Fully reconfigurable silicon CMOS photonic lattice filters

Abstract

We present reconfigurable CMOS-compatible silicon-photonic lattice-filters consisting of Mach-Zehnder structures and ring resonators configured as high resolution bandpass and notch filter shapes. Arbitrary filter synthesis and CMOS-compatible fabrication process are also discussed.

Full text

Page 1

Fully Reconfigurable Silicon CMOS Photonic Lattice Filters
Nicolas K. Fontaine(1), Salah Ibrahim(1), Stevan Djordjevic(1), Binbin Guan(1), Tiehui Su(1),
Stanley Cheung(1), Runxiang Yu(1), Ryan P. Scott(1), Andrew T. Pomerene(2), Liberty Gunter(2),
Steve Danziger(2), Zhi Ding(1), K. Okamoto(1), and S. J. B. Yoo(1)
(1) Department of Electrical and Computer Engineering, University of California, Davis, CA 95616,USA,
sbyoo@ucdavis.edu
(2) BAE Systems North America, 1300 N. 17th Street, Suite 1400, Arlington, VA 22209, USA
andrew.pomerene@baesystems.com

Abstract We present reconfigurable CMOS-compatible silicon-photonic lattice-filters consisting of
Mach-Zehnder structures and ring resonators configured as high resolution bandpass and notch filter
shapes. Arbitrary filter synthesis and CMOS-compatible fabrication process are also discussed.

Introduction
All-optical processing of RF and microwave
signals impressed on optical carriers using
reconfigurable optical filters can obtain higher
operation bandwidth, tuning bandwidths, and
lower power consumption than their pure
electrical counterparts1. To match the diverse
signal processing capabilities of electrical
systems, an optical system must be easily
reconfigurable, must have control over a large
number of zeros and poles in the filter transfer
function, and must achieve high resolution
(~50 MHz). The lattice filter construction2 builds
high-order filters with many poles and zeros by
cascading many identical unit cells, each cell
providing a single pole and zero. Here, we
demonstrate reconfiguration of single- and four-
cell photonic lattice filters on the silicon-on-
insulator (SOI) platform consisting of tunable
couplers and ring resonators into filter shapes
including those with pure finite impulse response
(FIR) and infinite impulse response (IIR).
Design and Fabrication
Fig. 1(a) presents the lattice filter unit cell3 which
provides a fully controllable pole and zero to the
filter transmission function by controlling four
parameters: the ring phase shift, the coupling
strength between the upper waveguide and the
ring, the lower waveguide phase shifter, and the
output coupler. Fig. 1(b) shows the four unit cell
lattice filter with four tuneable poles and zeros.
Rapid reconfiguration of the filter is possible
using phase shifters that utilize the free-carrier
plasma dispersion effect from current injection
into p-i-n diodes (< 20 ns measured switching
time) at the expense of some additional insertion
loss from free carrier absorption. The tuneable
couplers are 2×2 Mach-Zehnder interferometers
with phase shifters in each arm.
In the lattice filter construction, the two inputs
and outputs of adjacent unit cells are coupled
together. Within a unit cell, the upper arm
contains a waveguide coupled to a ring which
produces both a pole and zero. The pole is fully
defined by the ring coupler (pole magnitude) and
ring phase shifter (pole phase), however the
zero is not controllable. In FIR/IIR filter design,
the pole magnitude which ranges between zero
(FIR) and one (IIR) for passive devices is
inversely proportional to the minimum spectral
feature. The pole magnitude of the unit cell is
equal to the power transmittance of one round-
trip through the ring and is ?10??/???1 ? ?? 
where ? is the excess losses in dB and ? is the
power coupling coefficient of the ring coupler.
IIR functionality occurs when the ring stores light
(i.e., near 0% coupling out of the ring or a pole
with near unity magnitude). In this case, the ring
Fig. 1: Design of (a) single unit cell filter and (b) four unit cell filter. (c) Narrow waveguide design for the directional
couplers, bends, and p-i-n diodes. (d) Wide waveguide design for low-loss waveguiding in the ring (8.7 mm path
length). (e) Fabricated four unit cell filter and single unit cell filter.
1
2
3
4
5
6
Out 1
Out 2
(a) Single Unit Cell Filter
Ring Phase Shifter
6
Ring Coupler
3
Output
Coupler
5
Lower Waveguide
Phase Shifter
Input
Coupler
In 1
In 2
1
2
4
3 μm Wide
Waveguide
0.5 μm Wide
Waveguide
Input
Coupler
In 1
In 2
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
Unit Cell 1Unit Cell 2 Unit Cell 3Unit Cell 4
Out 1
Out 2
SiO2
Si
SiO2
3 μm
100 ps Lattice Constant
(c) Narrow Waveguide (d) Wide Waveguide
0.5 μm
(b) 4-Unit Cell
Filter
0.25 μm
0.25 μm
13 mm6 mm
(e)
ECOC 2010, 19-23 September, 2010, Torino, Italy
978-1-4244-8535-2/10/$26.00 ©2010 IEEE
Tu.4.C.3

Page 2

outputs an impulse every T seconds (i.e., the
propagation delay time of the ring or lattice
constant) which extends the filter impulse
response much longer than the delay provided
by the physical path length of the device.
Realizing a pure FIR filters requires 100%
coupling into and out of the ring (i.e., a pole with
a magnitude of zero). In this case, the ring
simply acts as a delay line and the impulse
response duration is no longer than T.
The unspecified zero provided by the upper
waveguide and ring is adjustable to a specified
value by properly combining the signals on the
upper and lower waveguides. A recursive
algorithm3-4 determines the settings for the
output couplers’ and lower waveguide’s phase
shifters to adjust the zero to the desired location.
Realizing pole magnitudes near unity require
low excess losses in the ring (i.e., ? is 0).
Therefore, Fig. 1(c,d) show the two types of
silicon rib-waveguides used in the unit-cell
design: a 3-µm wide waveguide for reduced
waveguide losses and nonlinearities in the
straight sections of the ring and a 0.5-µm wide
waveguide for single-mode confinement, strong
lateral evanescent coupling, and low current
tuning operation5. The chips are fabricated at
the BAE Systems CMOS foundry6 using a
procedure similar to Ref. 7. In addition, these
devices use angled facets to reduce facet
reflections and heat-isolating trenches to reduce
thermal crosstalk between electrodes. The 0.5-
µm waveguide losses are ~0.3 dB/cm.
Measurements and Tuning of a Single Cell
Fig. 2(a) shows the measured optical intensity
and phase transmission of a single-unit cell filter
tuned to form a 400-MHz wide bandpass filter.
The data are obtained using a frequency-
domain swept coherent interferometer8-9 that
enables simultaneous complex spectral
transmission measurements across 10 nm with
100 dB dynamic range and an update rate of
10 Hz. The filter’s free spectral range (FSR) is
10 GHz which is the inverse of T. The thin lines
overlaid in Fig. 2(a) depict the best-fit filter
containing only the pole and zero shown in
Fig. 2(b) obtained using the MATLAB system
identification toolbox10. The excellent match
between the measurement and fit indicates that
the unit cell provides only a single pole and zero
without undesired features or behaviours.
Fig. 2(c) shows the extracted pole magnitude
of the filter versus separate sweeps of the drive
current to each phase shifter in Fig. 1(a). To
obtain maximal reconfigurability of the unit cell,
the pole magnitude must tune from zero to one
to convert the filter from a pure FIR filter to a
FIR/IIR filter with vanishingly narrow spectral
features. In addition, simple reconfiguration
requires low crosstalk between the phase
shifters. In our devices, the pole magnitude does
not change when driving the lower waveguide
phase shifter and the output coupler (i.e.,
electrodes 1 and 4) which is strong evidence of
low crosstalk. The wide measured pole tuning
range of 0.1 to 0.93 (? between 13% and 99%)
is demonstrated by injecting current into the ring
coupler (electrodes 2 and 3). The maximum pole
magnitude is inversely proportional to the total
ring round-trip loss due to both the necessary
coupling to the upper waveguide and unwanted
excess losses. Therefore, assuming that excess
loss limits the maximum pole magnitude, a
worst-case estimate of the ring round trip loss is
0.6 dB. Controlling the ring phase shifter
(electrode 6) changes the pole angle over
2π rad for 13 mA of injected current; however
the pole magnitude decreases due to undesired
free-carrier absorption. Nevertheless, the unit
cell provides a large pole tuning range due to
the low-loss waveguides and tunable couplers.
Fig. 2: Single unit-cell filter tuning examples. (a) Measurement (thick lines) and single pole/zero fit (thin lines) of
the H21(ω) complex transmission. (b) Extracted pole and zero values of the fit. (c) Pole magnitude versus current
as the different electrodes indicated in Fig. 1(a) are separately tuned. Bandpass filter (d) with weak resonance
and (e) strong resonance. The impulse response is the inverse Fourier transform of the spectral transmission.
123462, 3
Time (500 ps/div)
H21(ω)
Impulse Response
(20 dB/div)
Frequency (10 GHz/div)
Transmission
(10 dB/div)
(d) Bandpass with small pole (FIR)
Pole Magnitude = 0.11
h21(t)
Time (500 ps/div)
Frequency (10 GHz/div)
H11(ω)
h11(t)
H11(ω)
h11(t)
H21(ω)
h21(t)
(e) Bandpass with large pole (IIR)
Pole Magnitude = 0.87
Impulse Response
(20 dB/div)
Transmission
(10 dB/div)
Frequency (10 GHz/div)
(c) Parameter Extraction
Transmission
(10 dB/div)
Phase
(1 rad/div)
Pole Magnitude
Current (mA)
015 015 015 015 015 0 10
1
0
Electrode Number
(a) Measurement and Fitted Transmission
Re
Im
(b) Pole-Zero Fit
Pole
Zero
Unit Circle
Tu.4.C.3

Page 3

upper input port to the upper and lower outputs
[H11(ω) and H21(ω)] for a near-FIR filter (pole
magnitude is 0.11) and a FIR/IIR filter with a
large pole magnitude near 0.9 optimized for the
H21(ω) transmission. The filter tuning procedure
is as follows: first, use the ring coupler and ring
phase shifter to set the pole magnitude and
phase. Then, use the lower waveguide phase
shifter and output coupler to position the zero’s
magnitude and phase. The ideal FIR filter’s
impulse response [top of Fig. 2(d)] should
contain two peaks of equal magnitude; the first
peak from the light in the lower waveguide and
the second peak from the ring delayed by T.
Here, the unwanted impulses contribute to less
than 1% of the transfer function. The spectral
transmission is sinusoidal for both outputs and if
the input coupler were
transmissions could be identical with large
extinction (e.g., an ideal delay interferometer).
Fig. 2(e) shows an high-resolution 400-MHz
wide bandpass filter (-3 dB bandwidth) created
using a pole with magnitude near 0.9. Here, the
impulse response rolls off at about 1 dB per
impulse and extends beyond 5 ns before
reaching the noise floor of the measurement.
The insertion loss of the FIR/IIR filter is 10 dB
lower than the pure FIR filter due to
compensation of the excess losses in the ring
with the output coupler. These examples
demonstrate full functionality of the unit cell.
Fig. 2(d,e) show the transmission from the
adjusted, both
Four Unit Cell Filters
Fig. 3 displays the transmission of the four unit
cell filter under three configurations. Fig. 3(a)
shows the case with no tuning. Each dip in the
transmission corresponds to a pole. The phase
of the pole sets its location within the FSR.
Since three dips are visible within a single FSR,
two poles have the same phase. Arbitrary filter
shapes require control of the 26 phase shifters.

reconfigured to have a flat-top bandpass shape
with sharp rolloff for the H21(ω) transmission.
The filter has an extinction of 30 dB and its flat-
top -1 dB bandwidth is 600 MHz. Fig. 3(c) shows
a filter optimized for the H11(ω) transmission. As
expected, the four unit cell devices achieved
more complex and versatile filter shapes when
compared to the single unit cell filter.
Fig. 3(b) shows the filter manually
Conclusions
We demonstrated reconfiguration of high order
photonic lattice filters fabricated on a CMOS-
compatible SOI platform. The high-quality
transmission of the four-unit cells shows strong
potential for eight- and even 16 unit-cell
structures. Inherent <20 ns switching speed,
development of automatic routines, and fast
digital-to-analog converters will enable sub-
100 ns switching between arbitrary filter shapes.
Such rapidly reconfigurable filters are highly
desirable in microwave-photonic
processing, optical communications, and optical
label switching.
signal
References
1 E. M. Dowling et al. J. Lightw. Technol. 12,
471 (1994).
2 B. Moslehi et al. Proc. IEEE 72, 909 (1984).
3 K. Jinguji et al. J. Lightw. Technol. 13(1995).
4 C. K. Madsen et al., Optical filter design and
analysis, John Wiley & Sons (2001).
5 L. Zhou et al. Proc. LEOS '09, WN5, (2009).
6 L. C. Kimerling et al. Proc. of SPIE
6125(2006).
7 S. Ibrahim et al. Proc. OFC'10, OWJ5, (2010).
8 K. Takada et al. Opt. Lett. 31, 323 (2006).
9 D. K. Gifford et al. Appl. Opt. 44, 7282 (2005).
10 L. Ljung et al., System identification: theory
for the user, Prentice-Hall Englewood
Cliffs, NJ (1987).
This work was supported in part by DARPA MTO Si-
PhASER project Grant No. HR0011-09-1-0013
Fig. 3: Four unit-cell filter tuning examples. (a) No tuning. Bandpass filters optimized for (b) H21(ω) with a, −1 dB
bandwidth of 600 MHz, −3 dB bandwidth of 1 GHz, and −20 dB bandwidth of 2.8 GHz. (c) H11(ω) optimized
transmission with −1 dB bandwidth of 1.8 GHz, −3 dB bandwidth of 2.5 GHz and −20 dB bandwidth of 7.6 GHz.
H11(ω)
h11(t)
H21(ω)
h21(t)
H11(ω)
h11(t)
H21(ω)
h21(t)
Time (500 ps/div)
H11(ω)
h11(t)
H21(ω)
h21(t)
Time (500 ps/div)
Frequency (10 GHz/div)
Time (500 ps/div)
Frequency (10 GHz/div)Frequency (10 GHz/div)
Impulse Response
(20 dB/div)
Transmission
(10 dB/div)
Impulse Response
(20 dB/div)
Transmission
(10 dB/div)
Impulse Response
(20 dB/div)
Transmission
(10 dB/div)
(a) No Tuning
(b) Filter Optimized for H21(ω)
(c) Filter Optimized for H11(ω)
Tu.4.C.3