Professional Documents
Culture Documents
moelastic damping, QTED, of MEMS beams and is found to B. Lifshitz and Roukes’s theory
give good agreement with experimental results for simple- The Zener theory was improved in Ref. 3 by using the
harmonic oscillations in the flexural mode.9,10 When a beam beam theory.17 The equation of motion for a beam under
is flexed, one side of the beam is in tension and the other side thermoelastic damping is given by3
冉 冊
is in compression. The side in compression gets slightly
warmer and the side in tension gets slightly cooler due to the 2U 2 2U
A + EI + E␣IT = 0 共3兲
coupled nature of the thermal and mechanical domains.16 A t2 x2 x2
temperature gradient is formed across the beam giving rise to
heat flow 共for nonzero thermal conductivity兲 and this heat where is the density of the beam, A and I are the cross-
flow is an irrecoverable energy loss that limits the quality sectional area and the mechanical contribution to the mo-
factor of the beam.1 ment of inertia of the beam, respectively, and U is the dis-
placement of the beam in the y direction. The x axis is
defined along the length of the beam and the y and z axes are
A. Zener’s theory along the thickness and the width direction of the beam,
respectively. The term IT is the thermal contribution to the
Zener’s theory is valid for thin rectangular beams under-
beam’s moment of inertia 共measure of the thermal stress in
going simple-harmonic vibrations in the flexural mode. Ac-
the beam兲 and is given by
cording to this theory, the quality factor QTED due to ther-
moelastic damping is given by1,2
IT = 冕 A
y dy dz 共4兲
E ␣ 2T 0 z z E ␣ 2T 0
−1
QTED = = ⌬E , ⌬E = , where = T − T0 is the change in the temperature from the
C p 1 + 共 z兲 2 1 + 共 z兲 2 C p
ambient temperature T0. The linearized heat equation 共as-
共1兲 suming Ⰶ T0兲 along the y direction 共temperature gradients
along the other directions are assumed to be negligible兲 is
where is the density, E is the Young’s modulus, ␣ is the given by3
冉 冊
coefficient of thermal expansion, and C p is the specific heat
2 ⌬ E 2U
under constant pressure of the beam material. T0 is the am- = 2 +y . 共5兲
bient temperature 共the beam is unstrained and unstressed at t y ␣ t x2
this temperature兲 and is the angular frequency of excita- The coupled thermoelastic equations 共3兲–共5兲 are solved as-
tion. z is the relaxation time 共the time necessary for a tem- suming simple-harmonic vibrations in Ref. 3 by setting
perature gradient to relax兲 of the first mode of vibration of
the beam and is given by U共x,t兲 = U1共x兲eit, 共x,y,t兲 = 1共x,y兲eit . 共6兲
The temperature profile along the beam’s cross section is
b2 calculated by using the heat equation 共5兲 and the computed
z = 2 共2兲
temperature is substituted into the equation of motion 共3兲 to
obtain a frequency-dependent elastic modulus E 共see Ref. 3
where is the thermal diffusivity of the beam material and b for details兲,
is the beam thickness. When the angular frequency of exci- E = E兵1 + ⌬E关1 + f共兲兴其 共7兲
tation is such that z = 1, the energy loss 关measured by
−1
QTED in Eq. 共1兲兴 is maximum as the beam flexes and allows where
just enough time for the temperature gradient to relax before
the beam changes direction 共the time period of oscillation
matches the relaxation time兲. This peak value of energy dis-
f共兲 = 冋
24 b
b 2
3 3 − tan
b
2
冉 冊册 , =i 冑
i .
共8兲
冉 冊
no time to relax 共the beam flexes back and forth faster than
the heat can flow兲 and very little energy is dissipated. This is 6 6 sinh + sin
−1
QTED = ⌬E − 共9兲
called the adiabatic regime. On the other hand when the an- 2 3 cosh + cos
gular frequency is much smaller than z−1, no temperature
gradient is formed because the temperature gradient is pro- where
portional to the strain rate, which is low due to the low
angular frequency, and very little energy is dissipated. This
frequency range is called the isothermal regime. Zener’s
=b 冑
2
. 共10兲
−1
theory works quite well for simple beams but is not suitable In deriving the expression for the quality factor QTED in the
for microstructures with complex geometries. Besides, it is Zener theory 关Eq. 共1兲兴 the temperature profile in the beam
also based on the assumption that the motion of the micro- was expressed in terms of the transverse thermal eigenmodes
structure is simple harmonic. 共see Refs. 1 and 2 for details兲. On the other hand, in the
144305-2
THEORY OF THERMOELASTIC DAMPING IN… PHYSICAL REVIEW B 74, 144305 共2006兲
Lifshitz and Roukes theory 关Eq. 共9兲兴 the temperature profile solve the coupled thermoelastic equations 关Eqs. 共12兲 and 共5兲兴
in the beam was computed explicitly 共without expanding it in by assuming
terms of the thermal eigenmodes兲. As a result, the Lifshitz
NT NT
and Roukes theory is found to be more accurate than the
Zener theory for predicting the quality factor due to ther- U共x,t兲 = 兺 UN共x兲e
N=0
iNt
, 共x,y,t兲 = 兺 N共x,y兲eiNt ,
N=0
moelastic damping in rectangular beams undergoing simple-
harmonic vibrations in the flexural mode.3 共13兲
冑
moelastic damping from that predicted by the classical
theory 共both the Zener and the Lifshitz and Roukes theories兲 iN
N = i . 共15兲
of TED. The overall quality factor Q of a MEMS device is
given by
The thermal contribution to the beam’s moment of inertia IT
−1
Q−1 = QTED + Q−1 −1
fluid + Qother 共11兲 can be computed from Eqs. 共14兲, 共15兲, and 共4兲, 关noting
where QTED, Q fluid, and Qother are the quality factors due to 0共x , y兲 = 0兴 as
冕
thermoelastic damping, fluid damping, and other sources of
damping 共for example, anchor damping兲, respectively. For IT = y dy dz
MEMS beam resonators, typically Q−1 fluid is the most domi- A
nant term in Eq. 共11兲 under normal atmospheric conditions
冕
NT
and determines the overall quality factor Q. At low atmo-
spheric pressures, the value of Q−1fluid becomes negligible and
= 兺
N=1 A
y N共x,y兲eiNtdy dz
−1
QTED 共typically present at all pressures兲 becomes dominant NT
and hence limits the overall quality factor Q of the system.8 ⌬ EI 2UN共x兲 iNt
The effect of the electrostatic actuation force on the quality = 兺
N=1 ␣
关1 + f共N兲兴
x2
e , 共16兲
factor due to thermoelastic damping QTED is studied in this
paper by using two approaches: 共i兲 an improved form of the where I = wb3 / 12 is the mechanical moment of inertia of the
classical theory 共the Zener and Lifshitz and Roukes theories beam 共of width w and thickness b兲 and f共N兲 is given by
of TED兲 and 共ii兲 a coupled physical level simulation.
A. Modified theory
f共N兲 =
24
b3N3
冋 bN
2
− tan
bN
2
冉 冊册 . 共17兲
The theory of thermoelastic damping presented in Sec. II Substituting the expression for the displacement U共x , t兲 from
is modified to predict the value of QTED in electrostatic Eq. 共13兲 and the expression for IT 关Eq. 共16兲兴 into Eq. 共12兲
MEMS under complex nonlinear oscillations. Under electro- gives
static actuation, in the presence of fluid and thermoelastic
damping, the equation of motion for a MEMS beam is given NT NT
by − A 2
兺 N UN共x兲e2 iNt
+ ic 兺 NUN共x兲eiNt
冉 冊
N=0 N=0
2U U 2 2U ⑀wV2
A 2 + c + 2 EI 2 + E␣IT = Fe = NT N
t t x x 2共g − U兲2 4UN共x兲 iNt T
+ EI 兺 e + EI⌬ E 兺 关1
共12兲 N=0 x4 N=1
where c is the fluid damping coefficient, ⑀ is the dielectric 4UN共x兲 iNt ⑀w共Vdc + Vaceit兲2
冉 冊
+ f共N兲兴 e = 2.
constant of the surrounding medium, V is the applied volt- x4 NT
age, g is the gap in the undeformed state between the beam 2 g− 兺 UN共x兲eiNt
and the ground electrode, Fe is the electrostatic force per unit N=0
length, and w is the width of the beam. A voltage of the form 共18兲
V = Vdc + Vaceit is considered, where the real and the imagi-
nary parts of eit correspond to a cosinusoidal and sinusoidal From Eqs. 共12兲 and 共18兲 the bending moment17 of the
ac excitation, respectively. In the modified theory, we first MEMS beam, M, is given by
144305-3
SUDIPTO K. DE AND N. R. ALURU PHYSICAL REVIEW B 74, 144305 共2006兲
2U
M = EI + E␣IT
x2
NT
2UN共x兲 iNt
= EI 兺 e
N=0 x2
NT
2UN共x兲 iNt
+ EI⌬E 兺 关1 + f共N兲兴 e
N=1 x2
= M + MT
M
共19兲
FIG. 1. Illustration of thermal-electrical-fluidic-mechanical cou-
where M M and M T are the mechanical and the thermal con- pling in an electrostatic MEMS through an example—a fixed-fixed
tributions to the bending moment, respectively. The third and MEMS beam over a ground plane: the deformed structure with the
fourth terms on the left-hand side of Eq. 共18兲 correspond to various forces acting on it when it is moving downward.
2M M / x2 and 2M T / x2, respectively. The stress in the
冋兺 冉 册
MEMS beam is given by17
冊
NT
6 6 sinh N + sin N
My M y M y M T −1
QTED = ⌬E − N¯N2 共¯max
2
兲−1
= = + = M + T 共20兲 N=1 N2 N3 cosh N + cos N
I I I
共26兲
where M and T are the mechanical and the thermal
stresses, respectively, with T significantly smaller than M where ¯N is the magnitude of N 共N can be complex兲, ¯max
共see Ref. 1兲. The strain in the MEMS beam is given by17 is the maximum value of the expression 兺N=0
NT
NeiNt evalu-
ated over a time period 2 / , and
2U M
=y
x2
=
E
. 共21兲
N = b 冑 N
2
. 共27兲
Using the expressions for the stress and strain in the MEMS
beam, the elastic potential energy stored in the system, E pot, When only one harmonic 共NT = 1兲 is considered in Eq. 共26兲,
is given by20 i.e., for simple-harmonic motions of the beam, ¯max = ¯1, and
the expression for QTED given by the modified theory 关Eq.
E pot = 冕 冕
V
dV
0
d ⬇ 冕 冕 V
dV
0
M d, 共22兲
共26兲兴 reduces to the expression for QTED given by the
Lifshitz and Roukes theory 关Eq. 共9兲兴.
冕 冖
for the dynamic analysis of electrostatic MEMS in the pres-
ence of fluid and thermoelastic damping. The coupling be-
⌬ETED = dV Td 共23兲
V
tween the electrical, fluidic, and mechanical domains in elec-
trostatic MEMS, in the absence of thermoelastic damping, is
where 养 signifies integration over a complete time period of described in detail in Refs. 15, 18, and 22. In the case of
vibration. The quality factor due to thermoelastic damping, thermoelastic damping, an additional damping force is
QTED, of the MEMS beam can then be computed as2 present because of the temperature gradient in the micro-
structure due to its deformation 共see Fig. 1兲.
1 ⌬ETED Figure 1 shows a typical MEM device—a deformable
−1
QTED = 共24兲
2 共E pot兲max fixed-fixed MEMS beam over a fixed ground plane. A poten-
tial difference V is applied between the two conductors. The
where 共E pot兲max is the maximum elastic potential energy applied voltage gives rise to an electrostatic force which de-
stored in the MEMS beam during one complete time period forms the beam. When the beam deforms, the charge redis-
of vibration. Expressing UN共x兲 as21 tributes on the surface of the conductors, and, consequently,
NT the resultant electrostatic force and the deformation of the
i.e., U共x,t兲 = 共x兲 兺 NeiNt , 共25兲
beam also change. The surrounding fluid displaced by the
UN共x兲 = N共x兲,
N=0
motion of the beam exerts a fluid damping force on the
beam. The deformation of the beam also causes stress inho-
where 共x兲 is the most dominant mode of vibration of the mogeneities within the beam which gives rise to temperature
beam at the given frequency of excitation and N denotes the fluctuations 共the side in compression gets warmer and the
vibration amplitude, a closed-form expression for the quality side in tension gets cooler兲 and hence a thermoelastic damp-
factor QTED for the MEMS beam can be obtained as 共see the ing force. A self-consistent final state is reached for a given
Appendix for details兲 time step where the sum of all the forces 共the inertial force,
144305-4
THEORY OF THERMOELASTIC DAMPING IN… PHYSICAL REVIEW B 74, 144305 共2006兲
the mechanical stiffness force, the fluid and thermoelastic ductivity of the material. e is the sum of the diagonal com-
damping forces, and the electrostatic force兲 is zero. The pro- ponents of E and the overdots in Eq. 共35兲 indicate the de-
cess is repeated for each time step. rivatives with respect to time. A Newmark scheme with an
A two-dimensional 共2D兲 large-deformation thermoelastic implicit trapezoidal rule 共see, for example, Ref. 26 for de-
analysis is performed for computing the mechanical defor- tails兲 is used to solve the dynamical system posed in Eqs.
mation and the temperature profile in the microstructure on 共28兲–共35兲. Numerical discretization is done by using the
which the external electrostatic and fluidic forces act. The finite-cloud method 共FCM兲 共see Refs. 27 and 28 for details
transient governing equations for the mechanical deforma- on the FCM兲.
tion using a Lagrangian description are given by23,24 The 2D governing equation for the electrostatic analysis
can be written in a boundary integral form in the Lagrangian
ü = · 共FS兲 in ⍀, 共28兲 frame 共see Ref. 29 for details兲:
兩u兩t=0 = G0 in ⍀, 共31兲
冕
d⍀
„q共Q兲…J共Q兲d⌫Q = CT , 共37兲
兩u̇兩t=0 = V0 in ⍀, 共32兲
where ⍀ is the mechanical domain, ⌫g is the portion of the J共Q兲 = 关T共Q兲 · C共Q兲T共Q兲兴1/2 , 共38兲
boundary on which Dirichlet boundary conditions 共e.g., dis- where ⑀ is the dielectric constant of the medium, is the
placement兲 are specified, and ⌫h is the portion of the bound- electrostatic potential, and is the electrostatic surface
ary on which Neumann boundary conditions 共electrostatic charge density. P and Q are the source and field points in the
and fluidic pressures兲 are specified. is the material density initial configuration corresponding to the source and field
in the undeformed 共initial兲 configuration, and u, u̇, and ü are points p and q in the deformed configuration, and G is the
the displacement, velocity, and acceleration vectors, respec- Green’s function. In two dimensions, G(p共P兲 , q共Q兲) =
tively. F = I + u is the deformation gradient, and S is the −ln兩p共P兲 − q共Q兲兩 / 2, where 兩p共P兲 − q共Q兲兩 is the distance be-
second Piola-Kirchhoff stress given by24
tween the source point p共P兲 and the field point q共Q兲. CT is
E␣ the total charge of the system and C is an unknown variable
S = CE − I = S M + ST 共33兲 which can be used to compute the potential at infinity. T共Q兲
1−
is the tangential unit vector at field point Q and C共Q兲 is the
where S M and ST are the mechanical and thermal compo- Green deformation tensor. Equations 共36兲–共38兲 are solved
nents of S, C is the material tensor, E = 共FTF − I兲 / 2 is the using the boundary-cloud method 共BCM兲 共see Refs. 30 and
Green-Lagrangian strain, and is the Poisson ratio 共I is the 31 for details on the BCM兲 to obtain the distribution of sur-
identity matrix兲. = T − T0 is the change in the temperature face charge density on the conductors. The electrostatic
from the ambient temperature T0. Equations 共29兲 and 共30兲 are pressure normal to the surface of the conductors is given by
the displacement and the surface traction boundary condi-
2
tions, respectively. G is the prescribed displacement, and N Pe = . 共39兲
is the unit outward normal vector in the initial configuration. 2⑀
H is the surface traction vector given by Reynold’s squeeze film equation 共RSFE兲 is used to compute
H = 共Pe − P f 兲JF−TN 共34兲 the fluid or air damping pressure acting on the MEM struc-
ture. The RSFE is applicable for structures where a small gap
where Pe and P f are the surface electrostatic and fluidic pres- between two plates opens and closes with respect to time,32
sures, respectively, and J is the determinant of F. P = FS is like, for example, the MEMS fixed-fixed beam resonator
the first Piola-Kirchhoff stress tensor. Equations 共31兲 and over a ground plane shown in Fig. 1. A Lagrangian form of
共32兲 are the initial conditions for displacement and velocity, the isothermal Reynold squeeze film equation for compress-
respectively. G0 and V0 are the initial displacement and ve- ible slip flow is given by18,33
locity, respectively. The temperature profile T in the micro-
structure is obtained by solving a Lagrangian form of the 共P f h兲
F−T · 关共1 + 6K兲h3 P f F−T P f 兴 = 12 , 共40兲
heat equation,24,25 t
E␣ where h is the gap between the movable structure 共for ex-
·h= Tė + C pṪ in ⍀ 共35兲 ample the fixed-fixed beam in Fig. 1兲 and the ground plane
1−
of the MEM device 共the same as the fluid film thickness兲, P f
where is the material density in the undeformed 共initial兲 is the fluid pressure under the structure, and is the viscosity
configuration, h = JF−1q is the referential heat flux vector,25 of the surrounding fluid. K = / h is the Knudsen number,
and q is the heat flux in the deformed configuration, which where is the mean free path of the surrounding fluid. Nu-
can be computed as q = kF−T T, where k is the thermal con- merical discretization of the RSFE is done by using the
144305-5
SUDIPTO K. DE AND N. R. ALURU PHYSICAL REVIEW B 74, 144305 共2006兲
冕 冉兺 冊
2
1 M
E pot = w s eij dA 共41兲
A i,j=1 2 ij
冕 冖冉兺 冊
ratio ␦ for five cantilever beams 共Ref. 9兲 in vacuum. f d = 1 / 共2z兲 is
2
the characteristic damping frequency.
⌬ETED = w sTijdeij dA 共42兲
A i,j=1 small excitation force was applied to prevent any nonlinear
where 养 signifies integration over a complete time period of oscillations in the system.9 The damping ratio is independent
vibration and sTij are the components of the thermal stress ST. of the excitation force for very small magnitudes of the ex-
Equations 共24兲, 共41兲, and 共42兲 are used to compute the qual- citation force in the linear regime. Figure 2 shows the varia-
ity factor due to thermoelastic damping in the MEMS de- tion in the damping ratio of the resonators with the resonant
vices in Sec. IV. frequency f. Comparisons between the experimentally mea-
sured data, the numerical simulations, and the modified
IV. RESULTS
theory are shown in Fig. 2. The resonant frequency of each
beam 共f兲 is normalized with respect to the characteristic
Two different sources of nonlinearity in the electrostatic damping frequency f d = 1 / 共2z兲 of the beam. A dc bias of
force per unit length, Fe, can be identified from the beam 0.1 V along with an ac bias of 0.01 V is used in the simula-
equation for electrostatic MEMS 关Eq. 共12兲兴 and they are 共i兲 tions and the modified theory to generate an electrostatic
Fe ⬀ V2 and 共ii兲 Fe ⬀ 1 / 共g − U兲2. These nonlinearities in Fe can actuation force in the linear regime.
give rise to complex nonlinear vibrations in the MEMS beam Figure 3 shows the comparison between the measured and
共discussed later and also presented in Refs. 14, 18, and 19兲. computed 共both the numerical simulations and the modified
The effect of these nonlinear vibrations on QTED is studied in theory兲 quality factor QTED for a set of silicon fixed-fixed13
this section using the modified theory and the physical level beams in vacuum at T0 = 300 K. The length of the fixed-fixed
analysis, for MEMS fixed-fixed beams of different geom- beam was changed to obtain different resonant frequencies
etries. The results obtained from the classical theory of ther- and the thickness was kept constant at 5 m. The five ex-
moelastic damping are also discussed. perimental data points in Fig. 3 were obtained from
5-m-thick fixed-fixed beams of lengths 700 m 共at its first
A. Model validation and third resonant frequencies, 80.3 and 490 kHz, respec-
tively兲, 500 m 共at its third resonant frequency 720 kHz兲,
The physical level simulation tool is used to simulate a set
and 200 m 共at its first and third resonant frequencies, 911
of MEMS cantilever beams9 to determine the thermoelastic
damping ratio ␦ = 1 / 共2QTED兲 at the resonant frequency for TABLE I. Cantilever beam dimensions 共from Ref. 9兲.
T0 = 300 K. The dimensions of the five different cantilever
beams 共corresponding to the five experimental data points in Length Thickness Width z
Fig. 2兲 and their relaxation times 共z兲 computed using Eq. 共2兲 Beam No. 共m兲 共m兲 共m兲 共s兲
are shown in Table I.
The beams are made of silicon and have the following Beam 1 395 10 54 0.1136
materials properties:9 ␣ = 2.6⫻ 10−6 K−1, E = 170 GPa, Beam 2 192 10 54 0.1136
= 0.3, = 2330 kg m−3, C p = 712 J kg−1 K−1, k = C p Beam 3 183 15 27 0.255
= 148 W m−1 K−1, where k is the thermal conductivity. Beam 4 204 17.5 27 0.3478
The resonators 共cantilever beams兲 were placed in a Beam 5 190 17.5 54 0.3478
vacuum chamber which eliminates gas damping and a very
144305-6
THEORY OF THERMOELASTIC DAMPING IN… PHYSICAL REVIEW B 74, 144305 共2006兲
and 5.05 MHz, respectively兲. The beams have the following the applied voltage are dominant. The effect of the different
material properties:13 ␣ = 2.6⫻ 10−6 K−1, E = 130 GPa, modes of operation 共ac, dc, and mixed mode兲 on thermoelas-
= 0.28, = 2330 kg m−3, C p = 712 J kg−1 K−1, k = C p tic damping in MEMS is studied for a fixed-fixed MEMS
= 148 W m−1 K−1. A dc bias of 0.1 V along with an ac bias of beam 共denoted as beam A兲 at T0 = 300 K. The fixed-fixed
0.01 V is used in the numerical simulations and the modified MEMS beam 共beam A兲 considered is made from silicon and
theory to generate an electrostatic actuation force in the lin- has dimensions 200⫻ 5 ⫻ 10 m3 共length⫻ thickness
ear regime. The numerically simulated values of QTED and ⫻ width兲, placed 1 m over a ground plane. The material
those obtained from the modified theory are close to the properties used are the same as those used for the cantilever
experimental data for both the cantilever beams and the beams in Sec. IV A and the beam’s resonant frequency is
fixed-fixed beams as shown in Figs. 2 and 3. f 0 = 1.08 MHz. The relaxation time of the beam is z
= 0.0284 s 关computed using Eq. 共2兲兴 giving a characteristic
damping frequency f d = 1 / 共2z兲 共the frequency where TED
B. Effect of the Fe Ê V2 nonlinearity
is maximum, i.e., QTED is minimum兲 of 5.6 MHz. The beam
When the applied voltage V is small 共compared to the is simulated in air at 1 atm 共 = 1.82⫻ 10−5 kg/ ms and the
pull-in voltage of the MEMS beam兲, the displacement U mean free path = 0.064 m兲. Figure 4 shows the variation
Ⰶ g and the nonlinearity due to Fe ⬀ 1 / 共g − U兲2 is negligible in QTED with the excitation frequency f = / 共2兲 for the
关see Eq. 共12兲兴. However, the nonlinearity due to Fe ⬀ V2 can fixed-fixed beam A, under dc operations 共Vdc = 1 V and Vac
still be present at such small voltages and give rise to differ- = 0.0001 V, r → 0兲 obtained using the Zener theory, the
ent values of QTED than that predicted by the classical theory modified theory, and the numerical simulations.
of thermoelastic damping.1,3 Considering an applied voltage The fluid damping coefficient 共needed in the modified
of the form V = Vdc + Vac sin共t兲, where Vdc is the applied dc theory兲 c = 0.0205 N s / m2 is obtained from the linearized
bias and Vac sin共t兲 is the sinusoidal ac voltage, the electro- Reynold squeeze film theory34 for beam A. The value of
static force per unit length Fe will have both the first and the QTED obtained from the Zener theory, the modified theory,
second harmonics of the exciting angular frequency due to and the numerical simulation are very close as shown in Fig.
the V2 nature, i.e., 4 共the numerical simulation results are typically more accu-
冉 冊
rate due to its 2D analysis as compared to the theories, which
⑀w 2 Vac2 2
Vac are based on 1D analysis of the system13兲. Under dc opera-
Fe ⬇ V + + 2V V
dc ac sin共 t兲 − cos共2t兲 .
2g2 dc 2 2 tions at small voltages where the nonlinearity due to Fe
共43兲 ⬀ 1 / 共g − U兲2 is negligible, the motion of the beam is simple
harmonic at the excitation frequency f. In this case, the Ze-
Defining r = Vac / 共Vdc + Vac兲, r = 1 共when Vdc = 0兲 implies ac ner theory and the Lifshitz and Roukes theory 共compared
operation 共when the second harmonic of the exciting angular with the Zener theory in the inset of Fig. 4兲 match the modi-
frequency is dominant兲 and r → 0 共when Vac Ⰶ Vdc兲 implies fied theory and are very close to the numerical simulations.
dc operation 共when the first harmonic is dominant兲. For in- The pull-in voltage of the fixed-fixed beam is found to be
termediate values of r between 0 and 1 共mixed-mode opera- 120 V, thereby indicating that the applied voltages are very
tion兲 both the first and the second harmonic components of small. The inset of Fig. 4 shows some minor variations be-
144305-7
SUDIPTO K. DE AND N. R. ALURU PHYSICAL REVIEW B 74, 144305 共2006兲
FIG. 5. 共Color online兲 Variation of QTED with the excitation FIG. 6. 共Color online兲 Variation in QTED with the excitation
frequency f for the MEMS fixed-fixed beam A under ac operations frequency f for the MEMS fixed-fixed beam A 共resonant frequency
共r = 1兲 obtained from the different theories and numerical simula- is f 0 = 1.08 MHz兲 under mixed-mode operation at r = 0.5 共Vdc
tion. Characteristic damping frequency is f d = 2.8 MHz, where QTED = 0.5 V, Vac = 0.5 V兲, obtained from the different theories and nu-
is minimum. merical simulation.
tween the Zener theory and the Lifshitz and Roukes theory at modified theory and the numerical simulations indicate the
higher frequencies which is consistent with the observations formation of a spike in the downward direction in QTED
made in Ref. 3. Note that the modified theory is identical to around f 0 / 2 共f 0 = 1.08 MHz is the resonant frequency of the
the Lifshitz and Roukes theory when only one harmonic beam兲, no such predictions are made by the Zener or the
共NT = 1兲 is considered in Eq. 共26兲, i.e., for simple-harmonic Lifshitz and Roukes theory. The formation of the spike in
motions of the beam. QTED around f 0 / 2 can be explained from the variation in the
For ac operations 共Vdc = 0 V and Vac = 1 V, r = 1兲, the value magnitudes of 1 and 2 关Eq. 共25兲兴 with the excitation fre-
of QTED obtained from the Zener and the Lifshitz and Roukes quency f for the fixed-fixed beam at Vdc = 0.5 V and Vac
theories evaluated at = 2 f does not match the predictions = 0.5 V 共r = 0.5兲. While the magnitude of the first harmonic
of the modified theory or the numerical simulations as shown 1 peaks at the resonant frequency of the system as expected,
in Fig. 5. However, the classical theories evaluated at 2 the second harmonic 2 becomes dominant 共larger than 1兲
= 2共2f兲 agree well with the modified theory 共and the results at f 0 / 2 as shown in Fig. 7. Since the quality factor QTED
are close to those of the numerical simulations兲 as the oscil- depends on the relative strengths of the different harmonics
lations are still simple harmonic but at twice the excitation present 关see Eq. 共26兲兴, maximum deviation between the clas-
frequency due to the V2 nature of Fe under ac operations sical theories 共based on the first harmonic兲 and the modified
from Eq. 共43兲. In this case, the characteristic damping fre- theory and numerical simulations is observed at f 0 / 2 where
quency is found to be half of that in the dc operation, i.e., the second harmonic is most dominant.
f d = 2.8 MHz as shown in Fig. 5. While minor variations can Similar observations were also made for another MEMS
be observed between the Zener model evaluated at 2 fixed-fixed beam 共beam B兲 of different geometry. The fixed-
= 2共2f兲 and the modified theory at higher frequencies in fixed beam B is made from silicon 共same material properties
Fig. 5, the value of QTED given by Lifshitz and Roukes as beam A兲 and has dimensions 80⫻ 2 ⫻ 10 m3 共length
theory evaluated at 2 = 2共2f兲 is found to be exactly the ⫻ thickness⫻ width兲, placed 0.5 m over a ground plane
same as the modified theory at all the frequencies. For and is simulated under similar ambient conditions as beam A.
mixed-mode operations 共intermediate values of r between 0 The relaxation time is z = 0.0045 s and the characteris-
and 1兲, the electrostatic force per unit length Fe contains both tic damping frequency is 35 MHz. The pull-in voltage of
the first and the second harmonic components of the exciting beam B is found to be 69 V dc and the resonant frequency is
frequency 关see Eq. 共43兲兴. As a result, the oscillations also f 0 = 2.72 MHz. The fluid damping coefficient for the beam B
have both these two frequency components in them, giving is c = 0.062 N s / m2 共used in the modified theory兲. Figure 8
rise to nonsimple harmonic oscillations. The value of the shows the variation in QTED with the excitation frequency f
quality factor due to thermoelastic damping, QTED, for such for the fixed-fixed beam B under ac operations 共r = 1兲 ob-
cases cannot be predicted correctly by the classical theories tained from the different theories and numerical simulation.
as shown in Fig. 6. Figure 9 shows the variation in QTED with the excitation
Figure 6 shows the variation in QTED with the excitation frequency f under mixed-mode operation at r = 0.5 共Vdc
frequency f for a mixed-mode operation at Vdc = 0.5 V and = 0.5 V, Vac = 0.5 V兲, obtained from the different theories and
Vac = 0.5 V, r = 0.5 in the fixed-fixed beam A. While the numerical simulation. Zener’s theory shows similar devia-
144305-8
THEORY OF THERMOELASTIC DAMPING IN… PHYSICAL REVIEW B 74, 144305 共2006兲
144305-9
SUDIPTO K. DE AND N. R. ALURU PHYSICAL REVIEW B 74, 144305 共2006兲
V. CONCLUSIONS
The formation of the spikes in Fig. 10 can be explained by
the presence of higher-order harmonics in the oscillations The nonlinear nature of the electrostatic actuation force is
introduced by the Fe ⬀ 1 / 共g − U兲2 nonlinearity at higher volt- found to change the nature of thermoelastic damping in elec-
ages 共close to pull in兲. Harmonic balance analysis of the trostatic MEMS significantly from that predicted by the clas-
beam equation 关Eq. 共12兲兴 shows the presence of several sical theory of thermoelastic damping developed by Zener
higher-order harmonics in the oscillations of beam A at and later improved by Lifshitz and Roukes. The nonlinearity
100 V dc and 5 V ac as shown in Fig. 11. The magnitude of due to the V2 nature of the electrostatic force is found to
the Mth harmonic  M 关Eq. 共25兲兴 is found to peak or spike at affect the thermoelastic quality factor QTED even at small
the first M superharmonic frequencies of f 0, i.e., at f = f 0 / N voltages 共far from pull in兲. At larger voltages 共closer to pull
for N = 1 , 2 , 3 , . . . , M. This in turn affects the quality factor in兲, the nonlinearity due to the 1 / 共g − U兲2 nature of the elec-
QTED at a given frequency as shown in Fig. 10 as the quality trostatic force is found to affect QTED. A modified theory is
factor QTED at a given frequency depends on the relative proposed for predicting thermoelastic damping in MEMS un-
strengths of the different harmonics N present 关from Eq. der arbitrary electrostatic actuation. The modified theory
共26兲兴.
The formation of spikes in the thermoelastic quality factor
QTED versus normalized excitation frequency f / f 0 plot at
larger voltages is also observed in beam B. Figure 12 shows
the variation in QTED with the normalized frequency f / f 0 for
the fixed-fixed beam B at 60 V dc 共close to the pull-in volt-
age of 69 V dc兲 and 5 V ac. The resonant frequency of the
beam B at 60 V dc is f 0 = 1.785 MHz. Figure 13 shows the
presence of several higher-order harmonics in the oscilla-
tions of the MEMS beam B at 60 V dc and 5 V ac resulting
in formation of the spikes in QTED.
The strength of the harmonics N decreases very rapidly
with increasing N 共see Ref. 18兲 and the quality factor QTED
computed using the modified theory is found to converge by
considering a finite number of harmonics. For example, the
number of harmonics used in the modified theory for both
beams A and B in this paper is NT = 10. Figures 10 and 12
show that the quality factor of the MEMS beams is affected
by the application of higher dc voltages not only at the su- FIG. 13. 共Color online兲 Amplitude variations of the first three
perharmonic frequencies but also at their resonant frequen- harmonics, 1, 2, and 3, with the normalized frequency f / f 0 for
cies, i.e., at f / f 0 = 1, which is the most commonly used exci- the fixed-fixed beam B at 60 V dc 共close to the pull-in voltage兲 and
tation frequency for microstructures. 5 V ac.
144305-10
THEORY OF THERMOELASTIC DAMPING IN… PHYSICAL REVIEW B 74, 144305 共2006兲
the quality factor. Although the quality factor QTED predicted 共A6兲
by the modified theory is close to the numerical simulation
Equating the coefficients of e0 to zero in Eq. 共A4兲 gives
predictions, typically the physical level or numerical simula-
tions based on the coupled 2D nonlinear electrical-thermal- ⑀wLVdc
2
mechanical-fluidic analysis are more accurate as they involve B̄R030 − 2gC̄R020 + Āg2R00 − = 0, 共A7兲
2
fewer assumptions compared to the modified theory.
from which a closed form expression for 0 can be obtained.
ACKNOWLEDGMENTS Equating the coefficients of eiNt for N = 1 , 2 , . . . , NT to zero
in Eq. 共A4兲 gives
This work is supported by the National Science Founda-
tion under Grants No. 0121616, No. 0217986, No. 0228390, ⑀wLVdcVac
and No. 0601479. 1 = ,
Āg R1 +2
B̄R120 + 2B̄R020 − 2C̄gR00 − 2C̄gR10
APPENDIX: DERIVATION OF QTED 共A8兲
After substituting Eq. 共25兲 into Eq. 共18兲, a harmonic bal-
ance analysis21 is performed to compute N for N ⑀wLVac
2
/2 − B̄R0021 − 2B̄R1021 + 2C̄gR121
2 = ,
= 0 , 1 , 2 , . . . , NT. The thermal contribution to the bending mo- Āg2R2 + B̄R220 + 2B̄R020 − 2C̄gR00 − 2C̄gR20
ment M T is negligible compared to the mechanical contribu-
tion to the bending moment M M .1 As a result, the vibration 共A9兲
冋 冉
amplitudes N of the different harmonics can be computed
Q−1
by neglecting the term 2M T / x2 in Eq. 共18兲 and by replacing
UN共x兲 by N共x兲 to obtain Q = − B̄ 兺 RNNQ/2−N/2
2
− 2B̄ R00¯⌫共Q兲
冉 冊 册
N=0
冊
NT NT Q−1 Q−1
− A2 兺 N2NeiNt + ic 兺 NNeiNt 共x兲 + 2C̄g 兺 RNNQ−N
N=0 N=0
+ 兺 RNN⌫共Q − N兲
N=1 N=1
N
4共x兲 T ⑀w共Vdc + Vaceit兲2 ⫻ 关Āg2RQ + B̄RQ20 + 2B̄R020 − 2C̄g共R00
兺 NeiNt =
冉 冊
+ EI 2.
x4 N=0 NT
2 g − 共x兲 兺 NeiNt + RQ0兲兴−1 for Q = 3, . . . ,NT 共A10兲
N=0
where ⌫共I兲 is the summation of the combination of all pos-
共A1兲 sible pairs JK for J + K = I and J ⬍ K and ¯⌫共I兲 is exactly the
The mode shape of vibration 共x兲 for a beam is given by26 same as ⌫共I兲 but does not include the term 0I in the sum-
mation.  P = 0 if P is a fraction.
共x兲 = cosh共qx兲 − cos共qx兲 + p关sinh共qx兲 − sin共qx兲兴 The maximum potential energy of the MEMS beam,
共A2兲 共E pot兲max, during one time period of the vibration, can be
computed using Eqs. 共21兲 and 共22兲 as
where p and q depend on the beam type 共cantilever or fixed-
fixed兲 and on the order of the mode and resonant frequency.
From Eq. 共A2兲 共E pot兲max = 冕 冕
V
dV
0
max
M d
4共x兲
x4
= q4共x兲. 共A3兲 = 冕 冕
V
dV
0
max
E d
冕 冕
Using Eq. 共A3兲, rearranging the terms, and integrating Eq. t/2 L
Ew
共A1兲 along the length of the beam gives = max
2
dx dy 共A11兲
冉兺 冊冋 冉 兺 冊
2 −t/2 0
NT NT 2
RNNeiNt Āg2 + B̄ NeiNt where w and L are the beam width and length, respectively,
and max is given by
册
N=0 N=0
冏 冏 冏兺 冏
NT
⑀wL共Vdc + Vaceit兲2 2U共x,t兲 2共x兲
NT
− 2C̄g 兺 NeiNt = 共A4兲 max = y =y NeiNt
N=0 2 x2 max x2 N=0 max
where L is the length of the beam, 2共x兲 ¯
=y max 共A12兲
RN = − A2N2 + icN + EIq4 , 共A5兲 x2
144305-11
SUDIPTO K. DE AND N. R. ALURU PHYSICAL REVIEW B 74, 144305 共2006兲
兺N=0
NT
NeiNt evaluated over a time period 2 / . Integrating
over the volume of the MEMS beam, V, the expression for
the peak potential energy becomes
d = d y 冉 2U共x,t兲
x2
冊
NT
2UN共x兲 iNt
冋冕 冉 冊 册
= i y 兺 Ne dt
EI¯max
2 L
2共x兲 2
N=0 x2
共E pot兲max = dx . 共A13兲
2 x2 N
0
2共x兲 T
= i y 兺 NNeiNtdt.
x2 N=0
共A16兲
The energy dissipated per period by thermoelastic damping
can be computed using Eq. 共23兲: From Eqs. 共A14兲–共A16兲, the energy dissipated per cycle by
thermoelastic damping, ⌬ETED, can be obtained as
⌬ETED = 冕 冖
V
dV Td = w 冕 冕冖t/2
−t/2
L
0
Td dx dy
⌬ETED = EI⌬E 冋冕 冉 冊 册 兺
L
2共x兲 2
dx
NT
0 x2 N=1
共A14兲
冑
T =
I N
N = b , 共A18兲
NT
2UN共x兲 iNt 2
= yE⌬E 兺 关1 + f共N兲兴 e
N=1 x2 which gives the expression for the quality factor QTED for the
MEMS beam as
冋兺 冉 册
N
冊
2共x兲 T
= yE⌬E 兺 N关1 + f共N兲兴eiNt .
x2 N=1
共A15兲 −1
QTED =
NT
⌬E
6
−
6 sinh N + sin N
N¯N2 关¯max
2
兴−1 .
N=1 N2 N3 cosh N + cos N
In Eq. 共A14兲, d can be written as 共A19兲
attle, WA, 1995 共IEEE, New York, 1995兲, p. 489. Materials 共Academic Press, New York, 1972兲.
8
A. Duwel, J. Gorman, M. Weinstein, J. Borenstein, and P. Ward, 21
A. T. Nayfeh and D. T. Mook, Nonlinear Oscillations 共John-
Sens. Actuators, A 103, 70 共2003兲. Wiley and Sons, New York, 1979兲.
9 22
T. V. Roszhart, in Proceedings of the Solid-State Sensor and Ac- S. K. De and N. R. Aluru, J. Micromech. Microeng. 16, 1705
tuator Workshop, Hilton Head Island, SC, 1990 共IEEE, New 共2006兲.
23
York, 1990兲, p. 13. D. S. Chandrasekharaiah and L. Debnath, Continuum Mechanics
10 K. Y. Yasumura, T. D. Stowe, E. M. Chow, T. Pfafman, T. W.
共Academic Press, New York, 1994兲.
Kenny, B. C. Stipe, and D. Rugar, J. Microelectromech. Syst. 9, 24
H. Parkus, Thermoelasticity 共Blaisdell Publishing Company, New
117 共2000兲. York, 1968兲.
11
V. T. Srikar and S. D. Senturia, J. Microelectromech. Syst. 11, 25
P. Yarrington and D. E. Carlson, Int. J. Eng. Sci. 14, 113 共1976兲.
499 共2002兲. 26 K. J. Bathe, Finite Element Procedures 共Prentice-Hall, Engle-
12
R. Abdolvand, H. Johari, G. K. Ho, A. Erbil, and F. Ayazi, J. wood Cliffs, NJ, 1995兲.
Microelectromech. Syst. 15, 471 共2006兲. 27 N. R. Aluru and G. Li, Int. J. Numer. Methods Eng. 50, 2373
13
S. Pourkamali, A. Hashimura, R. Abdolvand, G. K. Ho, A. Erbil, 共2001兲.
and F. Ayazi, J. Microelectromech. Syst. 12, 487 共2003兲. 28 X. Jin, G. Li, and N. R. Aluru, Comput. Struct. 83, 1366 共2005兲.
14
S. K. De and N. R. Aluru, Phys. Rev. Lett. 94, 204101 共2005兲. 29
G. Li and N. R. Aluru, J. Microelectromech. Syst. 11, 245 共2002兲.
15 S. K. De and N. R. Aluru, J. Microelectromech. Syst. 13, 737 30 G. Li and N. R. Aluru, Comput. Methods Appl. Mech. Eng. 191,
144305-12
THEORY OF THERMOELASTIC DAMPING IN… PHYSICAL REVIEW B 74, 144305 共2006兲
144305-13