nature electronics

Article https://doi.org/10.1038/s41928-023-01016-9

Thin-film image sensors with a pinned

photodiode structure

Received: 2 July 2022 Jiwon Lee 1,2,5 , Epimitheas Georgitzikis 1,5, Yannick Hermans1,
Nikolas Papadopoulos1, Naresh Chandrasekaran1, Minhyun Jin 1,3,
Accepted: 15 July 2023
Abu Bakar Siddik1,4, Florian De Roose 1, Griet Uytterhoeven1, Joo Hyoung Kim1,
Published online: 14 August 2023 Renaud Puybaret1, Yunlong Li 1, Vladimir Pejovic1, Gauri Karve1, David Cheyns1,
Jan Genoe 1,4, Paweł E. Malinowski1, Paul Heremans1,4 & Kris Myny 1,4
Check for updates

Image sensors made using silicon complementary metal–oxide–

semiconductor technology can be found in numerous electronic devices
and typically rely on pinned photodiode structures. Photodiodes based on
thin films can have a high absorption coefficient and a wider wavelength
range than silicon devices. However, their use in image sensors has been
limited by high kTC noise, dark current and image lag. Here we show that
thin-film-based image sensors with a pinned photodiode structure can
have comparable noise performance to a silicon pinned photodiode pixel.
We integrate either a visible-to-near-infrared organic photodiode or a
short-wave infrared colloidal quantum dot photodiode with a thin-film
transistor and silicon readout circuitry. The thin-film pinned photodiode
structures exhibit low kTC noise, suppressed dark current, high full-well
capacity and high electron-to-voltage conversion gain, as well as preserving
the benefits of the thin-film materials. An image sensor based on the organic
absorber has a quantum efficiency of 54% at 940 nm and read noise of 6.1e–.

Thin-film-based image sensors can offer high absorption over a broad The earliest implementation of CMOS image sensors used a passive
wavelength spectrum and can potentially be monolithically co-integrated pixel sensor structure16,17, which was uncompetitive with the domi-
with readout circuits made using complementary metal–oxide–semi- nant image sensor technology at the time—the charge-coupled device
conductor (CMOS) or thin-film transistors (TFTs). Thin-film devices (CCD). CCDs used a pinned photodiode (PPD) structure18, which can
based on organic semiconductors have, for example, demonstrated a fully deplete the photodiode storage and better transfer charges to the
ten- to hundred-fold higher absorption coefficient in the visible range readout circuit, resulting in improved image quality. In comparison,
than silicon1; devices based on colloidal quantum dots (QDs) are sensi- early CMOS image sensors had inferior noise performance due to fixed
tive in the short-wave infrared (SWIR) to mid-wave infrared regime2–5; pattern noise and low charge-to-voltage conversion gain19.
devices based on metal–organic perovskites can be used for the direct The development of active pixel sensors20 in CMOS image sensors
absorption of high-energy radiation6. Also, 8K resolution cameras using with an in-pixel buffer improved the photon-to-electron conversion
organic photodiodes (OPDs)7,8 and X-ray image sensors using organic gain, and retaining the advantages, compared with CCDs, of lower
or perovskite photodiodes have also been created9–11, as well as com- power consumption and compatibility with other CMOS technologies
mercial SWIR image sensors based on QDs12–15. However, applications for (enabling the co-integration of peripheral and readout circuits on the
thin-film-based image sensors are currently limited to those that are chal- same chip). However, they still exhibited worse image quality compared
lenging to achieve with silicon-based devices because thin-film-based with CCDs19,21. This was mainly due to the high dark current caused by
image sensors have inferior noise performance than silicon. the extension of the photodiode to the silicon interface and kTC noise

imec, Leuven, Belgium. 2Department of Photonics and Nanoelectronics, BK21 FOUR ERICA-ACE Center, Hanyang University ERICA, Ansan, Korea.
1

3
Dongguk University, Seoul, Korea. 4ESAT, KU Leuven, Heverlee, Belgium. 5These authors contributed equally: Jiwon Lee, Epimitheas Georgitzikis.
e-mail: leejiwon1@hanyang.ac.kr

Nature Electronics | Volume 6 | August 2023 | 590–598 590

Article https://doi.org/10.1038/s41928-023-01016-9

a TiN TiN to IGZO contact b ITO

HTL
Thin-film absorber

IGZO
Al2O3
Top contact PG TG FD
a a'
VDD

RST

IGZO Silicon wafer SF

SEL
Output

c PG TG PG TG PG TG PG TG

Signal

Reset state During integration Charge transfer Signal readout

FD FD FD FD

Fig. 1 | Structure of the thin-film PPD pixel. a,b, Simplified diagrams of two isolated between the pixels; all the pixels have a common indium tin oxide anode
adjacent TF-PPD pixels in the top view (a) and cross-sectional view (b) along the contact and the IGZO TFT is gated from the back side. The IGZO TFTs and OPDs
dashed line (a–a′), with an additional top contact. c, Operation diagram of the are processed on a 130 nm CMOS readout chip containing three transistors
TF-PPD pixel. In our implementation, the thin-film absorber layer and HTL are not per pixel.

generated during photodiode reset. Both these phenomena result from near-infrared wavelengths, or a QD photodiode (QDPD) sensitive to
the photodiode connecting directly to the metal contact of the in-pixel SWIR. In both cases, high linearity is measured for varying integration
source follower (SF) and reset (RST) transistor. This contact area and times, indicating the efficient transfer of charges, and the full deple-
the top silicon surface contain a high density of defect states. Addition- tion of the photodiode node prevents the generation of kTC noise and
ally, kTC noise is created when the charge storage node cannot be fully image lag. Compared with thin-film imagers with the same photodiode
depleted, which causes reset-level changes from frame to frame. The material and a conventional 3 T pixel structure, our 4 T TF-PPD pixel has
full-well capacity (FWC) of the photodiode—which determines the a two order of magnitude decrease in dark read noise (114.7e– to 6.1e–),
dynamic range and charge-to-voltage conversion gain of the sensor—is and a five times improvement in conversion gain.
also interlinked since the photodiode is itself part of the conversion
node. This introduces a trade-off: a large capacitance, for example, is TF-PPD pixel structure
required for high FWC and low capacitance favours high conversion Figure 1 shows our thin-film photodiode and TFT hybrid PPD structure
gain. These early implementations of CMOS image sensors also had in the top view (Fig. 1a) and cross-sectional view (Fig. 1b). Compared
high image lag because the reset-level potential of the photodiode with conventional thin-film-based pixel structures, where the photodi-
was affected by the state of the photodiode at the end of the previous ode is directly connected to the readout (Supplementary Fig. 1)7,8,13–15,27,
integration phase. an additional TFT structure is included. The TFT has two purposes: it
To overcome these issues, pixel sensors were extended from previ- operates as a transfer gate (TG) and it is part of the photogate (PG)22,28.
ous designs, which had been based on three transistor gates (referred An indium gallium zinc oxide (IGZO) semiconductor acts as the pho-
to as the ‘3 T’ pixel) to a four-gate design (‘4 T’ pixel) that incorporated todiode electron transport layer (ETL) as well as the channel layer for
the PPD structure19,22,23. This configuration uses a p+np photodiode the TG TFT due to its relatively high electron mobility29. The IGZO tran-
structure with a sandwiched n region, which is used as storage, being sistor integration scheme is based on a bottom-gate configuration.
separated from the readout node by an extra transistor. By fully deplet- Two types of TF-PPD device are demonstrated with different material
ing the storage node during the reset operation, the effect of the stacks, using either colloidal QDs or organic semiconductor as the
previous state (image lag) and kTC noise can be reduced. Also, the photoabsorption material.
separation of the storage node from the surface by the p+ region, and The source of the TFT is directly contacted to the photodiode to
from the metal contact by the additional transistor, minimizes the collect the generated electrons and the drain is connected to the read-
dark current, as the photodiode is isolated from the defective surface out transistors. The drain node is constituted as a floating diffusion (FD)
or contact region. The adoption of the PPD structure to CMOS image node where photogenerated charges are converted to a voltage level.
sensors allowed them to compete with CCD image sensors and they The gate of the TFT is controlled by external biasing (TG) to connect or
are now the standard technology for consumer imaging devices22–24. disconnect the photodiode and FD node. A PG underneath the source
Thin-film photodiodes currently suffer from the same problems node of the TFT is used to control the potential of IGZO and hence the
as early CMOS image sensors: high noise levels, high dark currents, field distribution in the photodiode. This structure can be understood
non-optimized gain versus FWC and image lag. A pixel structure using the as a back side-illuminated PG pixel structure. By realizing a back-gated
negative feedback operation during reset or storing reset-level voltage by PG structure, the photodiode materials can be directly deposited on
adding a capacitor in pixel have been implemented to suppress kTC noise top of the IGZO channel, and the incoming light will reflect on the metal
and image lag7,8,25,26, but the design is complex and still has a higher dark gate towards the photodiode to increase absorption in the thin-film
current and unoptimized conversion gain/FWC compared with CMOS active layers.
image sensors. Adding a capacitive transimpedance amplifier (CTIA) can Because of the large bandgap of IGZO, the depleted surface of
suppress the dark current and optimize the gain/FWC, but fails to eliminate the semiconductor near the gate dielectric does not increase the dark
kTC noise and requires a relatively large pixel and high static power13. current, which is the main drawback of the surface-channel silicon
In this Article, we report a thin-film-based image sensor with a PPD PG30. Therefore, the generated charges can be stored in the IGZO at
pixel structure (TF-PPD). We use either an OPD sensitive to visible to the interface to the gate dielectric, which enables high FWC by using

Nature Electronics | Volume 6 | August 2023 | 590–598 591

Article https://doi.org/10.1038/s41928-023-01016-9

a b 0 OPD QDPD

ITO
–2 –2.10 –2.10
–2.30
HTL
Infrared absorber Organic blend
–3

DNTPD

DNTPD
(organic or QD)

PolyTPD
–3.50 PbS QD Eg = 0.87 eV

E – EVAC (eV)
IGZO –4.03 –4.00 –4.03
–4
TiN TiN TiN PbS QD
PbS QD

Acceptor

Donor
ITO ITO
p-type
–4.54 –4.54 n-type
–5 –4.70 –4.70
Silicon
–5.20 –5.10 –5.20 –5.10
–5.30

IGZO

IGZO
–6

–7

–7.54 –7.54

c 10
–2
d 60 e 30
OPD OPD QDPD
–3
10 50
QDPD
Current density (A cm–2)

–4
10
EQE at –3 V (%)

EQE at –3 V (%)
40 20
–5
10
30
–6
10

–7
20 10
10

–8 10
10

–9
10 0 0
–4 –3 –2 –1 0 1 2 3 4 400 500 600 700 800 900 1,000 1,100 1,100 1,200 1,300 1,400 1,500 1,600 1,700

Bias (V) Wavelength (nm) Wavelength (nm)

f g h 10
–6
80
IGZO (12 nm) –7
VDS = 0.1 V
10
Dielectric Al2O3 (12 nm) VDS = 3.0 V
–8
10 60
Source Gate Drain
–9
Silicon 10

ID (nA)
ID (A)

–10
10 40
S D 10
–11

TiN –12
10 20
IGZO Al2O3
–13
10

G 10
–14
0
0.1 µm –2 –1 0 1 2 3

VGS (V)

Fig. 2 | Fabrication and characterization of TF-PPD pixel. a,b, Device structure OPD (d) and QDPD (e). f,g, Microscopy image of the TFT test structures (f) and
(a) and energy diagram (b) of the near-infrared-sensitive OPD and the SWIR- the corresponding cross-section schematic and TEM along the source–drain
sensitive QDPD. ETL and HTL denote the electron and hole transport layers, plane (g). h, Transfer characteristics of the TFT under VDS of 0.1 and 3.0 V.
respectively. c, Current–voltage curve under dark conditions. d,e, EQE for the

high-density gate capacitance. The photodiode can be negatively FD voltage to the output node and the select (SEL) transistor connects
biased by controlling the PG potential to successfully separate the the pixels to the subsequent readout chain. These three transistors are
generated charges and subsequently collect photoelectrons in the implemented on silicon.
IGZO layer. The PG bias also ensures that the collected photoelectrons Figure 1c demonstrates the operation of the proposed thin-film
are fully transferred to the FD node when TG is turned on following the PPD pixel. Immediately after the thin-film PPD reset, no charges remain,
lateral potential difference. The potential distribution of the proposed ensuring isolation from the previous frame and hence no kTC noise is
device is simulated using technology computer-aided design, which observed. During the integration time, the photogenerated electrons
confirmed the operation of the TG and PG (Supplementary Fig. 2). are stored in the IGZO on top of the PG with the help of the positive
In a PPD pixel, reset and signal are available almost simultane- bias applied to the PG as the TG is turned off. The stored charges are
ously, enabling on-chip correlated double sampling operation at the transferred to the FD node by turning the TG on and are read by the pixel
circuit level. In the case of a general 3 T pixel structure, reset and sig- electronics. The voltage difference with respect to the reset level, which
nal are output at the intervals of integration time, which means that is sampled right before the charge transfer, defines the intensity of the
the frame memory and per-pixel correction in the digital domain are detected light. The process of sampling the FD node twice—once with
required to remove the kTC noise, which greatly increases the system the reset level and once with the signal level—is called the correlated
complexity. A TF-PPD pixel has the same electrical pixel components double sampling operation and is a conventional technique to sup-
as a conventional silicon PPD pixel23 (Fig. 1b). The RST transistor sets press reset noise (kTC) from the FD node and fixed pattern noise from
the FD node to the reference potential, the SF transistor buffers the variation in the pixel transistors.

Nature Electronics | Volume 6 | August 2023 | 590–598 592

Article https://doi.org/10.1038/s41928-023-01016-9

a b
ITO

PG OPD + HTL
PG TG FD TG
TG
IGZO
Al2O3

Bottom electrode
FD
0.2 µm

20 nm

c d e 10
–7

Dark
10
–8 Light

Cathode current (A)

IGZO island –9
10

–10
10

–11
10

–12
10
Diode reverse bias
–13
10
–5 –4 –3 –2 –1 0 1 2 3 4 5
20 µm
0.2 µm Cathode voltage (V)

f 10
–9
g 10
–9
h 0.15
Reference OPD

–10 –10
10 10

Photocurrent (nA)
FD current (A)

FD current (A)

0.10

–11 –11
10 10
VTG = –4 V VPG = –4 V
VTG = –2 V VPG = –2 V 0.05
–12 VTG = 0 V –12 VPG = 0 V
10 10
VTG = 2 V VPG = 2 V VTG = 4 V
VTG = 4 V VPG = 4 V VPG = 4 V
Reference Reference
–13 –13
10 10 0
–5 –4 –3 –2 –1 0 1 2 3 4 5 –5 –4 –3 –2 –1 0 1 2 3 4 5 –5 –4 –3 –2 –1 0 1 2 3 4 5
VPG (V) VTG (V) VTG or VPG (V)

Fig. 3 | Processed TF-PPD test structures. a, Scanning electron microscopy an active area of 10 × 10 µm2. e, Corresponding I–V characteristics under dark
image of the processed TF-PPD pixel with OPD after dielectric deposition and and light conditions (530 nm, 1 mW cm−2). f, FD current versus VPG at different
patterning. b, TEM picture of the cross section of the fully processed TF-PPD with VTG values. g, FD current versus VTG at different VPG values. h, Photocurrent
OPD. c, Microscopic image of the TF-PPD test structures after IGZO deposition comparison with a reference OPD.
and patterning. d, Microscopic image of the reference OPD pixel array with

TF-PPD fabrication and characterization (Fig. 3d,e). For the case of the 10 µm pixel, the TG has a width and length
Two types of photodiode are used to demonstrate the operation of the of 7.6 µm and 0.4 µm, respectively, and the contact size connecting the
thin-film PPD architecture: an OPD and a QDPD (Fig. 2a–e). Both these IGZO layer to the drain metal is 0.3 × 0.3 µm2. The thicknesses of IGZO
are co-integrated with an IGZO TFT (Fig. 2f–h). For the OPD, a bulk and Al2O3 dielectric are designed to be 12 nm each. To enable a practical
heterojunction system is used, composed of a donor–acceptor blend readout, 500 unit PPD structures are connected in parallel. This paral-
with high spectral sensitivity from the visible regime up to 1,000 nm lel output node is then either connected to the parameter analyser to
(Supplementary Fig. 3a). For the QDPD, lead sulfide (PbS)-based col- measure the d.c. characteristics or to the off-chip transimpedance
loidal QDs are used as the absorber material. The QDs have a size of amplifier to test the transient response.
5.5 nm, corresponding to an absorption peak at 1,425.0 nm when in a Figure 3a shows a scanning electron microscopy image of the
colloidal solution phase (Supplementary Fig. 3b). To form a photodi- fabricated TF-PPD device and Fig. 3b shows a tunnelling electron
ode, a junction between the n-type and p-type QD layers is formed by microscopy (TEM) cross-section image of the device. The reference
appropriate ligand exchange engineering (Methods)31–33. Figure 2c–e photodiode was optimized on the TF-PPD test structure to yield a
shows the measured photodiode characteristics optimized to work thin-film stack (Fig. 2a). The aim was to assemble a photodiode stack
with IGZO as the ETL. with a low dark current density and still maintaining a high external
Two different test pixel sizes, namely, 7.5 × 7.5 µm 2 and quantum efficiency (EQE) over a wide wavelength range. IGZO is used
10.0 × 10.0 µm2, are designed. The size of the PG (which defines the as an ETL. The morphology of the IGZO ETL and the organic blend
photosensitive part of the device) occupies about 95% of the device deposited on top of it is characterized by atomic force microscopy,
footprint for the case of the 10 × 10 µm2 pixel. A conventional photo- which shows uniform films with a roughness of 2.2 nm and 0.9 nm,
diode pixel with an active area of 10 × 10 µm2 is included as a reference respectively (Supplementary Fig. 4). IGZO has already been shown

Nature Electronics | Volume 6 | August 2023 | 590–598 593

Article https://doi.org/10.1038/s41928-023-01016-9

a b c
RST 2.0 2.0

×500... OPD QDPD

TG 7.5 µm 10.0 µm
1.5 1.5
Integration time 10.0 µm 7.5 µm

Output (Me)

Output (Me)
RST
1.0 1.0

Slope (a.u.)
1.0
Cint
0.5

0.5 0 0.5
0 50 100
– PD active size (µm )
2

0 0
+ Output 0 0.5 1.0 1.5 0 0.5 1.0 1.5 2.0

Integration time (ms) Integration time (ms)

Anode PG TG

e f
d 5 OPD 7.5 µm
0.25 60
10.0 µm
Dark current 10.0 µm OPD
50 7.5 µm OPD
0 0.20
10.0 µm QDPD
Linearity error (%)

40

Image lag (%)

Output (Me)
0.15
–5
30
2 QDPD 7.5 µm
10.0 µm 0.10
20
0
0.05
10
–2
0 0
0 0.5 1.0 1.5 2.0 0 1 2 3 4 0 0.1 0.2 0.3 0.4 0.5

Output (Me) Integration time (ms) Transfer time (ms)

Fig. 4 | Transient operation of the test TF-PPD structures. a, Test setup. level for the OPD and QDPD by calculating the least-squares linear regression.
b,c, Measured number of output electrons versus integration time for the OPD e, Dark current for the 10 µm pixel QDPD for increasing the integration time.
(b) and QDPD (c). d, Calculated linearity error from 10% to 90% of the saturation f, Image lag for the different photodiodes and pixel sizes.

to be an effective ETL for OPDs and in combination with PbS QDs for the operation of PG and TG can be confirmed. The sensitivity of the
creating photoconductors34,35. device (Fig. 3h) reaches that of the reference diode, which proves that
Here we incorporate IGZO in a p–n junction with PbS QDs to form the proposed device does not cause a loss of sensitivity.
a photodiode configuration. Due to the dual role of IGZO as the ETL The transient response of the developed device is measured with
and TG, it is crucial that all the photogenerated electrons are efficiently off-the-shelf CTIA and field-programmable gate array integrated on
transferred. Thus, in both cases, a staggered photodiode band align- a printed circuit board (Methods). As shown in Fig. 4a, 500 TF-PPD
ment design is implemented (Fig. 2b). The IGZO layer composition and devices are connected in parallel to the CTIA and the charges collected
energy levels are characterized by X-ray photoelectron spectroscopy in the PG are transferred to the input node of the CTIA. The output of
and ultraviolet photoelectron spectroscopy (Supplementary Figs. 5 the CTIA is monitored by an oscilloscope. The reset level is sampled
and 6, respectively). By the addition of an organic hole transport layer right after turning off the RST transistor, followed by the signal-level
(HTL), we can minimize the dark current and achieve a dark current sampling right after TG is turned off. The output, representing the
density of 2.6 × 10−8 and 1.1 × 10−7 A cm−2 at −3 V for the case of OPD and amount of incident light signal, is calculated by subtracting the reset
QDPD, respectively (Fig. 2c). An EQE above 40% was attained from 600 level from the signal level, as in a correlated double sampling opera-
to 970 nm under an applied bias of −3 V (Fig. 2d) for the OPD, result- tion. By changing the integration time, a linear input-to-output relation
ing in a photodiode sensitive in the visible to near-infrared region is measured as expected for the case of both OPD- and QDPD-based
(Supplementary Fig. 7a). Moreover, EQE of 27% could be achieved at a TF-PPDs (Fig. 4b–e), indicating that the generated charges are suc-
wavelength of 1,450 nm for the case of the QDPD enabling SWIR imaging cessfully transferred and readout.
(Supplementary Fig. 7c). The developed back-gated IGZO transistor To verify the full depletion of the TF-PPD after reset (which pre-
with back source/drain contact is measured to have a threshold voltage vents kTC noise and the effect of the previous frame), the image lag is
of −0.5 V, subthreshold swing of 90 mV dec−1 and on–off ratio higher measured (Fig. 4f). After illuminating the device under test for a single
than 104 (Fig. 2f–h). frame by flashing a light-emitting diode (LED), it is kept under dark for
The fabricated OPD-based TF-PPD is measured under constant the subsequent frames36. The remaining charges after the illuminated
TG bias (VTG) and varying the PG bias (VPG) (Fig. 3f), and under constant frame are transferred during subsequent frames. If the photodiode is
VPG and varying VTG (Fig. 3g). From Fig. 3f, the operation of the TG can fully depleted, there will be no charges transferred in the subsequent
be confirmed by the observation that when the TG voltage is low, the frames. The test result presented in Fig. 4f shows that the remaining
photodiode is not connected to the FD, so the photocurrent is low; charges decrease with a longer transfer time (‘TG-on time’). The number
when the TG voltage is high, the photocurrent is high. As evident from of charges transferred during the subsequent frame is normalized to
Fig. 3g, when TG is fully open, the sensitivity of the device increases with the number filled in the illuminated frame. At TG-on time of 50 µs, for
VPG and eventually saturates to the same sensitivity level as the refer- the case of the OPD-based TF-PPD, the image lag is measured to be 3.5%
ence diode. Even when the PG is not connected to the FD node, the cur- and 1.5% for 10.0 and 7.5 µm pixels, respectively. From this measure-
rent is higher than in the dark because the node also has a small parasitic ment, it can be concluded that during the transfer operation, most
photodiode effect that responds to light. From this d.c. measurement, of the charges in the PG are transferred to the FD node, so that the PG

Nature Electronics | Volume 6 | August 2023 | 590–598 594

Article https://doi.org/10.1038/s41928-023-01016-9

a b

c RST d e
Shutter 40 60

TG
Read Integration time 50

Quantum efficiency (%)

Reset sample 30
Signal sample 40

Output (ke)
20 30
RST (V)

3.3
0
20
10
TG (V)

10
3.0

0 0 0
0 0.5 1.0 1.5 2.0 400 500 600 700 800 900 1,000
3.0
Output (V)

Integration time (ms) Wavelength (nm)

Reset level
2.5
Output Signal level
2.0
0 0.01 0.02 0.03 0.04 0.05
Time (ms)

f 200 g 100 h Conventional 3 T

4 T TF-PPD (same conditions)
TF-PPD TF-PPD
Conv. 3 T Conv. 3 T
80
150
Number of pixels

Number of pixels

60
100
40

50
20

0 0
0 50 100 150 200 250 300 0 20 40 60 80 100
–
Read noise (e ) Dark current (nA cm–2)

Fig. 5 | TF-PPD image sensor demonstration. a, Test camera and processed quantum efficiency versus wavelength (e). f,g, Measured pixel statistics of the
image sensor. b, Captured uncorrected 64 × 64 image with TF-PPD imager (left) dark read noise (f) and dark current (g). h, Image compared with the conventional
and 180 × 180 image with a commercial camera (right). c, Operation timing 3 T-based imager.
diagram and oscilloscope output. d,e, Output versus integration time (d) and

empties almost completely. However, some charges are still left after operation mode. Figure 5c shows the operation timing diagram and
relatively long transfer times and the QD-based TF-PPD showed even direct output of the TF-PPD pixel in the test mode. The output voltage
worse image lag performance. The remaining image lag, even account- drops as the TG is turned on, proportional to the number of charges trans-
ing for the mobility of IGZO (around 10 cm2 V−1 s−1), is believed to be ferred from PG to FD. Figure 5d shows that the image sensor has a linear
caused by traps in the channel and the 200 nm gap between the PG and response, with a measured conversion gain of 58 µV per e–, about five
TG in the current design. The unbiased-gap region can cause a potential times higher compared with the conventional 3 T photodiode pixel (11 µV
barrier, which prevents the full depletion of the device37,38. Further per e–) structure27. This can be attributed to the isolation of the photode-
research is required to identify the cause of the remaining image lag. tector area from the FD region, which results in a lower FD capacitance.
In addition to the kTC noise reduction due to the full depletion of
Image sensor array demonstration the PG, the high conversion gain further improves the input-referred
A prototype image sensor array with the TF-PPD pixel structure was noise by suppressing noises from the subsequent readout chain. The
fabricated using a 130 nm node CMOS process. A back-gated IGZO TFT input-referred dark read noise is reduced to 6.1e– (compared with
is integrated on top of the fully processed silicon readout integrated 114.7e– for a conventional 3 T pixel with the same photodetector mate-
circuit, followed by processing of the optimized OPD. Figure 5a shows rial) due to the TF-PPD operation with only the minimum additional
an example prototype image sensor and camera demonstration. The complexity of one added TFT (TG) (Fig. 5f). Even with the addition of the
array consists of 64 × 64 pixels with a 5 µm pixel pitch. Figure 5b shows an in-pixel component, the proposed TF-PPD-based 4 T pixel is beneficial
image taken with the prototype image sensor. The operation of the for pixel scaling by enabling pixel transistor sharing (for example, for
TF-PPD based imager chip is confirmed by measuring the output of the RST, SF and SEL transistors), a technique widely used for conven-
the pixel SF without going through the readout chain by enabling a test tional silicon CMOS image sensors23,39–41. The quantum efficiency is

Nature Electronics | Volume 6 | August 2023 | 590–598 595

Article https://doi.org/10.1038/s41928-023-01016-9

Table 1 | Measured characteristics of TF-PPD (this work) compared with published state of the art

PD Photodetector/pixel structure Pixel Conversion FWC (e) Dark Dynamic Dark Notes (indicated
Stack pitch gain (µV e–1) read range current by * and **)
(µm) noise (dB) (nA cm−2)
(e)

This work OPD TF-PPD/4 T 5 58.1 25,000* 6.1 72.3 14.4 *FD capacitance
limited
3 T reference OPD Photodiode/3 T 5 11 90,000 114.7 57.9 60.1
2018 Panasonic7 OPD Photodiode/3 T + active 3 45,000 8.6 74.4
feedback
2021 STMicroelectronics15 QDPD Photodiode/3 T 1.62 10,000 21.3 53.4 250* *@60 °C
2020 imec27 QDPD Photodiode/3 T 5 8.3 470,000 33* 83.1** 300 *Interframe digital
CDS (×2 readout)**
Dual conversion
gain
2021 SWIR Vision Systems13 QDPD Photodiode/CTIA 15 550,000 65* 78.5 8 *Interframe digital
CDS (×2 readout)
2019 Sony45 OPD Photodiode/3 T 2 6.5* *Interframe digital
CDS (×2 read), 27e
@ single read
2021 Emberion46 QDPD Photodiode/CTIA 20 200 100* *Dark current @
0 °C, −0.86 V
2022 HiSilicon47 QDPD Photodiode/direction injection 15 10.0/1.42* 350,000 110 70.1 17.8 *Dual conversion
gain (high/low)
2015 NHK48 Silicon PPD/4 T 2.45 80 10,005 7.68 62.3 0.13* *@42 °C
2017 Sony 49
Silicon PPD/4 T 5.86 30.3 30,450 4.62 76.4
2021 Sony50 Silicon PPD/4 T 4.16 87.5 60,000 1.18 94.1* *Dual conversion
gain

measured to be 54% at 940 nm, which is comparable with the value from thin-film photodetectors into commercial image sensors, and retain-
a discrete photodiode. Dark current is reduced from 60.1 nA cm−2 of ing the advantages—such as higher absorption coefficients and wider
the conventional 3 T pixel to 14.4 nA cm−2 (Fig. 5g), due to the isolation wavelength range—of thin-film absorbers over silicon-based sensors.
of the photodiode from the high-defect silicon contact region, which
contributes about 10 fA to the dark current42. The improvement can be Methods
noticed when comparing the images captured under the same condi- Device fabrication
tions using the 4 T and 3 T pixel sensors (Fig. 5h and Supplementary The silicon substrates including TiN bottom contacts and Al2O3 dielec-
Fig. 11). The lower noise and dark current, combined with the higher tric for the photodiode, TFT and TF-PPD passive devices were fabricated
conversion gain, results in a brighter image with better image quality. in a 200-mm-node semiconductor line. The CMOS readout electronics
Table 1 summarizes the measured performance of the proposed was fabricated using a 130-nm-node line. Al2O3 was deposited using
TF-PPD image sensor and compares it with published state-of-the-art atomic layer chemical vapour deposition. IGZO was deposited by
thin-film- and silicon-based image sensors. The TF-PPD image sensor sputtering using a target with an In:Ga:Zn ratio of 1:1:1 at room
compares favourably against the conventional approaches, mainly temperature with an oxygen flow ratio (in O2 + Ar sputtering gas
because of its low dark read noise and low dark current. The relatively mixture) of 6%. After deposition, IGZO was patterned by photolithog-
low FWC is not limited by the TF-PPD itself but by the small FD capaci- raphy and wet etching with oxalic acid, and then annealed in FOG
tance of the current implementation. The FWC could be extended atmosphere (forming gas, 5% H2 in N2) at 180 °C for 30 min. Next, an
by using dual conversion gain techniques, which are widely used for organic absorber was deposited by spin coating using a donor–accep-
conventional CMOS image sensor designs43,44. The FWC of the TF-PPD tor blend solution. After drying, a film thickness of about 120 nm was
is expected to be about five times higher than that of the 3 T pixel struc- achieved. For the QD absorber, films were deposited by spin coating a solu-
ture due to higher metal–oxide–semiconductor capacitance than the tion of PbS QDs in octane (40 mg ml−1) at 2,500 r.p.m. for 15 s. Solid-state
photodiode junction capacitance. ligand exchange was performed by soaking the deposited film in a
ligand solution, followed by spin coating to dry the film. The film was
Conclusions then washed with a pure ligand solvent twice to remove the excess
We have reported a thin-film-based PPD image sensor that can offer cor- ligands. For the n-type film, ZnI2–3-mercaptopropionic acid mixture
related double sampling operation by fully transferring charges from the was used as the ligand, whereas for the p-type film, 1,4-benzenedithiol
PG to the FD node. The image sensors are created by integrating either was used. The deposition, ligand exchange and washing processes
an OPD or a colloidal QDPD with a PG structure made of IGZO TFTs on were repeated multiple times until the required thickness was achieved.
top of CMOS readout circuits. The TF-PPDs show reduced kTC noise, These steps were performed in a N2-filled glovebox. Then, the film
increased conversion gain and reduced dark current compared with was annealed at 80 °C for 10 min under N2. After the formation of
state-of-the-art thin-film imagers along with conserving the absorption the p-type film, the substrates were placed overnight in a dry-air
advantages of thin-film photodiodes. Our PG structure could potentially box to enhance p doping, and afterwards returned to the glovebox
be integrated with various other thin-film materials, such as perovs- for the remaining fabrication steps. A PolyTPD layer was formed by
kites, that are currently limited by pixel architectures. The thin-film spin coating a 6 mg ml−1 solution in chlorobenzene at 1,000 r.p.m.
PPD structure addresses a key obstacle to the widespread adoption of for 60 s. The film was then annealed at 80 °C for 20 min. For the

Nature Electronics | Volume 6 | August 2023 | 590–598 596

Article https://doi.org/10.1038/s41928-023-01016-9

HTL, 100 nm of N1,N1′-(biphenyl-4,4′-diyl)bis(N1-phenyl-N4,N4-di-m- 8. Nishimura, K. et al. Advanced features of layered-structure

tolylbenzene-1,4-diamine) (DNTPD) was deposited by thermal evapora- organic-photoconductive-film CMOS image sensor: over
tion using a Kurt J. Lesker vacuum system. To reduce the contact resist- 120 dB wide dynamic range function and photoelectric-
ance, a 10 nm doped DNTPD layer with 5% F6TCNNQ was added on top. conversion-controlled global shutter function. Jpn. J. Appl. Phys.
The ratio of DNTPD and F6TCNNQ was controlled by their evaporation 57, 1002B4 (2018).
rates. Finally, the ITO top transparent electrode was deposited by sputter- 9. Basiricò, L. et al. Direct X-ray photoconversion in flexible organic
ing using an Angstrom Engineering physical vapour deposition system. thin film devices operated below 1 V. Nat. Commun. 7, 13063
(2016).
Film characterization 10. van Breemen, A. J. J. M. et al. Curved digital X-ray detectors. npj
The absorption spectra were measured using a Shimadzu UV-1601 Flex. Electron. 4, 22 (2020).
UV-VIS spectrophotometer. The film morphology was inspected using 11. Tsai, H. et al. A sensitive and robust thin-film X-ray detector using
a Bruker atomic force microscope in the tapping mode in ambient air. 2D layered perovskite diodes. Sci. Adv. 6, eaay0815 (2020).
The thickness profiles of the different films were measured using a 12. Goossens, S. et al. Broadband image sensor array based on
Bruker DektakXT profiler. For the TEM cross sections, specimen prepa- graphene–CMOS integration. Nat. Photon. 11, 366–371 (2017).
ration was performed by focused ion beam with a Helios 450 system and 13. Gregory, C., Hilton, A., Violette, K. & Klem, E. J. D. 66‐3: invited
the inspection was done with a Titan G2 60-300 TEM system. paper: colloidal quantum dot photodetectors for large format
NIR, SWIR, and eSWIR imaging arrays. SID Symp. Dig. Tech. Pap.
Device characterization 52, 982–986 (2021).
For the photodiode characterization, current–voltage (I–V) curves 14. Pejovic, V. et al. Infrared colloidal quantum dot image sensors.
were measured in a glovebox with an Agilent 4156 semiconductor IEEE Trans. Electron Devices 69, 2840–2850 (2022).
parameter analyser in the dark. The EQE spectra were recorded on a 15. Steckel, J. S. et al. 1.62 µm global shutter quantum dot image
Stanford Research System model SR830 lock-in amplifier unit coupled sensor optimized for near and shortwave infrared. In 2021 IEEE
with a monochromator and a 500 W xenon lamp. The signal was first International Electron Devices Meeting (IEDM) 23.4.1–23.4.4 (IEEE,
amplified by a Stanford Research System SR570 low-noise current 2021).
preamplifier. A calibrated silicon photodiode with a known spectral 16. Weckler, G. P. Operation of p-n junction photodetectors in a
response was used as a reference. The transient photoresponse, pho- photon flux integrating mode. IEEE J. Solid-State Circuits 2, 65–73
tocurrent linearity and capacitance–voltage measurements were per- (1967).
formed using a Paios all-in-one system by Fluxim. 17. Dyck, R. H. & Weckler, G. P. Integrated arrays of silicon
The TFT measurements were performed in air using an Agilent photodetectors for image sensing. IEEE Trans. Electron Devices 15,
4156 semiconductor parameter analyser. 196–201 (1968).
The TF-PPD passive devices were characterized by a custom-made 18. Teranishi, N., Kohono, A., Ishihara, Y., Oda, E. & Arai, K. No image
printed circuit board probe card (Supplementary Fig. 10). The printed lag photodiode structure in the interline CCD image sensor. In
circuit board integrated an off-the-shelf CTIA (Texas Instruments 1982 International Electron Devices Meeting 324–327 (IRE, 1982).
ACF2101) and field-programmable gate array (Xilinx Artix-7). The 19. Fossum, E. R. CMOS image sensors: electronic camera-on-a-chip.
output of the CTIA is connected to the oscilloscope to be recorded. The IEEE Trans. Electron Devices 44, 1689–1698 (1997).
device was illuminated by a Thorlabs M940L3 940 nm LED or M530L4 20. Chamberlain, S. G. Photosensitivity and scanning of silicon image
530 nm LED, which were modulated by pulse signals from a Thorlabs detector arrays. IEEE J. Solid-State Circuits 4, 333–342 (1969).
DC2200 LED driver. 21. Fry, P. W., Noble, P. J. W. & Rycroft, R. J. Fixed-pattern noise in
photomatrices. IEEE J. Solid-State Circuits 5, 250–254 (1970).
Data availability 22. Fossum, E. R. Active pixel sensors: are CCDs dinosaurs? in
The data that support the plots within this paper and other findings of Charge-Coupled Devices and Solid State Optical Sensors III
this study are available from the corresponding author upon reason- Vol. 1900 (ed Blouke, M. M.) 2–14 (SPIE, 1993).
able request. 23. Fossum, E. R. & Hondongwa, D. B. A review of the pinned
photodiode for CCD and CMOS image sensors. IEEE J. Electron
References Devices Soc. 2, 33–43 (2014).
1. Rand, B. P. et al. The impact of molecular orientation on 24. Fossum, E. R., Mendis, S. & Kemeny, S. E. Active pixel sensor with
the photovoltaic properties of a phthalocyanine/fullerene intra-pixel charge transfer. US patent 5,471,515 (1995).
heterojunction. Adv. Funct. Mater. 22, 2987–2995 (2012). 25. Stephan, B. et al. A vertically integrated high resolution active
2. Georgitzikis, E. et al. Integration of PbS quantum dot photodiodes pixel image sensor for deep submicron CMOS processes. In Proc.
on silicon for NIR imaging. IEEE Sens. J. 20, 6841–6848 (2020). 1999 IEEE Workshop on Charge-Coupled Devices and Advanced
3. Georgitzikis, E. et al. Optimization of charge carrier extraction in Image Sensors 64 (1999).
colloidal quantum dots short-wave infrared photodiodes through 26. Schulze Spuentrup, J. D. et al. Thin film on CMOS active pixel
optical engineering. Adv. Funct. Mater. 28, 1804502 (2018). sensor for space applications. Sensors 8, 6340–6354 (2008).
4. Leemans, J. et al. Colloidal III–V quantum dot photodiodes for 27. Lee, J. et al. Imaging in short-wave infrared with 1.82 μm pixel pitch
short‐wave infrared photodetection. Adv. Sci. 9, 2200844 quantum dot image sensor. In 2020 IEEE International Electron
(2022). Devices Meeting (IEDM) 16.5.1–16.5.4 (IEEE, 2020).
5. Ackerman, M. M., Tang, X. & Guyot-Sionnest, P. Fast and sensitive 28. Blanksby, A. J., Loinaz, M. J., Inglis, D. A. & Ackland, B. D. Noise
colloidal quantum dot mid-wave infrared photodetectors. performance of a color CMOS photogate image sensor. In
ACS Nano 12, 7264–7271 (2018). International Electron Devices Meeting, IEDM Technical Digest 47,
6. Li, Z. et al. Halide perovskites for high-performance X-ray 205–208 (IEEE, 1997).
detector. Mater. Today 48, 155–175 (2021). 29. Arora, H. et al. Amorphous indium-gallium-zinc-oxide as electron
7. Nishimura, K. et al. An 8K4K-resolution 60 fps 450ke–-saturation- transport layer in organic photodetectors. Appl. Phys. Lett. 106,
signal organic-photoconductive-film global-shutter CMOS image 143301 (2015).
sensor with in-pixel noise canceller. In 2018 IEEE International 30. Janesick, J. R., Elliott, T., Collins, S. & Blouke, M. M. Scientific
Solid-State Circuits Conference (ISSCC) 82–84 (IEEE, 2018). charge-coupled devices. Opt. Eng. 26, 692–714 (1987).

Nature Electronics | Volume 6 | August 2023 | 590–598 597

Article https://doi.org/10.1038/s41928-023-01016-9

31. Voznyy, O. et al. A charge-orbital balance picture of doping 47. Liu, J. et al. A near-infrared colloidal quantum dot imager with
in colloidal quantum dot solids. ACS Nano 6, 8448–8455 monolithically integrated readout circuitry. Nat. Electron. 5,
(2012). 443–451 (2022).
32. Pejović, V. et al. Photodetectors based on lead sulfide quantum 48. Funatsu, R. et al. A 33 Mpixel 60 fps CMOS image sensor with
dot and organic absorbers for multispectral sensing in the visible 32-column shared high-speed column-parallel SAR ADCs. In 2015
to short‐wave infrared range. Adv. Funct. Mater. 32, 2201424 IEEE International Solid-State Circuits Conference—(ISSCC) Digest
(2022). of Technical Papers 58, 1–3 (IEEE, 2015).
33. Bi, Y. et al. Infrared solution-processed quantum dot solar cells 49. Oike, Y. et al. 8.3 M-pixel 480 fps global-shutter CMOS image
reaching external quantum efficiency of 80% at 1.35 µm and Jsc in sensor with gain-adaptive column ADCs and chip-on-chip
excess of 34 mA cm−2. Adv. Mater. 30, 1704928 (2018). stacked integration. IEEE J. Solid-State Circuits 52, 985–993
34. Kramer, I. J. et al. Efficient spray-coated colloidal quantum dot (2017).
solar cells. Adv. Mater. 27, 116–121 (2015). 50. Okada, C. et al. A 50.1-Mpixel 14-bit 250-frames/s back-illuminated
35. Choi, H. T. et al. Zero-dimensional PbS quantum dot-InGaZnO film stacked CMOS image sensor with column-parallel kT/C-canceling
heterostructure for short-wave infrared flat-panel imager. ACS S&H and ΔΣADC. IEEE J. Solid-State Circuits 56, 3228–3235 (2021).
Photon. 7, 1932–1941 (2020).
36. Xu, Y. & Theuwissen, A. J. P. Image lag analysis and photodiode Acknowledgements
shape optimization of 4T CMOS pixels. In Proc. International We would like to acknowledge the support from the Mitsubishi
Image Sensor Workshop 153–157 (2013). Chemical Corporation for providing the organic semiconductor
37. Belmonte, A. et al. Capacitor-less, long-retention (>400s) materials used in this study.
DRAM cell paving the way towards low-power and high-density
monolithic 3D DRAM. In 2020 IEEE International Electron Devices Author contributions
Meeting (IEDM) 28.2.1−28.2.4 (IEEE, 2020). J.L. and E.G. designed, performed and analysed the experiments. P.H.,
38. Han, L., Yao, S. & Theuwissen, A. J. P. A charge transfer model P.E.M. and K.M. analysed the experiments. N.P. and F.D.R. designed the
for CMOS image sensors. IEEE Trans. Electron Devices 63, 32–41 experiments. M.J. and J.H.K. performed the experiments. Y.H., A.B.S.,
(2016). D.C., V.P. and G.U. performed the photodiode analysis and fabrication.
39. Young Chan, K. et al. 1/2-inch 7.2 Mpixel CMOS image sensor with N.C., R.P., Y.L. and G.K. performed the fabrication. All authors have
2.25/spl mu/m pixels using 4-shared pixel structure for pixel-level reviewed the paper.
summation. In 2006 IEEE International Solid State Circuits
Conference—Digest of Technical Papers 39, 1994–2003 (IEEE, Competing interests
2006). The authors declare no competing interests.
40. Kim, H. et al. 5.6 A 1/2.65in 44 Mpixel CMOS image sensor with
0.7 µm pixels fabricated in advanced full-depth deep-trench Additional information
isolation technology. In 2020 IEEE International Solid-State Supplementary information The online version contains supplementary
Circuits Conference—(ISSCC) 104–106 (IEEE, 2020). material available at https://doi.org/10.1038/s41928-023-01016-9.
41. Mori, M., Katsuno, M., Kasuga, S., Murata, T. & Yamaguchi, T.
A 1/4in 2 M pixel CMOS image sensor with 1.75 transistor/pixel. Correspondence and requests for materials should be addressed to
In 2004 IEEE International Solid-State Circuits Conference (IEEE Jiwon Lee.
Cat. No. 04CH37519) 47, 110–111 (IEEE, 2003).
42. Kim, J. H. et al. Detailed characterization of short-wave infrared Peer review information Nature Electronics thanks Konstantin
colloidal quantum dot image sensors. IEEE Trans. Electron Devices Stefanov and the other, anonymous, reviewer(s) for their contribution
69, 2900–2906 (2022). to the peer review of this work.
43. Huggett, A., Silsby, C. & Cami, S. A dual-conversion-gain video
sensor with dewarping and overlay on a single chip. In 2009 IEEE Reprints and permissions information is available at
International Solid-State Circuits Conference—Digest of Technical www.nature.com/reprints.
Papers 52–53,53a (IEEE, 2009).
44. Oh, Y. et al. A 0.8 μm nonacell for 108 megapixels CMOS image Publisher’s note Springer Nature remains neutral with regard to
sensor with FD-shared dual conversion gain and 18,000e– full-well jurisdictional claims in published maps and institutional affiliations.
capacitance. In 2020 IEEE International Electron Devices Meeting
(IEDM) 16.2.1–16.2.4 (IEEE, 2020). Springer Nature or its licensor (e.g. a society or other partner) holds
45. Togashi, H. et al. Three-layer stacked color image sensor with exclusive rights to this article under a publishing agreement with
2.0-μm pixel size using organic photoconductive film. In 2019 the author(s) or other rightsholder(s); author self-archiving of the
IEEE International Electron Devices Meeting (IEDM) 16.6.1–16.6.4 accepted manuscript version of this article is solely governed by the
(IEEE, 2019). terms of such publishing agreement and applicable law.
46. Allen, M., Bessonov, A. & Ryhänen, T. 66‐4: invited paper:
graphene enhanced QD image sensor technology. SID Symp. Dig. © The Author(s), under exclusive licence to Springer Nature Limited
Tech. Pap. 52, 987–990 (2021). 2023

Nature Electronics | Volume 6 | August 2023 | 590–598 598