Download Single-photon imaging at 20000 frames/s

3024
OPTICS LETTERS / Vol. 30, No. 22 / November 15, 2005
Single-photon imaging at 20, 000 frames/ s
F. Zappa, S. Tisa, and S. Cova
Dipartimento Elettronica e Informazione, Politecnico di Milano, p.za Leonardo da Vinci 32, 20133 Milano, Italy
P. Maccagnani
CNR-IMM Sezione di Bologna, Via Gobetti 101, 40129 Bologna, Italy
R. Saletti and R. Roncella
Dipartimento Informazione, Università di Pisa, Elettronica, Informatica, Telecomunicazioni, via Caruso,
56122 Pisa, Italy
Received June 21, 2005; accepted July 15, 2005
A complete two-dimensional imaging system based on a silicon monolithic array of 60 single-photon counters
is presented. The fabricated solid-state array is rugged and operates at low voltages. Detection efficiency is
higher than 40% in the visible range, and cross talk among 50 ␮m pixels is lower than 10−4. The complete
system provides a maximum throughput of 20 kframes/ s with truly parallel readout and nanosecond gating, thanks to the use of an integrated active quenching circuit for each pixel of the array. We report optical
and electrical characterizations of the whole imaging system. © 2005 Optical Society of America
OCIS codes: 040.5160, 040.6070, 130.0250, 120.1880, 130.6010, 130.6750.
Many imaging applications demand single-photon
sensitivity, a high frame rate, truly parallel readout,
low intrinsic noise, and fast gating. For instance, this
sensitivity is needed in laser-induced fluorescence of
biological samples and organic materials, and for
photon migration and in vivo monitoring of cerebral
activity and tissue oxygenation. In astrophysics,
single-photon sensitivity is needed to study the fast
evolution of optical counterparts in high-energy
gamma ray bursts and to correct atmospheric aberration, through adaptive optics.1
By now, single-photon avalanche diodes (SPADs)
have proved to be the best single-pixel candidates for
the detection of ultrafaint and very fast optical
signals.2 In applications where many more pixels are
needed, many research groups have resorted to the
use of multiple commercially available modules.3,4
Only recently, monolithic arrays of SPADs made in
standard complementary metal-oxide semiconductor
technologies have been proposed in the literature,
but usually these technologies suffer from a small
area, high noise, afterpulsing, and low detection efficiency. For instance, in Ref. 5 an array of 32 SPADs
with a 6.4 ␮m diameter, an average noise of
100 counts/ s, and just 17% efficiency at 500 nm was
reported. Moreover, pixel readout is often
multiplexed,6 thus complicating the use of SPADs in
many real photon-starving applications while simplifying any issue of cross talk, which is instead tough
to solve when parallel readout is a must.
In this Letter we present a 60-pixel monolithic array of fully parallel photon counters operating at low
voltage (below 30 V) and with high frame rates. Applications are planned in adaptive optics, for the auxiliary telescopes of the European Southern Observatory (ESO),1 in fluorescence measurements, and in
fast-transient imaging. The array has a circular geometry dictated by ESO requirements and also features nanosecond gating.
Figure 1 shows the SPAD array (SPADA) system,
0146-9592/05/223024-3/$15.00
which consists of a 60-element monolithic sensor, detection electronics, a real-time photon-counting and
data processing board, and a computer for system
control, data uploading, and postprocessing. Each element is a SPAD detector, biased above its breakdown voltage, where a generated carrier can ignite a
self-sustaining avalanche multiplication process.2
Each element’s behavior is radically different from
that of standard avalanche photodiodes, which are
operated below breakdown, in the linear analog multiplication regime. In fact, a single photon absorbed
within a SPAD can trigger an avalanche current of
tens or hundreds of milliamps. Ignitions that are not
due to photons but are caused by carrier thermal generation are called dark counts and represent the detector’s intrinsic noise. To detect a subsequent photon, the avalanche current must be quenched as soon
as possible by lowering the SPAD bias below breakdown and then resetting it to its quiescent condition
after a short dead time; such tasks are performed by
use of proper quenching electronics.7
The 60 SPADA pixels are placed at the field focus of
Fig. 1. (Color online) Architecture of the SPADA imaging
system.
© 2005 Optical Society of America
November 15, 2005 / Vol. 30, No. 22 / OPTICS LETTERS
Fig. 2. SPADA chip (left) with 60 elements, each with four
independent pixels (right).
Fig. 3. Detection board, assembled with twelve 5 channel
iAQC-modules and the SPADA sensor.
Fig. 4. (Color online) Dark-counting rate of all sixty
50 ␮m diameter pixels of a SPADA chip at different overvoltages and at room temperature. The elements are ordered by increasing dark counts.
a 60-subaperture segmented lenslets array (Fig. 2,
left). The overall chip is 18 mm⫻ 18 mm and contains
four sets of detectors, with diameters of 20, 35, 50,
and 75 ␮m (Fig. 2, right). The input optics facing the
array chip is aligned and focused onto one set of pixels, chosen by the user depending on the tenable
dark-counting rate. The blind (active) region among
the elements can be substantially avoided by use of
3025
commercially available lenslet arrays8 that focus the
signal onto the 60 active areas.
The detection board is built around the activequenching circuit (AQC) concept, introduced by researchers at Politecnico di Milano, developed through
various generations,2,7 and used in many laboratories
and applications (e.g., in 14 adaptive optics systems
deployed at the ESO Observatory1). For the SPADA
sensor we developed a novel integrated circuit, containing five iAQC, with the structure depicted in Ref.
9. Each chip is assembled in a small package with
overall dimensions of 1 cm⫻ 1.4 cm. Twelve packages
are needed for the 60 pixels of the array and are assembled with the rest of the detection board,10 which
includes differential I/O, temperature control, and
bias adjustment electronics. A picture of the entire
detection head is shown in Fig. 3.
The pixels of the fabricated array chip have a low
breakdown voltage (25 V) and show low power dissipation during avalanche (a few tens of mW at
Fig. 5. Cross talk among 50 ␮m pixels placed at increasing distance, at 5 V overvoltage.
Fig. 6. Detection efficiency of SPADA pixels.
Fig. 7. Images acquired by the SPADA at two levels of
static illumination, with 500 ␮s time slots.
3026
OPTICS LETTERS / Vol. 30, No. 22 / November 15, 2005
Fig. 8. Sequential frames acquired by the SPADA: a
bright spot is moving along the detector (10 ms for each
frame).
Fig. 9. Fast-transient imaging at 20 kframes/ s: the pixels
are illuminated by a uniform 1 ms period saw-tooth modulated light.
10 Mcount/ s). Therefore in most applications it is not
necessary to cool the array, unless an extremely low
dark-counting rate is a must. The dark-counting rate
of 50 ␮m pixels is plotted in Fig. 4. Elements with
very low dark counts (below 1 kcount/ s at 5 V overvoltage) represent 40% of the array, and this percentage increases to 80% when the detector is cooled to
−15° C. The measured afterpulsing2 of 50 ␮m pixels
is 0.5% at 5 V overvoltage and with a 50 ns hold-off
time.
Since all pixels are contemporaneously enabled
and the readout of the array is fully parallel, a complete investigation of cross talk is a must. Spurious
couplings could be both electrical (i.e., triggering of
iAQCs caused by interactions among adjacent channels) and optical (i.e., avalanches ignited in a pixel by
photons generated during an avalanche in a neighboring pixel). We carefully avoided electrical cross
talk by means of deep diffusions that electrically isolate each pixel, and by using an iAQC layout (separation of shielding and paths). Then we measured the
residual (optical) cross talk through correlation measurements (Fig. 5). With pixels placed at distances
larger than 5 mm, the SPADA residual cross talk is
lower than 10−4 and is negligible in most applications, since a detector should be triggered on average
10,000 times before causing spurious ignition of a
neighboring pixel.
The measured detection efficiency (Fig. 6) shows a
broad maximum around 500 nm and also a convenient response in the near infrared. We used a deposited 120 nm thick SiO2 layer over the silicon that acts
as a built-in antireflection coating and yields a detection efficiency about 25% higher than for an abrupt
silicon–air interface.
We tested the entire system, composed of the imaging array together with detection, counting, and processing electronics. At first, we acquired static weak
visible radiation, within time slots of 500 ␮s. In Fig.
7, more illuminated (whiter) pixels acquired, on average, 9 counts (left) and 400 counts (right), yielding to
detected photon fluxes of 18 and 800 kphoton/ s, respectively, compared with a background (black pixels) of 6 kcounts/ s. Then, we acquired a dynamic image, based on a focused spot moving at 20 cm/ s.
Figure 8 shows four snapshots, one taken every 10
ms during the motion of the light source. The light
was then modulated by a triangular wave with a 1
ms period. The frame rate was set to 20 kHz, i.e.,
with 50 ␮s integration windows. In Fig. 9, the peaks
correspond to a measured photon rate of
1.2 Mphotons/ s, while the background (dark counts
plus ambient stray light) sets at an average of ⬃4
counts per each window, i.e., at 80 kcounts/ s.
In conclusion, the reported single-photon counter
array represents major progress in the imaging of
fast and weak optical transients. Both the detection
efficiency (53% at 520 nm) and the diameter of the
pixels 共50 ␮m兲 are much better than those obtained
so far. The maximum frame rate topped
20 kframes/ s. The system is now being integrated
into adaptive optics and fast transient applications.10
Other application-specific arrays with different layouts and (or) numbers of pixels are easily feasible.
F. Zappa’s e-mail address is [email protected].
References
1. D. Bonaccini, F. Rigaut, A. Glindemann, G. Dudziak,
J.-M. Mariotti, and F. Paresce, in Proc. SPIE 3353, 224
(1998).
2. S. Cova, M. Ghioni, A. Lotito, I. Rech, and F. Zappa, J.
Mod. Opt. 51, 1267 (2004).
3. H. Dautet, P. Deschampes, B. Dion, A. D. MacGregor,
D. MacSween, R. J. McIntyre, C. Trottier, and P. Webb,
Appl. Opt. 32, 3894 (1993).
4. “Photon Counting Detector Module, PDM Series,”
datasheet, Micro Photon Devices, http://www.microphoton-devices.com.
5. A. Rochas, M. Gosch, A. Serov, P. A. Besse, R. S.
Popovic, T. Lasser, and R. Rigler, IEEE Photon.
Technol. Lett. 15, 963 (2003).
6. C. Niclass, A. Rochas, P. A. Besse, and E. Charbon, in
Proceedings of the IEEE 2004 Custom Integrated
Circuits Conference, (IEEE, 2004), pp. 705–708.
7. S. Cova, A. Longoni, and G. Ripamonti, IEEE Trans.
Nucl. Sci. 29, 599 (1982).
8. M. Rossi and I. Kallioniemi, in Diffractive Optics and
Micro-Optics, Vol. 75 of OSA Trends in Optics and
Photonics Series (Optical Society of America, 2002), pp.
108–110.
9. F. Zappa, A. Lotito, A. C. Giudice, S. Cova, and M.
Ghioni, IEEE J. Solid-State Circuits 38, 1298 (2003).
10. F. Zappa, S. Tisa, S. Cova, P. Maccagnani, D. Bonaccini,
G. Bonanno, M. Belluso, R. Saletti, and R. Roncella, in
Proc. SPIE 5490, 1200 (2004).