhr, M. Brune, J.-M. Raimond, and S. Haroche, 2007,
URL http://ieeexplore.ieee.org/xpls/abs_all.jsp? Nature 448(7156), 889, ISSN 0028-0836, URL http://dx.
isnumber=31006&arnumber=1439774&count=243&index= doi.org/10.1038/nature06057.
195. Guillaud, J., and M. Mirrahimi, 2019, Phys. Rev. X
Gambetta, J., A. Blais, M. Boissonneault, A. A. Houck, D. I. 9, 041053, URL https://link.aps.org/doi/10.1103/
Schuster, and S. M. Girvin, 2008, Phys. Rev. A 77, 012112, PhysRevX.9.041053.
URL http://dx.doi.org/10.1103/PhysRevA.77.012112. Gustafsson, M. V., T. Aref, A. F. Kockum, M. K.
Gambetta, J., A. Blais, D. I. Schuster, A. Wallraff, L. Frunzio, Ekstram, G. Johansson, and P. Delsing, 2014, Sci-
J. Majer, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, ence 346(6206), 207, URL http://www.sciencemag.org/
2006, Phys. Rev. A 74(4), 042318, URL http://pra.aps. content/346/6206/207.abstract.
org/abstract/PRA/v74/i4/e042318. Gyenis, A., P. S. Mundada, A. Di Paolo, T. M. Hazard, X. You,
Gambetta, J., W. A. Braff, A. Wallraff, S. M. Girvin, and D. I. Schuster, J. Koch, A. Blais, and A. A. Houck, 2019,
R. J. Schoelkopf, 2007, Phys. Rev. A 76(1), 012325. arXiv e-prints , arXiv:1910.07542eprint 1910.07542.
Gambetta, J. M., A. A. Houck, and A. Blais, 2011a, Phys. Hadfield, R. H., 2009, Nat. Photon. 3(12), 696, ISSN 1749-
Rev. Lett. 106(3), 030502. 4885, URL http://dx.doi.org/10.1038/nphoton.2009.
Gambetta, J. M., F. Motzoi, S. T. Merkel, and F. K. Wilhelm, 230.
2011b, Phys. Rev. A 83(1), 012308. Haikka, P., Y. Kubo, A. Bienfait, P. Bertet, and K. Mølmer,
Gardiner, C. W., 1986, Phys. Rev. Lett. 56, 1917, URL https: 2017, Phys. Rev. A 95, 022306, URL https://link.aps.
//link.aps.org/doi/10.1103/PhysRevLett.56.1917. org/doi/10.1103/PhysRevA.95.022306.
Gardiner, C. W., and M. J. Collett, 1985, Phys. Rev. A 31(6), Haroche, S., 1992, Fundamental Systems in Quantum Optics
3761. (Elsevier, New York), chapter Cavity quantum electrody-
Gardiner, C. W., and P. Zoller, 1999, Quantum Noise 2nd ed namics.
(Springer, Berlin Heidelberg). Haroche, S., and D. Kleppner, 1989, Physics Today 42, 24,
Gilchrist, A., K. Nemoto, W. J. Munro, T. Ralph, S. Glancy, URL http://dx.doi.org/10.1063/1.881201.
S. L. Braunstein, and G. Milburn, 2004, Journal of Optics Haroche, S., and J.-M. Raimond, 2006, Exploring the Quantum:
B: Quantum and Semiclassical Optics 6(8), S828. Atoms, Cavities, and Photons (Oxford University Press,
Giovannetti, V., S. Lloyd, and L. Maccone, 2004, Science 306, New York, USA).
1330. Hatridge, M., S. Shankar, M. Mirrahimi, F. Schackert, K. Geer-
Giustina, M., M. A. M. Versteegh, S. Wengerowsky, J. Hand- lings, T. Brecht, K. M. Sliwa, B. Abdo, L. Frunzio, S. M.
steiner, A. Hochrainer, K. Phelan, F. Steinlechner, J. Kofler, Girvin, R. J. Schoelkopf, and M. H. Devoret, 2013, Sci-
J.-A. Larsson, C. Abellán, W. Amaya, V. Pruneri, et al., ence 339(6116), 178, URL http://www.sciencemag.org/
2015, Phys. Rev. Lett. 115, 250401, URL http://link.aps. content/339/6116/178.abstract.
org/doi/10.1103/PhysRevLett.115.250401. Havlı́ček, V., A. D. Córcoles, K. Temme, A. W. Harrow,
Glaetzle, A., K. Hammerer, A. Daley, R. Blatt, and P. Zoller, A. Kandala, J. M. Chow, and J. M. Gambetta, 2019, Nature
2010, Optics Communications 283(5), 758 . 567(7747), 209.
Glazman, L. I., and G. Catelani, 2020, arXiv e-prints , Heeres, R. W., P. Reinhold, N. Ofek, L. Frunzio, L. Jiang,
arXiv:2003.04366eprint 2003.04366. M. H. Devoret, and R. J. Schoelkopf, 2017, Nature
Gleyzes, S., S. Kuhr, C. Guerlin, J. Bernu, S. Deleglise, Communications 8, 94, URL https://doi.org/10.1038/
U. Busk Hoff, M. Brune, J.-M. Raimond, and S. Haroche, s41467-017-00045-1.
2007, Nature 446(7133), 297, ISSN 0028-0836, URL http: Heeres, R. W., B. Vlastakis, E. Holland, S. Krastanov, V. V.
//dx.doi.org/10.1038/nature05589. Albert, L. Frunzio, L. Jiang, and R. J. Schoelkopf, 2015,
Goldstein, H., C. P Poole, and J. L Safko, 2001, Classical Me- Phys. Rev. Lett. 115, 137002, URL http://journals.aps.
chanics (3rd Edition) (Addison Wesley), ISBN 0201657023. org/prl/abstract/10.1103/PhysRevLett.115.137002.
Göppl, M., A. Fragner, M. Baur, R. Bianchetti, S. Filipp, J. M. Heinsoo, J., C. K. Andersen, A. Remm, S. Krinner, T. Wal-
Fink, P. J. Leek, G. Puebla, L. Steffen, and A. Wallraff, ter, Y. Salathé, S. Gasparinetti, J.-C. Besse, A. Potočnik,
2008, J. Appl. Phys. 104(6), 113904, URL http://dx.doi. A. Wallraff, and C. Eichler, 2018, Phys. Rev. Applied
 74

10, 034040, URL https://link.aps.org/doi/10.1103/ inen, S. Simbierowicz, J. Hassel, et al., 2019, Phys. Rev.
PhysRevApplied.10.034040. Lett. 122, 080503, URL https://link.aps.org/doi/10.
Helmer, F., M. Mariantoni, E. Solano, and F. Marquardt, 1103/PhysRevLett.122.080503.
2009, Phys. Rev. A 79, 052115, URL http://link.aps. İmamoğlu, A., 2009, Phys. Rev. Lett. 102(8), 083602, URL
org/doi/10.1103/PhysRevA.79.052115. http://link.aps.org/abstract/PRL/v102/e083602.
Hensen, B., H. Bernien, A. E. Dreau, A. Reiserer, N. Kalb, Inomata, K., Z. Lin, K. Koshino, W. D. Oliver, J. Tsai, T. Ya-
M. S. Blok, J. Ruitenberg, R. F. L. Vermeulen, R. N. mamoto, and Y. Nakamura, 2016, Nat. Commun. 7, 12303,
Schouten, C. Abellan, W. Amaya, V. Pruneri, et al., URL http://www.nature.com/articles/ncomms12303.
2015, Nature 526, 682, URL http://dx.doi.org/10.1038/ Jeffrey, E., D. Sank, J. Y. Mutus, T. C. White, J. Kelly,
nature15759. R. Barends, Y. Chen, Z. Chen, B. Chiaro, A. Dunsworth,
Higginbotham, A. P., P. S. Burns, M. D. Urmey, R. W. Pe- A. Megrant, P. J. J. O’Malley, et al., 2014, Phys. Rev. Lett.
terson, N. S. Kampel, B. M. Brubaker, G. Smith, K. W. 112, 190504, URL http://link.aps.org/doi/10.1103/
Lehnert, and C. A. Regal, 2018, Nature Physics 14(10), PhysRevLett.112.190504.
1038. Johansson, J., S. Saito, T. Meno, H. Nakano, M. Ueda,
Hofheinz, M., H. Wang, M. Ansmann, R. C. Bialczak, K. Semba, and H. Takayanagi, 2006, Phys. Rev. Lett. 96(12),
E. Lucero, M. Neeley, A. D. O’Connell, D. Sank, J. Wen- 127006 (pages 4), URL http://link.aps.org/abstract/
ner, J. M. Martinis, and A. N. Cleland, 2009, Nature 459, PRL/v96/e127006.
546, ISSN 0028-0836, URL http://dx.doi.org/10.1038/ Johnson, B. R., M. D. Reed, A. A. Houck, D. I. Schuster,
nature08005. L. S. Bishop, E. Ginossar, J. M. Gambetta, L. DiCarlo,
Hofheinz, M., E. M. Weig, M. Ansmann, R. C. Bialczak, L. Frunzio, S. M. Girvin, and R. J. Schoelkopf, 2010, Nat.
E. Lucero, M. Neeley, A. D. O’Connell, H. Wang, J. M. Phys. 6, 663.
Martinis, and A. N. Cleland, 2008, Nature 454(7202), Johnson, J. E., E. M. Hoskinson, C. Macklin, D. H.
310, ISSN 0028-0836, URL http://dx.doi.org/10.1038/ Slichter, I. Siddiqi, and J. Clarke, 2011, Phys. Rev.
nature07136. B 84, 220503, URL http://link.aps.org/doi/10.1103/
Hoi, I.-C., A. F. Kockum, T. Palomaki, T. M. Stace, PhysRevB.84.220503.
B. Fan, L. Tornberg, S. R. Sathyamoorthy, G. Johans- Josephson, B. D., 1962, Physics Letters
son, P. Delsing, and C. M. Wilson, 2013, Phys. Rev. Lett. 1(7), 251, URL http://www.sciencedirect.
111(5), 053601, URL http://link.aps.org/doi/10.1103/ com/science/article/B6X44-47RBX8S-5/2/
PhysRevLett.111.053601. 31fc71946bd983d4a8fd06a69e777173.
Hoi, I.-C., A. F. Kockum, L. Tornberg, A. Pourkabirian, G. Jo- Kamal, A., J. Clarke, and M. H. Devoret, 2011, Nat. Phys.
hansson, P. Delsing, and C. M. Wilson, 2015, Nat Phys 7(4), 311, ISSN 1745-2473, URL http://dx.doi.org/10.
URL http://dx.doi.org/10.1038/nphys3484. 1038/nphys1893.
Hoi, I.-C., T. Palomaki, J. Lindkvist, G. Johansson, P. Delsing, Kandala, A., A. Mezzacapo, K. Temme, M. Takita, M. Brink,
and C. M. Wilson, 2012, Phys. Rev. Lett. 108(26), 263601. J. M. Chow, and J. M. Gambetta, 2017, Nature 549(7671),
Hoi, I.-C., C. M. Wilson, G. Johansson, T. Palomaki, 242, ISSN 0028-0836, URL http://dx.doi.org/10.1038/
B. Peropadre, and P. Delsing, 2011, Phys. Rev. Lett. nature23879.
107(7), 073601, URL http://link.aps.org/doi/10.1103/ Kannan, B., D. Campbell, F. Vasconcelos, R. Winik, D. Kim,
PhysRevLett.107.073601. M. Kjaergaard, P. Krantz, A. Melville, B. M. Niedziel-
Holland, E. T., B. Vlastakis, R. W. Heeres, M. J. Reagor, ski, J. Yoder, T. P. Orlando, S. Gustavsson, et al., 2020,
U. Vool, Z. Leghtas, L. Frunzio, G. Kirchmair, M. H. De- arXiv:2003.07300 .
voret, M. Mirrahimi, and R. J. Schoelkopf, 2015, Phys. Kapit, E., 2015, Phys. Rev. A 92, 012302, URL http://link.
Rev. Lett. 115, 180501, URL http://link.aps.org/doi/ aps.org/doi/10.1103/PhysRevA.92.012302.
10.1103/PhysRevLett.115.180501. Kerman, A. J., 2013, New J. Phys. 15(12), 123011, URL
Houck, A. A., J. A. Schreier, B. R. Johnson, J. M. Chow, http://stacks.iop.org/1367-2630/15/i=12/a=123011.
J. Koch, J. M. Gambetta, D. I. Schuster, L. Frunzio, M. H. Khaneja, N., T. Reiss, C. Kehlet, T. Schulte-Herbrüggen, and
Devoret, S. M. Girvin, and R. J. Schoelkopf, 2008, Phys. S. J. Glaser, 2005, Journal of Magnetic Resonance 172(2),
Rev. Lett. 101(8), 080502, URL http://link.aps.org/ 296, ISSN 1090-7807, URL http://www.sciencedirect.
abstract/PRL/v101/e080502. com/science/article/pii/S1090780704003696.
Houck, A. A., D. I. Schuster, J. M. Gambetta, J. A. Schreier, Kimble, H., 1994, Structure and dynamics in cavity quantum
B. R. Johnson, J. M. Chow, L. Frunzio, J. Majer, M. H. electrodynamics (Academic Press), pp. 203–.
Devoret, S. M. Girvin, and R. J. Schoelkopf, 2007, Nature Kimble, H. J., 1998, Physica Scripta T76, 127.
449, 328. Kimble, H. J., 2008, Nature 453(7198), 1023, ISSN 0028-0836.
Houck, A. A., H. E. Türeci, and J. Koch, 2012, Nat. Phys. Kirchmair, G., B. Vlastakis, Z. Leghtas, S. E. Nigg, H. Paik,
8(4), 292, ISSN 1745-2473, URL http://dx.doi.org/10. E. Ginossar, M. Mirrahimi, L. Frunzio, S. M. Girvin, and
1038/nphys2251. R. J. Schoelkopf, 2013, Nature 495(7440), 205, ISSN 0028-
Hu, L., Y. Ma, W. Cai, X. Mu, Y. Xu, W. Wang, Y. Wu, 0836, URL http://dx.doi.org/10.1038/nature11902.
H. Wang, Y. P. Song, C. L. Zou, S. M. Girvin, L.-M. Duan, Kjaergaard, M., M. E. Schwartz, J. Braumüller, P. Krantz,
et al., 2019, Nature Physics 15(5), 503. J. I-Jan Wang, S. Gustavsson, and W. D. Oliver, 2019,
Hutchings, M., J. Hertzberg, Y. Liu, N. Bronn, G. Keefe, arXiv e-prints , arXiv:1905.13641eprint 1905.13641.
M. Brink, J. M. Chow, and B. Plourde, 2017, Physical Re- Knill, E., and R. Laflamme, 1997, Phys. Rev. A 55, 900, URL
view Applied 8(4), 044003, URL https://link.aps.org/ https://link.aps.org/doi/10.1103/PhysRevA.55.900.
doi/10.1103/PhysRevApplied.8.044003. Knill, E., R. Laflamme, and G. J. Milburn, 2001, Nature
Ikonen, J., J. Goetz, J. Ilves, A. Keränen, A. M. Gunyho, 409(6816), 46, ISSN 0028-0836.
M. Partanen, K. Y. Tan, D. Hazra, L. Grönberg, V. Vester- Koch, J., T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schuster,
 75

J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and R. J. Phys. Rev. Lett. 77(1), 198, URL http://link.aps.org/
Schoelkopf, 2007, Phys. Rev. A 76(4), 042319, URL http: doi/10.1103/PhysRevLett.77.198.
//link.aps.org/abstract/PRA/v76/e042319. Lalumière, K., J. M. Gambetta, and A. Blais, 2010, Phys. Rev.
Kochetov, B. A., and A. Fedorov, 2015, Phys. Rev. B A 81, 040301, URL http://link.aps.org/doi/10.1103/
92, 224304, URL http://link.aps.org/doi/10.1103/ PhysRevA.81.040301.
PhysRevB.92.224304. Lalumière, K., B. C. Sanders, A. F. van Loo, A. Fe-
Kollár, A. J., M. Fitzpatrick, and A. A. Houck, 2019, Nature dorov, A. Wallraff, and A. Blais, 2013, Phys. Rev.
571(7763), 45. A 88, 043806, URL http://link.aps.org/doi/10.1103/
Kono, S., K. Koshino, Y. Tabuchi, A. Noguchi, and Y. Naka- PhysRevA.88.043806.
mura, 2018, Nature Physics ISSN 1745-2481, URL https: Lamb, W. E., and R. Retherford, 1947, Phys. Rev. 72,
//doi.org/10.1038/s41567-018-0066-3. 241, URL http://link.aps.org/doi/10.1103/PhysRev.
Kono, S., Y. Masuyama, T. Ishikawa, Y. Tabuchi, R. Ya- 72.241.
mazaki, K. Usami, K. Koshino, and Y. Nakamura, 2017, Lambert, N. J., A. Rueda, F. Sedlmeir, and H. G. Schwefel,
Phys. Rev. Lett. 119, 023602, URL https://arxiv.org/ 2019, Advanced Quantum Technologies , 1900077.
abs/1702.06004. Lamoreaux, S. K., K. A. van Bibber, K. W. Lehnert, and
Korotkov, A. N., 2001, Phys. Rev. B 63, 115403, URL https: G. Carosi, 2013, Phys. Rev. D 88, 035020, URL https:
//link.aps.org/doi/10.1103/PhysRevB.63.115403. //link.aps.org/doi/10.1103/PhysRevD.88.035020.
Koshino, K., K. Inomata, T. Yamamoto, and Y. Nakamura, Landauer, R., 1991, Physics Today 44, 23.
2013, Phys. Rev. Lett. 111, 153601, URL https://link. Landig, A. J., J. V. Koski, P. Scarlino, U. C. Mendes, A. Blais,
aps.org/doi/10.1103/PhysRevLett.111.153601. C. Reichl, W. Wegscheider, A. Wallraff, K. Ensslin, and
Koshino, K., Z. Lin, K. Inomata, T. Yamamoto, and Y. Naka- T. Ihn, 2018, Nature 560(7717), 179.
mura, 2016, Phys. Rev. A 93, 023824, URL https://link. Lang, C., D. Bozyigit, C. Eichler, L. Steffen, J. M. Fink,
aps.org/doi/10.1103/PhysRevA.93.023824. A. A. Abdumalikov Jr., M. Baur, S. Filipp, M. P. da Silva,
Koshino, K., and Y. Nakamura, 2012, New J. Phys. 14(4), A. Blais, and A. Wallraff, 2011, Phys. Rev. Lett. 106(24),
043005, ISSN 1367-2630, URL http://stacks.iop.org/ 243601, URL http://prl.aps.org/abstract/PRL/v106/
1367-2630/14/i=4/a=043005. i24/e243601.
Krantz, P., M. Kjaergaard, F. Yan, T. P. Orlando, S. Gustavs- Lang, C., C. Eichler, L. Steffen, J. M. Fink, M. J. Woolley,
son, and W. D. Oliver, 2019, Applied Physics Reviews 6(2), A. Blais, and A. Wallraff, 2013, Nat. Phys. 9(6), 345, ISSN
021318. 1745-2473, URL http://www.nature.com/nphys/journal/
Krastanov, S., V. V. Albert, C. Shen, C.-L. Zou, R. W. Heeres, v9/n6/full/nphys2612.html.
B. Vlastakis, R. J. Schoelkopf, and L. Jiang, 2015, Phys. Lauk, N., N. Sinclair, J. Covey, M. Saffman, M. Spiropulu,
Rev. A 92, 040303, URL http://link.aps.org/doi/10. C. Simon, et al., 2020, Quantum Science and Technology .
1103/PhysRevA.92.040303. Law, C. K., and J. H. Eberly, 1996, Phys. Rev. Lett. 76, 1055,
Krinner, S., S. Storz, P. Kurpiers, P. Magnard, J. Heinsoo, URL https://link.aps.org/doi/10.1103/PhysRevLett.
R. Keller, J. Lütolf, C. Eichler, and A. Wallraff, 2019, EPJ 76.1055.
Quantum Technology 6(1), 2. Lecocq, F., L. Ranzani, G. A. Peterson, K. Cicak, R. W.
Ku, J., X. Xu, M. Brink, D. C. McKay, J. B. Hertzberg, Simmonds, J. D. Teufel, and J. Aumentado, 2017, Phys.
M. H. Ansari, and B. L. T. Plourde, 2020, arXiv e-prints , Rev. Applied 7, 024028, URL https://link.aps.org/doi/
arXiv:2003.02775eprint 2003.02775. 10.1103/PhysRevApplied.7.024028.
Kubo, Y., F. R. Ong, P. Bertet, D. Vion, V. Jacques, D. Zheng, Leek, P. J., M. Baur, J. M. Fink, R. Bianchetti, L. Steffen,
A. Dréau, J.-F. Roch, A. Auffeves, F. Jelezko, J. Wrachtrup, S. Filipp, and A. Wallraff, 2010, Phys. Rev. Lett. 104(10),
M. F. Barthe, et al., 2010, Phys. Rev. Lett. 105(14), 140502. 100504, URL http://prl.aps.org/abstract/PRL/v104/
Kuhr, S., S. Gleyzes, C. Guerlin, J. Bernu, U. B. Hoff, i10/e100504.
S. Delèglise, S. Osnaghi, M. Brune, J.-M. Raimond, Leek, P. J., S. Filipp, P. Maurer, M. Baur, R. Bianchetti, J. M.
S. Haroche, E. Jacques, P. Bosland, et al., 2007, Applied Fink, M. Göppl, L. Steffen, and A. Wallraff, 2009, Phys.
Physics Letters 90(16), 164101, URL https://doi.org/10. Rev. B 79, 180511, URL http://link.aps.org/abstract/
1063/1.2724816. PRB/v79/e180511.
Kurpiers, P., P. Magnard, T. Walter, B. Royer, M. Pechal, Leggett, A. J., 1980, Progress of Theoretical Physics Supple-
J. Heinsoo, Y. Salathé, A. Akin, S. Storz, J. C. Besse, ment 69, 80.
S. Gasparinetti, A. Blais, et al., 2018, Nature 558(7709), Leggett, A. J., 1984a, in Essays in theoretical physics. In
264. honour of Dirk ter Haar., edited by W. E. Parry (Pergamon,
Kurpiers, P., T. Walter, P. Magnard, Y. Salathe, and A. Wall- Oxford), pp. 95–127.
raff, 2017, EPJ Quantum Technology 4(1), 8, URL http: Leggett, A. J., 1984b, Phys. Rev. B 30, 1208, URL http:
//dx.doi.org/10.1140/epjqt/s40507-017-0059-7. //link.aps.org/doi/10.1103/PhysRevB.30.1208.
Kyriienko, O., and A. S. Sorensen, 2016, Physical Review Lehmberg, R. H., 1970, Phys. Rev. A 2(3), 883, URL http:
Letters 117(14), 140503, ISSN 1079-7114, URL http://dx. //link.aps.org/doi/10.1103/PhysRevA.2.883.
doi.org/10.1103/PhysRevLett.117.140503. Leonhardt, U., 1997, Measuring the Quantum State of Light
Lachance-Quirion, D., Y. Tabuchi, A. Gloppe, K. Usami, and (Cambridge University Press, United Kingdom).
Y. Nakamura, 2019, Applied Physics Express 12(7), 070101. Leonhardt, U., and H. Paul, 1993, Phys. Rev. A 48, 4598.
Lachance-Quirion, D., Y. Tabuchi, S. Ishino, A. Noguchi, Leppäkangas, J., M. Marthaler, D. Hazra, S. Jebari, R. Albert,
T. Ishikawa, R. Yamazaki, and Y. Nakamura, 2017, Science F. Blanchet, G. Johansson, and M. Hofheinz, 2018, Phys.
Advances 3(7), URL http://advances.sciencemag.org/ Rev. A 97, 013855, URL https://link.aps.org/doi/10.
content/3/7/e1603150. 1103/PhysRevA.97.013855.
Laflamme, R., C. Miquel, J. P. Paz, and W. H. Zurek, 1996, Lescanne, R., S. Deléglise, E. Albertinale, U. Réglade,
 76

T. Capelle, E. Ivanov, T. Jacqmin, Z. Leghtas, and E. Flurin, Lehnert, 2011, Phys. Rev. Lett. 106(22), 220502.
2019a, arXiv e-prints , arXiv:1902.05102eprint 1902.05102. Mallet, F., F. R. Ong, A. Palacios-Laloy, F. Nguyen, P. Bertet,
Lescanne, R., M. Villiers, T. Peronnin, A. Sarlette, M. Delbecq, D. Vion, and D. Esteve, 2009, Nat. Phys. 5, 791, ISSN
B. Huard, T. Kontos, M. Mirrahimi, and Z. Leghtas, 2019b, 1745-2481, URL http://dx.doi.org/10.1038/nphys1400.
arXiv e-prints , arXiv:1907.11729eprint 1907.11729. Malnou, M., D. A. Palken, L. R. Vale, G. C. Hilton, and K. W.
Leung, D. W., M. A. Nielsen, I. L. Chuang, and Y. Yamamoto, Lehnert, 2018, Phys. Rev. Applied 9, 044023, URL https://
1997, Phys. Rev. A 56(4), 2567, URL http://link.aps. link.aps.org/doi/10.1103/PhysRevApplied.9.044023.
org/doi/10.1103/PhysRevA.56.2567. Manucharyan, V. E., E. Boaknin, M. Metcalfe, R. Vijay, I. Sid-
Leung, N., Y. Lu, S. Chakram, R. K. Naik, N. Earnest, R. Ma, diqi, and M. Devoret, 2007, Phys. Rev. B 76, 014524.
K. Jacobs, A. N. Cleland, and D. I. Schuster, 2019, npj Manucharyan, V. E., J. Koch, L. I. Glazman, and M. H.
Quantum Information 5(1), 18. Devoret, 2009, Science 326(5949), 113, URL http://www.
Liu, L., F.-Z. Guo, S.-J. Qin, and Q.-Y. Wen, 2017, Scien- sciencemag.org/cgi/content/abstract/326/5949/113.
tific Reports 7, 42261, URL http://dx.doi.org/10.1038/ Mariantoni, M., E. P. Menzel, F. Deppe, M. A. Araque Ca-
srep42261. ballero, A. Baust, T. Niemczyk, E. Hoffmann, E. Solano,
Liu, Y., and A. A. Houck, 2016, Nat Phys advance online A. Marx, and R. Gross, 2010, Phys. Rev. Lett. 105(13),
publication, , ISSN 1745-2481, URL http://dx.doi.org/ 133601.
10.1038/nphys3834. Mariantoni, M., H. Wang, R. C. Bialczak, M. Lenander,
Liu, Y.-x., L. Wei, J. Johansson, J. Tsai, and F. Nori, 2007, E. Lucero, M. Neeley, A. D. O’Connell, D. Sank, M. Weides,
Physical Review B 76(14), 144518. J. Wenner, T. Yamamoto, Y. Yin, et al., 2011a, Nat. Phys.
van Loo, A., A. Fedorov, K. Lalumière, B. Sanders, 7(4), 287, ISSN 1745-2473, URL http://dx.doi.org/10.
A. Blais, and A. Wallraff, 2013, Science 342(6165), 1494, 1038/nphys1885.
URL http://www.sciencemag.org/content/early/2013/ Mariantoni, M., H. Wang, T. Yamamoto, M. Neeley, R. C. Bial-
11/13/science.1244324.abstract. czak, Y. Chen, M. Lenander, E. Lucero, A. D. O’Connell,
Loss, D., and D. P. DiVincenzo, 1998, Phys. Rev. A 57(1), 120, D. Sank, M. Weides, J. Wenner, et al., 2011b, Sci-
URL http://link.aps.org/abstract/PRA/v57/p120. ence 334(6052), 61, URL http://www.sciencemag.org/
Luthi, F., T. Stavenga, O. W. Enzing, A. Bruno, C. Dickel, content/334/6052/61.abstract.
N. K. Langford, M. A. Rol, T. S. Jespersen, J. Nygård, Marquardt, F., and C. Bruder, 2001, Phys. Rev. B 63(5),
P. Krogstrup, and L. DiCarlo, 2018, Phys. Rev. Lett. 054514.
120, 100502, URL https://link.aps.org/doi/10.1103/ Martinis, J. M., M. H. Devoret, and J. Clarke, 1985, Phys.
PhysRevLett.120.100502. Rev. Lett. 55, 1543, URL http://link.aps.org/doi/10.
Ma, R., B. Saxberg, C. Owens, N. Leung, Y. Lu, J. Simon, 1103/PhysRevLett.55.1543.
and D. I. Schuster, 2019, Nature 566(7742), 51. Martinis, J. M., S. Nam, J. Aumentado, K. Lang, and U. C.,
Ma, W.-L., M. Zhang, Y. Wong, K. Noh, S. Rosenblum, 2003, Phys. Rev. B 67, 094510, URL http://journals.
P. Reinhold, R. J. Schoelkopf, and L. Jiang, 2019, arXiv aps.org/prb/abstract/10.1103/PhysRevB.67.094510.
e-prints , arXiv:1911.12240eprint 1911.12240. Martinis, J. M., S. Nam, J. Aumentado, and C. Urbina, 2002,
Ma, Y., Y. Xu, X. Mu, W. Cai, L. Hu, W. Wang, X. Pan, Phys. Rev. Lett. 89(11), 117901, URL http://link.aps.
H. Wang, Y. P. Song, C. L. Zou, and L. Sun, 2020, Nature org/doi/10.1103/PhysRevLett.89.117901.
Physics ISSN 1745-2481, URL https://doi.org/10.1038/ Maser, A., B. Gmeiner, T. Utikal, S. Götzinger,
s41567-020-0893-x. and V. Sandoghdar, 2016, Nat. Photon. 10, URL
Macklin, C., K. O’Brien, D. Hover, M. E. Schwartz, http://www.nature.com/nphoton/journal/vaop/
V. Bolkhovsky, X. Zhang, W. D. Oliver, and I. Siddiqi, 2015, ncurrent/full/nphoton.2016.63.html.
Science 350(6258), 307, URL http://www.sciencemag. McDonald, D. G., 2001, Physics Today 54, 46.
org/content/350/6258/307.abstract. McKay, D. C., S. Filipp, A. Mezzacapo, E. Magesan,
Magesan, E., A. D. Gambetta, J. M. and. Córcoles, J. M. Chow, and J. M. Gambetta, 2016, Phys. Rev. Ap-
and J. M. Chow, 2015, Phys. Rev. Lett. 114, 200501, plied 6, 064007, URL http://link.aps.org/doi/10.1103/
URL http://journals.aps.org/prl/abstract/10.1103/ PhysRevApplied.6.064007.
PhysRevLett.114.200501. McKay, D. C., C. J. Wood, S. Sheldon, J. M. Chow, and J. M.
Magesan, E., and J. M. Gambetta, 2018, ArXiv e-prints eprint Gambetta, 2017, Phys. Rev. A 96, 022330, URL https:
1804.04073. //link.aps.org/doi/10.1103/PhysRevA.96.022330.
Majer, J., J. M. Chow, J. M. Gambetta, J. Koch, B. R. Megrant, A., C. Neill, R. Barends, B. Chiaro, Y. Chen, L. Feigl,
Johnson, J. A. Schreier, L. Frunzio, D. I. Schuster, A. A. J. Kelly, E. Lucero, M. Mariantoni, P. J. J. O’Malley,
Houck, A. Wallraff, A. Blais, M. H. Devoret, et al., 2007, D. Sank, A. Vainsencher, et al., 2012, arXiv:1201.3384 URL
Nature 449(7161), 443, ISSN 0028-0836, URL http://dx. http://arXiv.org/abs/1201.3384.
doi.org/10.1038/nature06184. Menzel, E. P., F. Deppe, M. Mariantoni, M. A. Araque Ca-
Makhlin, Y., G. Schön, and A. Shnirman, 2001, Rev. Mod. ballero, A. Baust, T. Niemczyk, E. Hoffmann, A. Marx,
Phys. 73(2), 357. E. Solano, and R. Gross, 2010, Phys. Rev. Lett. 105(10),
Makhlin, Y., G. Scohn, and A. Shnirman, 1999, Nature 100401.
398(6725), 305, ISSN 0028-0836, URL http://dx.doi.org/ Metcalfe, M., E. Boaknin, V. Manucharyan, R. Vijay, I. Siddiqi,
10.1038/18613. C. Rigetti, L. Frunzio, R. J. Schoelkopf, and M. H. Devoret,
Malekakhlagh, M., A. Petrescu, and H. E. Türeci, 2017, Phys. 2007, Phys. Rev. B 76(17), 174516, URL http://link.aps.
Rev. Lett. 119, 073601, URL https://link.aps.org/doi/ org/abstract/PRB/v76/e174516.
10.1103/PhysRevLett.119.073601. Metelmann, A., and A. A. Clerk, 2015, Phys. Rev.
Mallet, F., M. A. Castellanos-Beltran, H. S. Ku, S. Glancy, X 5, 021025, URL http://link.aps.org/doi/10.1103/
E. Knill, K. D. Irwin, G. C. Hilton, L. R. Vale, and K. W. PhysRevX.5.021025.
 77

Mi, X., M. Benito, S. Putz, D. M. Zajac, J. M. Taylor, 398(6730), 786.

G. Burkard, and J. R. Petta, 2018, Nature URL http: Narla, A., S. Shankar, M. Hatridge, Z. Leghtas, K. M. Sliwa,
//dx.doi.org/10.1038/nature25769. E. Zalys-Geller, S. O. Mundhada, W. Pfaff, L. Frunzio,
Mi, X., J. V. Cady, D. M. Zajac, P. W. Deelman, and R. J. Schoelkopf, and M. H. Devoret, 2016, Phys. Rev.
J. R. Petta, 2017, Science 355(6321), 156, ISSN 0036- X 6, 031036, URL http://link.aps.org/doi/10.1103/
8075, URL http://science.sciencemag.org/content/ PhysRevX.6.031036.
355/6321/156. Nersisyan, A., S. Poletto, N. Alidoust, R. Manenti, R. Renzas,
Michael, M. H., M. Silveri, R. T. Brierley, V. V. Albert, C.-V. Bui, K. Vu, T. Whyland, Y. Mohan, E. A. Sete,
J. Salmilehto, L. Jiang, and S. M. Girvin, 2016a, Phys. Rev. S. Stanwyck, A. Bestwick, et al., 2019, arXiv e-prints ,
X 6, 031006, URL http://link.aps.org/doi/10.1103/ arXiv:1901.08042eprint 1901.08042.
PhysRevX.6.031006. Nielsen, M. A., and I. L. Chuang, 2000, Quantum Computation
Michael, M. H., M. Silveri, R. T. Brierley, V. V. Al- and Quantum Information (Cambridge University Press).
bert, J. Salmilehto, L. Jiang, and S. M. Girvin, Niemczyk, T., F. Deppe, H. Huebl, E. P. Menzel, F. Hocke,
2016b, arXiv:1602.00008 URL http://arxiv.org/abs/ M. J. Schwarz, J. J. Garcia-Ripoll, D. Zueco, T. Hümmer,
1602.00008. E. Solano, A. Marx, and R. Gross, 2010, Nat. Phys. 6, 772 .
Minev, Z. K., 2019, Catching and Reversing a Quantum Jump Nigg, S. E., H. Paik, B. Vlastakis, G. Kirchmair, S. Shankar,
Mid-Flight, Ph.D. thesis, Yale University. L. Frunzio, M. H. Devoret, R. J. Schoelkopf, and S. M.
Minev, Z. K., S. O. Mundhada, S. Shankar, P. Reinhold, Girvin, 2012, Phys. Rev. Lett. 108, 240502, URL http:
R. Gutiérrez-Jáuregui, R. J. Schoelkopf, M. Mirrahimi, H. J. //link.aps.org/doi/10.1103/PhysRevLett.108.240502.
Carmichael, and M. H. Devoret, 2019, Nature 570(7760), Niskanen, A. O., K. Harrabi, F. Yoshihara, Y. Nakamura,
200. S. Lloyd, and J. S. Tsai, 2007, Science 316(5825), 723.
Mirhosseini, M., E. Kim, V. S. Ferreira, M. Kalaee, Niskanen, A. O., Y. Nakamura, and J.-S. Tsai, 2006, Physical
A. Sipahigil, A. J. Keller, and O. Painter, 2018, Nature Review B 73(9), 094506.
Communications 9(1), 3706. O’Brien, K., C. Macklin, I. Siddiqi, and X. Zhang, 2014, Phys.
Mirhosseini, M., E. Kim, X. Zhang, A. Sipahigil, P. B. Dieterle, Rev. Lett. 113, 157001, URL http://link.aps.org/doi/
A. J. Keller, A. Asenjo-Garcia, D. E. Chang, and O. Painter, 10.1103/PhysRevLett.113.157001.
2019, Nature 569(7758), 692. Oelsner, G., C. K. Andersen, M. Rehák, M. Schmelz, S. Anders,
Mirrahimi, M., Z. Leghtas, V. V. Albert, S. Touzard, R. J. M. Grajcar, U. Hübner, K. Mølmer, and E. Il’ichev, 2017,
Schoelkopf, L. Jiang, and M. H. Devoret, 2014, New Journal Phys. Rev. Applied 7, 014012, URL http://link.aps.org/
of Physics 16(4), 045014, URL http://stacks.iop.org/ doi/10.1103/PhysRevApplied.7.014012.
1367-2630/16/i=4/a=045014. Ofek, N., A. Petrenko, R. Heeres, P. Reinhold, Z. Leghtas,
Mlynek, J., A. Abdumalikov, C. Eichler, and A. Wall- B. Vlastakis, Y. Liu, L. Frunzio, S. M. Girvin, L. Jiang,
raff, 2014, Nat. Commun. 5, 5186, URL http: M. Mirrahimi, M. H. Devoret, et al., 2016, Nature 536,
//www.nature.com/ncomms/2014/141104/ncomms6186/ 441, ISSN 1476-4687, URL http://dx.doi.org/10.1038/
full/ncomms6186.html. nature18949.
Monroe, C., R. Raussendorf, A. Ruthven, K. R. Brown, Oliver, W. D., and P. B. Welander, 2013, MRS bulletin 38(10),
P. Maunz, L.-M. Duan, and J. Kim, 2014, Phys. Rev. 816.
A 89, 022317, URL https://link.aps.org/doi/10.1103/ O’Malley, P. J. J., R. Babbush, I. D. Kivlichan, J. Romero,
PhysRevA.89.022317. J. R. McClean, R. Barends, J. Kelly, P. Roushan, A. Tran-
Mooij, J. E., T. Orlando, L. Levitov, L. Tian, C. H. ter, N. Ding, B. Campbell, Y. Chen, et al., 2016, Phys. Rev.
van der Wal, and S. Lloyd, 1999, Science 285(5430), X 6, 031007, URL http://link.aps.org/doi/10.1103/
1036, URL http://www.sciencemag.org/cgi/content/ PhysRevX.6.031007.
abstract/285/5430/1036. Ong, F. R., M. Boissonneault, F. Mallet, A. C. Doherty,
Motzoi, F., J. M. Gambetta, P. Rebentrost, and F. K. Wilhelm, A. Blais, D. Vion, D. Esteve, and P. Bertet, 2013, Phys.
2009, Phys. Rev. Lett. 103(11), 110501 (pages 4), URL Rev. Lett. 110, 047001, URL https://link.aps.org/doi/
http://link.aps.org/abstract/PRL/v103/e110501. 10.1103/PhysRevLett.110.047001.
Movshovich, R., B. Yurke, P. G. Kaminsky, A. D. Smith, A. H. Opremcak, A., I. V. Pechenezhskiy, C. Howington, B. G.
Silver, R. W. Simon, and M. V. Schneider, 1990, Phys. Christensen, M. A. Beck, E. Leonard, J. Suttle, C. Wilen,
Rev. Lett. 65, 1419, URL https://link.aps.org/doi/10. K. N. Nesterov, G. J. Ribeill, T. Thorbeck, F. Schlenker,
1103/PhysRevLett.65.1419. et al., 2018, Science 361(6408), 1239.
Müller, C., J. H. Cole, and J. Lisenfeld, 2019, Reports on Orlando, T. P., J. E. Mooij, L. Tian, C. H. van der Wal,
Progress in Physics 82(12), 124501. L. S. Levitov, S. Lloyd, and J. J. Mazo, 1999, Phys. Rev. B
Murch, K. W., S. J. Weber, K. M. Beck, E. Ginossar, and 60(22), 15398, URL http://link.aps.org/doi/10.1103/
I. Siddiqi, 2013a, Nature , 62URL http://arxiv.org/abs/ PhysRevB.60.15398.
1301.6276. Paik, H., A. Mezzacapo, M. Sandberg, D. McClure, B. Abdo,
Murch, K. W., S. J. Weber, C. Macklin, and I. Sid- A. Corcoles, O. Dial, D. Bogorin, B. Plourde, M. Steffen,
diqi, 2013b, Nature 502(7470), 211, ISSN 0028- A. Cross, J. Gambetta, et al., 2016, Physical Review Letters
0836, URL http://www.nature.com/nature/journal/ 117(25), 250502, ISSN 1079-7114, URL http://dx.doi.
v502/n7470/full/nature12539.html. org/10.1103/PhysRevLett.117.250502.
Naik, R. K., N. Leung, S. Chakram, P. Groszkowski, Y. Lu, Paik, H., D. I. Schuster, L. S. Bishop, G. Kirchmair, G. Cate-
N. Earnest, D. C. McKay, J. Koch, and D. I. Schus- lani, A. P. Sears, B. R. Johnson, M. J. Reagor, L. Frunzio,
ter, 2017, Nature Communications 8, 1904, URL https: L. I. Glazman, S. M. Girvin, M. H. Devoret, et al., 2011,
//arxiv.org/abs/1705.00579. Phys. Rev. Lett. 107, 240501, URL http://link.aps.org/
Nakamura, Y., Y. A. Pashkin, and J. S. Tsai, 1999, Nature doi/10.1103/PhysRevLett.107.240501.
 78

Palomaki, T. A., J. W. Harlow, J. D. Teufel, R. W. Sim- Puri, S., S. Boutin, and A. Blais, 2017, npj Quantum Informa-
monds, and K. W. Lehnert, 2013a, Nature 495(7440), tion 3(1), 18, ISSN 2056-6387, URL https://www.nature.
210, ISSN 0028-0836, URL http://dx.doi.org/10.1038/ com/articles/s41534-017-0019-1.
nature11915. Puri, S., L. St-Jean, J. A. Gross, A. Grimm, N. E. Frattini,
Palomaki, T. A., J. D. Teufel, R. W. Simmonds, and K. W. P. S. Iyer, A. Krishna, S. Touzard, L. Jiang, A. Blais,
Lehnert, 2013b, Science , URL http://www.sciencemag. S. T. Flammia, and S. M. Girvin, 2019, arXiv e-prints ,
org/content/early/2013/10/02/science.1244563. arXiv:1905.00450eprint 1905.00450.
Pashkin, Y. A., T. Yamamoto, O. Astafiev, Y. Nakamura, Ralph, T. C., A. Gilchrist, G. J. Milburn, W. J. Munro, and
D. V. Averin, and J. S. Tsai, 2003, Nature 421(6925), 823. S. Glancy, 2003, Physical Review A 68(4), 042319.
Pechal, M., L. Huthmacher, C. Eichler, S. Zeytinoğlu, A. A. Rau, I., G. Johansson, and A. Shnirman, 2004, Phys. Rev. B
Abdumalikov Jr., S. Berger, A. Wallraff, and S. Filipp, 2014, 70(5), 054521, URL http://link.aps.org/abstract/PRB/
Phys. Rev. X 4, 041010, URL https://link.aps.org/doi/ v70/e054521.
10.1103/PhysRevX.4.041010. Reagor, M., C. B. Osborn, N. Tezak, A. Staley, G. Prawiroat-
Petersson, K. D., L. W. McFaul, M. D. Schroer, M. Jung, modjo, M. Scheer, N. Alidoust, E. A. Sete, N. Didier, M. P.
J. M. Taylor, A. A. Houck, and J. R. Petta, 2012, Nature da Silva, E. Acala, J. Angeles, et al., 2018, Science Advances
490(7420), 380, ISSN 0028-0836, URL http://dx.doi.org/ 4(2), URL http://advances.sciencemag.org/content/4/
10.1038/nature11559. 2/eaao3603.
Pfaff, W., C. J. Axline, L. D. Burkhart, U. Vool, P. Reinhold, Reagor, M., W. Pfaff, C. Axline, R. W. Heeres, N. Ofek,
L. Frunzio, L. Jiang, M. H. Devoret, and R. J. Schoelkopf, K. Sliwa, E. Holland, C. Wang, J. Blumoff, K. Chou,
2017, Nat. Phys. URL https://www.nature.com/nphys/ M. J. Hatridge, L. Frunzio, et al., 2016, Phys. Rev.
journal/vaop/ncurrent/full/nphys4143.html. B 94, 014506, URL http://link.aps.org/doi/10.1103/
Pietikäinen, I., S. Danilin, K. S. Kumar, A. Vepsäläinen, D. S. PhysRevB.94.014506.
Golubev, J. Tuorila, and G. S. Paraoanu, 2017, Phys. Rev. Reagor, M. J., 2015, Superconducting Cavities for Circuit
B 96, 020501, URL https://link.aps.org/doi/10.1103/ Quantum Electrodynamics, Ph.D. thesis, Yale University.
PhysRevB.96.020501. Reed, M. D., L. DiCarlo, B. R. Johnson, L. Sun, D. I. Schuster,
Pioro-Ladrière, M., T. Obata, Y. Tokura, Y. S. Shin, T. Kubo, L. Frunzio, and R. J. Schoelkopf, 2010a, Phys. Rev. Lett.
K. Yoshida, T. Taniyama, and S. Tarucha, 2008, Nature 105(17), 173601.
Physics 4(10), 776. Reed, M. D., B. R. Johnson, A. A. Houck, L. DiCarlo, J. M.
Place, A. P. M., L. V. H. Rodgers, P. Mundada, B. M. Chow, D. I. Schuster, L. Frunzio, and R. J. Schoelkopf,
Smitham, M. Fitzpatrick, Z. Leng, A. Premkumar, J. Bryon, 2010b, Appl. Phys. Lett. 96(20), 203110 (pages 3), URL
S. Sussman, G. Cheng, T. Madhavan, H. K. Babla, et al., http://link.aip.org/link/?APL/96/203110/1.
2020, arXiv:2003.07300 . Reinhold, P., S. Rosenblum, W.-L. Ma, L. Frunzio,
Planat, L., R. Dassonneville, J. P. Martı́nez, F. Foroughi, L. Jiang, and R. J. Schoelkopf, 2019, arXiv e-prints ,
O. Buisson, W. Hasch-Guichard, C. Naud, R. Vijay, arXiv:1907.12327eprint 1907.12327.
K. Murch, and N. Roch, 2019, Phys. Rev. Applied Rempe, G., H. Walther, and N. Klein, 1987, Phys. Rev. Lett.
11, 034014, URL https://link.aps.org/doi/10.1103/ 58(4), 353.
PhysRevApplied.11.034014. Rice, P., and R. Brecha, 1996, Optics communications 126(4),
Planat, L., A. Ranadive, R. Dassonneville, J. Puertas Martı́nez, 230.
S. Léger, C. Naud, O. Buisson, W. Hasch-Guichard, D. M. Richer, S., and D. DiVincenzo, 2016, Phys. Rev. B
Basko, and N. Roch, 2020, Phys. Rev. X 10, 021021, 93, 134501, URL http://link.aps.org/doi/10.1103/
URL https://link.aps.org/doi/10.1103/PhysRevX.10. PhysRevB.93.134501.
021021. Richer, S., N. Maleeva, S. T. Skacel, I. M. Pop, and D. Di-
Plastina, F., and G. Falci, 2003, Phys. Rev. B 67(22), 224514. Vincenzo, 2017, Phys. Rev. B 96, 174520, URL https:
Poletto, S., J. M. Gambetta, S. T. Merkel, J. A. Smolin, J. M. //link.aps.org/doi/10.1103/PhysRevB.96.174520.
Chow, A. D. Córcoles, G. A. Keefe, M. B. Rothwell, J. R. Rigetti, C., and M. Devoret, 2010, Phys. Rev. B 81(13),
Rozen, D. W. Abraham, C. Rigetti, and M. Steffen, 2012, 134507.
Phys. Rev. Lett. 109, 240505, URL http://link.aps.org/ Rigetti, C., J. M. Gambetta, S. Poletto, B. L. T. Plourde,
doi/10.1103/PhysRevLett.109.240505. J. M. Chow, A. D. Córcoles, J. A. Smolin, S. T. Merkel,
Potočnik, A., A. Bargerbos, F. A. Y. N. Schröder, S. A. J. R. Rozen, G. A. Keefe, M. B. Rothwell, M. B. Ketchen,
Khan, M. C. Collodo, S. Gasparinetti, Y. Salathé, et al., 2012, Phys. Rev. B 86(10), 100506, URL http://
C. Creatore, C. Eichler, H. E. Türeci, A. W. Chin, link.aps.org/doi/10.1103/PhysRevB.86.100506.
and A. Wallraff, 2018, Nature Communications 9(1), Ristè, D., M. Dukalski, C. A. Watson, G. de Lange,
904, ISSN 2041-1723, URL https://www.nature.com/ M. J. Tiggelman, Y. M. Blanter, K. W. Lehnert, R. N.
articles/s41467-018-03312-x. Schouten, and L. DiCarlo, 2013, Nature 502(7471), 350,
Pozar, D. M., 2011, Microwave Engineering (John Wiley & ISSN 0028-0836, URL http://www.nature.com/nature/
Sons, Inc.), 4th ed. edition. journal/v502/n7471/full/nature12513.html.
Pozar, D. M., 2012, Microwave engineering (Wiley & Sons, Ristè, D., S. Poletto, M.-Z. Huang, A. Bruno, V. Vesterinen,
Hoboken), ISBN 978-0-470-63155-3. O.-P. Saira, and L. DiCarlo, 2015, Nat Commun 6, 6983,
Preskill, J., 2018, Quantum 2, 79, ISSN 2521-327X, URL URL http://dx.doi.org/10.1038/ncomms7983.
https://doi.org/10.22331/q-2018-08-06-79. Roch, N., M. E. Schwartz, F. Motzoi, C. Macklin, R. Vi-
Purcell, E. M., 1946, Phys. Rev. 69, 681. jay, A. W. Eddins, A. N. Korotkov, K. B. Whaley,
Puri, S., and A. Blais, 2016, Phys. Rev. Lett. 116, 180501, M. Sarovar, and I. Siddiqi, 2014, Phys. Rev. Lett.
URL https://link.aps.org/doi/10.1103/PhysRevLett. 112, 170501, URL http://link.aps.org/doi/10.1103/
116.180501. PhysRevLett.112.170501.
 79

Rol, M. A., F. Battistel, F. K. Malinowski, C. C. Bultink, B. M. andp.201200261.

Tarasinski, R. Vollmer, N. Haider, N. Muthusubramanian, Schneider, A., J. Braumüller, L. Guo, P. Stehle, H. Rotzinger,
A. Bruno, B. M. Terhal, and L. DiCarlo, 2019, Phys. Rev. M. Marthaler, A. V. Ustinov, and M. Weides, 2018, Phys.
Lett. 123, 120502, URL https://link.aps.org/doi/10. Rev. A 97, 062334, URL https://link.aps.org/doi/10.
1103/PhysRevLett.123.120502. 1103/PhysRevA.97.062334.
Rol, M. A., L. Ciorciaro, F. K. Malinowski, B. M. Tarasinski, Schoelkopf, R., A. Clerk, S. Girvin, K. Lehnert, and M. De-
R. E. Sagastizabal, C. C. Bultink, Y. Salathe, N. Haandbaek, voret, 2003, Quantum noise in mesoscopic physics (Kluwer
J. Sedivy, and L. DiCarlo, 2020, Applied Physics Letters Ac. Publ.), chapter Qubits as Spectrometers of Quantum
116(5), 054001. Noise, pp. 175–203.
Romanenko, A., R. Pilipenko, S. Zorzetti, D. Frolov, M. Awida, Schoelkopf, R. J., and S. M. Girvin, 2008, Nature 451, 664.
S. Belomestnykh, S. Posen, and A. Grassellino, 2020, Phys. Schreier, J. A., A. A. Houck, J. Koch, D. I. Schuster, B. R.
Rev. Applied 13, 034032, URL https://link.aps.org/ Johnson, J. M. Chow, J. M. Gambetta, J. Majer, L. Frunzio,
doi/10.1103/PhysRevApplied.13.034032. M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, 2008,
Romero, G., J. J. Garcı́a-Ripoll, and E. Solano, 2009, Phys. Phys. Rev. B 77, 180502.
Rev. Lett. 102(17), 173602. Schrieffer, J. R., and P. A. Wolff, 1966, Phys. Rev. 149,
Rosenblum, S., P. Reinhold, M. Mirrahimi, L. Jiang, L. Frun- 491, URL https://link.aps.org/doi/10.1103/PhysRev.
zio, and R. J. Schoelkopf, 2018, Science 361(6399), 266. 149.491.
Roy, A., and M. Devoret, 2016, Comptes Ren- Schuller, J. A., E. S. Barnard, W. Cai, Y. C. Jun, J. S. White,
dus Physique 17(7), 740 , ISSN 1631-0705, URL and M. L. Brongersma, 2010, Nature Materials 9(3), 193,
https://www.sciencedirect.com/science/article/pii/ URL https://doi.org/10.1038/nmat2630.
S1631070516300640. Schuster, D. I., 2007, Circuit Quantum Electrodynamics, Ph.D.
Roy, D., C. M. Wilson, and O. Firstenberg, 2017, Rev. Mod. thesis, Yale University.
Phys. 89, 021001, URL https://link.aps.org/doi/10. Schuster, D. I., A. A. Houck, J. A. Schreier, A. Wallraff,
1103/RevModPhys.89.021001. J. M. Gambetta, A. Blais, L. Frunzio, J. Majer, B. Johnson,
Roy, T., S. Kundu, M. Chand, A. M. Vadiraj, A. Ranadive, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, 2007,
N. Nehra, M. P. Patankar, J. Aumentado, A. A. Clerk, and Nature 445(7127), 515, ISSN 0028-0836, URL http://dx.
R. Vijay, 2015, Appl. Phys. Lett. 107(26), 262601, URL doi.org/10.1038/nature05461.
http://dx.doi.org/10.1063/1.4939148. Schuster, D. I., A. P. Sears, E. Ginossar, L. DiCarlo, L. Frunzio,
Royer, B., A. L. Grimsmo, A. Choquette-Poitevin, and J. J. L. Morton, H. Wu, G. A. D. Briggs, B. B. Buckley,
A. Blais, 2018, Phys. Rev. Lett. 120, 203602, URL https: D. D. Awschalom, and R. J. Schoelkopf, 2010, Phys. Rev.
//link.aps.org/doi/10.1103/PhysRevLett.120.203602. Lett. 105(14), 140501.
Royer, B., A. L. Grimsmo, N. Didier, and A. Blais, 2017, Quan- Schuster, D. I., A. Wallraff, A. Blais, L. Frunzio, R.-S. Huang,
tum 1, 11, URL http://quantum-journal.org/papers/ J. Majer, S. M. Girvin, and R. J. Schoelkopf, 2005, Phys.
q-2017-05-11-11/. Rev. Lett. 94(12), 123602, URL http://link.aps.org/
Ryan, C. A., B. R. Johnson, J. M. Gambetta, J. M. Chow, abstract/PRL/v94/e123602.
M. P. da Silva, O. E. Dial, and T. A. Ohki, 2015, Phys. Rev. Sete, E. A., A. Galiautdinov, E. Mlinar, J. M. Martinis, and
A 91, 022118, URL http://link.aps.org/doi/10.1103/ A. N. Korotkov, 2013, Phys. Rev. Lett. 110(21), 210501.
PhysRevA.91.022118. Sete, E. A., J. M. Gambetta, and A. N. Korotkov, 2014, Phys.
Sage, J. M., V. Bolkhovsky, W. D. Oliver, B. Turek, and P. B. Rev. B 89, 104516, URL http://journals.aps.org/prb/
Welander, 2011, Journal of Applied Physics 109(6), 063915, abstract/10.1103/PhysRevB.89.104516.
URL https://doi.org/10.1063/1.3552890. Shalm, L. K., E. Meyer-Scott, B. G. Christensen, P. Bierhorst,
Salathé, Y., P. Kurpiers, T. Karg, C. Lang, C. K. Andersen, M. A. Wayne, M. J. Stevens, T. Gerrits, S. Glancy, D. R.
A. Akin, S. Krinner, C. Eichler, and A. Wallraff, 2018, Phys. Hamel, M. S. Allman, K. J. Coakley, S. D. Dyer, et al., 2015,
Rev. Applied 9, 034011, URL https://link.aps.org/doi/ Phys. Rev. Lett. 115, 250402, URL http://link.aps.org/
10.1103/PhysRevApplied.9.034011. doi/10.1103/PhysRevLett.115.250402.
Samkharadze, N., G. Zheng, N. Kalhor, D. Brousse, A. Sam- Sheldon, S., E. Magesan, J. M. Chow, and J. M. Gambetta,
mak, U. C. Mendes, A. Blais, G. Scappucci, and L. M. K. 2016, Phys. Rev. A 93, 060302, URL http://link.aps.
Vandersypen, 2018, Science 359(6380), 1123. org/doi/10.1103/PhysRevA.93.060302.
Sandberg, M., C. M. Wilson, F. Persson, T. Bauch, G. Jo- Shen, J.-T., and S. Fan, 2005, Phys. Rev. Lett.
hansson, V. Shumeiko, T. Duty, and P. Delsing, 2008, Appl. 95(21), 213001, URL http://link.aps.org/doi/10.1103/
Phys. Lett. 92(20), 203501, URL http://link.aip.org/ PhysRevLett.95.213001.
link/?APL/92/203501/1. Shnirman, A., G. Schön, and Z. Hermon, 1997, Phys. Rev.
Sank, D., Z. Chen, M. Khezri, J. Kelly, R. Barends, B. Camp- Lett. 79(12), 2371.
bell, Y. Chen, B. Chiaro, A. Dunsworth, A. Fowler, Sillanpää, M. A., J. I. Park, and R. W. Simmonds, 2007,
and et al., 2016, Physical Review Letters 117(19), Nature 449(7161), 438, ISSN 0028-0836, URL http://dx.
190503, ISSN 1079-7114, URL http://dx.doi.org/10. doi.org/10.1038/nature06124.
1103/PhysRevLett.117.190503. da Silva, M. P., D. Bozyigit, A. Wallraff, and A. Blais, 2010,
Sathyamoorthy, S. R., L. Tornberg, A. F. Kockum, B. Q. Phys. Rev. A 82(4), 043804, URL http://pra.aps.org/
Baragiola, J. Combes, C. M. Wilson, T. M. Stace, and abstract/PRA/v82/i4/e043804.
G. Johansson, 2014, Physical Review Letters 112(March), Simons, R. N., 2001, Coplanar waveguide circuits, compo-
1, ISSN 00319007. nents and systems, Wiley Series in Microwave and Optical
Schleich, W., and J. A. Wheeler, 1987, Nature 326(6113), 574. Engineerging (Wiley Inter-Science), ISBN 0-471-16121-7.
Schmidt, S., and J. Koch, 2013, Ann. Phys. (Berlin) 525(6), Sirois, A. J., M. A. Castellanos-Beltran, M. P. DeFeo, L. Ran-
395, ISSN 1521-3889, URL http://dx.doi.org/10.1002/ zani, F. Lecocq, R. W. Simmonds, J. D. Teufel, and J. Au-
 80

mentado, 2015, Applied Physics Letters 106(17), 172603, Teufel, J. D., D. Li, M. S. Allman, K. Cicak, A. J. Sirois, J. D.
URL http://scitation.aip.org/content/aip/journal/ Whittaker, and R. W. Simmonds, 2011b, Nature 471(7337),
apl/106/17/10.1063/1.4919759. 204, ISSN 0028-0836, URL http://dx.doi.org/10.1038/
Sivak, V., N. Frattini, V. Joshi, A. Lingenfelter, nature09898.
S. Shankar, and M. Devoret, 2019, Phys. Rev. Applied Tey, M. K., Z. Chen, S. A. Aljunid, B. Chng, F. Huber,
11, 054060, URL https://link.aps.org/doi/10.1103/ G. Maslennikov, and C. Kurtsiefer, 2008, Nat. Phys. ad-
PhysRevApplied.11.054060. vanced online publication, , ISSN 1745-2481, URL
Sletten, L. R., B. A. Moores, J. J. Viennot, and K. W. Lehnert, http://dx.doi.org/10.1038/nphys1096.
2019, Phys. Rev. X 9, 021056, URL https://link.aps. Thompson, R. J., G. Rempe, and H. J. Kimble, 1992, Phys.
org/doi/10.1103/PhysRevX.9.021056. Rev. Lett. 68(8), 1132.
Slichter, C. P., 1990, Principles of Magnetic Resonance, 3rd Tinkham, M., 1996, Introduction to Superconductivity
ed. (Springer-Verlag, Berlin). (McGraw-Hill International Editions).
Slichter, D. H., R. Vijay, S. J. Weber, S. Boutin, M. Boisson- Tinkham, M., 2004, Introduction to superconductivity
neault, J. M. Gambetta, A. Blais, and I. Siddiqi, 2012, Phys. (Dover Publications, Mineola (N.Y.)), 2nd edition, ISBN
Rev. Lett. 109, 153601, URL http://link.aps.org/doi/ 0486435032.
10.1103/PhysRevLett.109.153601. Toida, H., T. Nakajima, and S. Komiyama, 2013, Phys. Rev.
Solgun, F., D. W. Abraham, and D. P. DiVincenzo, 2014, Lett. 110(6), 066802, URL http://link.aps.org/doi/10.
Phys. Rev. B 90, 134504, URL http://link.aps.org/doi/ 1103/PhysRevLett.110.066802.
10.1103/PhysRevB.90.134504. Touzard, S., A. Kou, N. E. Frattini, V. V. Sivak, S. Puri,
Solgun, F., D. P. DiVincenzo, and J. M. Gambetta, 2019, A. Grimm, L. Frunzio, S. Shankar, and M. H. Devoret,
IEEE Transactions on Microwave Theory and Techniques 2019, Phys. Rev. Lett. 122, 080502, URL https://link.
67(3), 928, ISSN 1557-9670. aps.org/doi/10.1103/PhysRevLett.122.080502.
Srinivasan, S. J., A. J. Hoffman, J. M. Gambetta, and A. A. Toyli, D. M., A. W. Eddins, S. Boutin, S. Puri, D. Hover,
Houck, 2011, Phys. Rev. Lett. 106(8), 083601. V. Bolkhovsky, W. D. Oliver, A. Blais, and I. Siddiqi, 2016,
Srinivasan, S. J., N. M. Sundaresan, D. Sadri, Y. Liu, J. M. Phys. Rev. X 6, 031004, URL http://link.aps.org/doi/
Gambetta, T. Yu, S. M. Girvin, and A. A. Houck, 2014, 10.1103/PhysRevX.6.031004.
Phys. Rev. A 89, 033857, URL http://link.aps.org/doi/ Tripathi, V., M. Khezri, and A. N. Korotkov, 2019, Phys.
10.1103/PhysRevA.89.033857. Rev. A 100, 012301, URL https://link.aps.org/doi/10.
Stockklauser, A., P. Scarlino, J. V. Koski, S. Gasparinetti, 1103/PhysRevA.100.012301.
C. K. Andersen, C. Reichl, W. Wegscheider, T. Ihn, Tsang, M., and C. M. Caves, 2012, Phys. Rev. X
K. Ensslin, and A. Wallraff, 2017, Phys. Rev. X 7, 2, 031016, URL https://link.aps.org/doi/10.1103/
011030, URL https://journals.aps.org/prx/abstract/ PhysRevX.2.031016.
10.1103/PhysRevX.7.011030. Tuckett, D. K., S. D. Bartlett, and S. T. Flammia, 2018, Phys.
Strand, J. D., M. Ware, F. Beaudoin, T. A. Ohki, B. R. Rev. Lett. 120, 050505, URL https://link.aps.org/doi/
Johnson, A. Blais, and B. L. T. Plourde, 2013, Phys. Rev. 10.1103/PhysRevLett.120.050505.
B 87, 220505, URL http://link.aps.org/doi/10.1103/ Tuckett, D. K., S. D. Bartlett, S. T. Flammia, and B. J. Brown,
PhysRevB.87.220505. 2019a, arXiv preprint arXiv:1907.02554 .
Strauch, F. W., P. R. Johnson, A. J. Dragt, C. J. Lobb, Tuckett, D. K., A. S. Darmawan, C. T. Chubb, S. Bravyi,
J. R. Anderson, and F. C. Wellstood, 2003, Phys. Rev. Lett. S. D. Bartlett, and S. T. Flammia, 2019b, Phys. Rev.
91(16), 167005, URL http://link.aps.org/doi/10.1103/ X 9, 041031, URL https://link.aps.org/doi/10.1103/
PhysRevLett.91.167005. PhysRevX.9.041031.
Sun, L., A. Petrenko, Z. Leghtas, B. Vlastakis, G. Kirch- Turchette, Q. A., N. P. Georgiades, C. J. Hood, H. J. Kimble,
mair, K. M. Sliwa, A. Narla, M. Hatridge, S. Shankar, and A. S. Parkins, 1998, Phys. Rev. A 58, 4056, URL
J. Blumoff, L. Frunzio, M. Mirrahimi, et al., 2014, Nature https://link.aps.org/doi/10.1103/PhysRevA.58.4056.
511, 444, URL http://www.nature.com/nature/journal/ Varcoe, B. T. H., S. Brattke, M. Weidinger, and H. Walther,
v511/n7510/full/nature13436.html. 2000, Nature 403(6771), 743, ISSN 0028-0836, URL http:
Sundaresan, N. M., Y. Liu, D. Sadri, L. J. Szőcs, D. L. Un- //dx.doi.org/10.1038/35001526.
derwood, M. Malekakhlagh, H. E. Türeci, and A. A. Houck, Vepsäläinen, A., A. H. Karamlou, J. L. Orrell, A. S. Do-
2015, Phys. Rev. X 5, 021035, URL http://link.aps.org/ gra, B. Loer, F. Vasconcelos, D. K. Kim, A. e. J. Melville,
doi/10.1103/PhysRevX.5.021035. B. M. Niedzielski, J. L. Yoder, S. Gustavsson, J. A. For-
Takita, M., A. Corcoles, E. Magesan, B. Abdo, M. Brink, maggio, et al., 2020, arXiv e-prints , arXiv:2001.09190eprint
A. Cross, J. M. Chow, and J. M. Gambetta, 2016, Physical 2001.09190.
Review Letters 117(21), 210505, ISSN 1079-7114, URL Verney, L., R. Lescanne, M. H. Devoret, Z. Legh-
http://dx.doi.org/10.1103/PhysRevLett.117.210505. tas, and M. Mirrahimi, 2019, Phys. Rev. Applied
Tavis, M., and F. W. Cummings, 1968, Phys. Rev. 170(2), 11, 024003, URL https://link.aps.org/doi/10.1103/
379. PhysRevApplied.11.024003.
Terhal, B. M., J. Conrad, and C. Vuillot, 2020, arXiv e-prints Viennot, J. J., M. R. Delbecq, M. C. Dartiailh, A. Cottet, and
, arXiv:2002.11008eprint 2002.11008. T. Kontos, 2014, Phys. Rev. B 89(16), 165404, URL http:
Tesche, C. D., 1987, Japanese Journal of Applied Physics //link.aps.org/doi/10.1103/PhysRevB.89.165404.
26(S3-2), 1409, ISSN 0021-4922. Vijay, R., M. H. Devoret, and I. Siddiqi, 2009, Rev. Sci. In-
Teufel, J. D., T. Donner, D. Li, J. W. Harlow, M. S. Allman, strum. 80(11), 111101 (pages 17), URL http://link.aip.
K. Cicak, A. J. Sirois, J. D. Whittaker, K. W. Lehnert, and org/link/?RSI/80/111101/1.
R. W. Simmonds, 2011a, Nature 475(7356), 359, ISSN 0028- Vijay, R., D. H. Slichter, and I. Siddiqi, 2011, Phys. Rev. Lett.
0836, URL http://dx.doi.org/10.1038/nature10261. 106(11), 110502.
 81

Vion, D., A. Aassime, A. Cottet, P. Joyez, H. Pothier, Wendin, G., 2017, Rep. Prog. Phys. 80, 106001, URL https:
C. Urbina, D. Esteve, and M. H. Devoret, 2002a, Science //doi.org/10.1088/1361-6633/aa7e1a.
296, 886. Wenner, J., Y. Yin, Y. Chen, R. Barends, B. Chiaro,
Vion, D., A. Aassime, A. Cottet, P. Joyez, H. Pothier, E. Jeffrey, J. Kelly, A. Megrant, J. Mutus, C. Neill,
C. Urbina, D. Esteve, and M. H. Devoret, 2002b, Science P. OḾalley, P. Roushan, et al., 2014, Phys. Rev.
296, 886. Lett. 112(21), 210501, URL http://link.aps.org/doi/
Vissers, M. R., J. Gao, D. S. Wisbey, D. A. Hite, C. C. Tsuei, 10.1103/PhysRevLett.112.210501.
A. D. Corcoles, M. Steffen, and D. P. Pappas, 2010, Applied White, T. C., J. Y. Mutus, I.-C. Hoi, R. Barends, B. Camp-
Physics Letters 97(23), 232509, URL https://doi.org/10. bell, Y. Chen, Z. Chen, B. Chiaro, A. Dunsworth, E. Jeffrey,
1063/1.3517252. J. Kelly, A. Megrant, et al., 2015, Applied Physics Let-
Vlastakis, B., G. Kirchmair, Z. Leghtas, S. E. Nigg, ters 106(24), 242601, URL http://scitation.aip.org/
L. Frunzio, S. M. Girvin, M. Mirrahimi, M. H. De- content/aip/journal/apl/106/24/10.1063/1.4922348.
voret, and R. J. Schoelkopf, 2013, Science 342, 6158, Winik, e. a., R., 2020, In preparation .
URL http://www.sciencemag.org/content/early/2013/ Winkler, R., 2003, Quasi-Degenerate Perturbation Theory. In:
09/25/science.1243289. Spin–Orbit Coupling Effects in Two-Dimensional Electron
Vollmer, M., 2004, Physics Education 39(1), 74, URL http: and Hole Systems. Springer Tracts in Modern Physics, vol
//stacks.iop.org/0031-9120/39/i=1/a=006. 191. (Springer, Berlin).
Vool, U., and M. Devoret, 2017, International Journal of Cir- Wiseman, H., and G. Milburn, 2010, Quantum Measurement
cuit Theory and Applications 45(7), 897, ISSN 1097-007X, and Control (Cambridge University Press).
cta.2359, URL http://dx.doi.org/10.1002/cta.2359. Wong, C. H., and M. G. Vavilov, 2017, Phys. Rev. A
Vool, U., S. Shankar, S. O. Mundhada, N. Ofek, A. Narla, 95, 012325, URL http://link.aps.org/doi/10.1103/
K. Sliwa, E. Zalys-Geller, Y. Liu, L. Frunzio, R. J. PhysRevA.95.012325.
Schoelkopf, S. M. Girvin, and M. H. Devoret, 2016, Phys. Wustmann, W., and V. Shumeiko, 2013, Phys. Rev. B
Rev. Lett. 117, 133601, URL https://link.aps.org/doi/ 87(18), 184501, URL http://link.aps.org/doi/10.1103/
10.1103/PhysRevLett.117.133601. PhysRevB.87.184501.
Wallraff, A., D. I. Schuster, A. Blais, L. Frunzio, R.-S. Huang, Xiang, Z.-L., S. Ashhab, J. Q. You, and F. Nori, 2013, Rev.
J. Majer, S. Kumar, S. M. Girvin, and R. J. Schoelkopf, 2004, Mod. Phys. 85, 623, URL http://link.aps.org/doi/10.
Nature 431, 162, URL http://www.nature.com/nature/ 1103/RevModPhys.85.623.
journal/v431/n7005/full/nature02851.html. Yamamoto, T., K. Inomata, M. Watanabe, K. Matsuba,
Wallraff, A., D. I. Schuster, A. Blais, L. Frunzio, J. Majer, T. Miyazaki, W. D. Oliver, Y. Nakamura, and J. S. Tsai,
M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, 2005, 2008, Appl. Phys. Lett. 93(4), 042510, URL http://dx.
Phys. Rev. Lett. 95, 060501, URL http://link.aps.org/ doi.org/10.1063/1.2964182.
abstract/PRL/v95/e060501. Yamamoto, T., M. Neeley, E. Lucero, R. C. Bialczak, J. Kelly,
Wallraff, A., A. Stockklauser, T. Ihn, J. R. Petta, and A. Blais, M. Lenander, M. Mariantoni, A. D. O’Connell, D. Sank,
2013, Phys. Rev. Lett. 111, 249701, URL http://prl.aps. H. Wang, M. Weides, J. Wenner, et al., 2010, Phys. Rev. B
org/abstract/PRL/v111/i24/e249701. 82(18), 184515, URL http://link.aps.org/doi/10.1103/
Walls, D. F., and G. J. Milburn, 2008, Quantum Optics PhysRevB.82.184515.
(Springer Verlag, Berlin), 2 edition. Yamamoto, T., Y. A. Pashkin, O. Astafiev, Y. Nakamura, and
Walter, T., P. Kurpiers, S. Gasparinetti, P. Magnard, A. Po- J. S. Tsai, 2003, Nature 425(6961), 941.
tocnik, Y. Salathé, M. Pechal, M. Mondal, M. Oppliger, Yan, F., S. Gustavsson, A. Kamal, J. Birenbaum, A. P. Sears,
C. Eichler, and A. Wallraff, 2017, Phys. Rev. Applied D. Hover, T. J. Gudmundsen, D. Rosenberg, G. Samach,
7, 054020, URL https://journals.aps.org/prapplied/ S. Weber, J. L. Yoder, T. P. Orlando, et al., 2016, Nat.
abstract/10.1103/PhysRevApplied.7.054020. Commun. 7, 12964.
Wang, C., C. Axline, Y. Y. Gao, T. Brecht, Y. Chu, Yan, F., P. Krantz, Y. Sung, M. Kjaergaard, D. L. Campbell,
L. Frunzio, M. H. Devoret, and R. J. Schoelkopf, T. P. Orlando, S. Gustavsson, and W. D. Oliver, 2018,
2015, Applied Physics Letters 107(16), 162601, URL Phys. Rev. Applied 10, 054062, URL https://link.aps.
http://scitation.aip.org/content/aip/journal/apl/ org/doi/10.1103/PhysRevApplied.10.054062.
107/16/10.1063/1.4934486. Yang, C.-P., S.-I. Chu, and S. Han, 2003, Phys. Rev.
Wang, C. S., J. C. Curtis, B. J. Lester, Y. Zhang, Y. Y. A 67, 042311, URL http://link.aps.org/doi/10.1103/
Gao, J. Freeze, V. S. Batista, P. H. Vaccaro, I. L. Chuang, PhysRevA.67.042311.
L. Frunzio, L. Jiang, S. M. Girvin, et al., 2019a, eprint Yin, Y., Y. Chen, D. Sank, P. J. J. O’Malley, T. C. White,
1908.03598. R. Barends, J. Kelly, E. Lucero, M. Mariantoni, A. Megrant,
Wang, Z., S. Shankar, Z. Minev, P. Campagne-Ibarcq, C. Neill, A. Vainsencher, et al., 2013, Phys. Rev. Lett. 110,
A. Narla, and M. Devoret, 2019b, Phys. Rev. Applied 107001.
11, 014031, URL https://link.aps.org/doi/10.1103/ Yoshihara, F., T. Fuse, S. Ashhab, K. Kakuyanagi, S. Saito,
PhysRevApplied.11.014031. and K. Semba, 2016, Nat Phys advance online publi-
Ware, M., B. R. Johnson, J. M. Gambetta, T. A. Ohki, cation, , ISSN 1745-2481, URL http://dx.doi.org/10.
J. M. Chow, and B. L. T. Plourde, 2019, arXiv e-prints , 1038/nphys3906.
arXiv:1905.11480eprint 1905.11480. You, J. Q., X. Hu, S. Ashhab, and F. Nori, 2007, Phys. Rev.
Wei, K. X., I. Lauer, S. Srinivasan, N. Sundaresan, D. T. B 75, 140515, URL https://link.aps.org/doi/10.1103/
McClure, D. Toyli, D. C. McKay, J. M. Gambetta, and PhysRevB.75.140515.
S. Sheldon, 2019, arXiv e-prints , arXiv:1905.05720eprint You, J. Q., and F. Nori, 2003, Phys. Rev. B 68(6), 064509.
1905.05720. Yurke, B., 1987, J. Opt. Soc. Am. B 4(10), 1551, URL http:
Wen, C., 1969, IEEE Trans. Microwave Theory Tech. 17, 1087. //josab.osa.org/abstract.cfm?URI=josab-4-10-1551.
 82

Yurke, B., 2004, Input-Output theory (Springer), chapter 3. Zhang, H., S. Chakram, T. Roy, N. Earnest, Y. Lu, Z. Huang,
Yurke, B., L. R. Corruccini, P. G. Kaminsky, L. W. Rupp, A. D. D. Weiss, J. Koch, and D. I. Schuster, 2020, arXiv e-prints
Smith, A. H. Silver, R. W. Simon, and E. A. Whittaker, , arXiv:2002.10653eprint 2002.10653.
1989, Phys. Rev. A 39(5), 2519. Zheng, H., and H. U. Baranger, 2013, Phys. Rev. Lett.
Yurke, B., and J. S. Denker, 1984, Phys. Rev. A 29(3), 1419. 110, 113601, URL http://link.aps.org/doi/10.1103/
Yurke, B., P. G. Kaminsky, R. E. Miller, E. A. Whittaker, PhysRevLett.110.113601.
A. D. Smith, A. H. Silver, and R. W. Simon, 1988, Phys. Zhong, L., E. P. Menzel, R. D. Candia, P. Eder, M. Ihmig,
Rev. Lett. 60(9), 764. A. Baust, M. Haeberlein, E. Hoffmann, K. Inomata, T. Ya-
Zagoskin, A., and A. Blais, 2007, 63(4), 215. mamoto, Y. Nakamura, E. Solano, et al., 2013, New J. Phys.
Zeytinoğlu, S., M. Pechal, S. Berger, A. A. Abdumalikov Jr., 15(12), 125013, URL http://stacks.iop.org/1367-2630/
A. Wallraff, and S. Filipp, 2015, Phys. Rev. A 91, 043846, 15/i=12/a=125013.
URL http://journals.aps.org/pra/abstract/10.1103/ Zhu, G., D. G. Ferguson, V. E. Manucharyan, and J. Koch,
PhysRevA.91.043846. 2013, Phys. Rev. B 87, 024510, URL http://link.aps.
org/doi/10.1103/PhysRevB.87.024510.