In the format provided by the authors and unedited.

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

SUPPLEMENTARY INFORMATIONDOI: 10.1038/NPHOTON.2017.75

NATURE PHOTONICS | www.nature.com/naturephotonics 1

Supplementary Materials for

Broadband image sensor array based on graphene-CMOS integration

Stijn Goossens1,*, Gabriele Navickaite1,*, Carles Monasterio1,*, Shuchi Gupta1,*, Juanjo Piqueras1, Raul Perez1,

Gregory Burwell1, Ivan Nikitskiy1, Tania Lasanta1, Teresa Galan1, Eric Puma1, Alba Centeno3, Amaia Pesquera3, Amaia Zurutuza3, Gerasimos Konstantatos1,2,†, Frank Koppens1,2,†

1 ICFO-Institut de Ciencies Fotoniques, The Barcelona Institute of Science and Technology, 08860 Castelldefels

(Barcelona), Spain. 2 ICREA – Institució Catalana de Recerça i Estudis Avancats, Lluis Companys 23, Barcelona, Spain.

3Graphenea SA, 20018 Donostia-San Sebastian, Spain * These authors contributed equally to this work

correspondence to: gerasimos.konstantatos@icfo.eu , frank.koppens@icfo.eu

Supplementary Methods

ROIC post processing

Graphene channel fabrication

The post processing of the ROIC die involves a transfer of a sheet of CVD grown graphene

on copper (supplied by Graphenea S.A. and cut to the size of the ROIC die) using a wet

transfer process followed by pixel patterning using a photoresist mask and reactive ion

etching using an argon and oxygen plasma. Contact patterning and metallization was done

using a lift-off process. Figure S1 shows a Raman map of the resulting graphene pixels.

PbS QD layer deposition

The synthesis of PbS nanocrystals was carried out under inert conditions using a Schlenk line

as previously described in literature.1,2 The final PbS oleate-capped nanocrystals were

dispersed in toluene for device fabrication. The PbS layers were spin casted in a layer-by-

layer process followed by a ligand exchange with EDT (1,2-ethanedithiol).

Prototype digital camera description

The graphene-quantum dot digital camera was operating at room temperature in

ambient conditions. The ROIC die was first bonded to a LCC chip carrier and clamped into a

custom rig that provides the electrical connections to a control system that comprises the

following blocks (Figure S2):

(1) Power supply unit to provide external bias voltages, including readout circuitry and pixel

supplies;

(2) Digital sequencing unit to generate all clock and control signals responsible for the sensor

timing, e.g. array row selection, exposure time definition, or shutter operation; and

(3) 12-bits digitizer to convert the sensor output signal to digital values.

these three blocks are controlled by an embedded computer via a PXI (PCI eXtensions for

Instrumentation) bus, and interface the image sensor through a dedicated PCB board. The

camera allows operating and reading out the image sensor at 50 frames per second.

The user interface is directly connected to the embedded computer, which runs both

acquisition and analysis software packages. The acquisition software is written in C# while

all analysis is performed using Python. Both embedded computer and PXI modules are off-

the-shelf components from National Instruments.

Image capturing

The visible image Fig 2c was captured in reflection mode by illuminating a target

image with a desklamp fitted with a 6.5 W 3000K LED bulb. The lamp was directed to a

picture placed in front of the lens. A f/2.8 objective of 50 mm focal length was used to

project the image on the image sensor. The irradiance behind the objective was ~1·10-4

W/cm2.

The objects for the infrared images (Fig 2b,d,e,g,i) were illuminated with a 1000W,

3200K incandescent lamp yielding an irradiance behind the objective of ~10-4 W/cm2. We

placed a 1100nm long pass filter in the optical path for the images that were taken purely

with infrared light. The lamp was directed to 3 dimensional objects placed in front of the lens.

A f/2.9 objective of 75 mm focal length was used to project the image on the image sensor.

The image capture process consists of collecting a sequence of ~100 frames at ~14

fps. As the pixel resistance uniformity and drift were not optimized, we used a chopper to

modulate the light at 1 Hz which enabled us to obtain the dark current of the pixels, hence the

signal we plot is dV=Vout,light - Vout,dark. To correct for photo-response non-uniformity we

obtained a white reference map by placing a RESTAN target or a blank paper at the end of an

image capturing sequence.

In Figure S3 we summarize the detailed setup for capturing each of the images in

Figure 2.

A stage fog machine generated the fog in Figure 2e,f. This fog is much denser than

outdoor fog. The silicon wafer that we used in figure 2g,h was standard, low-doped silicon.

Data processing

First we corrected drift for the captured time sequence per pixel by selecting two

periodic points, fitting a line and subtracting that line from the data. Then we performed a

FFT to obtain the power spectral density (PSD) versus frequency of the signal. From the PSD

we extracted the square root of the level at 1 Hz for each pixel. We performed this for both

the image and the white reference data. By calculating dVimage/dVwhite for each pixel we

corrected for light response non-uniformities. Pixels with dVimage > dVwhite were marked as

non-working. Those pixels were patched with an infilling algorithm that takes the mean of

the surrounding, working pixels. Finally we performed a median filter to get rid of salt and

pepper noise.

Optical system description

The power dependencies in Figs. 3d and 4d in the main text were measured by

attenuating the light of a fiber-coupled laser with a digital variable attenuator before it was

coupled into an integrating sphere that uniformly illuminates the image sensor (or reference

detector). In between the image sensor and the integrating sphere we placed a chopper. For

the visible image sensor we used a 633 nm fiber coupled diode laser source and for the

infrared image sensor a fiber coupled super continuum laser with AOTF module. The data

acquisition was performed in the same way as for the image capturing.

For the acquisition of the spectra in Figs. 4a,b in the main text a NKT supercontinuum

laser with AOTF module was used. The laser was fiber-coupled to a collimator (0.98 mm

beam diameter) that was directly illuminating the sample. Between the collimator and the

device a chopper was placed. For each wavelength we acquired a time sequence of frames

from which we extracted dV.

Reference photodetectors

Reference photodetectors were made on standard n++ Si / 285 nm SiOx substrates

using a wet transfer technique of CVD graphene (obtained from Graphenea S.A.) followed by

pixel patterning using a photoresist mask and etching with an argon and oxygen plasma.

Contact patterning and metallization was done using a lift-off process. The quantum dots and

quantum dot layer application details are the same as for the ROIC process (see above).

Night-glow measurement

For the night-glow measurement shown in Fig. 4C of the main text, we used a

reference photodetector with an area of 1 mm². The device was mounted in a setup without

lens, operating at room temperature in ambient conditions. The detector was pointing to the

sky at an angle of ~30 degrees. For obtaining a high signal-to-noise ratio we modulated the

incoming light at 9.9 Hz with a chopper and read the light signal using a lock-in technique

based on a software FFT. We used long-pass filters on top of the detectors for selecting

spectral bands. In Figure S4 we show a schematic layout of the setup

Supplementary Notes

Signal path image sensor

In Figure S5 we show the full functional diagram of the image sensor. From the

control circuitry we can operate the VDD , VSS and VREF voltages. In practice we used the

controllable compensating resistor for tuning the operating regime for each pixel.

Optimization of pixel conductance

Controlling the pixel dark conductance is essential to provide a proper matching

between photosensitive and blind pixels, which maximizes the photo-signal amplitude at the

imager output. Besides, homogeneous dark conductance within pixels is required to achieve

good spatial uniformity. Figure S6 shows how the dark signal of a pixel varies when the

pixel resistance changes, by tuning the compensating resistor. The optimum operating point

corresponds to the case when the output dark signal is 0V. A change of ±1 kΩ around the

optimum value reduces the photo-signal by more than 50%.

Electro-optical characterization

Resistance maps

We have fabricated different ROIC dies. In Figure S7 we show resistance data

obtained from three different ROIC dies covered with graphene and colloidal quantum dots.

Die 1 was used to obtain the images shown in the main text in Fig. 2c. As we transferred the

CVD graphene sheets by hand on the ROIC dies, alignment was not always perfect, hence the

pixelated (active) area of the ROIC was partially (>85%) covered with graphene.

Yield calculation

The image sensor has in total 111,744 pixels. When characterizing the image sensor,

we detect pixels that have a conductive graphene channel by sweeping the Rcomp resistor value

and recording Vout. If Vout crosses 0 V the graphene channel is conductive. This is the case for

94,983 pixels for the visible and NIR image sensor, which gives a graphene coverage of 85%.

A large part of the 15% of non-working pixels can be attributed to the area that is not covered

with graphene. If we select a rectangular area of 255 x 345 (87,975) pixels within the area

that is covered with graphene we find 87,834 pixels that have a Vout that crosses 0 V.

Therefore, the yield of the area with transferred graphene is 99.8 %, close to unity.

Noise characterization

We performed a noise analysis of the graphene quantum dot hybrid photodetectors by

recording a time trace of the current under constant source-drain bias in the dark. Taking the

Fourier transform of the time trace yields a typical noise spectrum as plotted in Fig. S8 with a

1/frequency dependence. Moreover, we find that the noise scales linearly with source-drain

bias and inversely with the area of the graphene channel. These observations are typical for

1/f noise in graphene 3.

To make a proper comparison to the pixels in the image sensors, we show in Fig. S8 a

noise spectrum obtained from a reference hybrid graphene quantum dot photodetector with a

similar area (308 µm2) as the pixels in the ROIC (255 µm2). At 1 Hz, the normalized power

spectral density SI/I2=6·10-10 Hz-1. The factor ß = (SI/I

2)(W*L) (defined at f=10 Hz) was

introduced by Stolyarov et al. 4. We observe ß = 1.8·10-8 µm2/Hz. Improvements in device

processing can reduce the 1/f noise such that lower noise equivalent irradiances can be

obtained. In graphene encapsulated with hBN ß=5·10-9 µm2/Hz has been observed 4.

To filter out the majority of 1/f noise, a lock-in scheme detection can be implemented:

the higher the modulation frequency, the lower the noise. The data presented in Fig. 3D in the

main text in purple was obtained using a lock-in detection with a modulation at 100 Hz,

yielding an NEI of 9·10-10 W/cm2 for a pixel of 384 µm2. The measurement of the night glow

(Fig. 4C in the main text) was performed at a lock-in frequency of 9.9 Hz. In an image sensor

however, a lock-in scheme is not practical and a broadband read–out is most often used. This

system integrates all noise up to the cut-off frequency of the read-out system: the area under

the data in Fig. S8.

Performance summary

In Table S1 we summarize the performance parameters extracted from Fig 3D and 4D

in the main text. We remark that the graphene-QD reference detectors exhibit a response time

below 1 ms, and this time response can in principle also be achieved for the imager, by

optimizing the design.

Performance projection

In Table S2 we calculated the noise equivalent irradiance (NEI) and specific

detectivity for various cases of an optimized graphene pixel read-out system and compared

those values to commercially available silicon CMOS and InGaAs image sensors. First for

modulation-type read-out which involves modulating the light signal and also for a standard

broadband amplifier system. We assumed 1/area scaling of the 1/f noise in graphene, which

we also verified experimentally 3. The NEI and detectivity were calculated for our sensors at

a read-out speed of 60 fps. We can improve the mobility of graphene (and hence the signal of

the detector) by more than a factor 10 by optimizing the substrate of the graphene (hBN for

example5).

The projected performance is comparable to the most recently available InGaAs

cameras. Those are not monolithic CMOS and are thus high cost (>15kE). Comparing to

uncooled InGaAs the projected performance of the graphene quantum dot hybrid imagers is

better. Especially for λ>1700 nm, the region where only extended InGaAs operates, the

detectivity of the graphene quantum dot hybrid sensors is at least an order of magnitude

better without the need of a power consuming four-stage thermo-electric cooler.

Pixel size and fill factor

The image sensors used for obtaining the data represented in the main text were based

on a graphene pattern that was optimized for the specific ROIC but sub-optimal in terms of

fill factor. In Figure S9a we show a schematic of the pixel design. As the source and drain

contacts in the off-the-shelve ROIC were placed in the corners and the resistance of the

graphene needed to be in the range 20-100 kΩ we patterned the graphene in an s-shape. The

fill factor for the image sensors used in the paper was less than 35%.

In a custom ROIC design we can increase the fill factor to almost 100%. For the large

pixel in Figure S9b the fill factor is 95%. Moreover, we can decrease the pixel size to 3x3

µm2 by relying on state of the art sub 100 nm precision lithography processes. For the design

in Figure S9c the fill factor is 93%.

References

1. Mihi, A., Beck, F. J., Lasanta, T., Rath, A. K. & Konstantatos, G. Imprinted Electrodes

for Enhanced Light Trapping in Solution Processed Solar Cells. Adv. Mater. 26, 443–

448 (2014).

2. Ip, A. H. et al. Hybrid passivated colloidal quantum dot solids. Nat. Nanotechnol. 7,

577–582 (2012).

3. Balandin, A. A. Low-frequency 1/f noise in graphene devices. Nat. Nanotechnol. 8,

549–55 (2013).

4. Stolyarov, M. A., Liu, G., Rumyantsev, S. L., Shur, M. & Balandin, A. A. Suppression

of 1/f noise in near-ballistic h-BN-graphene-h-BN heterostructure field-effect

transistors. Appl. Phys. Lett. 107, 23106 (2015).

5. Banszerus, L. et al. Ultrahigh-mobility graphene devices from chemical vapor

deposition on reusable copper. 1–6 (2015).

6. Tissot, J.-L. et al. High-performance uncooled amorphous silicon video graphics array

and extended graphics array infrared focal plane arrays with 17-μm pixel pitch. Opt.

Eng. 50, 61006 (2011).

7. Sony IMX377 CMOS image sensor. (2016). Available at: http://www.sony-

semicon.co.jp/products_en/IS/sensor2/img/products/IMX377CQT_ProductSummary_

v1.5_20150414.pdf.

8. Andanta FPA-640x512-TE2 InGaAs imager. (2012). Available at:

http://www.andanta.de/pdf/andanta_fpa640x512-te2.pdf.

9. Sensors Unlimited Micro-SWIR 640CSX Camera. Available at:

http://www.sensorsinc.com/images/uploads/documents/640CSX_commercial.pdf.

10. Theuwissen, A. J. P. CMOS image sensors: State-of-the-art. Solid. State. Electron. 52,

1401–1406 (2008).

Supplementary Figures

Figure S1 Raman map of the 2D-peak intensity (2682 cm-1) of an area of the ROIC with

patterned graphene on top (before depositing colloidal quantum dots).

Digitizer

Power Supply Unit

Digital Sequencing Unit

PXI embedded Computer

PXI BUS

PXI BUS

PXI BUS

IF PCB

Image Sensor

Figure S2 Block diagram of the camera prototype (optics not included).

Illumination 1 kW 3200 K

incandescent lamp

with 1100 nm long

pass filter

1 kW 3200 K

incandescent lamp

6.5 W 3000K LED

bulb

1 kW 3200 K

incandescent lamp

with 1100 nm long

pass filter

Object Apple and pear Box of apples Paper with image

Lena printed in

black and white

Glass with water

Lens system f/2.9, f=75 mm,

single lens

f/2.9, f=75 mm,

single lens

f/2.8, f=50 mm,

objective

f/2.9, f=75 mm,

single lens

Image sensor -

object distance

[cm]

100 100 60 100

Irradiance on

image sensor

[W/cm2]

1·10-4 1·10-4 1·10-4 1·10-4

Image sensor

wavelength

range [nm]

300-1850 300-1850 300-1000 300-1850

Figure S3 Overview of experimental details for obtaining the different images. The rightmost column

represents the setup for images e, g and i.

Figure S4 Schematic layout of the night-glow measurement setup. The image on the right is taken of

the sky with a standard silicon image sensor.

...Analog

Output(s)

AnalogOutput(s)

Capacitive TransImpedance

Amplifier

Sampling & Hold

BLIND PIXEL

Co

mp

ensa

tin

gR

esis

tor

Co

lum

n S

elec

tor

CDSCDS

Correlated Double

Sampling

OutputDriver

VREF

VSS

VDD

Row select

ACTIVE PIXELS (x288)

x388

Figure S5 Image sensor functional diagram. The active pixel is switched in a row-by-

row fashion. The two transistors in the circuit regulate the internal resistor biasing6.

Figure S6. Variation of a typical pixel’s output voltage as a function of the pixel dark

resistance, obtained by scanning an internal compensating resistor placed in series with the

photosensitive pixels inside the ROIC chip (Rcomp). The horizontal axis refers to resistance variation

with respect to a dark value of 28 kΩ.

1. S-shaped graphene 2. Meander shaped

graphene

3. Meander shaped

graphene

Bef

ore

ca

lib

rati

on

Aft

er c

ali

bra

tio

n

Res

ista

nce

his

tog

ram

s

Figure S7 Pixel resistance maps for three different ROIC dies with graphene. In the upper row we plot the

resistance before calibration with the compensation resistor (Rpixel) and in the middle row the values after

calibration (Rpixel + Rcomp). In grey we plot the pixels that do not show conductance. The bottom row represents

histograms for the different ROIC dies before (red) and after calibration (blue). The graphene channels in the

pixels of the ROIC dies in column 2 and 3 were etched in a meander shape to obtain larger channel resistance

values as is visible from the median values in the histograms (40 and 55 kOhm). For ROIC 3 the calibration was

not as effective, because the devices suffered from hysteresis.

Figure S8 Normalized noise power spectral density for a reference hybrid graphene quantum dot

photodetector on a n++ Si, 285 nm SiOx substrate sensitized with quantum dots with an exciton peak

of 1670nm (area=308 µm2, Vbackgate=0 V, Vsd=0.1 V) The graphene pixels of the ROIC used for Figure

2C,D in main text are 288 µm2. The dashed line has a slope of -1, showing the 1/f nature of the noise

in the device. The grey zone is the total noise for a read-out system with a bandwidth of 10 Hz . The

noise for a lock-in measurement at a modulation frequency of 10 Hz is indicated with the vertical

arrow.

Column

Row

1 388

1

288

Rpix

el [

k

]

5

10

15

20

25

30

35

40

Column

Row

1 388

1

288

Rpix

el [

k

]

10

20

30

40

50

60

70

80

Column

Row

1 388

1

288

Rpix

el [

k

]

10

20

30

40

50

60

70

80

Column

Row

1 388

1

288

Rpix

el [

k

]

0

5

10

15

20

25

30

35

40

Column

Row

1 388

1

288

Rpix

el [

k

]

0

10

20

30

40

50

60

70

80

Column

Row

1 388

1

288

Rpix

el [

k

]

0

10

20

30

40

50

60

70

80

10 15 20 25 30

500

1000

1500

2000

2500

3000

3500

Resistance [k]

Counts

Rpixel

Rpixel

+ calibrated Roffset

10 20 30 40 50 60 70 80

500

1000

1500

2000

2500

3000

3500

Resistance [k]

Counts

Rpixel

Rpixel

+ calibrated Roffset

50 100 150

500

1000

1500

2000

2500

3000

3500

Resistance [k]

Counts

Rpixel

Rpixel

+ calibrated Roffset

Gr substrate Gr substrateGraphene

So

urc

e

Dra

in

1 µm

35

µm

1 µm

35 µm

Gr substrateGraphene

So

urc

e

Dra

in

100 nm

3 µ

m3 µm

100 nm

Source

Drain

35 µm

Current designDesign with

optimized fill factorScaled down pixel

35

µm

a b c

Figure S9 Different pixel designs. Please note that the different elements in the drawings are not to

scale.

Supplementary Tables

Parameter Units Graphene-QD CMOS imager

Graphene-QD

reference detector

Graphene-QD CMOS imager

Graphene-QD reference detector

Exciton peak nm 920 1670 1580

Illumination nm 633 1670 1550

Pixel size (WxL) µm2 35x35 48x8 35x35 1000x1000

NEI W/cm2 < 10-8 9·10-10 <8·10-7 10-9

Detectivity D* Jones >6·1010 4·1012 >6·108 > 1012

Dynamic Range dB >55 > 80 >32 > 80

Response time ms 100 ms (limited by

ROIC design)

10 ms 100 ms (limited by

ROIC design)

< 1 ms

Table S1. Summary of imager and reference detector performance.

Gr-QD CMOS Modulation-type

read-out

Gr-QD CMOS Broadband read–out

Silicon CMOS

High performance

InGaAs

non-CMOS

Typical InGaAs

non-CMOS

Extended InGaAs non-CMOS

Parameter Units standard substrate, uncooled, small pixel

optimized substrate, uncooled, small pixel

standard substrate, uncooled, small pixel

optimized substrate, uncooled, small pixel

Uncooled, Small pixel

pitch, smartphone

type

Cooled, Sensors unlimited Micro SWIR 640CSX,

Uncooled, large pixel size

4 stage thermo-electric cooler, large pixel size

Wavelength

range

nm 300-2500 300-1100 700-1700 700-1700 1000-2500

Pixel pitch µm <3 <3 12.5 12.5 30

Power con-

sumption

[mW] t.b.d. (for the current 100k Pixel ROIC it was 211 mW)

400 7 3258 (uncooled)-2500(packaged and cooled)9

85·103

Pixel fall

time

ms <1 <1E-4 <1E-4 <1E-4 <1E-4

Quantum

efficiency

% >50 >50 >65 >65 >65

Dynamic

Range

dB >80 <80 68 68 68

NEI* W/cm2 3·10-10 <2·10-11 2·10-9 <4·10-10 6·10-10 2.1·10-10 6·10-9 6.2·10-9

Detectivity Jones 1·1013 * >5·1013 * 6·1011 * >9·1012 * 4·1013 2.8·1013 4·1012 6·1010

Table S2 Performance projection for an optimized Graphene quantum dot CMOS image sensor. The

specific wavelength for which the values are given is 1550nm except for the silicon CMOS . We

compared the performance to a state of the art Si CMOS image sensor that can be found in

smartphones10, to a non-ITAR SWIR camera that is manufactured by Sensors Unlimited9 and to non-

cooled InGaAs and cooled extended InGaAs camera pixels (Xenics Xeva 2.5). *detectivity and NEI at

60 fps, for photoconductive detectors the NEP is determined by the built-in time constant of the

detectors.