Proceedings of the 45th IEEE Conference on Decision & Control WeA1 7.

3
Manchester Grand Hyatt Hotel
San Diego, CA, USA, December 13-15, 2006

Nonlinear Control of a MEMS Optical Switch

K. 0. Owusu, F. L. Lewis, B. Borovic, and A. Q. Liu

Abstract-This paper presents the development of a processing techniques applied to fabricate MEMS optical
dynamical model of a MEMS based optical switch. The model switches lead to high-volume, low cost production [2]. More
is verified using experimental results. An inherent weakness in importantly, MEMS technology allows large matrix switches
the use of the model for control design is the unavailability of to be monolithically integrated on a single chip [1].
position information at all times due to the saturated output With increasing complexity of modem optical networks,
characteristics. To counteract this problem, a nonlinear controlling and optimizing the operation of optical switches
observer is designed for the saturated output system to estimate with guaranteed performance, stability and reliability becomes
the states (both inside and outside of the saturation region) for a challenging task. Controlling the device requires the
dynamic control of the switch. The nonlinear observer is used establishment of its full models, including the optical,
to synthesize feedback linearization nonlinear control. The mechanical and electrical models. The optical models deal
central idea of the feedback linearization control is to with the relationship between the mirror position and the
compensate for the nonlinearity in the system dynamics. attenuation, while the mechanical and electrical models link
Simulation results show that the proposed nonlinear observer the driving voltage with the mirror position. Their
and controller design approaches have not only good tracking combination predicts the optical performances of the switch
ability and convergence but also extremely good parameter under electrostatic actuation.
variation robustness.
In the control of the optical switch, it is necessary to
I. INTRODUCTION
dynamically estimate the switch position and velocity. These
states cannot be easily obtained when the switch is near the
OPTICAL fibers in optical communication networks have fully opened or closed position due to the saturation in the
very wide bandwidth. Large scale matrix switches that available optical detector measurements. To overcome this
are mostly used in optical networks now are realized by obstacle, a full-order nonlinear state observer is constructed to
optical - electronic conversion / electronic switching / provide a reliable estimate of the optical switch position and
electronic - optical conversion (O-E-O). These switches are velocity based on the optical detector measurement of the
very expensive and have slow bit rates capacity than fiber output light intensity (power). The application of state
transmission lines due to the limited switching capacity of the observers in MEMS control is not new. In [4] a reduced-order
electronics involved. With the increased demand for observer is used to estimate the velocity for feedback control
bandwidth due to the rapid growth of the information of electrostatically actuated MEMS to eliminate "snap-
superhighway, it has become absolutely imperative to through". The observer design in this paper is effective for the
confront the bandwidth limitation problem. The obvious saturated output system and utilizes the Lipschitz property of
solution is to avoid the transitions between optical and the nonlinearities in the state and output equations of the
electronic transmitters. MEMS switch. Observer design for systems with Lipschitz
The optical switch is a device that switches an optical signal nonlinearities have been dealt with in [5] and [6]. These
from one optical fiber to another, without having to first designs assumed that the system output is linear. In this paper,
convert the optical signal into an electrical signal. This leads we present a simple nonlinear observer for a system with a
to the total exclusion of electronics and thus makes possible Lipschitz nonlinearity in the output characteristics.
the realization of all-optical networks (AON). The estimated state variables are then used to design a
Microelectromechanical systems (MEMS) technology has controller for the optical switch. The controller designed has
become a promising approach for optical switches. With this two components. A nonlinear control design approach called
approach, micro mirrors are fabricated to perform the feedback linearization [3] is applied to compensate the
switching function in an all optical medium. MEMS optical nonlinearity in the system dynamics and a linear tracking
switches are very attractive because of its small size and controller component which ensures that the switch position
weight. In addition, they are inherently fast, have low power tracks a desired reference trajectory generated by a desired
consumption [1] and have low insertion loss. The batch command generator.

This work was supported by NSF GOALI grant BES-0201773F. II. MEMS DEVICE MODEL AND DYNAMICS
K. 0. Owusu, F. L. Lewis and B. Borovic are with the Automation & The SEM micrograph of the MEMS optical switch is
Robotics Research Institute, University of Texas at Arlington, Fort Worth,
USA. (E-mail: lewis outa.edu, kowusu outa.edu) shown in Fig. 1 [9]. It consists of an electrostatic comb drive
A. Q. Liu is with the School of Electrical and Electronic Engineering, actuator, the suspension beam, a shuttle to which is attached a
Nanyang Technological University, Singapore.
1-4244-0171-2/06/$20.00 ©2006 IEEE. 597
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WeA1 7.3

shutter (micromirror) and optical fiber grooves. A voltage m = mmirror + 0.5mrigid + 2.74mbeam (4)
applied to the comb drive actuator generates an electrostatic
force that moves the micromirror into the optical fiber beam The effective mass of the system is calculated from the
path regulating the output power. device geometry as m = 2.35 x 10-9kg.
To model the system, it is assumed that the shuttle has only The stiffness of the suspension beam is assumed to be a
one degree of freedom. The mathematical model has three linear function of the position i.e. k(x) = kx and its
main components - mechanical, electrical and optical models. coefficient is given as [9]
The system can be described with a second order differential
equation with input, output and system nonlinearities as
k = 24EI /L3 = 2ET(BW)3/(BL)3 (5)
ir+±d(k,x)+k(x) = f(v,x); P=h(x) (1) where T is the thickness of the beam, E is the Young's
modulus of silicon (E 160GPa), BW is width of the
where m is the effective moving mass of the shuttle, d is the
suspension beams (BW 2pm), BL is length of suspension
damping, k is the stiffness of the suspension, f is the induced
beams (BL = 530um) and Iz moment of inertia around the
electrostatic force, P is the light intensity and x the shuttle
deflecting axis of the beam. Calculation gives k= 0. 6Nm-'.
position.
Damping is the parameter that is most difficult to
determine analytically, even through FEA. The reason lies in
the number of different mechanisms that causes it. Four
different mechanisms could contribute to damping; Couette
flow, Poiseuille flow, Stokes flow, and squeeze film
damping [1 ]. They can be summarized as
d(x, x) =(dix + d0 )x.
Poiseuille and Stokes flows are
difficult to estimate due to the complex geometry of the
device. Squeeze film damping is negligible. Couette
damping is given as

d(x,x) =(7A/dG)j=-[77n4x+x0)/dG]j=(dxx+do)* (6)

Fig. 1. SEM micrograph of the MEMS optical switch. where q = 18pPas is the viscosity of the air surrounding the
A. Electrical Model switch. The damping coefficients are calculated from (6) as
The electrical model considers the generation of the dx = 0.0363 and do = 5.4519 x 10-'.
The complete analytical model is given by
induced electrostatic force when a driving voltage is applied to
the actuator. The capacitance of the comb drive is calculated
as a sum of parallel capacitances among pairs of comb 2.35xIO-9+(o.o363x+5.4519X1O-7 (7)
,
electrodes. The total capacitance is given by [9] + 0.6x = 1.78x 10 -8V2
C. Optical Model
C(x) = c0A/dG = 2ne0T(x + xO)/dG (2)
The optical model relates the position of the mirror to the
light intensity attenuation. Fig. 2 shows the optical mirror and
where co = 8.854 x 10-12 Fm-' is the dielectric constant of the input and output single mode fibers (SMF). Light beam
free space, n is the number of movable comb fingers from the input SMF propagates to the mirror plane and is
(n=150), T is the thickness of the structural layer (T=35um), partially blocked by the mirror. It is then diffracted to the
dG is the length of the gap between fingers (dG = 2.6 um) receiving facet of the output SMF.
and xo is the overlapped length of fingers at no applied The model employed here is the Raleigh-Sommerfeld
voltage (xo= i5um). The electrostatic force is given as diffraction formula [10] which assumes a Gaussian
distribution for the fundamental mode of the light beam
f(v,x) = O.5v2 ac/ax = n0Tv2dG = kev2 (3) from the SMFs. The waist radius of the Gaussian beam
where v is the applied voltage and ke = 17.8nNJV2. coming from the input SMF is wo. After passing through a
distance zl, the waist radius w1 of the Gaussian beam is
B. Mechanical Model
To obtain the mechanical model of the system, the given w1 = wo1±+ z1 /ZR where ZR = zWO2/A denotes
as
effective moving mass m, the damping coefficient d, and the the Raleigh range and i the wavelength of light. In this
stiffness of the suspension k are estimated. The effective example, wo = 5.1,um, A= 1.55,m and z1 = lO,um.
mass of the movable structure can be expressed as [9]
598
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WeA1 7.3

The coupled power P (Pout/Pin) is obtained [10] as x lo-

P = [- Erf (4i(x - 0 )I WI)] (8)

.
Zil

input SMiE7 Mitr Output SIF

Voltage (V)

Fig. 2. Diagram of Optical Model [9]. Fig. 4. Static characteristics

where qo is the initial distance between the mirror and the

fiber axis, x is the displacement of the mirror and Erf is the
error function. In this set up po = 11.2 um. The relationship
between the power ratio and the position of the mirror is
shown in Fig. 3.
1 I 1 0
0.9 -L
0.8 - - - -_-
0.7 - - -
-X-

B_0
0.4 - L X Fig. 5. Open loop dynamic response of the optical switch obtained from
experiment.

This is compared with simulation results of the open loop

dynamic response obtained from the developed model by
Fi.. owr gansDisplacement.()
ato
o
applying a square-wave driving voltage of 5 - 30V. The
simulation results are shown in Fig. 6.
It is important to note that the attenuation curve (Fig. 3) is
saturated by the error function which makes it difficult to
reconstruct the states in the saturation region for control x0
2
design purposes.
D. Experimental Verification of the Model
The analytically obtained model is compared with
experimental results obtained from the MEMS optical switch
for accuracy. 02 0.005 0.01
Time (s)
0.015 0.02 0.0 25

The constant ke in (3) can be verified by conducting static

experiments i.e. 3~ 0 . Static conditions reduces (1) and
=
Fig. 6. Open loop dynamic response of the optical switch model
(3) to x (ke /k) V2. Different driving voltages are applied
=

across the comb drive and its corresponding deflections The results clearly show the underestimation of the damping
observed through an optical microscope. Experimentally and component. This necessitates an identification of accurate
analytically obtained static voltage-deflection curves are model parameters for control design purposes.
shown in Fig. 4. E. Identification ofAccurate Model Parameters
The difference between the curves may be due to likely An identification of accurate model parameters is carried
estimation errors of the gap between fingers dG and/or the out. The parameter identification procedure used in [12] is
width of the suspension beam BW used in calculating the employed yielding the accurate dynamical model
values of ke and k respectively.
For dynamic verification of the model, the open loop
dynamic response of the switch is obtained by applying a 2.35x10-9 +(0.0363x+1.6x1O-5)x
(9)
square-wave driving voltage of 5 - 30V and measuring the +0.6x =1.8xIO-9v2
light intensity (output power). The upper profile of Fig. 5
shows the applied voltage and the bottom profile shows the
measured output power.
599
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WeA1 7.3

III. NONLINEAR OBSERVER DESIGN FOR THE SATURATED The ratio in (16) is a maximum when Qo is the identity
OUTPUT MEMS OPTICAL SWITCH SYSTEM matrix [7]. Set Qo = I2, then a sufficient condition for Vobs
A. Observer Design to be negative is
The dynamical model of the optical switch can be described
by the nonlinear system ||L||k < (1/j6)[1/Amax (PO ) - Y] (17)

x= Ax+Bu±+ (x);y = Cx+O(x) (10) If (17) is satisfied, VObs (x) < O, and the
Vx X O
equilibrium point x = 0 is asymptotically stable and the
where A, B, C are known linear parts about an equilibrium observer accurately reconstructs the state estimates.
point, xo and 0(x) and 6(x) are known locally Lipschitz
nonlinearities at xo i.e., IV. NONLINEAR OBSERVER-CONTROLLER DESIGN
A. Feedback Linearization Controller
llo(x) - 0(x)11 ' 71lx - xll; 110(x) - 0(x)11 <'f,lJX - xl ( 1
Feedback linearization control is proposed to control the
switch position. An estimate of the switch states from the
Vx,4cD(xo,r) = {x c Rn lIX-_l1 < r}
with y and ,6 known
observer is needed for feedback linearization compensation.
Lipschitz constants. 0(x) is obtained by linearizing the With the nonlinear observer design discussed in section III, an
system output dynamics about xo. The nonlinearity in the integrated control scheme is presented to combine the
switch state dynamics can be regarded as Lipchitz, at least nonlinear controller and observer. Fig. 7 shows the block
locally, provided the operating range of x and x are diagram of the integration of the nonlinear controller and
guaranteed to be bounded. The pair (A, C) is observable. observer.
We create an observer with linear output injection of the
form
x=Ax+Bu+0(x() +L[Cx +0(k) -Cx- (x)] (12)

The estimation error dynamics are then seen to be given by

x =(A+LC)x +0(x) -0(X) +L[0(x) - 0(k)] (13)

where x = x - x is the observer estimation error.

B. Stability Analysis
If we choose (A + LC) to be Hurwitz, then for any
positive definite Q0, there exists a unique symmetric positive
definite P0 which satisfies the Lyapunov equation
(A+LC) P0 +PO(A+LC) =2Q.
Consider the Lyapunov function candidate
Fig. 7. Block Diagram of the integration of the Nonlinear Controller and
the Observer.
VobS(X)= 5Tp (14)

Taking its derivative along the trajectories of (13) and A smooth desired trajectory xd = [xd,xd,xd]T is
utilizing the Lipschitz property of 0(x) and 0(x) results in generated from a proper reference trajectory generator. The
objective of the controller is to provide proper control action
that enables the switch position to track the desired
b .[22min(Q)-2(7+flI L2)lmaL(I)] A12 (15) trajectory.
The dynamics of the optical switch is given by
where )min(Qo) and )umax(Po) are the minimum and maximum
eigenvalues of QO and PO respectively. Hence Vobs < 0 if x1 - (18)
(16)
X2- (1/m){[-kx1 -(dlx, +dO)X2]+keV2}
Amin (Q, ) /Ama-,, (P, ) > ()/ + 8 JIL 112 )

600
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WeA1 7.3

where the voltage term v2 is the control input. for the total system (observer and controller). Then
The dynamics in (18) can be put in the
formr = f(x)+ g(x)u. According to [8], any nonlinear V(x,e) < -bLx 12 ek [Ml e2
C (24)
system whose state space has dimension n = 2 can be
transformed into a linear system, via state feedback and where
change of coordinates, around a point xo, if and only if the
matrix (g(xO) adfg(xo)) has rank 2. This condition is 2- +,8llL )ma'(PO) -
2ma,(Pc lb L +
2
(25)
satisfied for the switch dynamics given in (18).
We choose a nonlinear feedback control law ofthe form
For V(x, e) to be negative, it is sufficient that the leading
v2= (m/ke){w -
(/m)[- i6 -
(d, + dj)X2]} (19)
principal minors of the 2x2 matrix M be positive. This
condition results in
where the term w guarantees the tracking of the desired
trajectory and x1, x2 are state estimates from the nonlinear
observer. Since it is desired to have the switch position xl, ..L. k < (1118)[11 Amax (PO ) ly I (26)
track a prescribed trajectory Xd, the term w in the control law
is specified as w = -a1e1- a01 where e x1 xd.
d llbllk< 2 [-( + 8IILII iax (Po)] /i max (Pc )-)/ (27)
We define the tracking error as el xl1 xd and
If (26) and (27) are satisfied, then V(x, e) <0
e2 = X2 - xd . The perturbed system tracking error dynamics
V x O, e 0 and the equilibrium point (x,e)= (0,0) is
is given by
asymptotically stable.
e=Ae bx + [0(x) 0(x)]
- (20) V. SIMULATION RESULTS
The performance of the proposed control scheme is
simulated on the MEMS optical switch. The reference
AC b =L
signal used is the output of the second order transfer
kl m alI-d o/m
Where =L and
-aO -al a-
function (r s + 1) driven by a pulse wave input. The
B. Stability Analysis choice of the time constant which determines the speed of
-c

To show the stability of the feedback linearization motion from the initial to the final position is limited by the
controller with estimated states, consider the Lyapunov constraint on the magnitude of the control input (Vmax =
function candidate for the controller 35 V). Zero initial velocity is assumed for both observer and
MEMS device to mimic the practical situation of the MEMS
device and the estimation process switching on at the same
VCfnK(e) =eTPce (21) time.
The time response of the system for a desired trajectory of
Select Ac to be Hurwitz, then for any positive definite Qc amplitude 10tm and 25ptm are shown in Fig. 8 and Fig. 9
there exists a unique symmetric positive definite Pc which respectively.
satisfies the Lyapunov equation ATPc + PcA = -2Qc. x 10-5
1.2,
Taking the derivative of Vc,n(e) along the trajectories of Reference
Plant

(20) and simplifying gives .------ Observer

08
Vc0n(e) < 2Lmin(Qc)ee +1
2 (22)
211e|12 112 max
(Pc)(lbll2 Y)
.2 06

04
where )jmin(Qc) and )umax(Pc) are minimum and maximum
eigenvalues of Q, and P, respectively. Since Q, is positive 0.2
definite, it is selected such that Qc = Qo = I2 .
Now consider the composite Lyapunov function
Time(s) x 10-3
V(x, e) = VcOn (e) + Vobs (x) (23) Fig. 8. Response ofthe system to a desired reference signal of amplitude
loQm.

601
45th IEEE CDC, San Diego, USA, Dec. 13-15, 2006 WeA1 7.3

x 105 x 105
2.5 2.5

2 2

E15
fi1.5- fi1.5-

.2_
on0 n 1

0.5 0.5

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Time(s) x 10-3 Time(s) x 10-3
Fig. 11. Observer model and feedback linearization has 20% error
Fig. 9. Response ofthe system to a desired reference signal of amplitude
25pm VI. CONCLUSION
It is observed that although the observer starts with different A MEMS optical switch has been developed. Its dynamical
initial conditions its states converge to the actual states very model has been established based on the electrical, mechanical
quickly and the tracking of the reference signal is almost and optical models of the device. The model is verified by
perfect. Also, the system performance is worse for small experimental results.
reference signals. A nonlinear observer was designed to estimate the dynamic
The response also shows a very fast response time with the variables of the optical switch. Feedback linearization
rise time and falling time less than 0.4ms. The overshoot controller with estimated states from a nonlinear observer has
above the desired reference signal amplitude and undershoot been introduced to improve the dynamic closed loop
below zero are almost negligible. There are no oscillations. performance of the optical switch. Simulations demonstrate
Now, we assume that the estimation of the model that even with large estimation error of the observer model
parameters used in the feedback linearization term and and feedback linearization term, the proposed control scheme
can still achieve the desired performance.
observer model have 20% error, specifically the parameters m,
dx, do, k, and ke were reduced by 20%. Fig. 10 and Fig. 11
REFERENCES
show the time response for a desired trajectory of amplitude
10tm and 25ptm respectively. [1] R. T. Chen, H. Nguyen and M. C. Wu, "A high-speed low-voltage
stress-induced micromachined 2x2 optical switch," IEEE Photonics
Technology Letters, Vol. 11, No. 11, 1999.
x
105 [2] C. R. Giles, V. Aksyuk, B. Barber, R. Ruel, L. Stulz, and D. Bishop,
Reference "A silicon MEMS optical switch attenuator and its use in light wave
Plant subsystems," IEEE Journal of Selected topics in Quantum Electronics,
1 - -- Obsermer Vol. 5, No. 1, pp. 18-25,1999.
[3] J-J. Slotine and W. Li, Applied Nonlinear Control. New Jersey:
0.8 - -T r - - Prentice Hall, 1991.
[4] D. H. S. Maithripala, J. M. Berg, and W. P. Dayawansa, "Nonlinear
dynamic output feedback stabilization of electrostatically-actuated
° 0.6 - _

MEMS," Proceedings of the CDC, Maui, HW, 2003.

0
[5] R. Rajamani, "Observers for lipschitz nonlinear systems," IEEE
Transactions on Automatic Control, Vol. 43, No. 3, March 1998.
[6] F. E. Thau, "Observing the state of nonlinear dynamic systems," Int. J.
02
Contr., Vol. 17, No. 3, 1973.
[7] R. V. Patel and M. Toda, "Quantitative measures of robustness in
multivariable systems," in Proc. American Control Conf., San
0
0 0.2 04 06 08 1 12 14 16 18 Francisco, CA, 1980, pp. TP8-A.
Time(s) x X3
10 [8] A. Isidori, Nonlinear Control Systems. London: Springer-Verlag,
1995.
[9] J. Li, Q. X. Zhang, A. Q. Liu, "Advanced fiber optical switches using
Fig. 10. Observer model and feedback linearization has 20% error deep RIE (DRIE) fabrication," Sensors and Actuators A: Physical,
Vol. 102, No. 3, 2003, pp 286-295.
A similar performance of the nonlinear observer-controller [10] A. Q. Liu, X. M. Zhang, C. Lu, F. Wang, C. Lu and Z. S. Liu,
design is observed even for very large modeling uncertainties. "Optical and mechanical models for a variable optical attenuator using
a micromirror drawbridge," J. Micromech. Microeng., Vol. 13, pp
This clearly demonstrates the parameter variation robustness 400-411, 2003.
of this proposed control scheme for the MEMS optical switch [11] S. D. Senturia, Microsystems Design. Kluwer Academic Publishers,
system. 2000, pp. 327.
[12] B. Borovic, C. Hong, A. Q. Liu, L. Xie and F. L. Lewis, "Control of a
MEMS optical switch," Proceedings of the CDC, Bahamas, 2004.

602