3B1.3
A NEW SINGLE PHOTON AVALANCHE DIODE IN CMOS HIGH-VOLTAGE
TECHNOLOGY
Z. Xiao1, D. Pantic2, and R. S. Popovic1
1
Institute of Microelectronics and Microsystems, Ecole Polytechnique Fédérale de Lausanne (EPFL),
Lausanne, SWITZERLAND
(Tel : +41-21-693-6742; E-mail: zhen.xiao@epfl.ch)
2
Department of Microelectronics, University of Niš, Niš, SERBIA
(Tel : +381-18-529-105; E-mail: panta@elfak.ni.ac.yu)
TRANSDUCERS & EUROSENSORS ’07
The 14th International Conference on Solid-State Sensors, Actuators and Microsystems, Lyon, France, June 10-14, 2007
Abstract: We report a new single photon avalanche diode (SPAD) implemented in a commercially
available high-voltage CMOS technology. The SPAD was designed with relatively low-doped layers to
form p-/n- junction, instead of commonly adopted p+/n- or n+/p- structures. We used the readily available
layers as given by the technology without any customization or post-processing. Careful design measures
were taken to ensure planar junction breakdown. Compared with a p+/n- diode, a p-/n- SPAD has relative
deep junction, wide depletion region, and thus improves probability of photon detection. The
measurement shows a maximum photon detection efficiency of 34.4%, and remains above 20% from
400nm to 620nm, whereas the dark count rate is only 50cps at room temperature with 5V excess voltage.
Keywords: Single Photon Detection, Avalanche Photodiode, Geiger mode, CMOS SPAD.
denoted as p+ or n+. A commonly adopted p+/nstructure [2-5] is shown in Fig. 1.
1. INTRODUCTION
Photodiodes operating in Geiger mode are
generally referred to as single photon avalanche
diodes (SPADs), since the absorption of a single
photon can initiate a strong avalanche current
pulse. They are of particular interest in
applications utilizing the quantum nature of light,
such as fluorescence lifetime imaging (FLIM),
fluorescence correlation spectroscopy, optical
quantum cryptography, and rangefinding based on
time-of-flight (TOF) of light.
Although solid-state SPADs have been available
for decades [1], they are generally fabricated in
customized processes, which results in bulky and
expensive devices. Another main drawback is
their incompatibility with integrated electronics,
and the difficulty of fabricating SPAD arrays.
Since a few years, SPADs based on CMOS
processes have been successfully implemented [25]. This greatly eases the integration of
electronics, and enables the compact array
implementation.
However, existing CMOS SPADs utilize
heavily-doped shallow junctions only. More
specifically, they are implemented with layers
used for the drain / source of CMOS transistors,
p+
n+
n+
p+
p-tub
p+
p-tub
deep n-tub
p-substrate
Fig. 1 Schematic of commonly adopted p+/nSPAD structure in CMOS technologies.
This structure limits the performances of
SPADs in several ways. Firstly, the photon
detection efficiency (PDE) is relatively low due to
the narrow depletion width and multiplication
region. Besides, the junction depth of p+/n- is too
shallow to detect efficiently red and near infra-red
photons. The typical spectral response of such a
device has peak efficiency in blue region and then
decays drastically. However, the low-energy
photons are of interest in many applications, for
example, fluorescence detection in bio-imaging.
Noise performance is another concern as the
heavily-doped CMOS layers are prone to have
generation and trapping centers. The likelihood of
tunneling-induced dark counts also increases with
1365
1-4244-0842-3/07/$20.00©2007 IEEE
3B1.3
the periphery. Moreover, a floating diffusion ring
is added to further reduce the electric field at the
corners.
p+
Shallow p-tub
p+
n+
n+
p+
Deep p-tub
deep n-tub
p-substrate
Fig. 2 Schematic of our proposed SPAD structure
using p-/n- junction in CMOS technology.
All these layers are used as given in a 0.35µm
CMOS technology which is designed for highvoltage electronics. The process parameters are
extracted from doping profile measurements, and
used for simulations in Silvaco’s ATHENA and
ATLAS tools. To ensure uniform breakdown at the
planar junction, various structures have been
studied to find a reliable guard ring design.
The simulation results of three different
structures are presented in Fig. 3.
2. DESIGN AND SIMULATION
TRANSDUCERS & EUROSENSORS ’07
The 14th International Conference on Solid-State Sensors, Actuators and Microsystems, Lyon, France, June 10-14, 2007
the doping levels. As the feature-size shrinks in
advanced CMOS technology, the dark count rate
and afterpulsing will be worsened if p+ or n+ is
still used to form SPAD junctions.
In this paper, we demonstrate a new SPAD
structure using relatively low-doped layers
available in commercial CMOS technology. With
p-/n- junction, we increase the junction depth and
extend the depletion region considerably, thus
improving the spectral response while ensuring
low dark counts.
The design concept and simulation of the
structure is elaborated in Section 2; Section 3
presents the experimental results of the
implemented SPAD.
Fig. 2 illustrates the concept of our proposed
SPAD structure. The cathode is a deep n-tub
isolating the device from p-substrate, and the
anode is now a combination of deep p-tub and
shallow p-tub. The shallow p-tub is sized smaller
than the deep p-tub, in order to reduce doping
concentration and hence the electrical field along
P2
P4
P6
(a)
(c)
(b)
Fig. 3 Simulations of 3 different SPAD structures with p-/n- junctions, namely P2, P4 and P6. Results are
shown in cross-sectional view with half planes only: (a) doping concentrations, (b) electrical fields, and
(c) current distributions.
1366
1-4244-0842-3/07/$20.00©2007 IEEE
3B1.3
corresponding to a peak current of 30A. The
persistent accumulated waveform shows a long
recharge time of 3.7s due to the large parasitic
capacitances in the hybrid setup.
Rq
50
Fig. 5 Setup of hybrid passive quenching
configuration. A persistent accumulated waveform
of avalanche current pulses is shown on the right.
Finally, the SPAD is operated with an active
quenching module developed by [6], where a
gated mode is available. This setup was then used
to perform the optical and dark count rate
measurements [7-8], in order to avoid false counts
due to afterpulses. The experimental results of the
optimized structure P6 are presented in Fig. 6~8.
With an active area of 5um in diameter and an
excess bias of 5V, the photon detection efficiency
(PDE) reaches the peak value of 34.4%. There is a
visible plateau from 450nm to 520nm, where the
PDE does not vary more than 1.5% from the
maximum. The efficiency remains above 20%
from 410nm to 640nm, and reaches 10% at 750nm
and 5% at 840nm. The falling-off towards red and
near IR is more gradual than those of CMOS
SPADs with p+/n- [2-5].
1.00E-06
1.00E-07
Vd (V)
30
35
40
45
50
Time (1µs/div)
Oscilloscope
After CMOS fabrication, the devices were first
measured statically. Experimental data of reverse
I-V characteristics, shown in Fig. 4, corroborated
the simulation results. As expected, P6 has the
highest breakdown voltage, 50V.
1.00E-05
Voltage
(5mV/div)
Vop
3. MEASUREMENT RESULTS
Id (A)
55
P2
P4
P6
35%
1.00E-08
30%
1.00E-09
25%
1.00E-10
PDE
TRANSDUCERS & EUROSENSORS ’07
The 14th International Conference on Solid-State Sensors, Actuators and Microsystems, Lyon, France, June 10-14, 2007
The first structure, namely P2, is constructed
with equally-sized deep p-tub and shallow p-tub.
The high electrical field is clearly observed at the
edge due to the high doping levels there, and the
current flows near the surface. In the second
structure, namely P4, the size of shallow p-tub is
reduced compared to the deep p-tub. It effectively
lowers the peripheral electrical fields, but the high
fields concentrate at the junction corners where
the curvature effect dominates. The current
distribution shows the localized breakdown
phenomenon. Finally, the optimal structure P6
adds a floating diffusion ring around to leverage
the electrical field. The current distribution proves
the planar junction breakdown.
One shall note that the depletion region is as
wide as 2.5m, and extends to both sides of the
junction at a depth of 0.8m. This is more than
twice the values of p+/n- junctions in the same
CMOS technology. However, this suggests that
the breakdown voltage is considerably increased
as well.
Fig. 4 Measured reverse characteristics of the
three implemented SPADs, P2, P4 and P6.
Ve=5V
Ve=4V
20%
15%
10%
5%
Device P6 was then tested in Geiger mode
operation, with a discrete quenching resistor of
250K. As shown in Fig. 5, the avalanche current
pulses are sensed through a 50 resistance. The
maximum signal value is about 15mv,
0%
370
420
470
520
570
620
670
720
770
820
870
920
970
Wavelength [nm]
Fig. 6 Measured photon detection efficiency of P6
with excess bias voltages of 4V and 5V.
1367
1-4244-0842-3/07/$20.00©2007 IEEE
3B1.3
Fig. 6 shows the typical exponential increase of
the dark count rate (DCR) versus the excess bias
voltage. The DCR is below 50cps at room
temperature (about 24C) with 5V excess voltage.
Such a low level of dark noise enables further
increase of the active area, and hence the fill
factor to further improve the performance.
4. CONCLUSIONS
A new SPAD structure using p-/n- junction was
successfully designed and implemented in a
CMOS technology. Without any technology
customization or post-processing steps, the
optimized SPAD exhibits favorable results in dark
count rate and photon detection efficiency, while
timing jitter remains comparable to existing p+/nSPADs. Thanks to the full CMOS compatibility, it
can be easily integrated with electronics to realize
large arrays of single photon detectors.
10
ACKNOWLEDGEMENTS
1
This work is supported by Swiss National
Science Foundation. The authors are grateful to
Cristiano Niclass for his contributions to
measurement setup, and Marc Lany for helpful
discussions.
Reverse Bias Voltage (V)
50.0
51.0
52.0
53.0
54.0
55.0
56.0
Fig. 7 Measured dark count rate of P6 at room
temperature of 24C.
Timing jitter was measured with a picosecond
laser source of 637nm. The passive quenching is
used to avoid extra jitter introduced by active
electronics. As shown in Fig. 7, a FWHM of 80ps
is achieved with 5V excess voltage.
Number of counts
1000
180
160
100
10
140
FWHM [ps]
TRANSDUCERS & EUROSENSORS ’07
The 14th International Conference on Solid-State Sensors, Actuators and Microsystems, Lyon, France, June 10-14, 2007
D C R (1 /s )
100
1
248.5n
120
249.0n
249.5n
Time (s)
250.0n
250.5n
100
80
60
50
51
52
53
54
55
56
Reverse Bias Voltage [V]
Fig. 8 Measured timing jitter of P6 in terms of
FWHM versus bias voltages. Insert: Timing
distribution at excess voltage of 5V.
REFERENCES
[1]
R. J. Mc Intyre, “Recent developments in
silicon avalanche photodiodes”, Measurement,
vol. 3, pp. 146-152, 1985.
[2]
A. Rochas et al., “Single Photon Detector
Fabricated in CMOS High Voltage Technology”,
Review of Scientific Instruments, vol. 74, pp.
3263-3270, 2003.
[3]
F. Zappa et al., “Monolithic CMOS
Detector Module for Photon Counting and
Picosecond Timing”, IEEE, ESSDERC, pp. 341344, Sept. 2004.
[4]
C. Niclass et al., “Design and
Characterization of a CMOS 3D Image Sensor
based on SPADs”, IEEE Journal of Solid-State
Circuits, pp. 1847-1854, Sep 2005.
[5]
C. Niclass et al., “A SPAD Array
Fabricated in 0.35um CMOS and based on an
Event-Driven Readout for TCSPC Experiments”,
Proc. of SPIE, Vol. 6372, Oct. 2006.
[6]
Id Quantique, www.idquantique.com.
[7]
A. Lacaita et al., “Single-Photon Detection
beyond 1µm: Performance of Commercially
Available InGaAs/InP Detectors”, Applied Optics,
vol.35, no. 16, pp. 2986-2996, 1996.
[8]
C. P. Morath et al., “Comparator-Based
Measurement Scheme for Dark-Count Rates in
SPADs”, IEEE Trans. Instrumentation and
Measurement, vol. 54, pp. 2020-2026, 2005.
1368
1-4244-0842-3/07/$20.00©2007 IEEE