Journal of Micromechanics and Microengineering

PAPER You may also like

- Low temperature capacitive
Modal analysis based equivalent circuit model and micromachined ultrasonic transducers
(cMUTs) on glass substrate
its verification for a single cMUT cell E Bahette, J F Michaud, D Certon et al.

- Evaluation of bias voltage modulation

sequence for nonlinear contrast agent
To cite this article: S P Mao et al 2017 J. Micromech. Microeng. 27 035001 imaging using a capacitive micromachined
ultrasonic transducer array
Anthony Novell, Mathieu Legros, Jean-
Marc Grégoire et al.

- Singulation for imaging ring arrays of

View the article online for updates and enhancements. capacitive micromachined ultrasonic
transducers
Chienliu Chang, Azadeh Moini, Amin
Nikoozadeh et al.

This content was downloaded from IP address 14.139.59.45 on 11/02/2022 at 07:03

 Journal of Micromechanics and Microengineering

J. Micromech. Microeng. 27 (2017) 035001 (17pp) doi:10.1088/1361-6439/aa53ce

Modal analysis based equivalent circuit

model and its verification for a single
cMUT cell
S P Mao1,2, X Rottenberg1, V Rochus1, P Czarnecki1, P Helin1, S Severi1,
B Nauwelaers2 and H A C Tilmans1
1
  imec, Kapeldreef 75, 3001 Leuven, Belgium
2
  Department of Electrical Engineering, (ESAT-TELEMIC), KU Leuven, 3001 Leuven, Belgium

E-mail: [email protected]

Received 20 June 2016, revised 21 November 2016

Accepted for publication 14 December 2016
Published 19 January 2017

Abstract
This paper presents the lumped equivalent circuit model and its verification of both
transmission and reception properties of a single cell capacitive micromachined ultrasonic
transducer (cMUT), which is operating in a non-collapse small signal region. The derivation of
this equivalent circuit model is based on the modal analysis techniques, harmonic modes are
included by using the mode superposition method; and thus a wide frequency range response
of the cMUT cell can be simulated by our equivalent circuit model. The importance of the
cross modal coupling between different eigenmodes of a cMUT cell is discussed by us for the
first time. In this paper the development of this model is only illustrated by a single circular
cMUT cell under a uniform excitation. Extension of this model and corresponding results
under a more generalized excitation will be presented in our upcoming publication (Mao
et al 2016 Proc. IEEE Int. Ultrasonics Symp.). This model is verified by both finite element
method (FEM) simulation and experimental characterizations. Results predicted by our model
are in a good agreement with the FEM simulation results, and this works for a single cMUT
cell operated in either transmission or reception. Results obtained from the model also rather
match the experimental results of the cMUT cell. This equivalent circuit model provides an
easy and precise way to rapidly predict the behaviors of cMUT cells.

Keywords: cMUT, SiGe-MEMS, equivalent circuit representation, cross modal coupling,

reception

(Some figures may appear in colour only in the online journal)

1. Introduction cMUTs are mainly used as arrays or a group of cells; in order

to predict the behaviors of large arrays for either optimized
Capacitive micromachined ultrasonic transducers (cMUTs) designs or new applications, the key step is the modelling and
are recognized as key enablers for many promising applica- its corresponding verification of a single cMUT cell [5–12].
tions such as medical diagnosis, ultrasonic therapy, chemical In the past two decades, the modelling of cMUTs is
or bio-sensing, ultrasonic flow sensor, acoustic manipulation, already well reported. As the modelling of cMUTs is a com-
obstacle detection, and acoustic telecommunication [2–4]. plicated multi-physics problem, using finite element method
In the meantime, these emerging applications using cMUTs (FEM) available in the commercial FEM software package
become more discriminating and more demanding from the (e.g. ANSYS, COMSOL) is an obvious choice [13–16].
design point of view. As a result, an accurate, fast and con- Individual cMUT cells, or element containing multiple cMUT
vincing modelling technique to predict both of the transmission cells, have been studied by FEM and compared to exper­
and reception characteristics of cMUTs is essential. Though, imental results obtained from optical characterizations [15] or

1361-6439/17/035001+17$33.00 1 © 2017 IOP Publishing Ltd  Printed in the UK

J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

acoustic measurements [16]. Although FEM poses no limita-

tion to the shape of a structure, when modelling a large group
of cMUT cells with the medium load, the FEM model may
become gigantically large. Solving of such model is very time
consuming, and a particularly long computational time is pre-
dictable [10]. Therefore, FEM often is not advised to study
many cMUT cells in a large group.
An alternative way, which partially has the advantage of
FEM but requires reduced computation, is the mixed finite Figure 1.  The sectional view of the conceptual design and
element/boundary element approach [10, 17–19]. It is a prom- operating principle of a single cMUT cell.
ising approach for simulating arbitrary cMUTs in an array.
circuit models, the derivation of our model is no longer lim-
However, this methodology always requires the discrete rep-
ited to the fundamental mode. A multi-mode simulation will
resentation of the plate using large matrices, and thus it is not
benefit from such a consistent description of eigenmodes in
convenient to be implemented and the computational cost is
our model. Using the mode superposition method, the total
also questionable. Especially, when the geometry of cMUTs is
response of the cMUT cell is a linear superposition of the indi-
regular, a quick and easy approach is preferred.
vidual response of all the modes. It is convenient to include the
Lumped equivalent circuit model is another widely used
effect of higher order modes, and thus a wide frequency range
approach for the modelling of cMUTs. In this model, the
simulation of cMUTs is achievable. Eccardt el al already pre-
distributed system is analytically described by lumped param­
sented the simulation of a cMUT cell with two eigenmodes by
eters in a circuit, which is usually considered as the Mason
using the mode superposition method. However, in his model,
equivalent circuit [20]. Initial models are based on the small
fitting of the radiation impedance is still needed in order to
signal 1D linear approximation, in which cMUTs are only
match the FEM results [29]. Moreover, cross modal coupling
simply approximated as two parallel plates: one is fixed, and
between different eigenmodes of a cMUT cell is discussed
the other is movable; the radiation impedance of a piston is
here for the first time, which is found importantly in the wide
adopted to represent the effect of the medium load [5–8]. Later
frequency range simulation of the cMUT cell.
on, these models have been further developed and improved.
The remaining portion of this paper is organized as fol-
Wygant et al [9] included an analytical model considering the
lows. In section 2, the conceptual design, operating principle
effect of the curved static deflection of the membrane. The
and build-up technology of our SiGeMEMS-based cMUTs are
nonlinearities of cMUTs due to a large signal excitation are
presented. Subsequently, as an illustration of our approach,
already investigated in literatures [7, 21]. Model for cMUTs
the equivalent circuit model of a single circular cMUT cell
operating in the collapse region has also been developed [22].
under a uniform excitation is derived in section  3. Both the
Oğuz et  al [23] showed a model which can investigate the
transmission and reception characteristics of a cMUT cell are
dynamics of cMUTs with mutual coupling of cells in an ele-
obtained from the equivalent circuit model, which are veri-
ment. It presents the equivalent circuit model can be used to
fied by FEM simulation in section 4. Section 5 presents the
solve some valuable problems, e.g. when a group of cMUT
experimental validation of the characteristics of a single cir-
cells are driven. In short, these models are relatively easy to be
cular cMUT cell.
used and can be considered as good candidates for the output
prediction of cMUTs in different operation regimes. However,
the development of these models is always limited in a cir- 2.  cMUT conceptual design and build-up
cular shape with a fixed velocity profile for the diaphragm,
which does not have the nature of the multi-mode simulation. The sectional view of the conceptual design and the operating
In this work we present a simple but also powerful equiva- principle of a single cMUT cell are shown in figure 1 [2]. A
lent circuit model, which is derived from the modal analysis cMUT cell consists of a clamped thin membrane (top elec-
techniques [24–28]. This model works in the small signal trode), a thin vacuum gap and a fixed buried bottom electrode.
operation regime of cMUTs and is capable of simulating When the device works in transmission (Tx mode), superim-
response over a wide frequency range, i.e. not limited to the posed DC and AC voltages are applied. The membrane vibrates,
frequency region in vicinity of the fundamental mode. In our and an outgoing ultrasound wave will be generated in the
model, cMUTs are formulated as a lumped system, which is medium load. Vice versa, when the device works in reception
easier to be implemented than the abovementioned discrete (Rx mode); the membrane will receive an incoming ultrasound
approach. In the meantime, the multi-mode simulation capa- wave, and a current or voltage change can be sensed.
bility also makes the scope of our model broader than other
conventional lumped equivalent circuit models. 2.1. imec’s SiGe cMUTs
As a typical lumped model, the distributed system is
described as a few lumped parameters in our equivalent cir- Figure 2 presents some examples of cMUT devices manufac-
cuit model. Parameters of each eigenmode in our model are tured by imec’s SiGeMEMS-based cMUT technology [30].
expressed in a generalized form, which are only linked to the A detailed cross-sectional view of one cMUT cell is shown
modal information (e.g. mode shape, frequency parameter) of in figure 2(a). The membrane is a composite of three different
the corresponding mode. Compared to other lumped equivalent materials: a 1.5 µm thick silicon oxide sealing layer, a 3.6 µm

2
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Figure 2.  cMUT devices: (a) schematic of the cross-sectional view of a cMUT cell; (b) SEM micrograph of a local cross-section of
a cMUT cell, showing the region in the dashed box of figure 2(a); (c) top view micrograph of circular cMUT cells (from left to right,
diameters are 80 µm, 100 µm and 120 µm respectively).

Table 1.  Material properties and thickness of the cMUT membrane;

information of SiO2, SiGe and SiC is from imec’s technology
handbook.
Property SiO2 SiGe SiC Equivalent

Thickness (µm) 1.5 3.6 0.4 5.5

Poisson’s ratio (—) 0.17 0.22 0.18 0.21
Young’s modulus (GPa) 64 120 150 93
Density (kg m−3) 2200 4775 3200 3969

of CMOS technology. Figure 2(c) is the top view micrograph

Figure 3.  Cross-section and geometry of a simplified circular of circular cMUT cells with different membrane dimensions,
cMUT cell. A cylindrical coordinate system is used for the showing the bond pads and membranes.
membrane,  is the radial coordinate, ϕ is the azimuth, and z is the
height.
2.2.  Simplified cMUT model
thick SiGe structural layer and a 400 nm thick SiC dielectric The real build-up of a cMUT cell in figure  2 is very com-
layer. The diameter of the cell membrane D is determined by plicated. In this work, several assumptions are taken in order
the spacing of the two anchors. The transducer gap, defined to clarify the derivation of our equivalent circuit model. The
by the (top electrode) SiGe membrane layer and the buried cross-section and geometry of a simplified circular cMUT
SiGe bottom electrode, is composed of a 200 nm vacuum gap cell is shown in figure  3. The real laminated membrane is
surrounded by two 400 nm thick SiC dielectric layers. The DC considered as one equivalent homogenous membrane. This
pad is directly connected to the SiGe structural layer, the AC homogenous membrane is made of the equivalent material and
pad is isolated from the SiGe structural layer and connected ideally clamped at the boundary. Material properties as used
to the bottom electrode of the device through the buried Al in this paper are listed in table  1. The equivalent thickness
electrode. Figure  2(b) is a cross-section SEM micrograph, h and the equivalent Poisson’s ratio v are obtained from the
showing the specific region of the device in the dashed box definitions in [31]. The equivalent Young’s modulus E for the
of figure 2(a). It presents the sealing oxide, SiGe membrane, flexural bending of the membrane is calculated by the equiva-
releasing hole, thin vacuum gap, anchor, buried Al electrode, Eh3
Si substrate etc. Potentially, the same built-up can be mono- lent flexural rigidity D = 12(1 − v 2 )
, which is also defined in
lithically integrated with a CMOS circuit by the MEMS on top [31]. The equivalent density ρ is a weighted average of the

3
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Figure 4.  The general equivalent circuit model of one single cMUT cell (the asterisk indicates the modification of the compliance as a
result of electromechanical coupling).

density of the laminated membrane. Assume that both of the Table 2.  Symbols and corresponding physical meanings of the
top and bottom electrode have a full coverage over the static quantities used in the equivalent circuit model.
area S of the gap (S= πa2, a is the radius of the memrbane). Physical domain Effort variables Flow variables
The transducer gap is composed of a vacuum gap surrounded
by two SiC dielectric layers, an equivalent thickness of the Electrical Voltage: V Current: I
vacuum gap is given by: Mechanical Modal force: F Velocity amplitude: u
  Acoustical Modal pressure: P Volume velocity: U
hi
d 0 = g0 +
(1) i
  εr
. ∑
and the modal damping Ri. Each mode i has an individual
Here d 0 is the equivalent thickness of the vacuum gap, g0 is modal mechanical radiation impedance Z mri, representing the
the initial vacuum gap, hi is the thickness of the ith dielectric effect of medium load applied on the membrane. Symbols and
layer, and the εir is the relative permittivity of the ith dielectric corresponding physical meanings of the quantities used in the
layer. The relative permittivity of the SiC dielectric layer is equivalent circuit model are summarized in table 2.
4.4, and the equivalent thickness of the vacuum gap will be This general equivalent circuit model presents both modes
382 nm. of operation, the Tx mode and the Rx mode, of the cMUT
cell. When the cMUT cell works in the Tx mode, the elec-
trical port is connecting to an alternating voltage source Vac
3.  Equivalent circuit model and the acoustical port will be a short circuit if there is no
external pressure load. In the Rx mode, the acoustical port is
3.1.  General equivalent circuit model
the incoming pressure Pin while the electrical port will be a
Equivalent circuit models of small signal electromechanical short circuit if the load impedance is very small. This general
systems are already developed by modal analysis techniques model does not rely on the geometry of the device, which can
in these two comprehensive papers [25, 26]. The general deal with a cMUT cell of arbitrary shapes in the small signal
equivalent circuit model of a single cMUT cell is presented in operating regime. In the following sections, as an illustration
figure 4. This circuit should have an infinite number of reso- of our approach, the equivalent circuit model will be derived
nating branches, and each branch of the circuit represents one based on a single circular cMUT cell. Apart from the different
excited eigenmode of the thin membrane. The equivalent cir- modal information, approach of dealing with other shapes is
cuit model represents three physical domains of the cMUT similar.
cell, these domains are coupled by means of two transformers.
For a particular ith mode, the transduction factor Γi and Aei of 3.2.  Fundamentals for a circular cMUT cell
the transformer represent the electromechanical transduction
 
factor and the effective area of the ith mode, respectively. In the The cMUT is usually biased by a DC bias V dc . Hence the
electrical domain, all of these branches share the same static total deflection of the membrane is a linear superposition
capacitance C0. The series RLC circuit comes from an analogy of the static deflection w0() as a result of the DC bias and
of the spring-mass-damper system in mechanics, which rep- the dynamic transverse deflection  w(, ϕ, t ) resulting from a
resents the vibrating membrane in absence of loading. In the dynamic load (figure 3). An approximate solution of the static
vicinity of the ith mode, the motional characteristics can be deflection w0() for a uniformly loaded circular cMUT cell is
evaluated as the modal mass Mi, the modal compliance 1/Ki   given by [32]:
4
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

⎛ ⎡ ⎤⎛
2 ⎞ 2 ⎞2
0.5

w0() = wpk ⎜1 − 2 ⎟ ≅ ⎢ ⎥ ⎜1 −  ⎟
2
ε0d 0V 2dca4 Analytical calculation
⎝ a ⎠ ⎢ 128D d 3 − 32 ε0V 2 a4 ⎥ ⎝ a2 ⎠
⎣ 0 dc ⎦ 21
0.4 COMSOL simulation

⎡ ⎤
⎢ ( )
Vdc 2
⎥ ⎛ 2 ⎞
2 0.3

2 ⎥ 0⎜ ⎟
=⎢
VPI
d 1 −
⎢ 128/β 2 − 32/21 Vdc ⎥ ⎝ a2 ⎠
( )

wpk/d0
⎣ VPI ⎦ 0.2

(2)
where the pull-in voltage of the circular cMUT cell is 0.1

D d 30 0.0
VPI = β .
ε0a4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

β is a constant dictated by boundary conditions; for a clamped Vdc/VPI

boundary condition, β is equal to 5.40 [9]. ε0 is the vacuum Figure 5.  Peak amplitude of the static deflection as a function of
permit­tivity, wpk is peak amplitude of the static deflection the applied DC bias.
w0(). The peak amplitude wpk is analytically calculated by
equation (2), which is compared to a COMSOL simulation in Assume the external load does not modify the mode shapes
figure 5. This approximate solution is accurate enough to be used a lot and make use of the orthogonality of mode shapes, a
until the applied DC bias Vdc is above 90% of the pull-in voltage governing equation can be built with the in vacuo mode shape
 
VPI. Similar to the treatment for the static deflection used in the φmn (, ϕ ) [26]:
1D model [7, 29], an effective gap d eff will be defined here:
Kmnwmn(t ) + Rmnw˙mn(t ) + Mmnw¨mn(t ) = ∬ ftmn (, ϕ, t )
⎡ Vdc 2 ⎤
⎢ 128/β − 13/7 VPI ⎥
d eff = d 0 − wavg = ⎢
2
( )  
φmn(, ϕ )dS. (6)
2⎥ 0
(3) d. 
⎢ 128/β 2 − 32/21 Vdc ⎥
⎣ V ⎦ ( )
PI
Here ftmn (, ϕ, t ) is the total external transverse force per unit
area applied on mode (m,  n ), such as the electrostatic force
Here wavg is the average static deflection of the plate, which is density, the reaction force density as a result of the medium
equal to one third of the peak amplitude of the static deflection load and the incoming pressure.
[9]. This effective gap assumption is mainly used to simply Though this methodology can deal with all kinds of excita-
the computation in the subsequent derivation of the equivalent tion, discussion in this paper will be limited to cMUTs with a
circuit model. uniform input (i.e. a uniform excitation voltage or a uniform
The dynamic transverse deflection w(, ϕ,  t ) of the mem- incoming pressure). Only axisymmetric modes (i.e. m = 0)
brane can be found from the equation of motion of a thin plate will be excited in this case, other modes (i.e. m ≠ 0) will not
[33]. Using modal analysis techniques, this dynamic trans- be excited. Therefore, the resonating branches in the equiv-
verse deflection can be written as a linear superposition of all alent circuit model will represent mode (0,  1), mode (0,  2),
the modal response: mode (0,  3) and so on. To simplify the notation and also the
discussion, the nodal diameter m will be neglected and corre­
w(, ϕ,  t ) = ∑w mn(t )φmn(, ϕ )
sponding symbols will be denoted in the following way (e.g.
⎧ ⎡ Jm(γmna ) ⎤ ⎫ φi  φ0n, i = n = 1,  2,  3 … ). In order to clearly depict our
∑ wmn(t )⎨αmn ⎢Jm(γmn) − Im(γmn)⎥ cos(mϕ )⎬
⎪ ⎪

=
⎩
⎪
⎣ Im(γmna ) ⎦ ⎭
⎪ approach, our equivalent model in this paper will be derived
(4) by the first three axisymmetric modes as an illustration. A
summary of the modal information for the first three axisym-
where m and n are the nodal diameter and the nodal circle
metric modes is shown in table 3, showing the signatures of
of a mode shape, respectively. wmn(t ) is the time dependent
the cMUT cell of a circular shape.
displacement amplitude of mode (m,  n ). φmn represents the nor­
malized mode shape of mode (m,  n ), which is normalized to the
maximum point of the mode shape by a dimensionless normal- 3.3.  Development of the equivalent circuit model for
ization factor αmn. Jm is the Bessel function of the first kind, and a circular cMUT cell
Im is the modified Bessel function of the first kind, both of these
In this section, derivation of the equivalent circuit model is from
functions are of order m. γmna is a dimensionless frequency
the viewpoint of the Tx mode. Parameters in the equivalent cir-
parameter of mode (m,  n ) dictated by boundary conditions.
cuit of the Rx mode are similar to those defined in the Tx mode.
The Eigen frequency of mode (m,  n ) of a plate is known
to be:
3.3.1. Electromechanical transduction.  In the small signal
D γ 2mn linear approximation, there will be Vac  Vdc and w(, t )  d 0
fmn =
(5) . −w0(). The electrostatic force applied on the membrane for
2π ρ h
the ith mode is given by equation (6):
5
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Table 3.  Summary of the modal information for the first three axisymmetric modes of a circular cMUT cell.

Mode (m, n) Mode (0, 1) Mode (0, 2) Mode (0, 3)

Mode shape φi()

γia 3.1962 6.3064 9.4395

αi 0.9472 1.0025 0.9999
∬S φi dS 0.3116 -0.1313 0.0736
χi = S
∬S φi2dS 0.1828 0.1018 0.0675
ψi = S

which is an effective area Aei that translates a force to a

Fei = ∬S fe (,  t )φi() dS pressure.
 

∬S
 
ε0V dcφi() ε0V 2dcφi()
≅ ∬  
S (d 0 − w0())2
V ac dS + ∬
S (d 0 − w0())3
w(, t )dS Aei =
(11) φi()dS = χiπa2.

(7)
where fe (, t ) is the variable electrostatic force per unit area
3.3.3. Electrical impedance.  In the electrical domain, the
[24, 25]. A linear transduction relationship between the alter-
nating voltage Vac and oscillating electrostatic force of the input static capacitance C0 of a circular cMUT cell is modified
ith mode Fei is obtained, denoting the first transformer in the by the static deflection w0() as a result of the DC bias, which
ith branch of the equivalent circuit model. The transduction is defined by:
  
factor of the transformer is recognized as the electromechan-
ε0 ε0πa2 arctan h wpk /d 0
ical transduction factor, which is defined by:
 
C0 = ∬ S d 0 − w0()
dS =
d0 wpk /d 0
. (12)
ε0V dcφi() εV
Γi =
(8)   ∬
S (d 0 − w0())
2
dS ≅ χiπa2 0 2 dc
d eff
S

Here, arctan h is an inverse hyperbolic tangent function.

where the electromechanical transduction factor Γi can be
obtained by the numerical integration itself. However, if the 3.3.4. Mechanical impedance.  Motional parameters of the
effective gap d eff defined in equation (3) is used as an assump- membrane, in absence of loading, are already defined in [26].
tion, Γi will be given in the form of an analytical expression. In The modal mass Mi is given by:
this case, the computation required each time will be greatly  
reduced and the effect of such assumption will be discussed
later. Making use of the orthogonality of the mode shape, we
Mi =
(13) ∬S
ρ h φi2()dS = ψiρ h .
S
can rearrange the second term of equation (7) and move it to the
left side of equation (6). A negative stiffness K ′i, representing The modal stiffness Ki can be calculated based on its original
definition in [26]. If there is no axial stress in the membrane,
the ‘spring softening’ as a result of the DC bias, can be defined:
as described herein, it can be computed by the resonance rela-
ε0V 2dcφi2() 2 ε0V dc
2 tionship of the harmonic oscillator:
K ′i = −
(9) ∬
S (d 0 − w0())
3
dS ≅ − ψiπa
d 3eff
.
Ki = (2πfi ) Mi = ψi γ i4D .
2
(14)
Here the similar d eff assumption for the electromechanical Note that, as a result of the electromechanical coupling, the
transduction factor Γi is used in the equation.
modal stiffness will be modified as K ∗i = Ki + K ′i. The modal
loss Ri can be obtained as follows:
3.3.2. Mechanical-acoustic transduction.  The modal force
projected on the mechanical structure from the acoustic side 2πfi Mi ρ hD
is given by equation (6): (15)
Ri = = ψi γ i2
Qi Qi
Fai =
(10) Piφi()dS = Piχiπa2. ∬S where, Qi is the intrinsic quality factor of the cMUT cell in
absence of loading. This intrinsic damping of the cMUT cell
Here Pi is the modal pressure in the acoustical domain. A includes the material damping, structural damping and so on.
mechanical-acoustical transduction factor can be defined, Usually, the Qi value of a cMUT cell is as high as hundreds or

6
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Normalized cross modal radiation impedance

Normalized self modal radiation impedance
0.25 (a) Real (Zmr11) 0.06 (b) Real (Zmr12)
Real (Zmr22) Real (Zmr13)
0.20 Real (Zmr33) 0.04 Real (Zmr23)
Imag (Zmr11) Imag (Zmr12)
Imag (Zmr22) Imag (Zmr13)
0.15 0.02
Imag (Zmr33) Imag (Zmr23)

0.10 0.00

0.05 -0.02

0.00 -0.04

0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
2ka 2ka

Figure 6.  Numerically calculated normalized mechanical radiation impedance of a circular cMUT cell as a function of 2ka (the
normalization factor is ρ0c0S; ρ0 is the density of the medium, and c0 is the speed of sound of the medium): (a) normalized self modal
radiation impedance Z mrii /(ρ0c0S ); (b) normalized cross modal radiation impedance Z mrij /(ρ0c0S ).

even thousands. Once the cMUT cell is loaded by the acoustic in which, ui and uj are the normal velocity amplitude of the
medium, the effect of the intrinsic damping will become ith and jth mode, respectively, here the superscript * means
negligible. In the simulation, a few hundred or an actual the conjugate. pi (), determined by the normal velocity ui ()
experimental value (e.g. a quality factor ~450 is measured in of the ith mode, is the surface acoustic pressure of the trans-
vacuum based on the devices as shown in figure 2(c)) can be ducer and can be calculated by the Rayleigh integral method.
chosen for Qi. When j is equal to i, Z mrij becomes the self modal radiation
impedance Z mrii of the ith mode, and hence the definition of
3.3.5. Mechanical radiation impedance.  The effect of the self modal radiation impedance is already included in equa-
medium load applied on the vibrating diaphragm is charac- tion (17). According to the acoustic reciprocity theorem, we
terized by the mechanical radiation impedance. The real part have Z mrij = Z mrji, this relationship holds when the velocity
of this radiation impedance is the radiation resistance, which distribution of the membrane has the same phase or only a π
represents the actual propagating pressure and provides an phase difference [37]. Note that, if this impedance is defined
additional damping to the device. The imaginary part of this in the acoustical domain as an acoustic radiation impedance
radiation impedance is the radiation reactance, which behaves Z ari, the reference velocity will be the volume velocity. A rela-
like an added mass vibrating with the diaphragm and lowers tionship of Z mri = Z ari A2ei will be easily obtained. In a circuit,
the resonance frequency of each mode [34]. In a mechani- it is equivalent to put the radiation impedance before or after
cal system, in absence of loading, there is no damping or the mechanical-acoustical transformer by using either the
the damping is negligible. There will be no power exchange mechanical radiation impedance Z mri or the acoustic radiation
between different modes by the orthogonality of mode shape. impedance Z ari, respectively [25]. However, when the volume
However, it is not the same case if there is acoustic radiation. velocity is zero (e.g. an antisymmetric mode), the acoustic
No mode is really isolated, and the cross modal coupling radiation impedance will become infinite by definitions.
due to the coexistence of other modes in the same structure Therefore, it is more robust to use the mechanical radiation
should be taken into account [35]. Correspondingly, the total impedance.
mechanical radiation impedance Z mri of the ith mode becomes Radiation impedance can be numerically calculated based
a summation of the self modal radiation impedance of the ith on the definition given by equation  (17) [38]. Self modal
mode and the total cross modal radiation impedance of the ith radiation impedance ( j = i) of the first three modes and the
mode that is the impedance of all the other jth modes ( j ≠ i) cross modal radiation impedance ( j ≠ i) between the first
applied to the ith mode [36–39]: three modes are shown in figures  6(a) and (b), respectively.
∞ In this work, velocity amplitude is used as the reference
Z mri = Z mrii +
(16)
Z mrijuj /ui ∑ velocity. The obtained radiation impedance should be divided
j=1 by χi2 if the reference velocity is average velocity [38]. This
j≠i
impedance data is not relied on a specific geometric dimen-
where, in the mechanical domain, cross modal radiation sion and regardless of the actual velocity amplitude. Once
impedance Z mrij between different modes in the the same the directivity of one eigenmode is known, the corresponding
structure can be defined by an analogy of the mutual radiation radiation impedance can be also obtained by the Bouwkamp’s
impedance between different cells: impedance theorem [1, 40]. Another approach of obtaining
the radiation impedance is by means of a quick simulation
1
Z mrij =
(17)
ui u∗j S
pi ()u∗j()dS ∬ in a FEM software package. The FEM method will be very
helpful when the numerical calculation of a certain geometry

7
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

is difficult, making our model feasible for arbitrary shapes

potentially.
In figure  6, maximum of the self modal radiation resist­
ance is approximately located at the position where the
corresponding mode resonances. As 2ka → ∞, the self modal
radiation resist­ance converges to be a constant and the reac-
tance vanishes. The cross modal radiation resistance and
reactance oscillate around zero as the 2ka value varies. Two
peaks of the cross modal radiation resistance are found at
the positions approximately correspond to the maximums
of the self modal radiation resistance of two modes. The
tendency for the cross modal radiation reactance is not very
evident, peaks approximately locate at the place where a
strongest interaction takes place between the self modal radia-
tion reactance of two modes. A smaller comparable area of
Z mr13 is found compared to the one of Z mr12 and Z mr23. As Figure 7.  Coordinates and geometry defined for the spatial pressure
the maximum value positions of the self modal radiation field calculation.
impedance Z mr11 and Z mr33 are further apart, the cross modal
coupling between mode (0, 1) and mode (0, 2) becomes weak.
Both of the cross modal radiation resistance and reactance 3 3
Vin Γiφi()
approach zero as 2ka → ∞, showing the cross modal cou-
pling vanishes if the 2ka value reaches a threshold value [23,
() =
u(18) ∑
i=1
uiφi() = − ∑ 1 ∗
i = 1 (iωMi + iω K i + Ri + Z mri )
.

41]. In actual use, these threshold values can be approxi-

The spatial pressure field of a cMUT cell can be numerically
mately described by the 2ka values (i.e. Z mr12 : 2ka > 25,
calculated by the surface mechanical velocity u() using the
Z mr13 : 2ka > 25, Z mr23 : 2ka > 25). Moreover, contributions
Rayleigh integral method [34]. Without loss of generality, a
of higher order modes (i.e. mode (0,  4), (0,  5), ...) in the fre-
Cartesian coordinate system is used for the spatial pressure
quency region of the first three modes are negligible; a brief
field calculation shown in figure 7. The spatial pressure gener-
explanation is given here. It has been proved that, the lower
ated by a circular cMUT cell is given by:
order mode is usually more likely to be excited than the higher
order mode and thus behaves a larger response (i.e. velocity e−i(k + iα)R
amplitude) [35]. Based on this fact, by equation  (16), it is
P (x , y , z ) =
(19)
S
∬
ikρ0u()
R
dS  
easy to conclude that the cross modal coupling effect from
in which
the higher order mode applied on the lower order mode is
relatively small. It is possible to set a cut-off frequency in the R = r→ − → = (x −  cos ϕ )2 + (y −  sin ϕ )2 + z 2 .
simulation; only the self radiation and the cross coupling of
those modes below this frequency are modelled. In this work, Here α is the acoustic attenuation coefficient of the medium, k
a cut-off frequency can be chosen between the resonance fre- is the wave number,  is the distance from the origin O to the
quency of mode (0,  3) and (0,  4). The first three modes will center A of the unit area, r is the distance from the origin O to
be considered in the modelling, and cross modal contributions the observation point B, R is the acoustic path from the center
of higher order modes to the first three modes are neglected. A of the unit area to the observation point B, and u() is the
normal velocity of the unit area.

3.4.  Output in the Tx mode 3.5.  Output in the Rx mode

When the cMUT cell works in the Tx mode, the electrical port When the cMUT cell works in the Rx mode, the membrane
is an alternating voltage Vac and the acoustical port is a short will pick up the incoming pressure and an output electric
circuit as shown in figure 4. The total volume velocity U can signal can be sensed in the electrical domain. Different from
be measured at the acoustical terminal, which is a summa- the Tx mode, as shown in figure 4, the incoming pres­sure Pin at
tion of all the volume velocity of each mode. However, this the acoustical port is the driving source in the Rx mode. If we
volume velocity does not include enough output informa- aim to measure the current coming out of the circuit (e.g. an
tion of the device (e.g. the directivity of the acoustic field), ideal transimpedance amplifier is used), the static capacitance
a more useful output of the equivalent circuit will be the will become disabled by a short circuit at the electrical port.
mechanical velocity of a cMUT cell. Using modal analysis In the frequency domain, the change of total current at this
techniques, this mechanical velocity is a linear superposition terminal will be:
of the velocity response of each mode. Given that the response
3 3
of the device can be written in the time harmonic form as ΓiAei Pin
ui(t ) = ui eiωt, where ω is the angular frequency, the surface I=
(20)
i=1
Ii =
i=1
∑i ω Mi +
∑ 1 ∗
K i + Ri + Z mri
.
mechanical velocity of a cMUT cell is given by: iω

8
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Figure 8.  A general simulation flowchart of the Tx mode.

In the equivalent circuit model shown as figure 4, we have

the relationship Ii = Γiui, and hence the total mechanical
radiation impedance Z mri of the ith mode defined in equa-
tion  (16) can be mathematically expressed by the modal
current in the Rx mode. In this paper, the current coming
out of the circuit is defined as a positive direction in the
Rx mode, otherwise a negative sign should be given for the
modal current Ii. Figure 9.  Schematic of the 2D axisymmetric COMSOL model of a
circular cMUT cell (not drawn to scale).

3.6.  Solving the model 4.1.  Tx mode verification

This equivalent circuit model can be simulated by a circuit 4.1.1.  COMSOL model of the Tx mode.  Figure 9 is the sche-
simulation software package (e.g. ADS, SPICE, Spectre) [23]. matic of a 2D axisymmetric FEM model of a circular cMUT
However, the mechanical radiation impedance introduces cell when it works in the Tx mode. This is a simplified model,
frequency dependent component in the circuit, which might but which can well validate our equivalent circuit model.
require further attentions. Alternatively, the response of the The cMUT cell is modelled by the ‘electromechanics phys-
circuit can be obtained with the help of a programming lan- ics (emi)’ module as defined in COMSOL. The membrane
guage (e.g. MATLAB, Python) or (Simulink) toolboxes for material is chosen to be a linear elastic dielectric. The mem-
block diagrams [10]. In this paper, a MATLAB (MATLAB brane is perfectly clamped at the boundaries. The vacuum gap
R2014a, USA) program is used to solve this equivalent circuit is represented by a free deformation thin layer which has an
model. equivalent thickness of d 0. An electromechanical interface is
To sum up, a general simulation flowchart of the Tx defined on the top boundary of the thin layer, and a zero pre-
mode using MATLAB programming is presented in figure 8. scribed mesh displacement is given on the bottom boundary
Firstly, ‘plate vibration calculation’ (module ①) provides the of this layer. The fixed mesh is used to represent the (ideally)
modal information. Transient voltage input will be converted fixed bottom electrode of the cMUT cell. In our simplified
to the frequency domain quantity by a fast Fourier trans- model, both of the membrane (i.e. top electrode) and bottom
form (FFT), and ‘equivalent circuit simulation’ (module ②) electrode are considered as a conductor with a given relative
is performed. The output mechanical velocity can be either permittivity of 1. An AC perturbation signal superimposed
converted back to the time domain quantity by an inverse fast on a DC voltage are applied on the boundary of the bottom
Fourier transform (IFFT), or which can be used as an input electrode, and the membrane will be grounded. ‘Pressure
of ‘Rayleigh integral calculation’ (module ③). Finally, the acoustics frequency domain physics (acpr)’ module is used
obtained frequency domain pressure can be converted back to represent the loaded medium, and a PML absorbing layer
to the time domain pressure by an IFFT. The way of dealing is used to eliminate the reflection of radiated wave. PML
with the Rx mode is similar. Examples of the frequency absorbing layer can be replaced by a spherical wave radia-
response and transient response in the Tx mode can be found tion boundary if necessary (e.g. a transient simulation). The
in figures 10 and 17, respectively. Examples of the frequency maximum element size of meshes in acpr is set to be a quarter
response in the Rx mode are presented in figure 13. of the smallest wavelength in a frequency domain sweep. The
interface between emi and acpr is a self-defined fluid–solid
coupled boundary, which is corresponding to a pressure load
4.  Model verification by FEM on emi and a normal acceleration on acpr, respectively.

In this section, the commercial FEM software package 4.1.2.  Natural frequency comparison.  A comparison of natu-
COMSOL (COMSOL 4.3, Burlington, USA) is used for the ral frequencies obtained by (analytical) equivalent circuit model
verification of this equivalent circuit model. and FEM simulation is shown in table 4, showing three cMUT

9
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Table 4.  A comparison of natural frequencies obtained by (analytical) equivalent circuit model and FEM simulation for cMUT cells with
three aspect ratios.
Percentage error (%)
fan − fFEM
Diameter D (µm) Aspect ratio (D /h ) Eigenmode (m, n ) Analytical (MHz) (fan ) FEM (MHz) (fFEM ) ( fFEM
× 100%)

60 11 (0, 1) 13.72 12.80 7.17

(0, 2) 53.40 43.60 22.48
(0, 3) 119.63 84.38 41.77
120 22 (0, 1) 3.44 3.37 1.81
(0, 2) 13.35 12.57 6.19
(0, 3) 29.91 26.59 12.47
240 44 (0, 1) 0.86 0.85 0.43
(0, 2) 3.34 3.29 1.58
(0, 3) 7.48 7.24 3.31

cells with different diameters. The largest mismatch between modal coupling can be not included in the equivalent circuit
equivalent circuit model and FEM simulation is found in the model. However, situations of those higher order harmonic
60 µm diameter cMUT cell, which has the smallest aspect ratio modes are a bit complicated. As discussed in the previous sec-
(diameter/thickness). From the table, it is easy to conclude that tion, the cross modal coupling vanishes when the 2ka value is
the smaller the aspect ratio and the higher the order of eigen- above the threshold value. In actual use, the center frequen-
mode, the larger the mismatch. If the aspect ratio is not large cies of those eigenmodes should fulfill this requirement. If we
enough (i.e. the plate is not ‘thin’ enough), the actual frequency carefully check this 2ka condition, neither the second nor the
parameters will be smaller than those obtained by the classi- third case meets this requirement and the cross modal cou-
cal thin plate theory [42, 43]. The actual natural frequency is pling should be taken into account. To conclude, it will be
smaller than the one predicted by the classical thin plate theory, helpful to make a simple analysis before choosing the suit-
and the mode shape will be changed correspondingly. If we able equivalent circuit model. The major steps are as follows:
modify the frequency parameter and the corresponding mode run the equivalent circuit model without the cross modal cou-
shape in the model, the mismatch due to the small aspect ratio pling; calculate the center frequencies and corresponding 2ka
can be overcome. However, the main purpose of this paper is values of all the modes; check whether these values meet the
to introduce our equivalent circuit model, only results of the threshold values described in the previous section or not. If the
classical thin plate theory will be used in the model. equivalent circuit model without the cross modal coupling is
good enough to be used as an approximation, the one without
4.1.3.  Frequency response verification.  A frequency response the cross modal coupling is preferred as a result of the sig-
verification is performed for the abovementioned three dif- nificantly predominate computing time efficiency than the one
ferent diameter cMUT cells, and results are presented in fig- with the cross modal coupling.
ure 10. The cMUT cell is operated in the Fluorinert (FC-84,
3M, USA). The speed of sound and density of FC-84 are 537 4.1.4. Limitations of the effective gap assumption.  In the
m s−1 and 1730 kg m−3, respectively. The applied DC bias is derivation of our model, an effective gap assumption is used
10 V, while the 240 µm diameter case has a reduced 5 V DC in order to provide an analytical expression for equations (8)
bias. The AC harmonic perturbation signal is fixed as 1 V in and (9). The effective gap is approximately defined by using
order to minimize the unwanted mismatch due to the large an average static deflection rather than a real curved shape of
harmonic deflection. The first two rows of figure 10 show the the static deflection. When the DC bias is small, this approx­
magnitude and phase of the peak velocity of the cMUT cells imation should work well since the membrane is nearly flat.
at the origin. The remaining two rows present the magnitude As the DC bias increases, an average static deflection cannot
and phase of the axial pressure at 100 µm distance from the well represent the large curved membrane. Mismatch due to
cMUT cell. It is noted that, results from the equivalent cir- this effective gap assumption will become more and more sig-
cuit model and COMSOL simulation match much better as nificant. Limitations and usage scope of our equivalent circuit
the increase of aspect ratios. Evidently, mismatch between our model as a result of the effective gap assumption are discussed
model and FEM simulation is mainly originated from the ini- here.
tial mismatch of natural frequencies. Figure 11 presents the electromechanical transduction
The differences, with or without the cross modal coupling factor Γi and the negative stiffness K ′i versus an applied DC
in the equivalent circuit model, attract our particular interest. bias. Curves are calculated based on the fundamental mode
For a conventional cMUT cell, the velocity response of the of the 240 µm diameter case. Note that, both of the electro-
fundamental mode is much larger than those of higher order mechanical transduction factor and the negative stiffness are
harmonic modes, and hence the cross modal coupling effect underestimated due to the d eff assumption. However, the d eff
from the higher order mode in the vicinity of the fundamental assumption has a negligible error until the applied DC bias
mode is small. Based on this fact, if only the frequency reaches 70–80% of the pull-in voltage. A similar tendency can
response near the fundamental mode is interested, the cross be also found in the one of mode (0, 2) and of mode (0, 3), and

10
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Diameter 60 µm Diameter 120 µm Diameter 240 µm

(aspect ratio: 11) (aspect ratio: 22) (aspect ratio: 44)
3.0E-3
Peak velocity at the center of the cMUT cell
1.0E-3 1.0E-2

Amplitude [m/s] 8.0E-4

2.5E-3
8.0E-3

2.0E-3
6.0E-4 6.0E-3
1.5E-3
4.0E-4 4.0E-3
1.0E-3
2.0E-4 2.0E-3
5.0E-4

0.0 0.0
0.0

0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6

4 4 4

3 3 3

2 2 2
Phase [Rad]

1 1 1

0 0 0

-1 -1 -1

-2 -2 -2

-3 -3 -3

-4 -4 -4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6

700 900
200

600 750
Amplitude [Pa]

160
500
600
Axial pressure at 100 µm distance

120 400
450
80 300
300
200
40
100 150

0 0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6

4 4 4

3 3 3

2 2 2
Phase [Rad]

1 1 1

0 0 0

-1 -1 -1

-2 -2 -2

-3 -3 -3

-4 -4 -4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6

Frequency [MHz] Frequency [MHz] Frequency [MHz]

Equivalent circuit simulation (without cross modal coupling)
Legend Equivalent circuit simulation (with cross modal coupling)
COMSOL simulation

Figure 10.  A comparison of frequency response results obtained by the equivalent circuit model and COMSOL simulation. The first two
rows present amplitude and phase of the peak velocity u(0) of the cMUT cell at the origin. The second two rows present amplitude and
phase of the axial pressure p(0, 0, 100 µ m) at 100 µm distance from the cMUT cell.

results will be not shown. It proves that this d eff assumption the findings of an underestimated electromechanical transduc-
will meet the requirement of most operation conditions of the tion factor and negative stiffness in figure 11. The mismatch
cMUT device. between this approximated model and COMSOL simulation
Figure 12 is the frequency response spectrum of velocity is acceptable even if the DC bias reaches 70% of the pull-in
amplitude in the vicinity of the fundamental mode of the 240 voltage. In practice, this model is already applicable to be used
µm diameter case. The cMUT cell is operated in FC-84. DC as a quick simulator of cMUTs with a considerable accuracy.
biases are swept from 30% to 70% of a 30 V pull-in voltage,
and the AC harmonic perturbation signal is fixed as 1 V. A 4.2.  Rx mode verification
decrease of the resonance frequency is found as an increase
of the applied DC bias, which is known as the ‘spring sof- When the cMUT cell works in the Rx mode, in principle,
tening’ phenomenon. The velocity amplitude and the amount the COMSOL model will be the same as the one shown in
of frequency shift from our model are both smaller than those figure 9. A main difference is that the incoming pressure will
obtained from the COMOSL simulation. It further confirms replace the AC perturbation signal as the harmonic excitation.

11
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

0.07
5x10 -5 0
30%, COMSOL
0.06
40%, COMSOL

Velocity amplitude [m/s]

4x10 -5
-1x10 3 50%, COMSOL
Numerical integration 0.05 60%, COMSOL
70%, COMSOL
deff assumption
3x10 -5 -2x10 3
30%, Model

K'1 [N/m]
0.04
Γ1 [N/V]

40%, Model
50%, Model
0.03 60%, Model
2x10 -5 -3x10 3
70%, Model
0.02

1x10 -5 -4x10 3
0.01

0 -5x10 3 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.20 0.25 0.30 0.35 0.40 0.45
Vdc / V PI Frequency [MHz]

Figure 11.  Effect of the numerical integration and d eff assumption Figure 12.  A comparison of results obtained by the equivalent
on the electromechanical transduction factor in equation (8) and the circuit model and COMSOL simulation. Frequency response results
negative stiffness in equation (9). Curves are calculated based on in the vicinity of the fundamental mode of the 240 µm diameter
the fundamental mode of the 240 µm diameter case. (‘Numerical case are shown. The AC perturbation voltage is fixed as 1 V, and the
integration’ denotes the integral form and ‘d eff assumption’ means DC bias is swept from 30% to 70% of the pull-in voltage.
the analytical form; both of results are only accurate when the
applied DC bias is smaller than 90% of the pull-in voltage.) 5.  Model validation by experiments

In this section, comparisons of the results obtained by the

However, it is found that the PML in COMSOL model does
equivalent circuit model and experimental characterizations
not offer a good absorption when the background pressure is
are presented. The 120 µm diameter circular cMUT cell as
used as the incoming pressure in this case. A significant error
shown in figure 2(c) is chosen for this experimental verifica-
will be obtained in FEM simulation. ‘Acoustic-solid interac-
tion, which is the circular cMUT device with the largest aspect
tion frequency domain physics (acsl)’ module is used as an
ratio as what we have.
alternative solution. It means the DC bias induced static elec-
trostatic force cannot be applied in the simulation. The 240
µm diameter cMUT cell, operated in FC-84, is modelled. A 5.1.  Optical characterization in the Tx mode
uniform 1 kPa pressure is applied on the acoustic-solid inter-
The commercial laser Doppler vibrometer (LDV) micro-system
action boundary as the incoming pressure, and a small 5 V
analyzer (MSA-500, Polytec GmbH Waldbronn, Germany) is
DC bias is chosen though it does not take effect in COMSOL
used to study the dynamic responses of the cMUT cell. The
simulation. If the incoming pressure is not uniform, a pressure
experimental set-up is already depicted in one of our previous
load in the form of corresponding function can be also applied
publications [44]. Measurements are performed in air and in
in the same manner.
FC-84 respectively. In the immersion experiment, the cMUT
The harmonic transverse deflection amplitude w() of
cell is operated in FC-84 liquid with a depth around 5 mm.
the membrane can be obtained from COMSOL simulation.
Correction is performed on the FC-84 experimental results,
Assume that the deformed membrane of the cMUT cell is
the measured displacement is divided by 1.261 which is the
one plate of a variable capacitance, the instantaneous current
refractive index of FC-84 [45]. Displacement of the device
coming out from this variable capacitance under a DC bias is
is measured rather than the velocity due to a 3 MHz upper
as follows:
frequency limit in the velocity measurement.
⎛ ε0 ε πa2 ⎞
IFEM = VDC 0 = iωVDC⎜
dC
∬
dS − 0  ⎟.
⎝ S d 0 − w() d0 ⎠
(21) 5.1.1.  Frequency response in air.  When cMUTs are operated
dt
in air, the frequency shift due to the acoustic mass is negligible.
The obtained current should also represent the output current The measured resonance frequencies can be approximately
in the equivalent circuit model when the static capacitance considered as natural frequencies of the device. Figure  14
is short. The current obtained by equations  (20) and (21), presents the displacement amplitude of the first four modes of
representing the results of the equivalent circuit model and a circular cMUT cell operated in air. The device is driven by a
COMSOL simulation respectively, are compared in figure 13. 10 V internal sweep voltage from the LDV instrument and an
The mismatch is significant if there is no cross modal cou- external 20 V DC bias. For those higher order modes, they are
pling considered in the model. Once the cross modal coupling not detectable any more by this instrument under such a driv-
is taken into account, good matches are found between the ing condition. Frequencies of these modes are also measured,
equivalent circuit model and COMSOL simulation. The slight and values in brackets are from the (analytical) equivalent cir-
mismatch of the third mode is due to the initial mismatch of cuit model. Results from equivalent circuit model are rather
the natural frequency. close to the measurement results. The mismatch is attributed

12
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Amplitude of the output current [nA]

(a) Model (without cross modal coupling) (b) Model (without cross modal coupling)

P h a s e o f t h e o u tp u t c u r r e n t [ R a d ]
150 Model (with cross modal coupling) 2 Model (with cross modal coupling)
COMSOL simulation COMSOL simulation

100 1

50 0

0 -1

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Frequency [MHz] Frequency [MHz]

Figure 13.  Output current of the 240 µm diameter cMUT cell: (a) amplitude; (b) phase. (The device is operated in FC-84, which is biased
at a 5 V DC bias with a 1 kPa incoming pressure.)

Mode (0, 1) Mode (1, 1)

Normalized displacement amplitude [a.u.b]

f01=3.39 (3.43) MHz f11=6.83 (7.14) MHz 1.0 Polytec measurement
Equivalent circuit model
0.8

0.6

0.4

0.2

Mode (2, 1) Mode (0, 2)

0.0
f21=10.98 (11.71) MHz f02=12.40 (13.35) MHz
0 2 4 6 8 10
Frequency [MHz]

Figure 15.  Frequency spectrum of the normalized displacement

amplitude of the cMUT cell. The device is operated in FC-84,
which is driven by a 10 V sine wave sweep and a 20 V DC bias.

very well in the vicinity of the fundamental mode. However,

the signal to noise ratio (SNR) of experimental results is not
Figure 14.  Root mean square (RMS) displacement amplitude and
enough to distinguish those harmonic modes when the cMUT
resonance frequency of the cMUT cell operated in air. (Frequency
values in brackets are obtained from the (analytical) equivalent cell is operated in FC-84.
circuit model.)
5.2.  Acoustical characterization in the Tx mode
to the small aspect ratio of this device, as already indicated
in table 4. Under the abovementioned driving condition (i.e. The test setup used for the transmission characterizations of
a uniform excitation), those antisymmetric modes will not be the cMUT cell is shown in figures 16(a) and (b). The cMUT
excited in an ideal axisymmetric device. These antisymmetric cell is put in a tank and immersed in FC-84, which is driven
modes are found in the experiments due to the imperfect sym- by a pulse superimposed on a (high) bipolar DC bias voltage.
metry of the real device. A needle type calibrated hydrophone (HNP1000, Onda Inc.,
USA) is used to detect the acoustic pressure at a few milli-
5.1.2.  Frequency response in FC-84.  Figure 15 shows the fre- meters from the transducer surface, and the exact distance is
quency spectrum of the normalized displacement amplitude of obtained from the transmitting time (figure 16(c)). The hydro-
the cMUT cell. The device is operated in FC-84, and the same phone signal is amplified by the pre-amplifier and recorded
driving condition as air is used. As the difference between the by an oscilloscope (TDS3014B, Tektronix, Inc., USA). A
equivalent circuit model with or without cross modal coupling Labview (Labview, National Instruments, USA) program is
is very small for this case, only the result from the equivalent used to control all the instruments and save the measured data
circuit model with cross modal coupling is shown here. The from the oscilloscope [44].
fundamental mode is found around 1.6 MHz. Results from A 15 V, 50 ns duration pulse superimposed on a 100 V
the Polytec measurement and equivalent circuit model match bipolar DC bias are applied to drive the cMUT cell in FC-84.

13
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Figure 16.  Transmission characterizations of cMUTs: (a) sketch of the acoustic characterization experiment; (b) photo of the experimental
set-up; (c) example of the experimental signal.

600
(a) 0 (b)
Transient waveform [Pa]

Frequency spectrum [dB]

400

-10
200
Mode (0, 1)
Mode (0, 2)
0 -20

-200
-30
-400
Model (without cross modal coupling) Model (without cross modal coupling)
Model (with cross modal coupling) Model (with cross modal coupling)
-600 Hydrophone measurement -40
Hydrophone measurement

8.0 8.5 9.0 9.5 10.0 2 4 6 8 10 12 14 16 18

Transmitting time [µs] Frequency [MHz]

Figure 17.  The transient pressure and corresponding frequency spectrum at 4.5 mm distance from the cMUT cell. The device is driven by a
15 V, 50 ns duration pulse and a 100 V bipolar DC bias.

Figure  17 shows the comparison of the results obtained by cross modal coupling is included in the model, the mismatch
the equivalent circuit model and hydrophone measurement. between experimental results and simulation results becomes
The transient pressure waveform with a transmitting time 8 much smaller. This is found in both transient waveform and
µs, corresponding to a measurement distance of 4.5 mm, is frequency spectrum.
shown in figure  17(a). Both of the main signal and ripples As our experiment has a lot of practical limitations and
match well. Figure 17(b) shows the corresponding frequency assumptions, the frequency spectrum does not completely
spectrum of this acoustic signal, which is already normalized match over the entire frequency range. Limitations and pos-
to the maximum amplitude. The fundamental mode (0,  1) of sible reasons which may cause this mismatch are summarized
this device is found at 1.6 MHz, and mode (0,  2) is found at as follows: (a) the sensitivity of our hydrophone is calibrated in
8 MHz, which are indicated by both of the equivalent circuit water within a frequency range of 1 MHz–20 MHz. However,
model and hydrophone measurement. Mode (0,  3) locates the effect of changing the medium on the hydrophone sensi-
around 14 MHz predicted by the equivalent circuit model, tivity is quite complicated. (b) Acoustic attenuation coefficient
however which is not recognizable in this figure. Once the of FC-72 (α = 35.2(f (MHz))1.56 dB m−1) from reference [46]

14
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

Figure 18.  Reception characterization of cMUTs: (a) schematic of this transimpedance amplifier (TIA); (b) photo of the experimental
set-up; (c) example of the measured signal; (d) applied Hanning window function on the measured signal in order to exclude the noise in
the time domain; (e) frequency spectrum of the measured and simulated signal both after applying the window.

is used as an approximation for FC-84. (c) The pulse generator amplifier and the experimental set-up are presented in
does not provide an ideal waveform to the device, which will figures 18(a) and (b), respectively. A high speed linear opera-
induce a different excitation power over the frequency range tional amplifier AD8001 is used, which has a bandwidth of
between the equivalent circuit model and the experiment. (d) 880 MHz. 1 nF and 100 nF capacitances are used in order to
The backside backing of the cMUT chip is needed in order increase the stability of the amplifier. A 10 kΩ resistance is
to eliminate the thickness resonance of the silicon substrate, chosen as the feedback resistance Rf of the transimpedance
it is found that this resonance always induce a dip around amplifier, the value of the resistance is a trade-off between
5.5 MHz in the frequency spectrum of our cMUT device. (e) the gain and the bandwidth of the output. Capacitance of the
The real cMUT device may have a lot of additional physical cMUT cell including the parasitic capacitance is 1.7 pF, which
phenomena which will all affect the results. For example, the is measured at 1 kHz by an LCR meter (E4980A, Agilent
vacuum gap is not ideally sealed with an internal pressure, Technologies, USA). Accordingly, the theoretical  −3 dB cut-
the membrane is not ideally clamped at the boundary and the off frequency of the gain will be around 90 MHz, which will
internal stress of the membrane is unneglectable. guarantee that in the interested frequency region the transim-
pedance amplifier is operating at a constant gain.
5.3.  Acoustical characterization in the Rx mode A commercial PVDF transducer (PA363, Precision Acoustics,
UK) is used as an acoustic source, which is initially calibrated
The output from the cMUT cell in the Rx mode can be at 11 mm distance using the hydrophone. During the experiment,
measured by two different approaches. One is the voltage mea- the cMUT chip will also be placed at the same distance from
surement, cMUTs are directly connected to an oscilloscope the transducer in order to endure that a same pressure will be
[28]. However, in this case, the measured result will be highly applied on the cMUT device. 10 cycles of Sine wave burst with
dependent on the connections. Parasitic capacitance, cables and a measured peak to peak voltage of 95 V is used to drive this
the internal impedance of oscilloscope should all be taken into transducer, and 1.68 MHz is chosen as the driving frequency
account in the equivalent circuit model. The other approach is for a maximum output from the cMUT device. An approxi-
the current measurement by using a transimpedance amplifier, mated 4 kPa incoming pressure is applied on the cMUT device,
which can convert the current to voltage with an amplification which is biased by a 30 V DC bias. Figure 18(c) shows the
gain. Static and parasitic capacitances are shorted by the small measured signal using the transimpedance amplifier, a burst
impedance of the transimpedance amplifier, and the measure- signal can be recognized after approximate 20 µs time delay
ment result is stable regardless of the connections. (corresponding to 11 mm distance). In order to clearly identify
In this paper, the current measurement scheme is used for this burst signal, a Hanning window function is applied on
the Rx mode validation. The schematic of the transimpedance the measured signal to exclude the noise in the time domain

15
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

(figure 18(d)). Frequency spectrum of the measured and simu- a uniform excitation which guarantees that no antisymmetric
lated signal both after applying the window function is shown eigenmodes will be excited in theory. If the excitation or the
in figure 18(e). Peak frequency of the spectrum is around 1.68 external load is non-uniform (e.g. cMUT cells are asymmetri-
MHz, which is the same as the applied pressure. It further cally actuated with a partial electrode [47] or cMUT cells are
confirms that this burst signal is the real output signal from operated in an array [48]), cMUT cells might tend to behave
the cMUT device, and the amplitude of this measured output antisymmetric modes. These cases can also be solved by our
signal is around 0.5 mV. As to the same operation condition, model by using generalized definitions in some parameters.
the simulated output current from the equivalent circuit is 135 However, the main purpose of this paper is to introduce the
nA. Correspondingly, 0.135 mV output voltage is obtained development of our equivalent circuit model and its valida-
after considering the gain of the transimpedance amplifier, tion. The extension of this model and the results of cMUT
which is almost four times smaller than the measured signal. cells under a more generalized excitation can be found in [1].
In fact, a precise experimental validation is difficult to be
implemented in this way. First, it requires a correct evalua-
Acknowledgments
tion of the pressure generated by the transducer, a hydrophone
with a point tip should be used in theory. Besides, the used
The authors should thank Dr Jeroen De Coster and Myriam
cMUT cell has a significantly smaller dimension than the one
Van De Peer for the help during the lab activities, and we also
of hydrophone, which may receive a larger pressure locally
thank Bruno Figeys and Guilherme Brondani Torri for the
than the one obtained from calibration. Moreover, the para-
fruitful discussions.
sitic effect of the whole measurement system is also difficult
to be evaluated, which can change the output behavior of the
transimpedance amplifier (e.g. a local peak of the gain due References
to the self-resonance). However, this experimental validation
already indicates that the equivalent circuit model can be used [1] Mao S P, Rottenberg X, Liu L W, Rochus V, Brondani G T,
to predict the output current of cMUTs in the Rx mode. The Haouari R, Nauwelaers B and Tilmans H A C 2016 A
equivalent circuit modelling is easy, fast and effortless, and general multi-mode lumped equivalent circuit model for
circular cMUT cells Proc. IEEE Int. Ultrasonics Symp.
the predicted result is comparable with the measurement. (IUS 2016) (Tours) (Piscataway, NJ: IEEE)
[2] Khuri-Yakub B T and Oralkan Ö 2011 Capacitive
micromachined ultrasonic transducers for medical imaging
6.  Discussions and conclusions and therapy J. Micromech. Microeng. 21 054004
[3] Park K K, Lee H, Kupnik M, Oralkan Ö, Ramseyer J,
Lang H P, Hegner M, Gerber C and Khuri-Yakub B T 2011
In this work we present a modal analysis based small signal Capacitive micromachined ultrasonic transducer (cMUT) as
equivalent circuit model to simulate cMUT cells. The equivalent a chemical sensor for DMMP detection Sensors Actuators
circuit model is derived based on a single circular cMUT cell. B 160 1120–7
As an illustration of our approach, only cMUTs under a uniform [4] Wong S H, Kupnik M, Watkins R D, Butts-Pauly K B
excitation are modelled in this paper. Cross modal coupling and Khuri-Yakub B T 2010 Capacitive micromachined
ultrasonic transducers for therapeutic ultrasound
between different eigenmodes of cMUTs is emphasized by us applications IEEE Trans. Biomed. Eng. 57 114–23
for the first time, and evidences of strong cross modal coupling [5] Ladabaum I, Jin X, Soh H T, Atalar A and Khuri-Yakub B T
are found in some cases. A multi-mode simulation of cMUTs is 1998 Surface micromachined capacitive ultrasonic
obtained by a correct mode superposition method. This model is transducers IEEE Trans. Ultrason. Ferroelectr.
verified by both of FEM simulation and experiments. Freq. Control 45 678–90
[6] Caronti A, Caliano G and Pappalardo M 2002 An accurate
Though the derivation of this model and its verification model for capacitive micromachined ultrasonic transducers
is only based on a circular shape in this paper, other shapes IEEE Trans. Ultrason. Ferroelectr. Freq. Control 49 159–68
can be simulated in a similar manner. cMUT cells with other [7] Lohfink A and Eccardt P-C 2005 Linear and nonlinear
geometry will have different parameters in the equivalent cir- equivalent circuit modelling of cMUTs IEEE Trans.
cuit. This appears in a different modal information and also a Ultrason. Ferroelectr. Freq. Control 52 2163–72
[8] Köymen H, Senlik M, Atalar A and Olçum S 2007 Parametric
new radiation impedance as a result of the different geometry. linear modelling of circular cMUT membranes in
In this paper, we only focus on the model of a single cMUT vacuum IEEE Trans. Ultrason. Ferroelectr. Freq. Control
cell, however potentially this model is not confined to one 54 1229–39
single cell. It is fully equipped with the capabilities to simulate [9] Wygant I O, Kupnik M and Khuri-Yakub B T 2008
a group of cMUT cells, if the mutual coupling effect between Analytically calculating membrane displacement and
the equivalent circuit model of a circular cMUT cell
eigenmodes of different cMUT cells is taken into account [23]. Proc. IEEE Int. Ultrasonics Symp. (IUS 2008) (Beijing)
There are also a few conditions of using this model. Our (Piscataway, NJ: IEEE) pp 2111–4
model is a linear small signal model, large signal models as [10] Satir S, Zahorian J and Degertekin F L 2003 A large-signal
shown in [10, 21] should be used when the excitation does not model for cMUT arrays with arbitrary membrane geometry
belong to the small signal category. Besides, classic ‘thin plate’ operating in non-collapsed mode IEEE Trans. Ultrason.
Ferroelectr. Freq. Control 60 2426–39
results are used in our model, as to those devices with a small [11] Certon D, Teston F and Patat F 2005 A finite difference model
aspect ratio, ‘thick plate’ results should be used for a better for cMUT devices IEEE Trans. Ultrason. Ferroelectr.
accuracy. Moreover, discussions in this paper are limited to Freq. Control 52 2199–210

16
J. Micromech. Microeng. 27 (2017) 035001 S P Mao et al

[12] Ronnekleiv A 2005 cMUT array modelling through free [29] Eccardt P-C, Wagner P and Hansen S 2006 Analytical models
acoustic cMUT modes and analysis of the fluid cMUT for micromachined transducers—an overview Proc. IEEE
interface through fourier transform methods IEEE Trans. Int. Ultrasonics Symp. (IUS 2006) (Vancouver) (Piscataway,
Ultrason. Ferroelectr. Freq. Control 52 2173–84 NJ: IEEE) pp 572–81
[13] Roh Y and Khuri-Yakub B T 2002 Finite element analysis of [30] Helin P, Czarnecki P, Verbist A, Bryce G, Rottenberg X and
underwater capacitor micromachined ultrasonic transducers Severi S 2012 Poly-SiGe-based cMUT array with high
IEEE Trans. Ultrason. Ferroelectr. Freq. Control 49 293–8 acoustical pressure Proc. IEEE Int. Conf. Micro Electro
[14] Lohfink A, Eccardt P-C, Benecke W and Meixner H 2003 Mech. Syst. (MEMS 2012) (Paris) (Piscataway, NJ: IEEE)
Derivation of a 1D cMUT model from FEM results for pp 305–8
linear and nonlinear equivalent circuit simulation Proc. [31] Hsueh C H, Luttrell C R and Becher P F 2006 Modelling of
IEEE Int. Ultrasonics Symp. (IUS 2003) (Honolulu) bonded multilayered disks subjected to biaxial flexure tests
(Piscataway, NJ: IEEE) pp 465–8 Int. J. Solids Struct. 43 6014–25
[15] Yaralioglu G G, Ergun A S and Khuri-Yakub B T 2005 [32] Ahmad B and Pratap R 2010 Elasto-electrostatic analysis of
Finite-element analysis of capacitive micromachined circular microplates used in capacitive micromachined
ultrasonic transducers IEEE Trans. Ultrason. Ferroelectr. ultrasonic transducers IEEE Sensors J. 10 1767–73
Freq. Control 52 2185–98 [33] Leissa A W 1969 Vibration of Plates (Washington,
[16] Caronti A, Savoia A, Caliano G and Pappalardo M 2005 DC: NASA)
Acoustic coupling in capacitive microfabricated ultrasonic [34] Kinsler L A, Frey A R, Coppens A B and Sanders J V 2000
transducers: modelling and experiments IEEE Trans. Fundamentals of Acoustics 4th edn (New York: Wiley)
Ultrason. Ferroelectr. Freq. Control 52 2220–34 [35] Hansen C, Snyder S, Qiu X J, Brooks L and Moreau D 2012
[17] Ballandras S, Wilm M and Daniau W 2005 Periodic finite Active Control of Noise and Vibration 2nd edn (Boca Raton,
element/boundary element modelling of capacitive FL: CRC Press)
micromachined ultrasonic transducer J. Appl. Phys. [36] Chang Y M and Leehey P 1979 Acoustic impedance of
97 034901 rectangular panels J. Sound Vib. 64 243–56
[18] Meynier C, Teston F and Certon D 2010 A multiscale [37] Sherman C H 1966 Analysis of acoustic interactions
model for array of capacitive micromachined ultrasonic in transducer arrays IEEE Trans. Sonics Ultrason.
transducers J. Acoust. Soc. Am. 128 2549–61 SU-13 9–15
[19] Boulmé A and Certon D 2015 Design of broadband linear [38] Porter D T 1964 Self- and mutual-radiation impedance and
micromachined ultrasonic transducer arrays by means beam patterns for flexural disks in a rigid plane J. Acoust.
of boundary element method coupled with normal mode Soc. Am. 36 1154–61
theory IEEE Trans. Ultrason. Ferroelectr. Freq. Control [39] Pritchard R L 1960 Mutual acoustic impedance between
62 1704–16 radiators in an infinite rigid plane J. Acoust. Soc. Am.
[20] Mason W P 1948 Electromechanical Transducers and Wave 32 730–7
Filters 2nd edn (New York: Van Nostrand) [40] Bouwkamp C J 1946 A contribution to the theory of acoustic
[21] Köymen H, Atalar A, Aydoğdu E, Kocabas C, Oğuz H, Olçum S, radiation Phillips Res. Rep. 1 251–77
Ozgurluk A and Ünlügedik A 2012 An improved lumped [41] Senlik M, Olçum S, Köymen H and Atalar A 2010 Radiation
element nonlinear circuit model for a circular cMUT cell impedance of an array of circular capacitive micromachined
IEEE Trans. Ultrason. Ferroelectr. Freq. Control 59 1791–9 ultrasonic transducers IEEE Trans. Ultrason. Ferroelectr.
[22] Olçum S, Yamaner F, Bozkurt A, Köymen H and Atalar A Freq. Control 57 969–76
2011 An equivalent circuit model for transmitting [42] Senjanović I, Hadžić N, Vladimir N and Cho D-S 2014
capacitive micromachined ultrasonic transducers in collapse Natural vibrations of thick circular plate based on the
mode IEEE Trans. Ultrason. Ferroelectr. Freq. Control modified Mindlin theory Arch. Mech. 66 389–409
58 1468–77 [43] So J and Leissa A W 1998 Three-dimensional vibrations
[23] Oğuz H, Atalar A and Köymen H 2013 Equivalent circuit- of thick circular and annular plates J. Sound Vib.
based analysis of cMUT cell dynamics in arrays IEEE 209 15–41
Trans. Ultrason. Ferroelectr. Freq. Control 60 1016–24 [44] Mao S P, Rottenberg X, Rochus V, Nauwelaers B and
[24] Tilmans H A C 1993 Micromechanical sensors using Tilmans H A C 2014 FEM simulation and measurement
encapsulated built-in resonant strain gauges PhD validation of a cMUT cell 15th Int. Conf. on Thermal,
Dissertation University of Twente, Enschede Mechanical and Multi-Physics Simulation and Experiments
[25] Tilmans H A C 1996 Equivalent circuit representation of in Microelectronics and Microsystems (EuroSimE) (Gent,
electromechanical transducers: I. Lumped-parameter April 2014) (Piscataway, NJ: IEEE) pp 1–7
systems J. Micromech. Mircoeng. 6 157–76 [45] Polytec 2005 Laser vibrometer measurements of objects
[26] Tilmans H A C 1997 Equivalent circuit representation of immersed in transparent fluids Polytec GmbH Laser
electromechanical transducers: II. Distributed-parameter Measurement Systems Application Note VIB-G-04
systems J. Micromech. Microeng. 7 285–309 [46] Strohm E M and Kolios M C 2011 Sound velocity and
[27] Rottenberg X, Erismis A, Czarnecki P, Helin P, Verbist A attenuation measurements of perfluorocarbon liquids using
and Tilmans H A C 2012 Consistent analytical model photoacoustic methods Proc. IEEE Int. Ultrasonics Symp.
for single and dual thickness capacitive micromachined (IUS 2011) (Orlando) (Piscataway, NJ: IEEE) pp 1–4
ultrasound transducers (cMUT) 13th Int. Conf. on Thermal, [47] You W 2013 Exploring multiple-mode vibrations of capacitive
Mechanical and Multi-Physics Simulation and Experiments micromachined ultrasonic transducers (cMUTs) PhD
in Microelectronics and Microsystems (EuroSimE) Dissertation University of British Columbia, Vancouver
(Cascais, April 2012) (Piscataway, NJ: IEEE) pp 1–6 [48] Berthillier M, Le Moal P, Lardiès J and Appliquée M
[28] Mao S P, Rottenberg X, Rochus V, Helin P, Verbist A, 2011 Dynamic and acoustic modelling of capacitive
Severi S, Nauwelaers B and Tilmans H A C 2014 Transmit micromachined ultrasonic transducers Proc. IEEE Int.
and receive of a cMUT cell: modelling and experiments Ultrasonics. Symp. (IUS 2011) (Orlando) (Piscataway, NJ:
Nanotech 2014 (Washington) vol 2 pp 81–4 IEEE) pp 608–11

17