Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
Lateral PIN-Photodetector in Commercial CMOS Technology
Operating at 1.25 Gbit/s and 850 nm
M. Jutzi, K. Eve, W. Vogel, D. Wiegner, M. Berroth
Institute for Electrical and Optical Communcation Engineering,
Pfaffenwaldring 47, 70550 Stuttgart, michael.jutzi@int.uni-stuttgart.de
A lateral PIN-photodetector in an unmodified 0.25 µm CMOS process is presented.
At –5 V bias the achieved photoresponsivity is 0.36 A/W and 0.37 A/W at 660 nm and
850 nm, respectively. Eye-diagram measurements demonstrate 1.25 Gbit/s operation
at 850 nm. At 660 nm and –5 V the rise time of the fast component of the detector
response is 109 ps.
1 Introduction
Optical interconnects and optical local area networking become attractive with increasing
bitrates. Since high volumes at low costs are required, CMOS is the technology of choice for
optoelectronic integrated circuits (OEICs).
An unmodified CMOS process can be used to fabricate photodetectors operating above
1 Gbit/s. Silicon with an indirect bandgap of 1.1 eV yields sufficient quantum efficiencies at
wavelengths of 850 nm and below. Since CMOS devices are implemented close to the surface
in an up to 2 µm thick region, the wavelength of the optical source should have a comparable
penetration depth. Using a wavelength of 660 nm instead of 850 nm the penetration depth
decreases from 16 µm to 3 µm [1]. Thus CMOS detectors and directly modulated lasers in the
visible are ideal candidates for low cost and short range optical communication.
In the literature bipolar silicon [2], twin-well CMOS [3], DRAM [4] and SOI [5] technologies
have been used to fabricate photodetectors. An excellent review of the state of the art can be
found in [6] and [7]. Further efforts are needed to realize detectors with a high bandwidth at
reasonable responsivities using an unmodified CMOS technology.
2 Device Cross Section and Layout
The photodiode has been processed in a 0.25 µm CMOS technology. The detector consists of
a coplanar contact pattern in a ground-signal-ground configuration and an interdigitated
metalization above the absorption region as shown in Fig. 1 and Fig. 2. Diffused n+ and p+
doped regions beneath the contact fingers and the lightly p-doped substrate form a lateral pinphotodiode. Photodetectors with two different finger spacings w = 4 µm and 10 µm have been
realized to study the influence of the geometry. All photodetectors have a sensitive area of
100×100 µ m2. The depth of the diffused highly doped regions is approximately 1.3 µm as
Fig. 3 illustrates. No specific anti-reflection coating is applied and no additional process steps
are performed to fabricate the device.
Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
hν
GND
Sig
GND
finger
contact
finger
contact
1 µm
1.3 µm
p+
n+
w = 4 µm,
w = 10 µm
lightly doped
substrate
Fig. 1: Top micrograph of the
photodetector
Fig. 2: Layout of the photodetector
Fig. 3: Schematic cross section
3 DC-Photoresponsivities
1e-2
0.45
A
A/W
photoresponsivity
current
The DC-photoresponsivity is measured on-wafer with a lensed, multimode fiber probe using
lasers with wavelengths of 660 nm and 850 nm.
1e-6
1e-8
1e-10
1e-12
-40
dark current
-36.33 dBm
-31.54 dBm
-27.19 dBm
-23.52 dBm
-30
0.35
0.3
660 nm, 4 µm finger spacing
660 nm, 10 µm finger spacing
850 nm, 4 µm finger spacing
850 nm, 10 µm finger spacing
0.25
-20
voltage
V
0
Fig. 4: IV-characteristics at 660 nm, pin diode with
4 µm finger spacing
0.2
-40
-30
-20
voltage
V
0
Fig. 5: Photoresponsivity versus bias voltage at 660 nm
and 850 nm for pin diodes with 4 µm and 10 µm finger
spacing
A linear dependence of the photocurrent on the incident optical power is deduced from Fig. 4.
Up to 40 V reverse bias no breakdown is observed. The devices exhibit dark currents densities
at -10 V of less than 0.1 mA/cm2.
A small finger spacing and a wavelength with a small penetration depth have to be chosen to
reduce the reverse bias necessary to attain saturation of the DC-photoresponsivity as can be
seen in Fig. 5. Since 850 nm optical radiation generates carriers deep in the substrate, the
collection efficiency and hence the photoresponsivity increases by applying high reverse bias
voltages. In contrast the responsivity at 660 nm saturates already at –5 V. Assuming total
absorption, the theoretical, maximum responsivities are calculated as R = q ⋅ λ (h ⋅ c ) ⋅ F with
q, λ, h, c, F denoting the unity charge, the wavelength, Planck’s constant, the speed of light
and the fill factor of the interdigitated contact pattern, respectively. This equation yields
0.43 A/W at 660 nm and 0.54 A/W at 850 nm for the 4 µm detector with F = 0.8 . Due to
coupling and surface reflection losses the measured responsivities are smaller than the
theoretical values.
Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
4 S-Parameter Characterization
The photodetector acts as an opto-electrical converter. The output reflection factor S22 and the
opto-electrical transfer function S21 are measured on-wafer at bias voltages from -40 V to
0.8 V. The measurement setup is outlined in Fig. 6. Port 1 of the vector network analyzer
(HP8510B) drives a directly modulated vertical cavity surface emitting laser (VCSEL)
(NewFocus1780) operating at 850 nm. Port 2 is connected on-wafer to the photodiode under
test with a CascadeMicrotech GSG-probe.
VNA
port 1
port 2
S21
PD
LD
multimode
fiber
S22
Fig. 6: Measurement setup for S-parameter characterization
4.1 S-Parameter Dark Small-Signal Equivalent Circuit
The output reflection factor S22 is measured for different bias points while the laser source is
switched off. The diode with 4 µm finger spacing exhibits a larger capacitance than the 10 µm
device as can be seen by comparing Fig. 7 and Fig. 8. Since the 4 µm diode has twice as many
fingers as the 10 µ m diode, the series resistance due to the metalization is lower. The series
resistance of the 4 µm diode is extracted from the IV-measurements and determined as 0.6 Ω
in contrast to 2.1 Ω for the 10 µm diode.
90°
90°
120°
150°
120°
60°
30°
-0.5 V
0.4 V
0.6 V
0.8 V
180°
330°
240°
-0.5 V
0.4 V
0.6 V
0.8 V
150°
0°
210°
60°
300°
270°
Fig. 7: Output reflection factor S22 in the frequency
range 0.1 to 20 GHz for bias voltages -0.5 V, 0.4 V,
0.6 V and 0.8 V, pin diode with 4 µm finger spacing
30°
180°
0°
330°
210°
240°
300°
270°
Fig. 8: Output reflection factor S22 in the frequency
range 0.1 to 20 GHz for bias voltages -0.5 V, 0.4 V,
0.6 V and 0.8 V, pin diode with 10 µm finger spacing
Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
S22 does not change significantly unless the diode is forward biased and hence dominated by a
large diffusion capacitance. For reverse operation the depletion capacitance stays almost
constant as expected for pin-diodes with a low doping of the intrinsic zone.
In order to deembed the intrinsic diode from the parasitics, an open and a short structure as
illustrated in Fig. 9 are characterized. The measurement results are presented in the Smithchart in Fig. 10. In the RF small-signal equivalent circuit of Fig. 11 the left part models the
pad parasitics and the center part additional, very small parasitics not included in the test
structures. The right part represents the intrinsic diode comprising a depletion capacitance Ci
and a parallel admittance Gi.
90°
120°
60°
150°
30°
open
short
180°
0°
210°
330°
240°
300°
270°
Fig. 9: Open and short structure
Fig. 10: Output reflection factor S22 of open and short
structure in the frequency range 0.1 to 20 GHz
First, the parameters of the pad equivalent circuit are determined from the S-parameter
measurements of the test structures. Then the remaining parameters are calculated. All
parameters are calculated by numerical optimization. The first guesses for the solution
algorithm are derived from characteristic points of the locus diagram of the output admittance.
The relative error between measured data and simulated S-parameters averaged over all
frequencies is typically below 3 %.
130
fF
L2
R1
110
R5
R2
pads
R3
C1
R4
additional
parasitics
Gi
Ci
intrinsic diode
intrinsic capacitance
L1
100
90
4 µm finger spacing
10 µm finger spacing
80
70
60
50
40
-40
-30
-20
bias voltage
V
0
Fig. 11: Small-signal equivalent circuit of the complete Fig. 12: Extracted capacitance of intrinsic diode versus
diode at darkness
bias voltage
Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
0.04
conformal mapping
extracted at -40 V
fF/µm^2
capacitance
0.03
0.025
0.02
0.015
0.01
0.005
0
2
4
6
8
finger spacing
µm
12
Fig. 13: Capacitance per µm2 calculated by the conformal mapping technique, finger width of 1 µm, finger
length of 95 µm
The extracted depletion capacitance displayed in Fig. 12 agrees well with conformal mapping
calculations for an interdigitated contact pattern. Applying the formulae of [8] the capacitance
per µm2 is plotted versus finger spacing in Fig. 13. The extracted capacitance of the 4 µm
detector is more bias dependent than of the 10 µm device, because the space charge region
moves deeper into the substrate. With exception of Gi the elements of the equivalent circuit of
Fig. 11 do not change significantly under illumination.
4.2 Opto-electrical Transfer Functions at 850 nm
An optical signal is applied to the detector and the opto-electrical transfer function S21 is
determined.
-40
-40
-50
-50
21
21
|S |
dB
|S |
dB
-55
-60
-65
1e8
-30 V
-20 V
-10 V
-5 V
-1 V
-55
-40 V
-20 V
-10 V
-5 V
-1 V
-60
Hz
frequency
1e10
Fig. 14: Magnitude of the opto-electrical transfer
function S21 in the frequency range 0.1 to 10 GHz
versus frequency measured for bias voltages
-40 V, -20 V, -10 V, -5 V and -1 V, photodetector with
4 µm finger spacing
-65
1e8
Hz
frequency
1e10
Fig. 15: Magnitude of the opto-electrical transfer
function S21 in the frequency range 0.1 to 10 GHz
versus frequency measured for bias voltages
-30 V, -20 V, -10 V, -5 V and -1 V, photodetector with
10 µm finger spacing
The opto-electrical 3 dB-bandwidth of the detectors is improved with increasing reverse
biases. Since the electric field penetrates deeper into the substrate for a smaller finger spacing,
a higher collection efficiency at moderate bias levels is achieved for the 4 µm photodetector.
At high bias voltages the collection efficiency of both types of detectors is comparable for low
frequencies. However, higher electric fields in the photodiode with the smaller finger spacing
yield faster drift velocities. The steep drop of the frequency response near 8 GHz originates
from the bandwidth limit of the VCSEL.
Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
The laser is an electro-optical converter with a gain of GE/O = 1.8 mW/V assuming that the
voltage is delivered to a 50 Ω load. The photodiode under test operates as an opto-electrical
converter with the gain GO/E . The relation between the electro-optical gain GE/O, the optoelectrical gain GO/E and S21 can be written as:




G

G

(1)
E
/
O
O
/
E
20 ⋅ log( S21 ) = 20 ⋅ log
+ 20 ⋅ log
= −41 dB


1W 
1V 


V 
W


144424443
-55 dB
If the maximum of S21 is approximately –41 dB as given in Fig. 14 and Fig. 15, the
photodiode has a voltage-power gain of:


G

(2)
20 ⋅ log O / E 1 V  = 14 dB
or
G O / E = 5.0 V/W


W

Since this opto-electrical gain refers to a 50 Ω environment, the corresponding responsivity is
0.1 A/W. The values in excess of 0.4 A/W determined in section 3 are larger, because
diffusion contributes more to the DC-responsivity.
5 Impulse Responses
Impulse response measurements provide information about the time constants of the
photoinduced carrier generation in the photodiode. In the measurement setup a bit pattern
generator (Anritsu MP1763C) drives a VCSEL laser source illuminating the photodiode. The
electrical output signal of the photodiode is amplified with a bandwidth of 6 GHz
(Minicircuits ZJL-6G). The waveform at the amplifier output is captured by an HP54750A
digital sampling oscilloscope.
The photodiode response comprises two time constants at each bias point, a small and a large
one. With increasing reverse bias the amplitude of the fast contribution becomes more
important, the corresponding time constant, however, stays approximately the same. The large
time constant decreases considerably as a higher reverse bias is applied. At 850 nm the
photoinduced carriers are generated deeper in the substrate than at 660 nm and travel further
to reach the n+ and p+ regions, as indicated in Fig. 16 and Fig. 17.
70
14
-30 V
-20 V
-10 V
-5 V
-1 V
0V
50
voltage
40
10
8
30
6
20
4
10
2
0
0
-10
0
20
40
60
time
80
ns
-40 V
-20 V
-10 V
-5 V
-1 V
0V
mV
voltage
mV
120
Fig. 16: Photodetector response to 660 nm VCSEL
being switched on for bias voltages –30 V, -20 V,
-10 V, -5 V, -1 V and 0 V, photodetector with 4 µm
finger spacing
-2
0
20
40
60
time
80
ns
120
Fig. 17: Photodetector response to 850 nm VCSEL
being switched on for bias voltages -40 V, -20 V,
-10 V, -5 V, -1 V and 0 V, photodetector with 4 µm
finger spacing
60
60
mV
mV
40
40
30
30
20
-40 V
-20 V
-10 V
-5 V
-1 V
0V
10
0
-10
0
voltage
voltage
Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
1
2
3
ns
-40 V
-20 V
-10 V
-5 V
-1 V
0V
20
10
0
-10
0
5
1
2
time
3
ns
5
time
Fig. 18: Photodetector response to 660 nm VCSEL
being switched on for bias voltages –40 V, -20 V,
-10 V, -5 V, -1 V and 0 V, photodetector with 4 µm
finger spacing
Fig. 19: Photodetector response to 660 nm VCSEL
being switched on for bias voltages -40 V, -20 V,
-10 V, -5 V, -1 V and 0 V, photodetector with 10 µm
finger spacing
A zoom into the rising edge of Fig. 16 is given in Fig. 18. The photodetector current follows
the relaxation oscillations of the VCSEL. The 10 % to 90 % rise time of the detector response
is 100 ps and 109 ps at bias voltages of -30 V and -5 V, respectively. The rise time of the
10 µm photodiode response depicted in Fig. 19 is slightly longer with 109 ps at –30 V.
6 Eye-Diagram Measurement at 850 nm
To demonstrate real life operation of the 4 µm photodiode, an eye-diagram measurement is
performed. The measurement setup is the same as for the impulse characterization. The bit
pattern generator produces a 31-bit wide pseudo random bit sequence at 1.25 Gbit/s. A first
approximation of the bit error rate is given by BER = exp(− Q2 2) (Q 2π ) , where Q is a quality
factor Q = (m1 − m 0 ) (σ1 + σ0 ) [9]. The average values of the ones and zeros are given by m1 and
m0 and σ1 and σ0 are the respective standard deviations. In the eye diagram of Fig. 20 the Qfactor can be estimated as Q § 5.7 resulting in a BER of about 6•10-9.
40
mV
30
6σ 1
20
m1-m0
0
-10
6 σ0
voltage
10
-20
-30
-40
0
0,4
0,8
1,2
ns
1,6
2
time
Fig. 20: Eye diagram at 1.25 Gbit/s for a 231-1 PRBS signal and 0 dBm optical power at 850 nm, integration time
100 s, 20 V reverse bias, photodetector with 4 µm finger spacing
Proceedings 7 th Workshop „Optics in Computing Technology“, 18./19. September 2002, Mannheim, pp. 55-62
7 Conclusion and Outlook
A 1.25 Gbit/s operation of lateral pin-CMOS-photodetectors is demonstrated at 850 nm.
Using shorter wavelengths in the visible range, even higher bitrates are expected. Directly
modulated VCSEL lasers at 660 nm will become commercially available soon and could
figure as low-cost optical sources. The optical and electrical characteristics of both, the
photodetector and the laser can be tested perpendicularly to the wafer surface, which presents
a cost advantage over edge-coupled devices. Further efforts are necessary to optimize the
layout of the photodetector to reduce substrate diffusion effects. A prototype of an integrated
version of a photodetector and a preamplifier is under test.
8 Acknowledgment
The authors thank T. Ballmann of the 4th Physical Institute of the University of Stuttgart for
providing the red VCSEL laser. The processing of the samples by Texas Instruments
Germany and the support of H. Parzhuber are gratefully acknowledged.
9 References
[1] D. F. Edwards: “Silicon (Si)/ Absorption constant of pure, unstrained, crystalline Si ”, Handbook of Optical
Constants of Solids, Academic Press, 1985, pp. 547-, Ed. by Palik et. al.
[2] H.H. Kim, R.G. Swartz, Y. Ota, T. K. Woodward, M.D. Feuer, W.L. Wilson: „Prospects for silicon
monolithic opto-electronics with polymer light emitting diodes“, IEEE Journal of Lightwave Technology,
Vol. 12, No. 12, Dec. 1994, pp. 2114 –2121
[3] H. Zimmermann, T. Heide, A. Ghazi: „Monolithic High-Speed CMOS Photoreceiver“, IEEE Photonics
Technology Letters, Vol. 11, No. 2, February 1999, pp. 254-255
[4] J. D. Schaub, D. M. Kuchta, D. L. Rogers, M. Yang, K. Rim, S. Zier, M. Sorna: „Multi Gbit/s, highsensitivity all silicon 3.3 V optical receiver using PIN lateral trench photodetector“, Optical Fiber
Communication Conference and Exhibit, 2001. OFC 2001 , Vol. 4 , 2001, pp. PD19 -P1-3
[5] S. M. Csutak, J. D. Schaub, W.E. Wu, J. C. Campbell: “High-speed monolithically integrated silicon optical
receiver fabricated in 130-nm CMOS”, IEEE Photonics Technology Letters, Vol. 14, No. 4 , April 2002, pp.
516 -518
[6] H. Zimmermann: „Integrated Silicon Optoelectronics“, Springer-Verlag 2000
[7] T. K. Woodward, A. V. Krishnamoorthy: „1-Gb/s Integrated Optical Detectors and Receivers in
Commercial CMOS Technologies“, IEEE J. of Selected Topics in Quantum Electronics, Vol. 5, No. 2,
March/April 1999, pp. 146-156
[8] S. S. Gevorgian, T. Martinsson, P. L. J. Linner, E. L. Kollberg: „CAD Models for Multilayered Substrate
Interdigital Capacitors”, IEEE Transactions on Microwave Theory and Techniques, Vol. 44, No. 6, June
1996, pp. 896-904
[9] G. Grau: Optische Nachrichtentechnik, Springer Verlag Berlin 1981, p. 253