LARGE SCALE MICRO-OPTIC SWITCHES

John Bowers
UCSB, Electrical & Computer Engineering Department, Santa Barbara, CA 93106
Calient Networks Inc., 25 Castilian Dr., Goleta, CA 93117
Plenary Talk at 2007 MicroOptics Conference
© 2007 Microoptics Group (OSJ/JSAP)

Abstract: The rapid growth of video and data transmission is driving the growth of fiber optic transmission and switching. Optical
switching is a low power, low cost solution to this need. We review approaches to optical switching and examine the performance
that is achievable. We review the interesting aspects of micro-optic design of the fiber collimator array, MEMS mirror deflection and
dimensions, and the overall switch performance such as insertion loss, return loss and polarization dependant loss. The testing results
from fabricated MEMS based micro-optic switches are compared to theoretical results.

1. Introduction
Video and data transmission are growing exponentially,
with a doubling time of one year or less[1]. This is
driven by widespread broadband and FTTH
deployments. The amount of electrical power required to
switch this information is growing exponentially and is
becoming a significant (>1%) amount of electrical power
consumed in many advanced countries[2]. Optical
switches are an important solution to this problem
because the power required to switch data is typically
1000 times less if high speed electrical switching and
OEO conversion are avoided. Fig. 1 illustrates a mesh
network utilizing a core of optical switches. Fig. 2. Tradeoff between capacity and granularity[5].

There are a variety of applications for optical switches in

fiber optic networks ranging from fiber switching to
waveband switching to wavelength switching. These
may be done in a single switch, or in multiple switches,
depending on the number of ports required. This
architecture is illustrated in Fig. 3. It shows the
separation of optical bypass and optical switching from
Fig. 1. Illustration of an optical switch core mesh network[3]. subwavelength switching in the service layer.
Fiber
Another advantage of using optical switches is the Switching
scalability to reconfigurable mesh networks, with the Waveband
resultant improvement in system reliability and Switching
flexibility. The tradeoffs between optical and electrical Wavelength
Switching Photonic
core switching is illustrated in Fig. 2. Optical switches (switching)
-Grooming
have a capacity of at least 32 Tbit/s (320 fibers at 100 -Wave shifting
Gbit/s each) requiring a power of on the order of 1 W, -Amplets TRANSPORT
-Xponders LAYER
and a size of 0.001 m3 [4]. This results in -OPM

capacity/power and capacity/size metrics on the order of SERVICE

32 Tbit/s/W and 32,000 Tbit/s/m3. These are illustrated DCS or
LAYER
OEO Switch
in Fig. 2, and are 100 to 1000 times better than electrical Electronic
core switching, but at the expense of switch granularity (grooming)
and switching speed. The important issue is whether the STS-n ATM IP Transparent Switching
(GBE, FC,Wave Services)
ability to switch packets or subwavelength switch is
worth the extra power, weight, size and cost. Fig. 3. Separation of transport switching (fiber, waveband and
wavelength from subwavelength service layer switching.
2. Optical Switch Scaling
Optical switches tend to scale well as transmission 1 10
4

capacity increases due to their transparent switching.

This is illustrated in Fig. 4. A switch that was installed
8 00 0
to carry 2.5 Gbit/s data can be upgraded to 10 Gbit/s and
to multiple wavelengths, resulting in a growth in 2D
capacity by three orders of magnitude without expense 6 00 0

as the transmission capacity increases and the switching

granularity required increases. This does not occur with 4 00 0

electrical core switches or routers.

10
2 00 0
2.4 Gb/s
1
3D
PXC/OEO Cost

t 2֩& 0
s
o 0.1 4֩& 0 20 40 60 80 1 00 1 20
C 8֩&
tr 16∪
∠&
o N u m b e r o f P o r ts
P 0.01
10 Gb/s 32∪
∠&
Fig. 6. Comparison of required number of switch elements in
128∪
∠&
2D and 3D switches.
0.001
2000 2002 2004 2006 2008 2010
Year
Fig. 4. Relative switch cost as the capacity is increased.

A wide variety of optical switch technologies have been

developed, and the tradeoff in speed and switch size is
illustrated in Fig. 5. Some technologies are quite fast,
with switching times of 1 ns or less, and are important
for packet switched systems[6-8]. Many of these utilize
passive arrayed waveguide routers with fast wavelength
tunable lasers. Other technologies have switching speeds
in the microsecond range, and are useful for burst
switched networks[9,10]. Microelectromechanical
system (MEMS) switches are slower, with switching
speeds in the millisecond range. Fig. 7. Schematic diagram of a 3D MEMS switch.

1000
In this remainder of this paper, we will focus on the
lse
micro-optic design of large optical 3D MEMS switches.
nn
a
We will present the issues and results of micro-optic
h
/C 3D MEMS design and simulation for a large scale MEMS based
st 100
r
o
P
f
micro-optic switch. We also compare our simulation
o Tunable
re
b 2D MEMS Wavelength
results with the measured results from a fabricated large
m
u
N Thermo- AOTF Liquid scale 360x360 MEMS optical switch.
10 Electro-optic
optic Crystal
2. Optical Design of Large Scale Micro-optic Switches
10 ms 10 µs 10 ns
The key optical design of the large-scale micro-optic
Fig. 5. Comparison of switch technologies in speed and size.
switch involves the MEMS mirror size and deflection
angle design. The insertion loss of the switching system
Most of the switch technologies illustrated in Fig. 5 are
depends on the fiber collimator array design, the switch
planar 2D switches. 2D switches are typically composed
size, and the MEMS mirror deflection angles
of 2x2 switches in a crossbar configuration, and so a
nonblocking switch requires N2 switches, as illustrated in
Diffraction typically dominates the insertion loss. The
Fig. 6. Planar switches are typically 32x32 or less,
MEMS mirror can be designed to be large enough so that
because of quadratic dependence on port count.
diffraction effect from MEMS mirror is small; however,
larger mirrors result in longer path length, higher
Recently, 3D switches (Fig. 7) have been developed to
vibration sensitivity and tighter mechanical alignment
solve this scaling bottleneck[12-14]. These approaches
tolerances. A better design is to make the MEMS mirrors
utilize just 2N switch elements for a nonblocking switch
to be just large enough such that the clipping or
with low loss (Fig. 6). Consequently, large switch arrays
diffraction introduced loss is well controlled.
of 320 ports and higher have been demonstrated using
MEMS technology[12-14]. 3D switches can also be
The fiber collimator array has a fiber array attached to a
implemented using individual collimators that can be
micro-lens array (Fig. 8). The shape of the micro-lens in
angled to point at each other.
the lens array is spherical. The error of the shape of 60000
micro-lens is very important for optical insertion loss. Measured
50000 Distribution
Figure 9 gives the measurement results of a silicon

Number of Paths
Theoretical
micro-lens shape together with a shape error plot. This 40000 Distribution
figure shows that the rms shape error is 19 nm, which
corresponding to a 0.04 λ wavefront error for 1550 nm 30000
wavelength. Such a small amount of wavefront error will 20000
contribute less than 0.4 dB insertion loss.
10000
0
0.4 0.8 1.2 1.6 2 2.4 2.8
Insertion Loss Bin (dB)

Figure 11. Theoretical insertion loss distribution and the

measured insertion loss distribution.

The calculated losses for all 129,600 paths of a 360x360

Fig.8. 400 beam collimator array. photonic switch are shown in Fig. 11. The measured
4 0 .05 losses are slightly higher and have a larger distribution,
3 partly due to connector variation.
Shape Error (um)

0 .04
2
Another critical parameter is the return loss. This is
Height (um)

1
0 .03 dominated by the return loss of the input collimator. A
typical distribution is shown in Fig. 12. Some
D

-1
0 .02
applications require worst case return losses of -55 dB,
-2
which has been demonstrated by angling interfaces of
0 .01
the input fiber block.
-3
140
-4 0
120
Number of measurements

0 100 200 3 00 400 500

A
x distance (um ) 100

Fig. 9. A typical micro-lens shape (red: design, blue: 80

experimental) and the shape error.
60

The optical beam pointing error is due to fiber position 40

error, the lens pitch error, and the focal length variations 20

of the micro-lens array. In practice, 98% of the beams 0

have less than 0.5 mrad pointing error. 30 35 40 45 50 55

Return Loss (dB)

Figure 10 shows the beam spot size and the beam waist Fig.12. Return loss of a 3D MEMS switch.
location plot for all the optical beams in a fabricated
fiber collimator array at 1550 nm wavelength. The beam The PDL of a switch can also be calculated as part of the
waist position error to be within ±1 mm, and the spot switch design. Fig. 13 shows a comparison of measured
size error to be ±15 um, which are sufficient to achieve a and experimental PDL distributions of an early switch.
low loss switch. Later designs have maximum PDL of 0.2 dB, but the
remnant PDL is more random.
(um (um)

230
30 0.5
Measurement PDL values for PXC (dB)
off Target

220
20
Target ) 0.4
10
210 B
(d
Diameter

2000
L 0.3
D
WaistSize

-10
190 P
d0.2
-20
180 e
Beam Spot

r
-30
170
u
s
a0.1
e
-40
160 M
-5
20 -3
22 -1
24 1
26 3
28 5
30 0.0
0.0 0.1 0.2 0.3 0.4 0.5
Beam Waist Location (mm)
BIN 1 Calculated
Expected PDL
values from model for(dB)
SMM (dB)
Figure 10. The fiber collimator array beam spot size and beam
waist location plot. Fig. 13. Comparison of measured and calculated PDL of an
early switch showing good correlation of calculated and
measured PDL.
Other important parameters are directivity, generally less Handbook of Massive Data Sets, J. Abello, P. Pardalos,
than -70 dB, and crosstalk. Crosstalk is dominated by and M. G. C. Resende, eds., pp. 47-93, Kluwer, 2002.
nearest neighbors. Mirrors farther away have crosstalk
[2] J. Baliga, R. Ayre, K. Hinton and R.S. Tucker,
contributions below -80 dB. Fig. 14a shows the loss
“Photonic Switching and the Energy Bottleneck”, PIS
distribution of the desired connection and Fig. 14b
2007 Photonics in Switching 2007, August 19-22 2007,
shows the crosstalk contributions of nearest neighbors,
San Francisco, California.
which are generally below -45 dB.
[3] G. Ellinas, E. Bouillet, R. Ramamurthy, J.-F.
Labourdette, S. Chaudhuri, and K. Bala, “Routing and
restoration architectures in mesh optical networks,” Opt.
Networking Mag., vol. 4, no. 1, pp. 91-106, 2003.
[4] X. Zheng, V. Kaman, S. Yuan, Y. Xu, O.
Jerphagnon, A. Keating, R. C. Anderson, H. N. Poulsen,
B. Liu, J. R. Sechrist, C. Pusarla, R. Helkey, D. J.
Blumenthal, and J. E. Bowers, “Three-Dimensional
Fig. 14. a) Insertion loss hill in angle space (degrees). Red is MEMS Photonic Cross-Connect Switch Design and
0-10 dB, and each color changes corresponds to 10 dB Performance,” Journal of Selected Topics in Quantum
increase, up to 80 dB. b) Crosstalk in angle space (degrees). Electronics, 9(2), 571-578, March (2003).
Red is -40 to -45 dB, and each color change corresponds to 5
dB increase, up to 80 dB. [5] OIDA Roadmap for Optical Networks: Access
and Core Networks (2005).
Switch Size Scaling
An important issue is scaling switches to larger sizes. [6] X. Song, N. Futakuchi, F. C. Yit, Z. Zhang, and Y.
Table 1 summarizes how the important design Nakano, "28-ps switching window with a selective area
parameters scale with N, the number of mirrors. MOVPE all-optical MZI switch," IEEE Photon. Technol.
Calculations and measurements indicate that Lett., 17, 1480-1482, 2005.
performance similar to that shown here should be [7] D. Wolfson, et al., “All-optical asynchronous
achievable for 1000x1000 switches. Loss does increase variable-length optically labeled 40 Gbps packet switch”.
for practical designs larger than 2000x2000. ECOC’05, PDP, 2005.
 
parameter symbol scaling [8] J. Gripp, et al., "IRIS optical packet router," J.
Beam radius at waist w0 N.5 Opt. Netw. 5, 589-597 (2006)
Beam radius at MEMS wm N.5 [9] S. J. B. Yoo, "Optical Packet and Burst Switching
mirror Technologies for the Future Photonic Internet," J.
Mirror diameter N.5 Lightwave Technol. 24, 4468-4492 (2006)
D
[10] C. Qiao and M. Yoo, "Optical Burst Switching
Mirror array area
A N2
(OBS)," J. High Speed Networks, vol. 8, 69-84 (1999).
Optical path length [11] L. Y. Lin, E. L. Goldstein, and R.W. Tkach,
L N “Free-space micromachined optical switches for optical
Active switching networking,” IEEE JSTQE 5(1), pp. 4–9, Jan./Feb. 1999.
volume V N3
[12] R. Helkey, S. Adams, J. Bowers, T. Davis, O.
Table 1. Dependence of important design parameters on the Jerphagnon, V. Kaman, A. Keating, B. Liu, C. Pusarla,
number of ports in the switch[15]. Y. Xu, S. Yuan, and X. Zheng, “Design of large scale,
MEMS based photonic switches,” Optical & Photonic
4. Conclusions News, p. 40-43 (May 2002).
In conclusion, we presented the optical model large scale
micro-optic switching using MEMS mirror array. The [13] Ming C. Wu, Olav Solgaard, and Joseph E. Ford,
micro-optic design of fiber collimator array, MEMS “Optical MEMS for Lightwave Communication”, J.
mirror deflection, and the overall switch performance LightWave Tech., 24(12), pp. 4433-4454, 2006.
such as insertion loss, polarization dependant loss are [14] V. A. Aksyuk, et al., “238x238 surface
theoretically simulated and agreed with the testing micromachined optical crossconnect with 2 dB
results from the fabricated 360x360 MEMS based micro- maximum loss,” OFC, PD FB9, March 2002.
optic switch.
[15] R. Helkey “Transparent Optical Networks with
5. References Large MEMS-Based Optical Switches”, Symposium on
Contemporary Photonics Technology, E2, 2005.
[1] K. G. Coffman and A. M. Odlyzko, “Internet
growth: Is there a “Moore's Law” for data traffic?",