You are on page 1of 27

Nonlinear Dynamics 31: 91117, 2003.

2003 Kluwer Academic Publishers. Printed in the Netherlands.


A Study of the Nonlinear Response of a Resonant Microbeam to
an Electric Actuation
M. I. YOUNIS and A. H. NAYFEH
Department of Engineering Science and Mechanics, MC 0219, Virginia Polytechnic Institute and State
University, Blacksburg, VA 24061, U.S.A.; E-mail: anayfeh@vt.edu
(Received: 21 February 2002; accepted: 1 October 2002)
Abstract. An investigation into the response of a resonant microbeam to an electric actuation is presented. A non-
linear model is used to account for the mid-plane stretching, a DC electrostatic force, and an AC harmonic force.
Design parameters are included in the model by lumping them into nondimensional parameters. A perturbation
method, the method of multiple scales, is used to obtain two rst-order nonlinear ordinary-differential equations
that describe the modulation of the amplitude and phase of the response and its stability. The model and the results
obtained by the perturbation analysis are validated by comparing them with published experimental results. The
case of three-to-one internal resonance is treated.
The effect of the design parameters on the dynamic responses is discussed. The results show that increasing
the axial force improves the linear characteristics of the resonance frequency and decreases the undesirable fre-
quency shift produced by the nonlinearities. In contrast, increasing the mid-plane stretching has the reverse effect.
Moreover, the DC electrostatic load is found to affect the qualitative and quantitative nature of the frequency-
response curves, resulting in either a softening or a hardening behavior. The results also show that an inaccurate
representation of the system nonlinearities may lead to an erroneous prediction of the frequency response.
Keywords: MEMS, resonator, primary resonance, forced vibration.
1. Introduction
Electrically actuated microbeams have been widely used and studied by the MEMS com-
munity. They form a major component in many micromechanical devices, such as capacitive
switches [1, 2], pull-in sensors [3], and resonant sensors [416]. In this paper, we study
resonant sensors using electrically actuated microbeams.
The fundamental natural frequency of a resonant microbeam is very sensitive to the axial
strain. External loads, such as pressure, temperature, force, and acceleration, apply axial
strains on a microbeam, leading to a shift in its natural frequencies. This shift is readily
converted to a digital signal, which is related to the physical quantity being measured. The
fundamental natural frequency also can be affected by other factors, such as squeeze-lm
damping [4] and the elasticity of the microbeam supports [5].
The electric load and the mechanical restoring force govern the behavior of a microbeam.
The electric load is composed of a DC polarization voltage and an AC voltage. The DC
component applies an electrostatic force on the microbeam, thereby deecting it to a new
equilibrium position, while the ACcomponent vibrates the microbeam around this equilibrium
position. The combined electric load has an upper limit beyond which the mechanical restoring
force can no longer resist its opposing force, thereby leading to the collapse of the microbeam.
This structural instability phenomenon is known as pull-in, and the critical voltage associated
92 M. I. Younis and A. H. Nayfeh
with it is called pull-in voltage. Younis et al. [6] showed that the pull-in voltage corresponds
to a saddle-node bifurcation. They calculated the static deection at each DC voltage and
found two solution branches: the larger solution is unstable and the smaller one is stable. As
the DC voltage increases, these branches get closer to each other, coalesce , and destroy each
other in a saddle-node bifurcation, corresponding to the pull-in voltage. The deection at pull-
in is approximately 57% of the airgap space [6, 7], which separates the microbeam from the
stationary substrate.
A microbeam restoring force is composed of three components: bending, axial force, and
mid-plane stretching due to immovable boundaries. The mechanical restoring force tends to
shift the natural frequencies to higher values, while the electrostatic force tends to shift the
natural frequencies to smaller values.
Anumber of studies have investigated the problem of a microbeam that is initially deected
by a DC electrostatic force and driven by an AC force. Zook et al. [8] reported experiment-
ally that increasing the driving AC voltage leads to an increase in the resonance frequency
(hardening behavior). Also, they observed a hysteresis that depends on the direction of the
frequency sweep.
Tilmans and Legtenberg [9] approximated the nonlinear resonance frequency as the fun-
damental natural frequency for small values of the AC forcing amplitude. For large-amplitude
AC excitations, they approximated the dynamic problem using Rayleighs energy method,
modied to account for electric forces and mid-plane stretching. They derived an equation that
governs the resonance frequency as a function of the excitation amplitude. They compared the
results obtained using this equation to their experimental results. Although they found a good
qualitative agreement, showing a hardening effect, the agreement was poor quantitatively.
They concluded that such a hardening behavior is more severe for high-quality factors, high
DC voltages, and high AC voltages, whereas large axial strains can decrease it.
Gui et al. [10] investigated the dynamic problem using Rayleighs energy method, modi-
ed to account for electric forces and mid-plane stretching. They neglected the applied axial
force. They derived an equation for the fundamental resonance frequency, which predicts a
hardening behavior. They also derived a criterion that denes a region of DC and AC voltages
for a hysteresis-free operation. They compared their theoretical and experimental results and
found good agreement.
Turner and Andrews [11] approximated the nonlinear resonance frequency of a microbeam
using a perturbation method. The problem was modeled using a spring-mass system with a
cubic nonlinearity representing mid-plane stretching. Using the method of harmonic balance,
they derived equations describing the microbeam resonance frequency for two separate cases.
In one case, they neglected the electrostatic force and included mid-plane stretching, and,
in the second case, they included the electrostatic force and neglected mid-plane stretching.
Using the equations obtained from both cases, they derived an equation that compensates for
the increase in the resonance frequency due to mid-plane stretching and the decrease in the
resonance frequency caused by the electrostatic force. The results showed that eliminating the
effect of the frequency dependence on the amplitude requires a very high DC voltage, which
may not be attainable.
Ayela and Fournier [12] experimentally studied the response of microbeams of different
geometric shapes to a general electric load composed of DC and AC components. They
made several plots to show variation of the resonance frequency with the excitation amplitude
for different axial loads. The experimental data were used to extract parameters that t the
equation of a spring-mass system with cubic nonlinearity and a linear electric force (i.e., the
Nonlinear Response of Resonant Microbeams to Electric Actuations 93
electric force is assumed to be independent of the gap distance). The experimental results
showed that, for the operating conditions used, some devices exhibited softening behavior,
whereas others exhibited hardening behavior. They concluded that a nonlinear behavior may
result from many distinct phenomena and that each case has to be studied separately. They
added that the nonlinear behavior of these devices may be investigated experimentally, but
there is no way of predicting it analytically.
Veijola et al. [13] modeled a microbeam using a spring-mass model with a cubic nonlin-
earity representing mid-plane stretching. They included the effect of the driving AC voltages,
but neglected the effect of the DC polarization voltages. They used the method of harmonic
balance to show that the nonlinearity of the electrostatic forces can cause a softening-type
behavior, whereas the mid-plane stretching can cause a hardening-type behavior. They con-
ducted an experiment, which showed a hardening behavior. They used the results to extract
values for the parameters involved in their model. They compared the experimental results
with those obtained using their model and found them to be in good agreement. However,
they did not give an explanation for observing a hardening rather than a softening behavior.
From the aforementioned overview, we note that the literature lacks a comprehensive
model that predicts the dynamic response of a microbeam to an electric excitation over general
operating conditions. Most existing models are formulated with a pre-assumed behavior based
on experimental observations, which limits the analysis to specic design parameters only.
For instance, the nonlinearity of the system is assumed to be cubic and positive to justify
the observed hardening behavior whereas the nonlinearity of the electric forces, which tends
to yield softening behavior, is ignored. In some models, the system nonlinearity is totally
ignored, thereby restricting the analysis to small values of AC voltages and DC polarization
voltages. Further, in many cases, the microbeam is modeled as a single-degree-of-freedom
system, thereby neglecting its mass distribution. The electric force is included incompletely
in some papers by neglecting one of its components. In other cases, the dependence of the
electric forces on the variable gap is ignored. In addition, most of the previous works do not
recognize the fact that a deected microbeam caused by DC polarization voltages has natural
frequencies different from those of a straight beam; thus, the models are valid only for very
low DC polarization voltages.
In a previous work [7], we introduced a nonlinear model for a microbeam, which accounts
for the initial deection due to a DC electrostatic force, mid-plane stretching, and an applied
axial load. The model does not account for the effects of the shear deformation and rotary
inertia. We analyzed the static deection of the microbeam and investigated its linear natural
frequencies and mode shapes when actuated by a DC electrostatic force. Here, we expand on
that work by considering the microbeam response to a general electric load composed of both
DC and AC components. We model the microbeam as a distributed-parameter system. The
equations are nondimensionalized and the design parameters of the resonator are lumped into
nondimensional parameters, which are then used to develop a general equation that describes
the dynamics of the microbeam. We apply a perturbation method, the method of multiple
scales, to the general equation of motion and its associated boundary conditions, to obtain
an approximation of the response of the microbeam to a primary-resonance excitation. We
derive equations that describe the nonlinear resonance frequency, the amplitudes of periodic
solutions, and the stability of these solutions. We study the effect of the design parameters on
the nonlinear resonance frequency and the effective nonlinearity of the system. We compare
the theoretical results with available experimental data. Then, we investigate the possibility
of activating a three-to-one internal resonance between the rst and second modes, which is
94 M. I. Younis and A. H. Nayfeh
Figure 1. A schematic drawing of a resonant microbeam.
the most likely case to occur during the normal operation of the resonator. The absence of
nonlinear interactions is necessary for a stable and reliable device.
2. Problem Formulation
We consider a resonant microbeam (Figure 1) actuated by an electric load composed of a
DC component (polarization voltage) V
p
and an AC component v(t ) and subject to a viscous
damping c per unit length.
We model the microbeam as a plate undergoing cylindrical bending under an applied load,
which is constant along the width of the plate. Typical dimensions of resonators violate the
basic assumptions of a beam model; the ratio of a typical microbeam length to width b is
< 10. The plate is clamped across its width and free across its long ends. We assume that the
transverse deection w(x, t ) of the plate is constant along the width, thus reducing the plate
static deection equation to [17]
EI
1
2

4
w
x
4
=
_
_
EA
2(1
2
)

_
0
_
w
x
_
2
dx +

N
_
_

2
w
x
2
, (1)
where x is the position along the plate length, A and I are the area and moment of inertia
of the cross section (A = bh and I = (1/12)bh
3
, where h is the microbeam thickness), E
is Youngs modulus, is Poissons ratio, and

N is the applied tensile axial force. The plate
boundary conditions are
w(0, y) = w(, y) = 0,
w(0, y)
x
=
w(, y)
x
= 0. (2)
Using the modied beam equation, we rewrite the equation of motion governing the transverse
deection of the microbeam under electric forces as
EI
1
2

4
w
x
4
+ A

2
w
t
2
+ c
w
t
=
_
_
EA
2(1
2
)

_
0
_
w
x
_
2
dx +

N
_
_

2
w
x
2
+
1
2

r
b
_
V
p
+ v(t )
_
2
(d w)
2
, (3)
where w(x, t ) is the deection of the microbeam, t is time, is the material density, d is the
gap width,
0
is the dielectric constant of vacuum, and
r
is the relative dielectric constant of
the gap medium to that of an air gap. The last term in Equation (3) represents the parallel-
plate electric forces [18] assuming a complete overlapping area between the microbeam and
Nonlinear Response of Resonant Microbeams to Electric Actuations 95
the stationary electrode. The microbeam is assumed to be clamped at both ends; hence the
boundary conditions are
w(0, t ) = w(, t ) = 0,
w(0, t )
x
=
w(, t )
x
= 0. (4)
For convenience, we introduce the nondimensional variables (denoted by stars)
w

=
w
d
, x

=
x

, t

=
t
T
, (5)
where T is a time scale, which is dened below. Substituting Equations (5) into Equations (3)
and (4) and dropping the stars, we obtain

4
w
x
4
+

2
w
t
2
+ c
w
t
= [
1
(w, w) + N]

2
w
x
2
+
2
(V
p
+ v(t ))
2
(1 w)
2
, (6a)
w(0, t ) = w(1, t ) = 0,
w(0, t )
x
=
w(1, t )
x
= 0, (6b)
where
(f
1
(x, t ), f
2
(x, t )) =
1
_
0
f
1
x
f
2
x
dx. (6c)
Equation (6a) is a nondimensional integral-partial-differential equation with linear and non-
linear terms as well as external and parametric excitation terms. The parameters appearing in
Equation (6a) are
c =
c
4
(1
2
)
EIT
,
1
= 6
_
d
h
_
2
, N =

N
2
(1
2
)
EI
,
2
=
6
0

4
(1
2
)
Eh
3
d
3
(7)
and T is chosen as
T =
_
bh
4
(1
2
)
EI
_
1/2
.
Typically, T is in the ranges of microseconds. The ranges of the rest of the nondimensional
parameters are c 100300,
1
0.110, N 10100, and
2
V
2
p
0.150.
The microbeam deection under an electric force is composed of a static component due to
the DC voltage, denoted by w
s
(x), and a dynamic component due to the AC voltage, denoted
by u(x, t ); that is,
w(x, t ) = w
s
(x) + u(x, t ). (8)
To calculate the static deection of the microbeam, we set the time derivatives and the AC
forcing term in Equation (6a) equal to zero and obtain
d
4
w
s
dx
4
= [
1
(w
s
, w
s
) + N]
d
2
w
s
dx
2
+

2
V
2
p
(1 w
s
)
2
. (9a)
96 M. I. Younis and A. H. Nayfeh
It follows from Equations (6b) and (8) that the boundary conditions are
w
s
= 0 and
dw
s
dx
= 0 at x = 0 and x = 1. (9b)
We generate the problem governing the dynamic behavior of the microbeam around the
deected shape by substituting Equation (8) into Equations (6) and using Equations (9) to
eliminate the terms representing the equilibrium position. To third-order in u, the result is

2
u
t
2
+ c
u
t
+

4
u
x
4
= (
1
(w
s
, w
s
) + N)

2
u
x
2
+ 2
1
(w
s
, u)
d
2
w
s
dx
2
+
2
2
V
2
p
(1 w
s
)
3
u +
1
(u, u)
d
2
w
s
dx
2
+ 2
1
(w
s
, u)

2
u
x
2
+
3
2
V
2
p
(1 w
s
)
4
u
2
+
1
(u, u)

2
u
x
2
+
4
2
V
2
p
(1 w
s
)
5
u
3
+
2
2
V
p
(1 w
s
)
2
v(t ) +
4
2
V
p
(1 w
s
)
3
v(t )u + , (10a)
u(0, t ) = u(1, t ) = 0,
u(0, t )
x
=
u(1, t )
x
= 0, (10b)
where the term involving v
2
(t ) is dropped because typically v
2
(t ) V
2
p
in resonant sensors.
We end this section with two notes regarding the choice of the boundary conditions and the
use of the linear damping in the model. First, the xed-xed end conditions are idealization
to actual cases which may have some elasticity [5]. Some structures employ step-up type
supports, which are not perfectly rigidly xed [19]. According to [19], the exibility of the
supports reduces the critical value of the buckling load to be less than that for ideal xed-
xed ends. Bouwstra and Geijselaers [5] mentioned that this exibility makes the resonance
frequencies lower than the predicted values for ideal xed-xed ends. The second note relates
to the so-called squeeze-lm damping. When a resonator moves too close to a stationary
substrate, additional damping forces, the squeeze-lm damping, increases due to the pressure
built up in the airgap [14]. Because typically resonant sensors are placed in a vacuum cavity in
low pressure under 0.2 mbar, the resonance frequencies are almost independent of the pressure
change [9, 15].
3. Linear Mode Shapes and Corresponding Natural Frequencies
We drop the nonlinear, forcing, and damping terms in Equation (10a) and obtain the linear
undamped eigenvalue problem

4
u
x
4
(
1

1
+ N)

2
u
x
2

2
2
V
2
p
(1 w
s
)
3
u 2
1
(w
s
, u)
d
2
w
s
dx
2
=

2
u
t
2
, (11a)
u = 0 and
u
x
= 0 at x = 0 and x = 1, (11b)
where
1
= (w
s
, w
s
). We solve Equations (11) for the undamped mode shapes and natural
frequencies under the static deection distribution w
s
(x). Assuming a harmonic motion in the
nth mode shape
u(x, t ) =
n
(x) e
i
n
t
, (12)
Nonlinear Response of Resonant Microbeams to Electric Actuations 97
where
n
(x) is the nth mode shape and
n
is the nth nondimensional natural frequency, we
reduce Equations (11) to

iv
n
(
1

1
+ N)

n

2
2
V
2
p
(1 w
s
)
3

n
2
1

2
w

s
=
2
n

n
, (13a)

n
=

n
= 0 at x = 0 and x = 1, (13b)
where
2
= (w
s
,
n
) and the prime denotes the derivative with respect to x.
4. Forced Vibration: Single-Mode Approximation
4.1. PERTURBATION ANALYSIS
We investigate the nonlinear vibrations of a clamped-clamped microbeam subject to an electric
load composed of DC and AC components. We analyze its nonlinear response to a primary-
resonance excitation of its rst mode because it is the case that is used in resonator applica-
tions. Still, the analysis is general and can be used to study nonlinear responses of the beam
to a primary resonance of any of its mode.
There are at least two approaches for determining the response of a distributed-parameter
system with quadratic and cubic nonlinearities. In the rst approach, called discretization
approach, one discretizes the governing integral-partial-differential equation and associated
boundary conditions, keeps enough modes in the discretization, and uses a perturbation method
to attack the discretized system [20]. In the second approach, called the direct approach, one
attacks directly the distributed-parameter problem. In this paper, we attack Equations (10) dir-
ectly using the method of multiple scales [20, 21] to determine a uniformly valid approximate
solution. To this end, we seek a second-order uniform solution of Equations (10) in the form
u(x, t ; ) = u
1
(x, T
0
, T
2
) +
2
u
2
(x, T
0
, T
2
) +
3
u
3
(x, T
0
, T
2
) + , (14)
where is a small nondimensional bookkeeping parameter, T
0
= t , and T
2
=
2
t . We note
that u(x, t ) does not depend on T
1
= t because, as shown below, secular terms arise at
O(
3
). In order that the nonlinearity balances the effects of the damping c and excitation v(t ),
we scale them so that they appear together in the modulation equations as follows:
2
c and

3
V
AC
cos(t ) where V
AC
is the magnitude of the applied AC voltage and is the excitation
frequency. Substituting Equation (14) into Equation (10a) and equating coefcients of like
powers of , we obtain
order
L(u
1
) = D
2
0
u
1
+ u
iv
1

1

1
u

1
Nu

1
2
1
(w
s
, u
1
)w

s

2
2
V
2
p
(1 w
s
)
3
u
1
= 0; (15)
order
2
L(u
2
) =
1
w

s
(u
1
, u
1
) + 2
1
(w
s
, u
1
)u

1
+
3
2
V
2
p
(1 w
s
)
4
u
2
1
; (16)
98 M. I. Younis and A. H. Nayfeh
order
3
L(u
3
) = 2D
0
D
2
u
1
cD
0
u
1
+ 2
1
w

s
(u
1
, u
2
) + 2
1
(w
s
, u
2
)u

1
+ 2
1
(w
s
, u
1
)u

2
+
1
(u
1
, u
1
)u

1
+ 2

F(x) cos(T
0
) +
6
2
V
2
p
(1 w
s
)
4
u
1
u
2
+
4
2
V
2
p
(1 w
s
)
5
u
3
1
, (17)
where

F(x) =

2
V
p
V
AC
(1 w
s
)
2
.
The boundary conditions at all orders are given by
u
j
= 0 and u

j
= 0 at x = 0 and x = 1, j = 1, 2, 3. (18)
The rst-order problem given by Equations (15) and (18) is identical to the linear ei-
genvalue problem, Equations (11). Because in the presence of damping the homogeneous
solutions corresponding to all modes that are not directly or indirectly excited decay with
time [21], the solution of Equations (15) and (18) is assumed to consist of only the directly
excited mode. Accordingly, we express u
1
as
u
1
= A(T
2
) e
iT
0
(x) +

A(T
2
) e
iT
0
(x), (19)
where A(T
2
) is a complex-valued function that is determined by imposing the solvability
condition at third order, the overbar denotes the complex conjugate, and and (x) are the
natural frequency and corresponding eigenfunction of the considered mode, respectively. The
eigenfunction (x) is normalized such that
_
1
0

2
dx = 1.
Substituting Equation (19) into Equation (16), we obtain
L(u
2
) = (A
2
e
2iT
0
+ 2A

A +

A
2
e
2iT
0
)h(x), (20)
where
h(x) =
1
(, )w

s
+ 2
1

(w
s
, ) +
3
2
V
2
p
(1 w
s
)
4

2
.
The solution of Equations (20) and (18) can be expressed as
u
2
=
1
(x)A
2
e
2iT
0
+ 2
2
(x)A

A +
1
(x)

A
2
e
2iT
0
, (21)
where
1
and
2
are the solutions of the boundary-value problems
M(
i
, 2
1i
) = h(x), (22)

j
= 0 and

j
= 0 at x = 0 and x = 1, j = 1, 2 (23)
and
ij
is the Kronecker delta and the linear differential operator M(, ) is dened as
M(, ) =
iv

2
+
1

2
1
(w
s
, )w

s

2
2
V
2
p
(1 w
s
)
3
. (24)
To describe the nearness of the excitation frequency to the fundamental natural fre-
quency , we introduce the detuning parameter dened by
= +
2
. (25)
Nonlinear Response of Resonant Microbeams to Electric Actuations 99
Substituting Equations (19), (21), and (25) into Equation (17) and keeping the terms that
produce secular terms only, we obtain
L(u
3
) =
_
i(2A

+ cA)(x) + (x)A
2

A +

F(x) e
iT
2
_
e
iT
0
+ cc + NST, (26)
where A

denotes the derivative of A with respect to T


2
, cc denotes the complex conjugate of
the preceding terms, and NST stands for terms that do not produce secular terms. The function
(x) is dened as
(x) =
G
q
(x) +
G
c
(x) +
E
q
(x) +
E
c
(x),
where

G
q
(x) = 2
1
w

s
(
1
, ) + 4
1
w

s
(
2
, ) +
_
2
1

1
+ 4
1

2
_
(w
s
, )
+ 2

1
(w
s
,
1
) + 4

1
(w
s
,
2
),

G
c
(x) = 3

1
(, ),

E
q
(x) =
6
2
V
2
p
(1 w
s
)
4
(2
2
+
1
) ,

E
c
(x) =
4
2
V
2
p
(1 w
s
)
5

3
.
The subscripts q and c denote quadratic and cubic nonlinear terms and the superscripts G and
E denote terms produced by the geometric and electric nonlinearities.
For a uniform second-order approximation, we do not need to solve for u
3
. Only, we need to
impose the solvability condition, which provides an equation for the function A. Because the
homogeneous problem governing u
3
has a nontrivial solution, the corresponding nonhomo-
geneous problem has a solution only if the right-hand side of Equation (26) is orthogonal
to every solution of the adjoint homogeneous problem governing u
3
[20]. We note that our
problem is self adjoint, the adjoints are given by (x) e
iT
0
. Multiplying the right-hand side
of Equation (26) with (x) e
iT
0
and integrating the result from x = 0 to x = 1, we obtain
the following solvability condition:
2i(D
2
A+ A) + 8SA
2

A F e
iT
2
= 0, (27)
where
S = S
G
q
+ S
G
c
+ S
E
q
+ S
E
c
,
S
G
q
=
1
8
1
_
0

G
q
dx,
S
G
c
=
1
8
1
_
0

G
c
dx,
S
E
q
=
1
8
1
_
0

E
q
dx,
100 M. I. Younis and A. H. Nayfeh
S
E
c
=
1
8
1
_
0

E
c
dx,
=
1
2
1
_
0
c
2
dx,
F =
1
_
0

F dx.
Here, S is the effective nonlinear coefcient of the system and S
m
n
denotes a nonlinear coef-
cient of source m(electric or geometric) and of order n (quadratic or cubic). If c is independent
of x, then = c/2. In general, c is a function of x.
Next, we express A in the polar form A = (1/2)a e
i
, where a and are real-valued
functions, representing, respectively, the amplitude and phase of the response. Substituting
for A in Equation (27), separating the real and imaginary parts, and letting = T
2
, we
obtain the following modulation equations:
a

= a +
F

sin , (28)
a

= a
Sa
3

+
F

cos . (29)
Substituting Equations (19) and (21) into Equation (14) and setting = 1, we express, to the
second approximation, the microbeam response to the external excitation as
u(x, t ) = a cos(t )(x) +
1
2
a
2
_

2
(x) + cos 2(t )
1
(x)
_
+ , (30)
where a and are governed by Equations (28) and (29).
It follows from Equation (30) that periodic solutions of Equations (10) correspond to
constant a and ; that is, the xed points (a
0
,
0
) of Equations (28) and (29). Thus, letting
a

= 0 in Equations (28) and (29) and eliminating


0
from the resulting equations, we
obtain the following frequency-response equation:
a
2
0
_

2
+
_

Sa
2
0

_
2
_
=
F
2

2
(31)
Equation (31) is an implicit equation for the amplitude a of the periodic response as a func-
tion of the detuning parameter (which is a representation of the excitation frequency), the
effective nonlinear coefcient S, the damping coefcient , and the amplitude of excitation
F. Solving Equation (31) for , we obtain
=
1
a
0
_
F
2

2

2
a
2
0
_
1/2
+
Sa
2
0

. (32)
Nonlinear Response of Resonant Microbeams to Electric Actuations 101
Recalling that =
2
, setting = 1, and noting that the amplitude a is maximum
when the expression inside the square root vanishes, we obtain the following equation for the
nonlinear resonance frequency:

r
= +
SF
2

2
. (33)
The parameters is related to the quality factor Q by
=

2Q
(34)
Substituting Equation (34) into Equation (33), we obtain

r
= +
4SQ
2

5
F
2
. (35)
Equation (35) relates the nonlinear resonance frequency
r
to the effective nonlinearity S
of the system, the amplitude F of the AC forcing, and the quality factor Q. Equation (35) can
be used to predict the resonance frequency for excitation frequencies much less than twice the
fundamental natural frequency. This because when 2 a principal parametric resonance
and a subharmonic resonance of order one-half will be activated.
The stability of the xed points (a
0
,
0
) is determined by examining the eigenvalues of the
Jacobian matrix of Equations (28) and (29) evaluated at the corresponding xed point [22].
The characteristic equation is

2
+ 2 +
_

2
+
_

3Sa
2
0

__

Sa
2
0

__
= 0. (36)
For asymptotically stable solutions, all of the eigenvalues must be in the left-half plane.
4.2. RESULTS
To describe the dynamic response of the microbeam, we need to determine the natural fre-
quency , the excitation amplitude F, the effective nonlinearity of the system S, and the
damping coefcients or Q.
As a rst step, we solved the boundary-value problem, Equations (9), for the deection
due to the DC electrostatic force. We did this iteratively using a shooting method to de-
termine the value of the integral (w
s
, w
s
). Using the converged solution w
s
, we solved
the boundary value problem, Equations (13), for the fundamental natural frequency and its
corresponding eigenfunction. We used a shooting method to iterate on
2
, , and until they
converged to within a predened tolerance. Next, we solved the boundary-value problems,
Equations (2224) to evaluate the functions
1
and
2
using a method similar to that used to
solve Equations (13). Finally, we evaluated the s, Ss, and F.
We begin by comparing the results obtained by using the present model with the experi-
mental results of Tilmans and Legtenburg [9] and Gui et al. [10]. Results from other papers,
which are mentioned in the introduction, are not used because either they were produced from
another device conguration [11, 12], for which our model does not apply, or the authors did
not give enough data about the specications of their devices [8, 13].
In Figure 2, we compare the normalized nonlinear resonance frequency
r
/ obtained
using Equation (35) with those obtained theoretically and experimentally by Tilmans and
102 M. I. Younis and A. H. Nayfeh
Figure 2. Comparison of the normalized nonlinear resonance frequency
r
/ calculated using our model (solid
lines) with those obtained theoretically (dashed lines) and experimentally by Tilmans and Legtenburg [9] for two
microbeams of lengths 210 m (diamonds) and 310 m (circles). The thin solid lines are based on the quality
factors of Tilmans and Legtenburg [9] and the thick solid lines are based on quality factors estimated using our
model.
Legtenburg [9] for two microbeams of lengths 210 and 310 m with h = 1.5 m, b =
100 m, d = 1.18 m, and subject to an axial load of 0.0009 Newton. The reported pull-
in voltages for the 210 and 310 m microbeams are 28 and 13.8 V, respectively. The DC
polarization voltage V
p
= 1 V for data points of the rst microbeam except for the rst two
points where V
p
= 2 V. For the second microbeam, V
p
= 2 V for all data points. The theory
of Tilmans and Legtenburg [9] is based on Rayleighs energy method modied to account for
electric forces and mid-plane stretching. The latter effect however was misrepresented, as we
show in Appendix A.
We show two sets of calculated results for each microbeam. The rst set, shown as thin
solid lines, uses the same values of Q reported by Tilmans and Legtenburg [9]. They extracted
the values of Q analytically using an equation derived from their model. The parameters
of this model were obtained by tting the predicted frequency-response curve at a low DC
voltage to the one obtained experimentally. These quality factors are Q = 592 and 151 for
the 210 and 310 m microbeams, respectively. As shown in Figure 2, these values give poor
results. Legtenburg and Tilmans [16] mentioned that the quality factor for the design of their
device varies across the wafer due to variations in the sealing pressure of the microbeam
encapsulation. Because such variations can lead to a wrong measurement of Q and due to the
difculty of measuring the system damping, in general, we determined the quality factors by
matching
r
/ obtained using Equation (35) at V
AC
= 0.6 V to the experimental value of
Tilmans and Legtenburg [9].
We obtained Q = 816.6 and 197 for the 210 and 310 m microbeams, respectively. The
results obtained using these values, shown as thick solid lines, are in excellent agreement with
the experimental results.
Nonlinear Response of Resonant Microbeams to Electric Actuations 103
Figure 3. Comparison of the hysteretic points obtained using our model (solid lines) with those obtained theoret-
ically (dashed line) and experimentally (circles) by Gui et al. [10]. The thin solid line is based on the quality factor
Q = 900 reported by Gui et al. [10] and the thick solid line is based on a quality factor Q = 1000 estimated using
our model.
In Figure 3, we show another comparison for the hysteretic points (points correspond to
the largest AC and DC voltages below which the response is single-valued and above which it
is multi-valued) predicted by our model with those obtained theoretically and experimentally
by Gui et al. [10] for a microbeam of length 210 m and h = 1.5 m, b = 100 m, d =
1 m, and Q = 900. The theoretical results of Gui et al. [10] were obtained using a modied
Rayleighs energy method that accounts for mid-plane stretching and electric forces. Mid-
plane stretching however was misrepresented, as we show in Appendix A.
In Figure 3, we show also two curves obtained using our theory corresponding to two dif-
ferent values of Q: the thick solid line was obtained using Q = 1000, which was determined
by matching our result at V
p
= 4 V to the experimental result of Gui et al. [10]. The thin solid
line was obtained using Q = 900, which is the value used by Gui et al. [10]. We estimated the
axial force to be

N = 0.00011 Newton by tting the theoretically obtained natural frequency
to the experimental value obtained by Gui et al. [10]. There is excellent agreement between
our results and the experimental results.
Figures 4 and 5 show the frequency-response curves (amplitude and phase responses)
corresponding to V
p
= 8 V in Figure 3. In both gures, thin lines are for V
AC
= 0.007 V
and thick lines are for V
AC
= 0.03 V. The unstable regions in the latter case are shown as
dashed lines. The results are obtained by solving Equations (31) and (36).
Next, we study the effect of the nondimensional design parameters on the nonlinear reson-
ance frequency. As indicated by Equation (35), the nondimensional design parameters affect
the nonlinear resonance frequency in two ways: changing the effective nonlinearity of the
104 M. I. Younis and A. H. Nayfeh
Figure 4. The amplitude of the dynamic response versus the frequency of excitation corresponding to V
p
= 8 V
in Figure 3.
Figure 5. The phase of the dynamic response versus the frequency of excitation corresponding to V
p
= 8 V in
Figure 3.
Nonlinear Response of Resonant Microbeams to Electric Actuations 105
Figure 6. Variation of the normalized nonlinear resonance frequency
r
/ with V
AC
for various values of the
nondimensional axial load N. The values of
1
,
2
V
2
p
, and Q are 3.7, 5.5, and 796, respectively.
system S and the fundamental natural frequency . We refer to Younis et al. [7] for the pull-in
voltages of the numerical data used in the following gures.
In Figure 6, we show the effect of varying the driving voltage AC on
r
/ for various val-
ues of the nondimensional axial load N. Increasing the driving voltage amplitude AC increase

r
/. On the other hand, increasing the axial force N at a constant V
AC
decreases
r
/ and
also increases its linear behavior. We note that negative values of N, which correspond to
compressive loads, have a greater effect on
r
/.
In resonant sensors, residual stresses are typically of the tensile type. They are induced
during the fabrication process. Legtenberg and Tilmans [16] mentioned that such stresses are
desirable to reduce the possibility of buckling. However, because resonant sensors may exhibit
temperature variations, which may induce compressive residual stresses, operation away from
the critical buckling load should be ensured. As an example, we calculated a critical buckling
load P
cr
= 13.3037 for the microbeam in Figure 2 of length 210 m using the formula
P
cr
=
4Eh
3
12L
2
(1
2
)
for a xed-xed beam.
In Figure 7, we show the effect of varying the driving voltage amplitude AC on
r
/ for
various values of
1
. It can be seen that, for values of
1
greater than 0.05, increasing AC
increases
r
/. In contrast, for values of
1
less than 0.05, increasing AC leads to a decrease
in
r
/. We also note that increasing
1
increases the nonlinear behavior of
r
/ and its
value at a constant value of AC.
In Figure 8, we show the effect of varying AC on
r
/ for various values of
2
V
2
p
. In-
creasing
2
V
2
p
at a constant V
AC
increases
r
/ up to a value of
2
V
2
p
35. Above this
value, the increase in
r
/ becomes smaller until it is reversed completely at
2
V
2
p
41,
106 M. I. Younis and A. H. Nayfeh
Figure 7. Variation of the normalized nonlinear resonance frequency
r
/ with V
AC
for various values of
1
.
The values of N,
2
V
2
p
, and Q are 8.7, 5.5, and 796, respectively.
Figure 8. Variation of the normalized nonlinear resonance frequency
r
/ with V
AC
for various values of
2
V
2
p
.
The values of
1
, N, and Q are 3.7, 8.7, and 796, respectively.
Nonlinear Response of Resonant Microbeams to Electric Actuations 107
Figure 9. Variation of the nonlinear coefcients with the axial load N. The values of
1
and
2
V
2
p
are 3.7 and
5.5, respectively.
where the nonlinear resonance frequency becomes highly sensitive to changes in the DC
polarization voltage. In fact, a slight increase in
2
V
2
p
beyond this value leads to a sharp drop,
which can even reach zero, in the nonlinear resonance frequency with increasing V
AC
. We
note that such a strange behavior occurs before reaching the pull-in limit, which in this case
occurs at
2
V
2
p
90. This result shows that the nondimensional parameter
2
V
2
p
changes
the dynamic response of the microbeam from a hardening to a softening response. This is
because, for this range of
2
V
2
p
, the electrostatic force, which tends to lower the resonance fre-
quency, drastically dominates the mid-plane stretching, which tends to increase the resonance
frequency.
4.3. DISCUSSION
In order to better understand how the nondimensional parameters N,
1
, and
2
V
2
p
affect
the effective nonlinearity of the system, we study the inuence of each parameter on the
nonlinear coefcients S
G
q
, S
G
c
, S
E
q
, S
E
c
, and S. In Figure 9, we show variation of these nonlinear
coefcients with increasing the axial load N. It shows that increasing N in the positive range
leads only to a slight linear decrease in the mid-plane stretching represented by S
G
c
, and hence
in the effective nonlinear coefcient S. However, this is not the major factor that leads to the
decrease in
r
/ with increasing N, which is shown in Figure 6. Increasing N also increases
, which according to Equation (35) decreases the shift in
r
/. This agrees with the results
of Figure 6, where the ratio
r
/ approaches unity for large N.
In Figure 10, we show variation of the nonlinear coefcients with
1
. Increasing
1
in-
creases linearly the mid-plane stretching coefcient S
G
c
, without affecting the other nonlinear
coefcients. The increase in S
G
c
, and so in S, increases
r
/, as indicated by Equation (35).
On the other hand, increasing
1
increases , which tends to decrease
r
/. The effect of
increasing S however dominates the later effect. This explains the increase in the normalized
108 M. I. Younis and A. H. Nayfeh
Figure 10. Variation of the nonlinear coefcients with
1
. The values of N and
2
V
2
p
are 8.7 and 5.5, respectively.
nonlinear resonance frequency with increasing
1
at a constant V
AC
, which is seen in Figure 7.
The effective nonlinear coefcient S starts with a negative value for very small values of
1
,
which explains the qualitative change observed in Figure 7 near this range of values.
In Figure 11, we show variation of the nonlinear coefcients with
2
V
2
p
. For small values
of
2
V
2
p
, the effective nonlinear coefcient S decreases with increasing
2
V
2
p
up to values near
40 where the quadratic electric termS
E
c
starts to decrease sharply, thereby dominating all other
nonlinear coefcients. At
2
V
2
p
41, the sign of S changes from positive, corresponding to
a hardening behavior, to negative, corresponding to a softening behavior, which explains the
qualitative change in the dynamic behavior observed in Figure 10. This is a signicant result;
it shows that using a spring-mass model with a cubic nonlinearity, which is the model usually
used in the literature, is inaccurate and might lead to wrong results. Such models neglect the
quadratic nonlinearity, which is due to the electrostatic force and the static deection. As
clearly shown in Figure 11, this nonlinearity becomes dominant and governs the dynamic
behavior beyond a critical value of
2
V
2
p
. Further, we note that assuming a single value for the
effective nonlinearity in the model may introduce another source of error because, as shown
in Figure 11, the effective nonlinearity is a strong function of the electrostatic force.
We note that the DC voltage needed to achieve a linear behavior in the nonlinear resonance
frequency (corresponding to S = 0) is relatively high (V
p
19V for the data of Figure 2)
and may not be attainable in resonant sensor applications. This result is in agreement with that
found by Turner and Andrews [11].
Next, we examine the sensitivity of the dynamic response of the microbeam to the quality
factor. In Figures 12 and 13, we show variation of the amplitude of the response a with the
detuning parameter for a microbeam with
1
= 3.7 and N = 8.7 and quality factors of Q =
200, Q = 500, and Q = 796. We use
2
V
2
p
=15 and 45 in Figures 12 and 13, respectively.
Both gures show that increasing the quality factor amplies the effect of the nonlinearity and
Nonlinear Response of Resonant Microbeams to Electric Actuations 109
Figure 11. Variation of the nonlinear coefcients with
2
V
2
p
. The values of
1
and N are 3.7 and 8.7, respectively.
Figure 12. Variation of the response amplitude a of a microbeam with the detuning parameter for
1
= 3.7,
N = 8.7, and
2
V
2
p
=15.
110 M. I. Younis and A. H. Nayfeh
Figure 13. Variation of the response amplitude a of a microbeam with the detuning parameter for
1
= 3.7,
N = 8.7, and
2
V
2
p
=45.
can change the dynamic state from a single-valued response, as in the Q = 200 curves, to a
multi-valued response, as in the Q = 500 and 796 curves. As can be noted from Figure 11,
the response curves exhibit a hardening behavior in Figure 12 and a softening behavior in
Figure 13.
5. Three-to-One Internal Resonance
5.1. PERTURBATION ANALYSIS
In this section, we consider modal interactions among the microbeam modes involving the
rst mode. It follows from Figure 14 that
1
is away from (1/2)
n
for any n. Hence, a two-
to-one internal resonance between
1
and
n
cannot be activated. However,
1
(1/3)
2
for
some range of V
p
, and hence we study the possibility of activating a 1:3 internal resonance
between the rst and second modes when the rst mode is excited with a primary resonance.
We apply the method of multiple scales directly to Equations (10). The analysis presen-
ted here is general and applicable to any two modes whose frequencies are in the ratio of
three-to-one. We seek a solution of Equations (10) in the form of Equation (14) and obtain
Equations (1518). Because in the presence of damping all modes that are not directly or
indirectly excited decay with time [16], the solution of Equations (15) and (18) is assumed to
consist of the two interacting modes; that is,
u
1
= A
n
(T
2
) e
i
n
T
0

n
(x) + A
m
(T
2
) e
i
m
T
0

m
(x) + cc, (37)
Nonlinear Response of Resonant Microbeams to Electric Actuations 111
Figure 14. Variation of the rst four natural frequencies, calculated by Younis et al. [7], of a microbeam with

2
V
2
p
for
1
= 3.7 and various applied axial loads.
where m and n denote the two modes being considered and the
j
are normalized such that
_
1
0

2
j
dx = 1. Substituting Equation (37) into Equation (16) yields
L(u
2
) = A
2
n
e
2i
n
T
0
h
1n
(x) + A
2
m
e
2i
m
T
0
h
1m
(x) + A
n

A
n
h
1n
(x) + A
m

A
m
h
1m
(x)
+ (A
n
A
m
e
i(
n
+
m
)T
0
+ A
n

A
m
e
i(
n

m
)T
0
)H
nm
(x) + cc, (38)
where
h
1i
(x) =
1
(
i
,
i
)w

s
+ 2
1

i
(w
s
,
i
) +
3
2
V
2
p
(1 w
s
)
4

2
i
,
H
ij
(x) = 2
1
(
i
,
j
)w

s
+ 4
1

i
(w
s
,
j
) +
6
2
V
2
p
(1 w
s
)
4

j
,
and the functional is dened in Equation (6c).
We express the solution of Equations (38) and (18) as
u
2
=
1n
(x)A
2
n
e
2i
n
T
0
+
1m
(x)A
2
m
e
2i
m
T
0
+
3
(x)A
n
A
m
e
i(
n
+
m
)T
0
+
4
(x)A
n

A
m
e
i(
n

m
)T
0
+
2n
(x)A
n

A
n
+
2m
(x)A
m

A
m
+ cc, (39)
where the
j
and
ij
are the solutions of the boundary-value problems
M(
1i
, 2
i
) = h
1i
(x), (40a)
M(
2i
, 0) = h
1i
(x), (40b)
112 M. I. Younis and A. H. Nayfeh
M(
3
,
n
+
m
) = H
nm
(x), (40c)
M(
4
,
n

m
) = H
nm
(x). (40d)
The operator M(, ) is dened in Equation (24). The boundary conditions for the
j
and

ij
are
= 0 and

= 0 at x = 0 and x = 1. (41)
Substituting Equations (37) and (39) into Equation (17) and considering the case
n
3
m
and either
n
or
m
, we obtain
L(u
3
) =
_
i
n
(2A

n
+ cA
n
)
n
(x) +
1n
(x)A
2
n

A
n
+
nm
(x)A
n
A
m

A
m
_
e
i
n
T
0
+
_
i
m
(2A

m
+ cA
m
)
m
(x) +
1m
(x)A
2
m

A
m
+
mn
(x)A
m
A
n

A
n
_
e
i
m
T
0
+
5
A
3
m
e
3i
m
T
0
+
6
A
n

A
2
m
e
i(
n
2
m
)T
0
+

F(x) e
iT
0
+ cc + NST, (42)
where A

j
is the derivative of A
j
with respect to T
2
and

1i
= 2
1
(
1i
,
i
)w

s
+ 4
1
(
2i
,
i
)w

s
+ 2
1
(w
s
,
1i
)

i
+ 4
1
(w
s
,
2i
)

i
+ 3
1
(
i
,
i
)

i
+ 2
1
(w
s
,
i
)

1i
+ 4
1
(w
s
,
i
)

2i
+
12
2
V
2
p
(1 w
s
)
5

3
i
+
6
2
V
2
p
(1 w
s
)
4
(
ii

i
+
2i

i
),

ij
= 2
1
(
3
,
j
)w

s
+ 2
1
(
4
,
j
)w

s
+ 4
1
(
i
,
2j
w

s
+ 2
1
(w
s
,
3
)

j
+ 2
1
(w
s
,
4
)

j
+ 4
1
(w
s
,
2j
)

i
+ 2
1
(w
s
,
j
)

3
+ 2
1
(w
s
,
j
)

4
+ 4
1
(w
s
,
i
)

2j
) + 2
1
(
j
,
j
)

i
+ 4
1
(
i
,
j
)

j
+
24
2
V
2
p
(1 w
s
)
5

2
j
+
6
2
V
2
p
(1 w
s
)
4
(
3

j
+
4

j
+ 2
2j

i
),

5
= 2
1
(
1m
,
m
)w

s
+ 2
1
(
1m
, w
s
)

m
+ 2
1
(w
s
,
m
)

1m
+
1
(
m
,
m
)

m
+
6
2
V
2
p
(1 w
s
)
4

1m

m
,

6
= 2
1
(
4
,
m
)w

s
+ 2
1
(
1m
,
n
)w

s
+ 2
1
(
4
, w
s
)

m
+ 2
1
(
1m
, w
s
)

n
+ 2
1
(w
s
,
m
)

4
+ 2
1
(
n
, w
s
)

1m
+
1
(
m
,
m
)

n
+ 2
1
(
n
,
m
)

m
+
6
2
V
2
p
(1 w
s
)
4

m
+
6
2
V
2
p
(1 w
s
)
4

1m

n
+
12
2
V
2
p
(1 w
s
)
5

2
m
.
To describe the nearness of
n
to 3
m
and to either
n
or
m
, we introduce the detuning
parameters
1
and
2
dened by

n
= 3
m
+
2

1
and =
i
+
2

2
. (43)
Nonlinear Response of Resonant Microbeams to Electric Actuations 113
Because the homogeneous part of Equations (42) and (18) has a nontrivial solution, the non-
homogeneous problem has a solution only if the right-hand side of Equation (42) is orthogonal
to every solution of the adjoint homogeneous problem governing u
3
[20]. We note that the
problem is self-adjoint, and hence the adjoints are given by
j
(x) e
i
j
T
0
. Multiplying the
right-hand side of Equation (42) by
n
(x) e
i
n
T
0
and
m
(x) e
i
m
T
0
, respectively, integrating
the results from x = 0 to x = 1, and using Equations (43), we obtain the following solvability
conditions:
2i
n
(D
2
A
n
+
n
A
n
) = 8S
nn
A
2
n

A
n
+ 8S
nm
A
n
A
m

A
m
+ 8
n
A
3
m
e
i
1
T
2
+ f
n
(x)
in
e
i
2
T
2
, (44)
2i
m
(D
2
A
m
+
m
A
m
) = 8S
mm
A
2
m

A
m
+ 8S
mn
A
m
A
n

A
n
+ 8
m
A
n

A
2
m
e
i
1
T
2
+ f
m
(x)
im
e
i
2
T
2
, (45)
where

i
=
1
2
1
_
0
c
2
i
dx,
f
i
=
1
_
0

F
i
dx,
S
ii
=
1
8
1
_
o

1i

i
dx, S
ij
=
1
8
1
_
o

ij

i
dx, i = j,

n
=
1
8
1
_
o

n
dx,
m
=
1
8
1
_
o

m
dx.
The S
ij
are nonlinear coefcients due to electric and geometric sources, and the
i
are the
nonlinear interaction coefcients between the nth and mth modes.
Expressing the A
i
in polar form and separating real and imaginary parts in Equations (44)
and (45), we obtain the modulation equations
a

n
=
n
a
n


n
a
3
m

n
sin
1
+
f
n

in

n
sin
2
, (46)
a
n

n
=
S
nn
a
3
n

S
nm
a
n
a
2
m


n
a
3
m

n
cos
1

f
n

in

n
cos
2
, (47)
a

m
=
m
a
m


m
a
n
a
2
m

m
sin
1
+
f
m

im

m
sin
2
, (48)
a
m

m
=
S
mm
a
3
m

S
mn
a
m
a
2
n


m
a
n
a
2
m

m
cos
1

f
m

im

m
cos
2
, (49)
114 M. I. Younis and A. H. Nayfeh
where
1
and
2
are dened as

1
=
1
T
2
3
m
+
n
,
2
=
2
T
2

in

n

im

m
.
Consequently, the microbeam dynamic response to second order can be expressed as
u(x, t ) = a
n
cos(
n
t +
n
)
n
(x) + a
m
cos(
m
t +
m
)
m
(x)
+
1
2
a
2
n
[cos 2(
n
t +
n
)
in
(x) +
2n
(x)]
+
1
2
a
2
m
[cos 2(
m
t +
m
)
im
(x) +
2m
(x)]
+
1
2
a
n
a
m
[cos((
n
+
m
)t +
n
+
n
)
3
(x)
+ cos((
n

m
)t +
n

m
)
4
(x)] + , (50)
where a
n
, a
m
,
n
, and
m
are governed by Equations (4649).
5.2. RESULTS
We considered the case in which the rst mode is directly excited with a primary resonance
(f
n
= 0) when
2
3
1
. We solved the boundary-value problem, Equations (9), for the static
deection. Then using this deection, we solved Equations (13) to determine the rst and
second natural frequencies and mode shapes. Using these results, we solved Equations (40)
and (41) for the s and then evaluated the s. Finally, we calculated the S
ij
,
i
, and f
m
.
We considered three cases. The rst case corresponds to
1
= 3.70,
2
V
2
p
= 35.30, N = 0,
and the rst and the second natural frequencies are
1
= 20.35 and
2
= 61. The second case
corresponds to
1
= 3.70,
2
V
2
p
= 5.50, N = 10, and the natural frequencies are
1
=
19.12 and
2
= 57.70. The last case corresponds to
1
= 3.70,
2
V
2
p
= 56.80, and N = 10,
and the natural frequencies are
1
= 21.60 and
2
= 64.20. For all three cases, the numerical
results show that the nonlinear interaction coefcients
i
vanish identically, which precludes
the possibility of activating this internal resonance. Hence, although the ratio between the
natural frequencies is close to three-to-one, the rst and second modes do not exchange energy.
This is because the rst mode is symmetric, the second mode is antisymmetric, and the static
deection is symmetric. And hence, the interaction coefcient
m
and
n
are identically zero.
6. Conclusions
We studied the nonlinear dynamic response of a microbeam that is actuated by a general
electric load subject to an applied axial load, accounting for mid-plane stretching. We used
a model that assumes operation at low pressures where the effect of squeeze-lm damping
on the resonance frequencies can be neglected. The model does not account for the effects
of shear deformation and rotary inertia. We used the method of multiple scales to determine
the response to a primary resonance excitation of the rst mode and obtained two nonlinear
rst-order ordinary-differential equations governing the amplitude and phase of the response.
We derived an equation that describes the nonlinear resonance frequency of the microbeam
as a function of the damping and the effective nonlinearity coefcient. This equation shows
that increasing the AC forcing and/or decreasing the damping leads to either an increase
Nonlinear Response of Resonant Microbeams to Electric Actuations 115
or a decrease in the nonlinear resonance frequency, depending on the sign of the effective
nonlinearity coefcient. We compared the nonlinear resonance frequency computed using our
theory to the experimental results available in the literature. We found excellent agreement.
We studied the effect of the nondimensional design parameters N,
1
, and
2
V
2
p
on the
normalized nonlinear resonance frequency
r
/. The results show that increasing N, the
axial force, improves the linear characteristics of
r
/ and decreases the frequency shift.
In contrast, increasing
1
, the mid-plane stretching, has the reverse effect on
r
/. On the
other hand,
2
V
2
p
affects the qualitative and quantitative nature of the effective nonlinearity
coefcient of the system. The sign of S changes from positive, corresponding to a hardening
behavior, to negative corresponding to a softening behavior. This is because the electric non-
linearity, mainly the quadratic, drastically increases in magnitude and overcomes the inuence
of the geometric nonlinearity. We note that most of the models used in the literature neglect
the effect of the electric nonlinearity and particularly the quadratic one. Instead, they assume
the nonlinearity of the system to be solely cubic and positive, which predicts a harding beha-
vior rather than the correct softening behavior. Therefore, failure to correctly account for the
nonlinearities in the system may lead to erroneous results. It is interesting to note that, for the
studied case, this reverse in behavior occurs at a value of
2
V
2
p
, which is about 46% of the
pull-in limit. Because at this value the coefcient S is equal to zero, the dynamic response is
theoretically linear. However, the simulation results show that
r
/ becomes highly sensitive
to any slight changes in
2
V
2
p
. This indicates that, although it is possible to compensate for the
effect of the electrostatic force with mid-plane stretching, operations near such conditions are
unstable and impractical. We believe more experimental work is needed to better understand
the system behavior near this region.
We applied the method of multiple scales to investigate possibility of activating a three-to-
one internal resonance between the rst and second modes, which if exists can adversely affect
the performance of the resonator. The analysis shows that these two modes are nonlinearly
uncoupled, and hence this internal resonance cannot be activated. This result enhances the
reliability of such a device.
In conclusion, the present nonlinear model provides an accurate prediction of the dynamic
behavior of microbeams, which linear models fail to explain. Unlike existing models in the
literature, the present nonlinear model is capable of simulating the mechanical behavior of
microbeams for general operating conditions and for a wider range of applied electric loads.
Further, by using the perturbation solution, one can easily derive analytical expressions that
present a clearer picture of the inuence of various design parameters, which is critical for
improving and optimizing designs.
Appendix A: Note on Energy Methods
We consider the microbeam shown in Figure 1 with zero AC and DC forces. To obtain the
equation that governs the microbeam motion, we set V
p
and v(t ) equal to zero in Equation (3),
assume the same boundary conditions as in Equation (4), and obtain
EI
1
2

4
w
x
4
+ bh

2
w
t
2
+ c
w
t
=
_
_
EA
2(1
2
)

_
0
_
w
x
_
2
dx +

N
_
_

2
w
x
2
, (A.1a)
116 M. I. Younis and A. H. Nayfeh
w(0, t ) = w(, t ) = 0 and
w
x

(0,t )
=
w
x

(,t )
= 0. (A.1b)
Equations (A.1) can be derived using the EulerLagrange equations from the Lagrangian
L = T U, (A.2)
where the potential U and kinetic T energies of the system are given by
U =
EI
2(1
2
)

_
0
_

2
w
x
2
_
2
dx +
1
2

_
0
_
w
x
_
2
dx
+
EA
8(1
2
)
_
_

_
0
_
w
x
_
2
dx
_
_
2
, (A.3a)
T =
1
2
bh

_
0
_
w
t
_
2
dx. (A.3b)
The total energy of the system is given by
H = U + T. (A.4)
Studying the same problem, Tilmans et al. [14] misrepresented the potential energy due to
mid-plane stretching by expressing it as
U
mid
=
EA
8

_
0
_
w
x
_
4
dx.
The correct representation however is the last term in Equation (A.3a). The theories of Tilmans
and Legtenburg [9] and Gui et al. [10] are based on this incorrect representation of the potential
energy.
References
1. Goldsmith, C. L., Yao, Z., Eshelman, S., and Denniston, D., Performance of low-loss RF MEMS capacitive
switches, IEEE Microwave and Guided Wave Letters 8, 1998, 269271.
2. Chan, E. K., Garikipati, K., and Dutton, W. R., Characteristics of contact electromechancis through
capacitance-voltage measurements and simulations, Journal of Microelectromechanical Systems 8, 1999,
208217.
3. Gupta, R. K. and Senturia, S. D., Pull-in time dynamics as a measure of absolute pressure, in Proceedings
of the IEEE Tenth Annual International Workshop on Microelectromechanical Systems: MEMS 97, Nagoya,
Japan, IEEE, New York, 1997, pp. 5159.
4. Andrews, M. K. , Turner, G. C., Harris, P. D., and Harris, I. M., A resonant pressure sensor based on a
squeezed lm of gas, Sensors and Actuators A36, 1993, 219226.
5. Bouwstra, S. and Geijselaers, B., On the resonance frequencies of microbridges, in Proceedings of the 6th
International Conference on Solid-State Sensors and Actuators (TRANSDUCERS 91), San Francisco, CA,
IEEE, New York, 1997, Vol. 2, pp. 11411144.
Nonlinear Response of Resonant Microbeams to Electric Actuations 117
6. Younis, M. I., Abdel-Rahman, E. M., and Nayfeh, A. H., A reduced-order model for electrically actuated
microbeam-based MEMS, Journal of Microelectromechanical Systems, to appear.
7. Younis, M. I., Abdel-Rahman, E. M., and Nayfeh, A. H., Static and dynamic behavior of an electrically ex-
cited resonant microbeam, in Proceedings of the AIAA 43rd Structures, Structural Dynamics, and Materials
Conference, Denver, CO, 2002, AIAA Paper No. 2002-1305.
8. Zook, J. D., Burns, D. W., Guckel, H., Sniegowski, J. J., Engelstad, R. L., and Feng, Z., Characteristics of
polysilicon resonant microbeams, Sensors and Actuators A35, 1992, 290294.
9. Tilmans, H. A. and Legtenberg, R., Electrostatically driven vacuum-encapsulated polysilicon resonators.
Part II. Theory and performance, Sensors and Actuators A45, 1994, 6784.
10. Gui, C., Legtenberg, R., Tilmans, H. A., Fluitman, J. H., and Elwenspoek, M.,Nonlinearity and hysteresis
of resonant strain gauges, Journal of Microelectromechanical Systems 7, 1998, 122127.
11. Turner, G. C. and Andrews, M. K., Frequency stabilization of electrostatic oscillators, in Digest of the 8th
International Conference on Solid-State Sensors and Actuators, Vol. 2, Stockholm, Sweden, S. Middelhoek
and K. Cammann (eds.), Elsevier, Amsterdam, 1995, Vol. 2, pp. 624626.
12. Ayela, F. and Fournier, T., An experimental study of anharmonic micromachined silicon resonators,
Measurement, Science and Technology 9, 1998, 18211830.
13. Veijola, T., Mattila, T., Jaakkola, O., Kiihamki, J., Lamminmki, T., Oja, A., Ruokonen, K., Sep, H.,
Seppl, P., and Tittonen, I., Large-displacement modeling and simulation of micromechanical electrostat-
ically driven resonators using the harmonic balance method, in the IEEE MTT-S International Microwave
Symposium Digest, Boston, MA, T. Perkins (ed.), IEEE, New York, 2000, Vol. 1, pp. 99102.
14. Tilmans, H. A., Elwespoek, M., and Fluitman, J. H., Micro resonant force gauges, Sensors and Actuators
A30, 1992, 3553.
15. Gui, C., Legtenberg, R., Elwenspoek, M., and Fluitman, J. H., Q-factor dependence of one-port encapsulated
polysilicon resonator on reactive sealing pressure, Journal of Micromechanics and Microengineering 5,
1995, 183185.
16. Legtenberg, R. and Tilmans, H. A., Electrostatically driven vacuum-encapsulated polysilicon resonators.
Part I. Design and fabrication, Sensors and Actuators A45, 1994, 5766.
17. Nayfeh, A. H., Nonlinear Interactions, Wiley, New York, 2000.
18. Grifths., D. J., Introduction to Electrodynamics, Prentice Hall, Englewood Cliffs, NJ, 1981.
19. Meng, Q., Mehregany, M., and Mullen, R., Theoretical modeling of microfabricated beams with elastically
restrained supports, Journal of Microelectromechanical Systems 2, 1993, 128137.
20. Nayfeh, A. H., Introduction to Perturbation Techniques, Wiley, New York, 1981.
21. Nayfeh, A. H. and Mook, D. T., Nonlinear Oscillations, Wiley, New York, 1979.
22. Nayfeh, A. H. and Balachandran B., Applied Nonlinear Dynamics, Wiley, New York, 1995.

You might also like