ACES JOURNAL, VOL. 21, NO.

1, MARCH 2006 9

Comparative Study of Analytical and Simulated

Doubly-Supported RF MEMS Switches for
Mechanical and Electrical Performance
Nickolas Kingsley, Guoan Wang, and John Papapolymerou
E-mail: [email protected], [email protected], [email protected]

Georgia Institute of Technology, School of Electrical and Computer Engineering, Atlanta, GA 30308

Abstract — Radio Frequency Microeletromechanical The switch works by deflecting the beam towards the
System (RF MEMS) switches are useful for providing low- bottom metal layer and causing an RF short circuit. The
loss switching elements in high frequency devices. Since inductive regions behave like springs and make it easier
these devices contain a mechanical and an electrical
component to their operation, predicting their performance
to deflect the beam. A spring constant can be determined
is not trivial. Computational analysis can be extremely which evaluates the amount of force necessary to deflect
complicated due to the large number of variables that need the beam a given distance. Changing the shape or
to be incorporated. Using a multi-physics simulation tool dimensions of the inductive region will increase or
seems like the only solution, but most simulators are decrease the spring constant. The capacitive regions are
optimized for only one engineering realm (i.e. mechanics or
electronics). Combining different engineering realms into
responsible for creating an electrostatic force between the
one simulated model will usually compromise the accuracy DC biased beam and the metal layer below it. This force
of the results. Often simulators cannot model a multi-realm is responsible for decreasing the “gap” between the metal
device at all. This paper offers a solution to this problem by layers. Changing the gap length, height, or the area of the
proposing a technique for combining computational analysis capacitive region will increase or decrease the
with simulation to determine the pull-down voltage and RF
characteristics of an RF MEMS switch. Measurement
electrostatic force necessary to deflect the beam.
results agree closely with the simulated results using this Across the capacitive region, the charge density in the
technique. metal should be uniform. Otherwise, the beam will not
deflect parallel to the bottom metal layer. Any skewing of
the beam caused by fabrication misalignment or non-
I. INTRODUCTION
symmetric inductive regions will result in a larger
RF MEMS switches have become a popular area of capacitance and a poor RF open circuit. As long as the
research in recent years due to their small size, low loss, switch is adequately thick (2-3 skin depths), made from a
good isolation, and low cost. Solid-state switches at high high-quality, highly conductive metal (copper or gold,
frequencies are lossy and cause more distortion. An usually), and properly aligned (to equalize the fringing
example of a doubly-supported (air-bridge type) electric fields on all sides) charge density in the metal
capacitive MEMS switch is shown in Figure 1. will be uniform. MEMS switches that are not deflecting
uniformly are usually caused by fabrication
misalignment, non-uniform metal thickness, or
contaminants in the capacitive region metal. The latter
two issues prevent the charge density from being uniform
by hampering the flow of electrons in the metal and can
be rectified by altering the fabrication recipe.
Electrically, the inductive and capacitive regions
behave as their name implies. Changes in these regions
will change the RF performance of the switch. The
dielectric layer provides high capacitance when the
switch is in the down state and is used to prevent stiction
between the two metal layers. A very thin layer (~2000Å)
of silicon nitride is typically used and generally has a
negligible effect on the mechanics of the switch. That is,
the bending of the beam is not directly effected by the
presence of the silicon nitride. However, electrons can
accumulate in this thin layer which can build up a large
Fig. 1. Rendering of RF MEMS switch in UP and DOWN state. enough charge to effect the electrostatic actuation of the
switch. Dielectric charging is especially pronounced in
silicon nitride layers that are deposited using Plasma

1054-4887 © 2006 ACES

10 ACES JOURNAL, VOL. 21, NO. 1, MARCH 2006

Enhanced Chemical Vapor Deposition (PECVD) because when devices with small feature sizes (i.e. RF MEMS
of the large number of atomic defects generated from the switches) are simulated in a complex environment (i.e.
plasma. Charging effects can be greatly reduced by when surrounded by an electric field) assumptions must
properly grounding the silicon nitride to prevent electron be made within the simulator and results will be
accumulation. This can be improved further by thermally compromised [9].
growing the dielectric layer instead of using PECVD [1]. Often, when multiple physical realms are involved in a
Modeling MEMS switches for optimal electrical and problem, the optimal solution method is to use a
mechanical performance can be a daunting task and is simulator to solve the problem in the more complicated
often substituted with a less accurate method. For realm and to combine those results manually with theory
instance, MEMS switches are often designed for optimal from the simpler realm. For the RF MEMS switch, we
electrical properties (such as a low RC time constant [2]) are combining a mechanical beam dynamics problem
or optimal mechanical properties (such as a low actuation with an electrostatic problem. The theory that deals with
voltage [3]). There are four popular inductive region the electrostatics of a capacitive region is well known and
configurations [4]. These designs, labeled 1-4, are shown straightforward, whereas the dynamics of a beam with
in Figure 2. complicated springs is much more difficult to solve.
Solving the problem in one simulation that couples the
two physical realms does not always give the most
accurate results because of assumptions and
simplifications used in the simulator. Instead, this paper
presents a straightforward method for modeling an RF
MEMS switch by simulating first in an optimized FEM
mechanical simulator then calculating the pull-down
voltage by using simple electrostatic equations. The
measured results match very closely with the results from
this method, which demonstrates its effectiveness.

II. MECHANICAL ANALYSIS OF RF MEMS

SWITCHES
Equations for predicting the bending of cantilever and
Fig 2. Switch designs 1-4. doubly-supported beams have been around for decades
[10]. Unfortunately, applying simplistic equations to
Deriving the equations for predicting MEMS switch complex MEMS devices can be cumbersome. The two
performance that utilizes these inductive and capacitive most important mechanical features of a MEMS switch
regions is difficult. Very general equations can be are the pull-down voltage and the deflection. Both of
investigated but the results can only be used as rough these quantities can be calculated by treating the MEMS
estimates [5-6]. Those who have tried predicting MEMS switch as a mechanical spring. In order to calculate the
switch behavior using only theory often report a pull-down voltage, one must equate the force pulling
discrepancy upwards of a factor of ten between predicted down on the beam by the electrostatic force between the
and measurement results [7]. Certainly design metal layers
optimization can not be done this way. Using simulation ε AV 2
software is the only way to take into account most of the f down = (1)
idiosyncrasies of device performance. However, it is not 2g 2
always possible, or effective, to use a simulator to predict and the force pushing up from the spring (Hooke’s Law)
mechanical performance due to an electrostatic force. [11],
RF MEMS switch feature sizes are often on the order
of λ/1000 or smaller. This is much smaller than the fup = − k ( g o − g ) . (2)
typical element size of a Finite Element Method (FEM)
or Finite-Difference Time-Domain (FDTD) simulator,
whose typical element sizes are λ/20 to λ/10, although For these equations, ε is the permittivity, A is the area, V
simulations with small feature sizes are still possible with is the voltage, k is the spring constant, go is the initial
these methods [8]. A Method of Moments (MOM) gap, and g is the evaluated gap. We can use these simple,
simulator could be used to model the small feature sizes, spatially independent equations since we know the
but if the switch is being simulated with other devices charge density (and therefore the force) is uniform across
(i.e. filters or antennas) or on a multilayer substrate then the capacitive region. It has been well documented that
an FEM simulator would be more accurate because of the for parallel plate electrostatic actuation, when the gap
improved cell size. Clearly there is a trade off. reduces to 2/3 of the original gap, the beam becomes
Alternatively, hybrid simulators have been investigated unstable and experiences a “pull-in” effect [11]. That is,
which attempt to utilize the advantages of both types of the MEMS switch does not deflect over the entire gap
simulation. No matter which type of simulator is used, according to the formula in (1). Instead, when the gap
KINGSLEY, WANG, PAPAPOLYMEROU: STUDY OF DOUBLY-SUPPORTED RF MEMS SWITCHES 11

reaches a certain threshold, namely 2/3 the original gap, springs in parallel add directly and springs in series add
the switch will snap down. Magnets experience the same as the inverse of the sum of the reciprocals. Therefore,
effect. As two magnets of opposite polarity are brought the effective spring constants for the four switch designs
closer together the attractive force is barely noticeable presented in this paper are:
until they reach a certain distance apart. At this point they
snap together and the force between them is great. Design 1
Equating (1) and (2) where the gap is 2/3 the original 32 EWH 3 (7)
gap and solving for the pull down voltage gives keff =
L3
8kg o3
VPI = . (3) Design 2
27ε A
1 1 1 1
= + +
The maximum deflection can also be calculated from keff km kn − m km
the spring constant by the equation [10]
km kn − m (8)
∴ keff =
δ = −F k (4) k m + 2k n − m
where δ is the deflection, F is the force pushing down on Design 3
the spring (in Newtons) and k is the spring constant. 1 1 1 1 1 1
The values for the permittivity, area, and gap can be = + + + +
keff km km kn − m km km
designed for and implemented in fabrication. The only
two unknowns for a given MEMS switch are the spring km kn − m
∴ keff = (9)
constant and the downward force. The spring constant k m + 4kn − m
can be derived for a meandered line by the equation [4]:
3 Design 4
 t 
Ew   1 1 1 1
 Lc  = + +
km = (5) keff 2km kn − m 2km
 
 2  km kn − m
L L  1 +ν  ∴ keff = (10)
1 + s   s  + 12 km + kn − m
Lc   Lc   w 
2

 1+   
 t   where km is the meandered spring constant given by (5)
and kn-m is the non-meandered spring constant given by
where w is the width of the meander, t is the thickness of (6). Substituting keff from (7)-(10) into (3) for k will give
the metal, v is the Poisson’s Ratio of the metal, Ls is the the theoretical pull down voltages.
overall width of the spring, and Lc is the distance from
the end of the spring to the start of the meander. These
dimensions are illustrated below. III. MECHANICAL SIMULATION OF RF MEMS
Before any complex mechanical simulations are
performed, it is necessary to validate the model. Careful
attention must be given to material properties, boundary
conditions, and the applied forces. One way to validate a
simulation model is to compare simulated values with
theoretical values for a simple case. If the results agree,
more complicated configurations can be simulated and
the results can be trusted.
Fig.3. Illustration of dimensions for (5).
A. Verification of Simulation Tool
For a non-meandered spring, the spring constant is The FEMLAB 3.0 static structural mechanics module
given by [12] from Comsol was used for the mechanical simulations.
32 EWH 3 FEMLAB is a multiphysics simulation tool, which is
knon − m = (6) commonly used in industry and university settings [13].
L3 The 3D MEMS switch structure with non-meandered
where E is the Young’s Modulus, W is the width, H is springs (Design 1) was simulated with a uniform force
the thickness, and L is the length. pushing down on the center capacitive region.
The effective spring constant, keff, for the entire MEMS The theoretical deflection profile can be determined by
switch can be determined by combining the simple spring taking advantage of spring superposition. This procedure
equations in a fashion similar to capacitors. That is,
12 ACES JOURNAL, VOL. 21, NO. 1, MARCH 2006

is demonstrated in the figure below for the distribution of

force, q.

Fig. 6. Comparison of Simulated and Analytical

displacement of non-meandered switch.

Since the simulation results agree closely with the

analytical results, it is safe to assume that the simulator
Fig. 4. Spring superposition. will be reasonably accurate for the more complicated
spring configurations. The simulated deflection profile of
the four switch designs is shown in Figure 7.
The deflection equation for a uniformly actuated beam
is given by [10]
x 2 ( L2 − 2 Lx + x 2 ) q
δ ( x) = (11)
2 EWH 3
where x is the position along the beam, L is the length of
the beam, and q is the force applied per length. These
parameters are exemplified in Figure 5.

Fig.7. 3D Deflection Profile of RF MEMS Switches.

B. Deriving Pull-down Voltage from Simulation
Using FEMLAB, it is easy to determine the force
necessary to deflect the MEMS switch a desired distance.
Fig. 5. Illustration of dimensions for (11). Ideally, it is necessary to deflect the MEMS switch the
The deflection equation for a partially actuated beam is same distance as the gap between the beam and the metal
given by [11]: layer below it (usually 1.5-3µm). The equation that
relates force to pull-down voltage in terms of the gap is
(12) given by [10]
 qbx 2
− ( 3L + 3a − 2 x ) for 0 ≤ x ≤ a
δ =  12 EI 2g 2 F
− q ( x 4 − 4 Lx 3 + 6 L2 x 2 − 4a 3 x + a 4 )
VPI = (14)
 24 EI
for a ≤ x ≤ L ε
where F is the force per area. This equation is derived
where a is the distance from the anchor that the force from the pull-down voltage in (3), where F incorporates
begins, b is the length of the beam that the force is the spring constant. Doing a unit analysis between (3)
applied to, and I is the moment of inertia given by [12]: and (14) will result in the same outcome, volts.
HW 3 Changing the force per area acting on the capacitive
I= (13) region until the deflection matches the gap will determine
12 the force. Although a guess-and-check method is
where H is the thickness and W is the width of the beam. necessary to determine the value, this can be performed
Figure 6 shows a plot of the deflection given by quickly using interpolation since force and deflection are
FEMLAB and the results from the superposition of (11) linearly related. This force can then be used in (14) to
and (12). calculate the pull-down voltage.
KINGSLEY, WANG, PAPAPOLYMEROU: STUDY OF DOUBLY-SUPPORTED RF MEMS SWITCHES 13

IV. ELECTRICAL SIMULATION OF RF MEMS

In addition to the mechanical performance of MEMS
switches, it is important to evaluate the RF
characteristics. The springs exhibit an inductance, the
actuation region exhibits a capacitance, and the metal
beam exhibits a resistance. All together, the beam
behaves like a series RLC circuit. These values can be
calculated within an order of magnitude by using
fundamental RLC equations. The resistance can be
calculated using [14]:
ρL (15)
R=
HW
where ρ is the metal resistivity and L is the length of the
beam. The capacitance can be calculated using [14]:
εA Fig. 8. Fabrication process for MEMS switches.
C= (16)
g
Knowing the resonant frequency from measurements, the
inductance can be calculated using [14]:

1000 (17)
L=
4π 2Cf 2
where f is the resonant frequency given in GHz, C is
given in pF, and L is calculated in nH. Papers have been
published which investigate elaborate circuit models for
MEMS switches [15-16]. However, if results within an
order of magnitude are suitable, these simple equations
are more than adequate.

V. MEASUREMENTS
All four switch designs were fabricated and measured
Fig. 9. SEM photos of fabricated switches.
to determine the actual pull-down voltage and resonance
frequency. The process steps are shown in Figure 8. The
coplanar waveguide (CPW) signal lines were fabricated
A. Comparison of Mechanical Analysis
by electron beam evaporating a titanium – gold (Ti-Au)
layer. Silicon nitride (Si3N4) was deposited using Plasma
Table 1 displays the comparison between the purely
Enhanced Chemical Vapor Deposition (PECVD) and
theoretical, the simulated method presented in this paper,
patterned using a Reactive Ion Etch (RIE). A sacrificial
and the measured pull-down voltage.
photoresist layer was spun on and hard baked. The
sacrificial layer was removed using a photoresist stripper TABLE I
and a carbon dioxide (CO2) critical point drying process COMPARISON OF THEORETICAL, SIMULATED, AND MEASURED VPI
was used to release the switches. Design Theoretical Simulated Measured Avg Avg %
Scanning electron pictures of two of the switches are Error Error
1 117.135V 127.5V 100V 13.75 11.97%
shown in Figure 9. 2 40.547V 38.4V 35V 3.85V 10.14%
Measurements were taken with Thru-Reflect Line 3 31.875V 27.8V 30V 2.98V 9.97%
(TRL) calibration to deembed the cable and connector 4 69.050V 72.8V 70V 2.35V 3.33%
losses.
The measurement results agree closely with the
theoretical and simulated results. The average error is
VI. RESULTS
within the measurement ramping tolerance (5V).
Results for the mechanical and electrical characteristics The theoretical results are generally within 5-8% of the
of the four spring designs are presented in the following simulated values. The small discrepancy is mainly due to
sections. Measurement results were taken for each the Poisson ratio of the metal, which the simulator takes
design. The measured pull-down voltage is within 5V of into account and theory does not [10,13]. The Poisson
the minimum pull-down voltage. Voltage ramping must ratio relates a change in the width as the length of the
be done quickly to minimize charge accumulation in the beam is increased. There is a small discrepancy between
underlying dielectric region.
14 ACES JOURNAL, VOL. 21, NO. 1, MARCH 2006

simulated and theoretical values due to simulator VII. CONCLUSION

meshing tolerances.
In this paper, four different RF MEMS switch designs
were analyzed using theory and simulations. By
B. Comparison of Electrical Analysis
combining mechanical simulation results with simple
electrostatic equations, a prediction for the pull-down
The switches were measured to determine the
voltage and RF performance was achieved. This
resonance frequency. This is shown in the figure below.
prediction was more accurate and much easier to
determine than using only theory or only simulations. To
verify our mechanical simulation model, it was shown
that for a simple switch geometry, the simulated
deflection closely matched the theoretical displacement
found by using spring superposition. A pull-down
voltage for each switch was determined by using the pull-
down force given by the mechanical simulator with an
equation that relates force to voltage. Measuring the
resonant frequency and calculating the resistance,
capacitance and inductance determined the electrical
circuit model. These RLC values can be used to design
other RF MEMS switches. Measurement results agreed
very well with predicted values, thus demonstrating that
Fig. 10. Resonance frequency of MEMS switches.
simulation results can be conveniently combined with
analytical results to achieve accurate predictions.
Using the measured resonance frequency and the
capacitance calculated from (16), the inductance can be
determined by (17). The resistance can be calculated ACKNOWLEDGEMENT
from (15). Table 2 shows the resonance frequency values
The authors wish to thank Holly Kingsley for her
and the calculated capacitance, inductance, and
insight on the mechanics of springs and for verifying this
resistance.
paper from a mechanical engineering perspective. The
TABLE II
authors would also like to thank Dr.Oliver Brand for
CAPACITANCE, INDUCTANCE, AND RESISTANCE OF RF MEMS inspiring this project.
Design Resonant C L R This work is partially supported by GEDC and
Frequency partially by NASA under contract #NCC3-1057.
1 22.8175GHz 2.2pF 22pH 0.3Ω
2 11.3625GHz 2.9pF 65pH 0.6Ω
3 12.1525GHz 2.8pF 60pH 0.5Ω REFERENCES
4 21.83GHz 1.9pF 28pH 0.2Ω [1] Campbell, Stephen, “The Science and Engineering of
Microelectronic Fabrication, Second Edition” Oxford
The measurement results were compared to a series University Press, 2001.
RLC circuit with the same values as Table 2 to verify the [2] S. Simion, “Modeling and design aspects of the MEMS
switch,” Semiconductor Conference, vol. 1, pp. 128, Oct.
model. One such comparison is shown in Figure 11.
2003.
[3] Pacheco, S.P.; Katehi, L.P.B.; Nguyen, C.T.-C., “Design
of low actuation voltage RF MEMS switch,”
IEEE MTT-S Microwave Symposium Digest., vol. 1, pp.
165-168, June 2000.
[4] Wang, G., Barstow, S., Jeyakumar, A., Papapolymerou, J.,
Henderson, C., “Low cost RF MEMS switches using
photodefinable mixed oxide dielectrics,”
IEEE MTT-S Microwave Symposium Digest, vol. 3, pp.
1633-1636, June 2003.
[5] Simion, S., “Modeling and design aspects of the MEMS
switch,” International Semiconductor Conference2003,
vol. 1, pp. 125-128, 2003.
[6] Mercier, D., Blondy, P., Cros, D., Guillon, P., “An
electromechanical model for MEMS switches,” IEEE
Microwave Symposium Digest, vol. 3, pp. 2123-2126, May
2001.
[7] Peroulis, D., Pacheco, S.P., Sarabandi, K., Katehi, L.P.B.,
Fig. 11. RLC Circuit vs. Measurement Results.
“Electromechanical considerations in developing low-
voltage RF MEMS switches,” IEEE Transactions on
These results agree very closely with each other. The Microwave Theory and Techniques, vol. 51, issue 1, pp.
electrical model is satisfactory. 259-270, Jan. 2003
KINGSLEY, WANG, PAPAPOLYMEROU: STUDY OF DOUBLY-SUPPORTED RF MEMS SWITCHES 15

[8] Bushyager, N., Lange, K., Tentzeris, M. and John Papapolymerou received
Papapolymerou, J., "Modeling and Optimization of RF- the B.S.E.E. degree from the
MEMS Reconfigurable Tuners with Computationally National Technical University of
Efficient Time-Domain Techniques", Proc. of the 2002
IEEE-IMS Symposium, pp. 883-886, June 2002. Athens, Athens, Greece, in 1993,
[9] Wang, Z., Jensen, B., Volakis, J., Saitou, K., Kurabayashi, the M.S.E.E. and Ph.D. degrees
K., “Analysis of RF-MEMS switches using finite element- from the University of Michigan,
boundary integration with moment method,” Antennas and Ann Arbor, in 1994 and 1999,
Propagation Society International Symposium, vol. 2, pp. respectively.
173-176, 22-27 June 2003.
[10] E.P. Popov, “Mechanics of Materials, Second Edition,”
From 1999-2001 he was an Assistant Professor at the
Prentice-Hall, Inc, 1976.
[11] Stephen Senturia, “Microsystem Design,” Kluwer Department of Electrical and Computer Engineering of
Academic Publishers, 2001. the University of Arizona, Tucson and during the
[12] James Gere, “Mechanics of Materials, Fifth Edition,” summers of 2000 and 2003 he was a visiting professor at
Thompson-Engineering, 2003. The University of Limoges, France. From 2001-2005 he
[13] Comsol, Inc. “FEMLAB Multiphysics Modeling,” was an Assistant Professor at the School of Electrical and
http://www.comsol.com/products/femlab/, November Computer Engineering of the Georgia Institute of
2004.
[14] David Pozar, “Microwave Engineering, Second Edition,”
Technology, where he is currently an Associate
John Wiley & Sons, Inc., 1999. Professor. His research interests include the
[15] Muldavin, J. and Rebeiz, G., “High-Isolation CPW MEMS implementation of micromachining techniques and
Shunt Switches – Part 1: Modeling,” IEEE Transactions MEMS devices in microwave, millimeter-wave and THz
on Microwave Theory and Techniques, vol. 48, no. 6, pp. circuits and the development of both passive and active
1045-1052, June 2000. planar circuits on semiconductor (Si/SiGe, GaAs) and
[16] Qian, J., Li, G., deFlavis, F., “A parametric model of organic substrates (LCP, LTCC) for System-on-a-Chip
MEMS capacitive switch operating at microwave
frequencies,” IEEE Microwave Symposium Digest, vol. 2, (SOC)/ System-on-a-Package (SOP) RF front ends.
pp. 1229-1232, June 2000. Dr. Papapolymerou received the 2004 Army Research
Office (ARO) Young Investigator Award, the 2002
National Science Foundation (NSF) CAREER award, the
Nickolas Kingsley received B.S. best paper award at the 3rd IEEE International Conference
(2002) and M.S. (2004) degrees in on Microwave and Millimeter-Wave Technology
electrical engineering from the (ICMMT2002), Beijing, China and the 1997 Outstanding
Georgia Institute of Technology, Graduate Student Instructional Assistant Award
where he is currently pursing a PhD. presented by the American Society for Engineering
He is currently researching the Education (ASEE), The University of Michigan Chapter.
integration and packaging of RF His student also received the best student paper award at
MEMS switches into various microwave devices on the 2004 IEEE Topical Meeting on Silicon Monolithic
liquid crystal polymer (LCP) and silicon substrates. He is Integrated Circuits in RF Systems, Atlanta, GA. He has
the recipient of the 2002 President’s Undergraduate authored or co-authored over 120 publications in peer
Research Award from Georgia Tech and the 2001 reviewed journals and conferences. He currently serves
Armada Award from Compaq Computer Corporation. He as the Vice-Chair for Commission D of the US National
is a student member of IEEE, IEEE APS, IEEE MTT-S, Committee of URSI and as an Associate Editor for IEEE
and Order of the Engineer. Transactions on Antennas and Propagation. During 2004
he was the Chair of the IEEE MTT/AP Atlanta Chapter.

Guoan Wang received his Bachelor

degree from Central South
University, his Master of Science in
Materials Science and Engineering
from Zhejiang University, and his
Master of Science in Electrical
Engineering from Arizona State
University. He is currently working toward the Ph.D
degree in Electrical Engineering at the Georgia Institute
of Technology. His research interests include: RF MEMS
switches with novel materials and their applications for
reconfigurable microwave circuits, integration of RF
circuitry onto CMOS grade silicon, and micromachining
techniques for microwave applications.