...

document

by user

on
Category: Documents
1

views

Report

Comments

Description

Transcript

document
A time-resolved 128128 SPAD camera for laser Raman spectroscopy
Yuki Maruyama1, Jordana Blacksberg2, and Edoardo Charbon1
1
2
Circuits and Systems, Delft University of Technology, Delft, the Netherlands
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 91109, USA
ABSTRACT
In this paper we present a time-gated single-photon avalanche diode (SPAD) array, the first of its kind to be integrated
with a newly developed time-resolved laser Raman spectrometer. Time-resolved Raman spectra from various highly
fluorescent minerals were successfully observed using our SPAD array; these spectra were obscured by an
overwhelming fluorescence background when measured using a traditional continuous wave green laser. The system has
photon detection efficiency (PDE) of 5 % at 5 V excess bias with on-chip microlenses. The dark count rate (DCR) of
this SPAD is 1.8 kHz at 5 V excess bias. However, thanks to the nanosecond scale time-gating, noise rate per frame is
effectively reduced to ~10-3 counts at 40 kHz laser repetition rate.
Keywords: Time-resolved laser Raman spectroscopy, CMOS image sensor, Single-photon avalanche diode, SPAD
1. INTRODUCTION
Raman spectroscopy is a nondestructive optical analysis technique that obtains structural and compositional information
from organic and inorganic samples. Thanks to its high identification performance, an in situ Raman spectrometer is
proposed for many future planetary surface missions, for example ExoMars and Max-C rovers [1]. However, in
continuous-wave (CW) laser Raman spectroscopy, significant background fluorescence often overwhelms the Raman
signature. To overcome this issue, fast time-resolved laser Raman spectroscopy has been developed using a streak
camera [2]. A time-resolved laser Raman spectrometer offers significant background reduction based on the temporal
difference between Raman signature (virtually instantaneous) and large background fluorescence (ns ~ ms). However,
such instruments require highly sensitive, high speed detectors such as intensified charge-coupled devices (iCCDs) or
streak cameras. Thus, time-resolved laser Raman spectroscopy has not been used commercially, in part due to cost, size,
and complexity.
Recently, a novel solution based on single-photon avalanche diodes (SPADs) has emerged as the technology of choice to
create compact, time-resolved image sensors in standard complementary metal-oxide semiconductor (CMOS) process
[3]. Nowadays, SPADs are available in standard nanometer scale CMOS technology [4],[5], which allows integration of
high-performance circuitry and parallel data processing [6].
This paper presents an all-digital, time-gated 128128 CMOS SPAD imager for time-resolved laser Raman spectroscopy.
Raman signatures observed in this study are comparable to those obtained with existing high-end technology based on a
streak camera, which requires high operating voltages, typically in the kV range. The time-gated SPAD imager offers
background fluorescence rejection with a significant reduction in size and power. This simplification of the timeresolved Raman spectrometer is a great advantage for in situ planetary exploration.
2. A TIME-GATED SPAD CAMERA
2.1 SPAD camera system
Figure 1 shows a block diagram of the proposed sensor and a simplified optical setup for time-resolved Raman
spectroscopy. A small fraction of the pulsed laser light is used as an optical trigger which is detected by a single SPAD.
Timing of the trigger signal is controlled by the external delay line to activate SPADs when the Raman spectrum is
projected on the image plane. A microphotograph of the SPAD imager and readout system are shown in Figure 2. The
chip was fabricated in 0.35 μm high-voltage standard CMOS technology. The pixel pitch is 25 μm and the total area
20.5 mm2. The close-up of the SPAD is shown as the figure inset. The time-gated, 128128 CMOS SPAD imager was
Updated 1 March 2012
first realized for on-chip fluorescence detection and fluorescence lifetime imaging microscopy (FLIM) in a time-gated
mode of operation [7]. The system was modified for time-resolved laser Raman spectroscopy.
Raman
spectrum
Optical delay line
Pulsed
laser
Delay canceller and buffer
Spectro
meter
FPGA
PC
USB 2.0
interface
Row decoder
Single SPAD
(optical trigger)
Delay line
USB
controller
Gating
signals
FIFO
memory
Control
signals
Reset signal generator
Mineral sample
128 x 128
SPAD array
Delay canceller and buffer
16bit counter / shift register
Column decoder
Figure 1: Block diagram of the proposed sensor and a simplified optical setup. Raman spectrum is projected on the SPAD
imager. A single SPAD is used to trigger the entire camera system. Each pixel’s state is stored in a 1-bit counter that is read
out in rolling shutter mode, accumulated and serialized on-chip, and send to the operation computer via a FIFO memory and
a USB communication module.
Delay canceller and Buffer
1cm
128 x 128
SPAD imager
USB 2.0 interface
Row decoder
128 x 128
SPAD array
SPAD
FPGA
10mm
Delay canceller and Buffer
16bit counter
shift register
Delay-line
0.5mm
Custom-made PCB
Column decoder
Power supplies
Spartan-3 board
Figure 2: A time-gated SPAD camera system. (a) Photograph of the sensor chip and a readout system. The chip, fabricated
in a 0.35 μm high-voltage standard CMOS process, has a total image plane area of 3.2 mm  3.2 mm. The pixel pitch is 25
μm. The inset shows close-up of the pixel without microlens.
2.2 Pixel architecture
Figure 3(a) shows a schematic diagram of the pixel. When the photon arrives at the image plane, the SPAD array is only
activated by properly controlled Trecharge followed by Tgate. The system has photon detection efficiency (PDE) of 5 % at
5 V excess bias with on-chip microlenses (CF=1.59). The median dark count rate (DCR) of this SPAD imager is 1.8
kHz at 5 V excess bias [8]. However, thanks to the nanosecond scale time-gating, noise rate per frame was reduced
Updated 1 March 2012
down to ~10-3 counts at 40 kHz laser repetition rate. A simplified timing diagram is shown in Figure 3(b). The trigger
signal enables Trecharge to activate the entire SPAD array. Then, signal Tgate allows the 1-bit counter to capture the
avalanche events caused by the Raman spectrum. The counter output is read out in rolling shutter mode every 409 μs.
Therefore, 16 excitation laser pulses reach the sample to cause Raman spectrum emission per frame at 40 kHz laser
repetition rate. Due to the low efficiency of the process, on average much fewer than 1 event per frame occur, thus
several tens or perhaps hundred frames are necessary to achieve accurate and statistically relevant measurements.
VDD
Vtopgate
VOP
VDD
Tspadoff
VDD
VDD
Trow_sel
Tgate
SPAD
Treset
Trecharge
(a)
Raman signal
Trigger
33 ns
Tspadoff
Trecharge
2 ns
Tgate
7 ns
Photon
SPAD anode
SPAD active
Counter input
Counter output
Trow_sel
Treset
Counter reset
Counter reset
Data readout
Data readout
1frame = 409 ms
(b)
Figure 3: Schematic diagram of the pixel. (a) Overall pixel schematics including a SPAD, a 1-bit counter and a readout
circuitry as well as Tspadoff, Trecharge and Tgate for time-gating. (b) Simplified timing diagram of the time-gated operation for
Raman spectroscopy.
2.3 Optical trigger module
Generally, passively q-switched solid state lasers have relatively large timing uncertainty. The pulsed laser used in this
study has ~ 1.5 μs timing jitter. Therefore, trigger signals must be created from each laser pulse to achieve nanosecond
scale gating. Figure 4(a) shows schematic diagram of the optical trigger module.
Updated 1 March 2012
ms order pulse
A
VOP
On-chip
Off-chip
B
SPAD
B
TQ
DL 2
A
DL 1
D
C
ns order trigger
D
C
(a)
t
(b)
SPAD
10mm
Inverter and
passive quenching transistor
0.5mm
(c)
Figure 4: The optical trigger module. (a) Schematic diagram of the optical trigger module. The delay line (DL1) controls
trigger width while the other delay line (DL2) controls the trigger timing. (b) Timing diagram of the trigger pulse
generation. Rising edge of the μs order pulse is trimmed down to the ns order trigger pulse. (c) Photograph of the optical
trigger chip and close-up of the single SPAD element. The chip was fabricated in 0.35 μm high-voltage standard CMOS
technology as well as the 128128 CMOS SPAD array.
A passively quenched SPAD output is buffered by two-stage inverters. Off-chip delay lines control the trigger width and
its timing to activate the SPAD camera when the Raman spectrum is arriving at the image plane. The rising edge of the
inverter output was trimmed by a NAND gate as shown in Figure 4(b). A SPAD dead time was set at 1 ms to minimize
afterpulses, which have negligible probability at the levels used during dead time. Figure 4(c) shows a microphotograph
of the trigger chip and close-up of the passively quenched SPAD as well as an integrated inverter.
3. DETECTOR PERFORMANCE
3.1 Noise and sensitivity
The sensor was first characterized in terms of its noise and sensitivity. To operate SPADs in Geiger mode, a reverse bias
voltage was applied to the SPADs in excess of its breakdown voltage of 19.1 V. Figure 5(a) shows the dark count rate
(DCR) cumulative probability plotted for different excess bias voltages (Ve) from 2.5 V to 4 V at room temperature.
The median DCR of this chip, 53 Hz with Ve=3 V, is superior to that measured in [9]. The proportion of noisy pixels
(DCR > 1 kHz) is 1.53 % with Ve=4 V, is significantly lower than that of [10]. We assume that the in-pixel 1-bit
counter plays a role in filtering afterpulses, since the memory stores only the first avalanche event in each frame. This
feature offers significant improvement in the image quality when the incident photon number is extremely low. Figure
5(b) plots the photon detection probability (PDP), as well as the photon detection efficiency (PDE) at different excess
biases. A peak PDP of 43.6 % at 475 nm was observed with Ve=4 V.
Updated 1 March 2012
Dark count rate (Hz)
105
104
103
Ve= 4.0V
102
3.5V
101
3.0V
100
2.5V
10-1
0
20
40
60
80
100
Percent of pixels (%)
(a)
50
2.3
Ve= 4.0V
1.8
30
20
10
1.4
3.5V
0.9
3.0V
PDE (%)
PDP (%)
40
0.5
2.5V
0
300
400
500
600
700
800
0
900 1000
Wavelength (nm)
(b)
Figure 5: Basic characteristics of the time-gated SPAD camera. (a) Cumulative dark count rate (DCR) for the pixel
population of the entire array (16,384 pixels). (b) Median photon detection probability (PDP) and photon detection
efficiency (PDE) of the entire array at different excess bias conditions without microlenses. PDE was calculated as PDP 
fill factor (4.5 %). All measurements were performed at room temperature.
3.2 On-chip microlens
For further improvement of PDE, on-chip microlenses were fabricated directly on the sensor surface. Figure 6(a) shows
the concentration factor (CF) map across the entire chip. Due to a misalignment that occurred during fabrication of the
microlenses, CF varies from 1 to 3 as shown in Figure 6(b). The measured median concentration factor is 1.59 which
improves the PDE of the SPAD camera. This issue will be solved by increasing the sensor size and optimizing the
microlens fabrication process. We believe that the microlens array can be improved to achieve its theoretical value of 15.
Updated 1 March 2012
0
1600
1400
2-3
1200
Row
1-2
0-1
Frequency
3-4
Median CF = 1.59
1000
25mm
800
600
400
200
127
0
0
0
127
Column
1
2
3
4
Concentration factor
(a)
(b)
Figure 6: The concentration factor (CF) of the on-chip microlens array. (a) The CF distribution across the entire chip. (b)
Histogram of the CF. The CF value varies from 1 to 3 due to the alignment and height issues. The inset shows a
microphotograph of circular shaped microlenses. Square microlenses (not shown here) have also been fabricated.
4. EXPERIMENTAL RESULTS
4.1 Experimental setup
Figure 7 shows block diagram of the simplified experimental setup. The time-resolved Raman spectroscopy instrument
was originally designed for detection using a streak camera [2]. Simple modifications to the setup allowed us to replace
the streak camera with our SPAD imager. Table 1 summarizes experimental parameters.
Focusing lens
Mirror
Spectrometer
Shutter
Dichroic
edge filter
Objective
lens
SPAD
imager
Optical
delay
Single SPAD
for trigger
Mirror
Mineral
Sample
Beam
expander
Pulsed
laser
Beam sampler
Figure 7: Block diagram of the simplified time-resolved Raman spectrometer setup.
Updated 1 March 2012
Mirror
Table 1. Experimental Parameters
Laser Wavelength
Laser Pulse Energy
Laser Pulse Width
Laser rep rate
Time per frame
Frames per measurement
Image plane size
Pixel size
SPAD on time
Trigger width
SPAD breakdown voltage
SPAD excess bias
SPAD dark count rate (DCR)
Microlens concentration factor
Photon detection efficiency (PDE)
Power consumption
532 nm
1 µJ/pulse
500 ps
40 kHz
409 µs
10,000 (typical)
3.2 mm  3.2 mm
25 µm  25 µm
7 ns
33 ns
19.1 V
5V
1830 Hz (at 5 V excess bias)
1.59 (median)
5 % (at 5 V excess bias)
360 mW (at Vdd = 2.5 V)
4.2 Time-resolved Raman spectra
Initial time-resolved Raman spectroscopy experiments were performed using natural calcite minerals since calcite has a
clear and intense Raman return. Figure 8 shows an example of the sensor output integrated over 10,000 frames at room
temperature. The Raman spectra were successfully observed with high signal-to-noise ratio (SNR) without any cooling
system, which is commonly used in CCD based CW Raman spectrometer designed for in situ planetary exploration
[11],[12].
As a demonstration of the large fluorescence rejection capabilities of this system, we measured the natural mineral
willemite which has very intense green fluorescence. In this mineral, Raman spectra are completely obscured by this
large fluorescence in the absence of time gating. Figure 9 shows the Raman spectra observed from willemite at different
time-gating window. Significant background fluorescence was suppressed by reducing the time-gating window from > 1
μs to < 33 ns. It should be noted that the minimum gating time that was achieved using this chip (33 ns) is not an
intrinsic limitation. This timing was chosen when the chip was originally designed for the FLIM application. Future
work will focus on optimizing this chip for time-resolved Raman, and we expect that sub-ns gating will be possible.
In addition to Raman spectroscopy, the proposed time-resolved laser spectrometer is also applicable to fluorescence
spectroscopy and Laser Induced Breakdown Spectroscopy (LIBS), providing complementary elemental information [13].
Thanks to the fluorescence background elimination effect of the time-domain filtering, a 532 nm pulsed laser can be used
for combined Raman, fluorescence and LIBS. Therefore, the volume/mass can be further more reduced compared to that
of combined Raman/LIBS instruments using two different lasers [12].
Updated 1 March 2012
Total counts
30000
25000
20000
15000
10000
5000
0
Column
60
30
0
64
Row
127
Figure 8: A time-gated SPAD camera output of natural calcite spectrum (false color). The gating time of this experiment
was 33 ns. A broadband grating (resolution: ~ 10 cm-1) was used to observe wide range of the Raman spectrum within 3.2
mm wide image plane. The wavenumber resolution will be improved in future work.
Counts (arb. units)
SPAD
>1 µs gate
200
SPAD
33 ns gate
300
400
500
600
700
800
900
1000 1100
Raman shift (cm-1)
Figure 9: Raman spectra from willemite at different time-gating windows using the proposed SPAD camera. Thanks to the
nanosecond scale gating, overwhelming fluorescence was successfully suppressed. Overwhelming fluorescence was not
suppressed by microsecond-gates or longer. The time-resolved imaging allows us to observe Raman spectra from highly
fluorescent sample which were impossible to extract the Raman signature using traditional CW Raman spectroscopy.
5. CONCLUSIONS
Time-resolved Raman spectra from highly fluorescent mineral samples were successfully observed using our SPAD
array; these spectra were obscured by an overwhelming fluorescence background when measured using a traditional
continuous wave green laser. The SPAD imager has various advantages compared to a traditional photocathode-based
imager system, such as smaller size, lighter weight, lower operation voltage, less power dissipation, and greater radiation
hardness. These valuable features will help us to develop portable, fully automated, time-resolved single-photon
Updated 1 March 2012
detectors. The integration of the SPAD imager with a laser Raman instrument can provide enhanced capability in
various fields such as mineralogy, archaeology, medical science, and planetary science, where rapid and non-destructive
material identification on a microscopic scale is required.
ACKNOWLEDGEMENTS
The time-gated CMOS SPAD imager development described in this work was performed at Delft University of
Technology. The time-gated laser Raman experiments were carried out at the Jet Propulsion Laboratory, California
Institute of Technology, under a contract with the National Aeronautics and Space Administration (NASA).
REFERENCES
[1] Grant, J.A., Westall, F., and the MEPEG 2R-iSAG team., “Tow rovers to the same site on Mars, 2018:
possibilities for cooperative science,” Astrobiology 10, 663-685 (2010).
[2] Blacksberg, J., Rossman, G.R., Glecker, A., “Time-resolved Raman Spectroscopy for in situ planetary
mineralogy,” Appl. Opt. 49(26), 4951-4962 (2010).
[3] Rochas, A., Gani, M., Furrer, B., Basse, P.A., and Popovic, R.S., “Single Photon Detector fabricated in a
Complementary Metal-oxide-semiconductor High-voltage Technology,” Rev. Sci. Instrum. 74, 3263-3270
(2003).
[4] Karami, M.A., Gersbach, M., Yoon, H., and Charbon, E., “A new single-photon avalanche diode in 90nm
standard CMOS technology,” Opt. Express, 18 (21), 22158-22166 (2010).
[5] Richardson, J.A., Webster, E.A.G., Grant, L.A., and Henderson, R.K., “Scalable Single-Photon Avalanche
Diode Structures in Nanometer CMOS Technology”, IEEE Trans. Electron Dev., 58 (7), 2018-2035 (2011).
[6] Veerappan, C., Richardson, J., Walker, R., Day-Uey Li, Fishburn, M.W., Maruyama, Y., Stoppa, D., Borghetti,
F., Gersbach, M., Henderson, R.K., Charbon, E., “A 160128 single-photon image sensor with on-pixel 55ps
10b time-to-digital converter”, Proc. ISSCC, 312-314 (2011).
[7] Maruyama, Y., and Charbon, E., “A time-gated 128128 CMOS SPAD array for on-chip fluorescence
detection,” Proc. IISW, 270-273 (2011).
[8] Blacksberg, J., Maruyama, Y., Charbon, E., Rossman, G.R., “Fast single-photon avalanche diode arrays for
laser Raman spectroscopy,” Opt. Lett. 36 (18), 3672-3674 (2011).
[9] Niclass, C., “Single-Photon Image Sensors in CMOS: Picosecond Resolution for Three-Dimensional Imaging,”
Ecole Polytechnique Fédérale de Lausanne, Ph.D. thesis dissertation, no. 4161 (2008).
[10] Stoppa, D., Mosconi, D., Pancheri, L., and Gonzo, L., “Single-Photon Avalanche Diode CMOS Sensor for
Time-Resolved Fluorescence Measurements,” IEEE Sens. J. 9 (9), 1084-1090 (2009).
[11] Ingley, R. Hutchinson, I.B., Edwards, H.G.M., Moral, A.G., Diaz, E., Ramos, G., Barcos, O., Canora, C.P., Rull,
F., Tato, C., Pool, P., “ExoMars Raman Laser Spectrometer Breadboard: Detector Design and Performance,”
Proc. SPIE 8152, 815215-815215-9 (2011).
[12] Ahlers, B., Hutchinson, I., Ingley, R., and Raman LIBS EBB team., “Combined Raman/LIBS Spectrometer
Elegant Breadboard – Built and Tested – and Flight Model Spectrometer Unit,” Proc. 7th Int. Conf. on Space
Optics (2008).
[13] Blacksberg, J., Maruyama, Y., Choukroun, M., Charbon, E., Rossman, G.R., “New Microscopic Laser-coupled
Spectroscopy Instrument Combining Raman, LIBS, and Fluorescence for Planetary Surface Mineralogy,” Proc.
43rd Lunar and Planetary Science Conference, 1510 (2012).
Updated 1 March 2012
Fly UP