You are on page 1of 11

Amir Heidari

Simulation Based Design of Disk


e-mail: AMIR0004@e.ntu.edu.sg
Resonator Biosensors Under
Yong-Jin Yoon
e-mail: yongjiny@ntu.edu.sg Fabrication Uncertainty
Hungsun Son
e-mail: hsson@ntu.edu.sg A high performance and cost effective biosensor is designed using a radial contour-mode
disk resonator (RCDR). This sensor measures tiny biological mass attached on a disk
School of Mechanical and vibrating at a high frequency, producing high quality of output signal. A series of analy-
Aerospace Engineering, sis and simulation models is developed to predict the mass sensitivity, dynamic stability,
Nanyang Technological University, and motional resistance of the RCDR biosensor with given geometry and signal input. In
639798 Singapore order to decrease motional resistance while keeping the fabrication cost low, a layer of
dielectric material is deposited within the capacitor gap. In designing the RCDR biosen-
sors, we employ Type I, II, and III robust design approach to design a device that is
Hae-Jin Choi1 insensitive to various types of uncertainty associated with the fabrication processes and
Assistant Professor analysis models. A mathematical construct, error margin index, is employed for this ro-
School of Mechanical Engineering, bust design. Traditional optimization and robust design approaches are separately for-
Chung-Ang University, mulated, solved, and compared. From the design results, we observe that the RCDR is a
Seoul, 156-756, South Korea promising bio-sensing device compared to the existing ones. [DOI: 10.1115/1.4006144]
e-mail: hjchoi@cau.ac.kr
Keywords: disk resonator, robust design, fabrication uncertainty, biosensor, simulation
based design, and bioMEMS

1 Introduction sensitivity needs to be further enhanced for practical biosensor


applications. Furthermore, majority of biochemical analyses still
A wide range of biosensors is being used for quantifying bio-
rely on expensive laboratory analysis with benchtop instruments.
logical and biochemical processes, which is very important for
These instruments are highly selective and sensitive, but the anal-
medical, biological, and biotechnological applications [1]. Biosen-
yses are typically performed in time and labor intensive manners;
sors may be classified into four main groups based on the princi-
thus limiting their applications [9]. Therefore, cost effective,
ples of transduction [2], which are thermal sensors, biomechanical
highly sensitive MEMS biosensors are still in great demand.
sensors (mass sensitive sensors), optical sensors, and electrochem-
ical sensors. Electrochemical sensors are used in confined space
applications for monitoring toxic and combustibles gases. Cata- 2 Radial Contour-Mode Disk Resonator (RCDR)
lytic thermal sensors are used for detecting flammable gases in a Based BioSensors
gas compound. Thermal or calorimetric sensors determine the
presence or concentration of a chemical by measuring the en- 2.1 Introduction to RCDR Biosensor. In this section, we
thalpy changes [3]. introduce a novel biosensor device based on a radial contour-
Depending on the types and applications of the biosensors, dif- mode disk resonator (RCDR) in order to meet the requirements of
ferent design requirements are used for the design of a high per- high performance biosensors described in the previous section.
formance biosensor. However, the general requirements of the Schematic and detail view of the RCDR biosensor is illustrated in
biosensors are higher sensitivity and selectivity, lower cost with Fig. 1.
high quality signals, higher efficiency of operation, and lower As shown in Fig. 1(a), the RCDR consists of a round disk anch-
detection and response time. Among the four groups of biosen- ored on the substrate, surround electrodes, and a ground plane.
sors, mass sensitive biomechanical sensors have high sensitivity, The disk is separated from the surrounding electrodes with the
wide measurement range, and fast response time, and can be fabri- 2 lm initial air gap that defines the capacitance of the electro-
cated with relatively simple processes. The biomechanical sensors mechanical transducer. The resonator is laterally driven with two
may detect tiny mass by measuring the change of resonance fre- electrodes in radial direction. A DC-bias voltage (Vp) is applied to
quency of an oscillator in the detection process. Some of the most the structure via the anchor, while two AC input voltages are
common mass sensitive devices are listed and compared in Table 1. applied to the input electrodes with 180 degrees of phase differ-
Details of the sensors are found in Refs. [4,5]. ence. These voltages result in a time-varying electrostatic force on
The issue of high mass sensitivity has been addressed by many the plate edges which makes it oscillate in its fundamental fre-
research groups using different techniques such as NEMS resona- quencies in the radial mode. The capacitive gap distances of the
tors, high frequency quartz crystal microbalances (QCMs), surface opposite side change with this frequency. Because of the electro-
acoustic waves (SAWs) and film bulk acoustic resonators static field in the gap, a time-varying current is induced in the out-
(FBARs) [6–8]. For instance, a 800 MHz FBAR sensor with put electrode. Geometric parameters of the RCDR sensor are
higher mass sensitivity of 2 KHzcm2/ng and minimum detectable illustrated in Fig. 1(b). These are disk diameter (D), disk thickness
mass of about 1 ng/cm2 is reported in Ref. [6]. However, the mass (t), stem diameter (SD), and the gap space between electrode and
disk (d0). The disk is separated from the surrounding electrodes
by a narrow air gap (d0) that plays the role of a capacitive electro-
1
Corresponding author. mechanical transducer of the device.
Contributed by the Design Automation Committee of ASME for publication in
the JOURNAL OF MECHANICAL DESIGN. Manuscript received September 28, 2010; final
When the radial vibration frequency approaches to the natural
manuscript received February 4, 2012; published online March 15, 2012. Assoc. Edi- frequency of the disk, the amplitude of the output signal signifi-
tor: David Gorsich. cantly increases due to the resonance effect. By sacrificing some

Journal of Mechanical Design Copyright V


C 2012 by ASME APRIL 2012, Vol. 134 / 041005-1

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Table 1 Comparison of mass sensitive biosensors [4,5,41]

Frequency Sensitivity
Device (MHz) (lm2=ng) Advantages Disadvantages

Bulk acoustic Thickness shear mode (TSM) 5–10 1.4 Robust nature, avail- Limited frequency
wave (BAW) ability, high mass range, low q-factor,a
resonators sensitivity fragility during
fabrication
Shear horizontal acoustic 25–200 1.5 Design to operate in __
plate mode (SH-APM) liquid
Surface acoustic Rayleigh 30–300 13.3 Working in gas __
wave (SAW) Shear horizontal surface 30–300 18 __ Higher energy loss in
resonators acoustic wave (SH-SAW) aqueous environment
Love wave 30–300 22.1 High sensitivity __
Flexural plate wave (FPW) 5–20 49 High q-factor, low __
energy loss in liquid
at a low resonant
frequency
a
Q-factor is a dimensionless parameter that compares the energy stored to the rate of energy dissipation. A higher Q indicates a lower rate of energy dissi-
pation. Generally Q is defined as: Q ¼ 2p  (Energy Stored)=( Energy dissipated per cycle).

degree of accuracy, an intuitive form of radial contour resonance


frequency of a disk resonator is obtained as [10]:
sffiffiffi
ka E
f0 ¼ (1)
R q

where f0 is the resonance frequency of disk resonator, R is the


disk’s radius, E is the elastic modulus, k is a parameter dependent
on Poisson’s ratio (k ¼ 0:342 for polysilicon), and a is a parame-
ter which depends on the order of the desired mode. It is 1, 2.62
and 4.17 for first, second and third modes, respectively [11]. The
material for disk resonator is considered to be silicon (E ¼ 169
GPa, q ¼ 2330 kg/m3,  ¼ 0.28). Therefore the mechanical reso-
nance frequency for the radial contour-mode of a disk is mainly
governed by its material properties and radius.
The RCDR has the advantages of high resonant frequency and
Q-factor over other resonators due to its unique dynamic behavior.
Firstly, since the disk resonator is anchored at its central node, the
Q-factor of the device is high, despite of its high stiffness. Sec-
ondly, the airflow damping force of the disk resonator is less than
that of diaphragm sensors because of its radial vibration.
The first application of RCDR was a signal filtering device in
electronics [11–13]. We believe aforementioned advantages of
RCDR are beneficial to the development of highly sensitive bio-
sensors. Existing biosensor technologies, such as BAW/SAW
devices listed in Table 1 and QCM based immunosensors [14,15],
offer fairly high mass sensitivity; however the RCDR is expected
to facilitate much higher sensitivity for mass detection because of
the aforementioned advantages. Currently, no biosensor has been
designed and developed using the RCDR.
The working principle of our RCDR biosensor is similar to that
of QCM based immunosensors, which is measuring the amount of
resonant frequency shift by a tiny mass change originated from an
immobilization process. Immobilization, as shown in Fig. 2, is a
process to capture biological entities of interest on the surface of a
measuring device, which is the top surface of the disk in our
device.
In the first step, Step A, a thin gold film (about 50 nm) is sput-
tered onto the disk surface in the reaction chamber before the
immobilization. This thin gold surface will make the biomaterial
immobilization easier. Step B is to immobilize primary antibody
Fig. 1 (a) Schematic view of a radial contour-mode disk reso- (phage) on top of the metal surface. In Step C, Casein is dropped
nator (RCDR), illustrating the two-port bias, excitation scheme on the surface to block the open space around the immobilized
and the important parameters; (b) top and cross sectional view bio-receptor. In Step D, when the surface is immersed in a test so-
of the RCDR biosensor lution, antigen (bacteria or virus) is captured by the antibodies.

041005-2 / Vol. 134, APRIL 2012 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


2.3 Approaches to the Challenges. In Sec. 2.2, we discuss
the problems in developing cost effective and high performance
RCDR biosensors. In this section, we discuss our approach taken
in this work to solve the problems.
First of all, we develop an FEM model to estimate detection
sensitivities of the RCDR biosensors, simulating the immobiliza-
tion process discussed in Sec. 2.1. This simulation model is
employed for robust design exploration to determine the RCDR
biosensor specifications. Currently, no work has been published
regarding this study. Modeling details are discussed in Sec. 3.1.
Secondly, a novel fabrication process is proposed to produce the
narrow gap in a cost effective manner while maintaining high out-
put signal quality from the RCDR sensor. The fabrication process
includes a deposition process of nonconductive material (silicon
nitride) layer between the electrodes. This layer will reduce the
gap size and decrease motional resistance of the device. To apply
the new fabrication process in designing the RCDR biosensor, a
circuit analysis model is developed. Analysis details are found in
Fig. 2 Schematic process of immobilizing antibody on a disk Sec. 3.2.
surface; (a) A thin gold film (about 50 nm) is sputtered onto the In order to consider the fabrication uncertainty in developing
disk surface; (b) The antibody (phage) is immobilized on top of the RCDR biosensor, we employ simulation-based robust design
the metal surface; (c) Casein is dropped on the surface to block approach. Applications of robust optimization for managing
the open space around the immobilized bio-receptor; (d) The uncertainties in electronic circuit fabrication process have been
sensor is immersed in a test solution; antigen (bacteria, virus . . .) investigated by a number of researchers [19–23]. However, in this
is captured by the antibody. work, our simulation and analysis models include ‘model structure
uncertainty’ (or ‘epistemic uncertainty’), [24] originated from the
lack of knowledge in simulation and analyses of the RCDR bio-
Due to this antibody interaction, the mass of the vibrating disk sensor. Specifically, complete knowledge about geometric condi-
will be changed and detected later via the shift of the resonant fre- tions for unstable dynamic behavior of the RCDR sensors is still
quency. The mass of the bio-entities can be estimated by meas- immature. Because of the model structure uncertainty, we employ
uring this frequency depression. Type III robust design in this design problem. Type III robust
In this work, we assumed that the biological material immobili- design is specifying adjustable ranges for control factors, which
zation will be implemented by a dip and dry process at this stage satisfy a set of output requirement targets and are insensitive to
of development. The sensor has to be immersed in liquid first to uncertainty in the system model [24]. Robust design of the RCDR
immobilize the bio-entities on the sensors’ surface. Before the biosensor is discussed in Sec. 4.
measurement in atmospheric pressure, it should be washed with In this section, we have discussed our plan for developing a
ultra pure water and dried in a nitrogen stream. The dip and dry high performance biosensor based on the RCDR. Challenges and
process may not be practical; however, researchers employ this approaches for the development are also discussed. In the next
approach for the validation of a new concept [16,17]. With further section, we discuss the simulation and analysis models of the
developments in the next stage (e.g., a secluded reaction chamber RCDR biosensor in detail.
inside the disk), our sensor will be able to operate in a liquid
environment.
3 Simulation and Analysis of RCDR Biosensors
2.2 Challenges. Several challenges must be overcome in
order to develop a high performance RCDR biosensor. First of all, 3.1 Simulation of Mass Sensitivity. Mass sensitivity (Sm) is
the immobilization and detection processes need to be modeled to a critical characteristic for biological mass sensors. A sensor with
accurately estimate the mass sensitivity of the RCDR, which is higher mass sensitivity allows us to measure tiny mass of bio-
the main performance criterion of a mass sensitive biosensor. Sec- entities much more accurately. In this paper, Sm is considered as
ondly, making a narrow gap (about 100 nm in the frequency filter the main performance indicator that we intend to maximize in our
application [11]) between the disk and electrodes (d0 in Fig. 1) to design. Mass sensitivity is degree of resonance frequency shift
achieve better output signal with less motional resistance is an ex- due to the mass change [25], and is defined as
pensive MEMS fabrication process. The fabrication process or the
device should be redesigned and modeled to realize cost effective Df 1
Sm ¼ lim (2)
and high performance biosensors. Dm!0 f0 Dm
Finally, the most important barrier in developing high perform-
ance and cost effective biosensors is the uncertainties originated where Dm is the added mass on the surface per unit area and
from the fabrication processes. These errors are originated from Df ¼ fL  f0 is the amount of resonance frequency shift in
the fabrication process of the RCDR, which include nonuniform response to the mass loading. f0 and fL are unloaded and loaded
deposition, mask design, dry and wet etching, and mask misalign- resonance frequencies, respectively. f0 and fL may be obtained by
ment in photolithography [18]. Due to these fabrication uncertain- either experiments or estimation models, but a computational
ties, real values of the geometric parameters shown in Fig. 1(a) model is required for the design later. In this work, we estimate
may be different from the intended ones. The tolerances due to those two resonance frequencies using computer simulation to
fabrication uncertainty are often from 10 to 15% of the designed consider all geometric parameters variations shown in Fig. 1 and
values in MEMS processes. Especially, d0 is one of the sensitive the loaded bio-entities.
parameters to the fabrication uncertainties; the tolerance of d0 A set of modal analyses is performed using an FEM model to
varies from 0.1 to 0.5 lm, depending on the materials (chrome, estimate the unloaded and loaded resonance frequencies. A per-
soda lime, or quartz) of the glass mask’s substrate. These inherent spective view of the loaded model for modal analysis is shown in
uncertainties may strongly affect the device performances because Fig. 3(a). As results of the FEM modal analysis of an unloaded
the tolerances in these processes are relatively large; however, disk (D ¼ 700 lm, t ¼ 40 lm and SD ¼ 40 lm), the first, second
increasing the process accuracy is often very expensive. and third resonance frequencies in radial mode are 8.1 MHz,

Journal of Mechanical Design APRIL 2012, Vol. 134 / 041005-3

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Fig. 3 Simulation model of the RCDR biosensor with the
mapped mesh in ANSYS, (a) The 3D model of the disk resonator
and the additional layer of biological entities on its surface, (b) Fig. 4 Harmonic analysis of the disk resonator; (a) relative dis-
Cross section of 3D model with material properties of the disk placement of Point A in radial and vertical directions (b) relative
resonator and biological entities [27] displacement of Point B in radial and vertical directions (Points
A and B are shown in Fig. 3(a))
21.45 MHz and 34.2 MHz, respectively, which agree with the
obtained resonance frequencies from Eq. (1). Then, the loaded capacitive [28]. The pressure reaches to 100  106 lN/lm2 when
resonance frequencies are calculated as follows. The average the design parameters, d 0 ¼ 2 lm, V p ¼ 80 V, vi ¼ 0:63V p and
thickness of a layer of antibody-antigen ranges from several nano- D ¼ 700 lm. Using the harmonic analysis, relative displacements
meters up to several micrometers, depending on the types of anti- at Point A and B within the frequency range are obtained (See
gen (bacteria, virus, etc.). In this simulation, a thin cylindrical- Fig. 3(a)) and plotted in Fig. 4. The out-of-plane component of
shape with the thickness of 100 nm is considered which is close to the vibration potentially originates from the nonzero Poisson ratio
the thickness of bacteria-phage layer [26]. This model added the of the disk material. The relative displacements at Points A and B
mass per unit area of 1.22  104 ng/(lm)2 to the resonator surface clearly indicate that out-of-plane mode vibration is ignorable
which used for calculation of Sm. The material properties of the compared with radial mode in all resonance frequencies. The rea-
layer are obtained from Ref. [27]. A cross section of the loaded son of different displacement profile in Point A comparing with
model is shown in Fig. 3(b) with material properties of disk struc- Point B is their locations on the disk. Point A is located where the
ture and additional layer of biological entities. As expected, loaded applied load is maximum. However, Point B is located in a posi-
resonance frequencies degrade (77341 Hz) when one layer of the tion that the load is zero and the displacement is only due to the
antibody and antigen is added to the resonator. A parametric study Poison ratio of the disk material. The double resonance shown in
has been performed for metamodeling mass sensitivity in Sec. 4.2. Fig. 4 only appears when there are one or more close overtones to
A set of harmonic analyses is also performed to check dynamic the main resonance frequency which does not appear in Point B.
stability of the RCDR biosensor, which is related to the amplitude
of out-of-plane vibrations when the disk is excited with the radial
load. The boundary conditions for these analyses are the same as 3.2 Analysis of Motional Resistance. The motional resist-
those of the modal analysis. However, as shown in Fig. 3(a), the ance (Rx) of the RCDR biosensor is another important characteris-
radial load is defined as a sinusoidal radial pressure tic since it is related to the required applied voltage. A lower Rx
ðPðtÞ ¼ P cosðxtÞÞ, on 60 deg of the half disk edge since the decreases the required supply voltage. Different researchers pro-
input electrode covers u1 ¼ 120 deg at one side of the disk as posed approaches to improve Rx [11,12]. Weinstein et al. used the
shown in Fig. 1. The applied electrostatic pressure has the ampli- nitride gap to reduce Rx for bar resonator [29]. Ayazi et al. also
tude of P ¼ 100  106 lN/lm2, and frequency range of 0–40 improved the motional resistance by proportional downscaling all
MHz. The pressure is calculated by F0 ¼ V p vi ðC0 =d0 Þ, where V p the dimensions (including the gap sizes) as well as the polariza-
is the applied bias voltage, vi is the alternating input voltage, d 0 is tion voltage [30]. Since Rx proportionally decreases with respect
the capacitive gap and C0 is the electrode-to-resonator overlap to d04 ½12, the best way is reducing the gap size (d0), which

041005-4 / Vol. 134, APRIL 2012 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Q ¼ 1012 =f0 for the silicon bulk mode resonators working in air
environment [31]. g1 and g2 are the electromechanical coupling
factors of the reduced gap derived as follows.
The initial capacitances of C10 and C20 (refer to Fig. 6) with sili-
con nitride (Si3N4) deposition are modeled as
 
M K1 K2
C0 ¼ e 0 A ; M ¼ 1 and 2 (4)
K1 d2 þ K2 d1

where A ¼ Run t, d2 ¼ d0  d1 is the total thickness of Si3N4,


and K1 and K2 are dielectric constants of the air and Si3N4, respec-
tively (K1 ¼ 1; K2 ¼ 7:8). When applying the AC and bias vol-
tages to the resonator, the time-varying input/output electrode-to-
resonator overlap capacitance is
Fig. 5 Schematic view of capacitor after deposition of Si3N4. K1 K2 CM0
The initial 2um air gap reduced with Si3N4 to improve the electri- Cn ðr Þ ¼ e0 A ¼  (5)
cal characteristic of the sensor. K1 d2 þ K2 ðd1  r Þ K2 r
1
K1 d2 þ K2 d1

where, CM 0 is the initial capacitance and r ðtÞ is the radial dis-


placement. Electromechanical coupling coefficients of the two
ports are
 
@cn ðr Þ
gn ¼ Vp ; where n ¼ 1 and 2 (6)
@r

and
0 12
@Cn ðr Þ K2 B 1 C
¼ CM ¼B C (7)
@r 0
K1 d2 þ K2 d1 @ K2 r A
1
K1 d2 þ K2 d1

If displacements are sufficiently small, Eq. (7) can be expanded to


obtain a more useful form:

@Cn ðr Þ K2  
¼ CM
0 ¼ 1 þ A1 r þ A1 r 2 þ    (8)
@r K 1 d2 þ K 2 d1
 2
K2 K2
where A1 ¼ , A2 ¼ 3 ;…
K 1 d2 þ K 2 d1 K 1 d2 þ K 2 d1
Fig. 6 Equivalent circuit of disk resonator and its electrodes For the present purpose of obtaining a linear model for the disk
using to calculate the motional resistance (Rx) resonator, @Cn ðr Þ=@r can be approximated by the first term in
Eq. (8)
requires an expensive fabrication process (such as E-Beam Li-
@Cn ðr Þ K2
thography). However, as mentioned in Sec. 2.3, we propose a  CM
0 (9)
method to decrease Rx without making the smaller d0. The idea is @r K 1 d2 þ K 2 d1
to fill the gap space by depositing a dielectric material, such as sil-
icon nitride (Si3N4), using chemical vapor deposition (CVD) pro- With Eqs. (3), (6), and (9), the motional resistance of the reduce
cess. Filling the capacitive gap with silicon nitride helps to gap RCDR is
achieve smaller gap distance in comparison with air gap counter-
parts [11]. A schematic figure of the reduced gap with silicon ðK1 d2 þ K2 d1Þ4 ceq
nitride deposition is shown in Fig. 5. The dielectric material helps Rx ¼   (10)
e20 A2 VP2 K22 K24
to reduce fabrication cost and improve the electrical characteris-
tics of the capacitor at the same time.
An analytical equation is derived and validated with a COM- The obtained results for motional resistance in Eq. (10) are plotted
SOL simulation model to calculate the motional resistance of the in Fig. 7. The dashed line shows the motional resistance of a reso-
reduced gap RCDR with a dielectric material layer. The RCDR nator with initial gap of 2 lm and different thicknesses for Si3N4.
shown in Fig. 1 is replaced with its equivalent electrical circuit as The solid line shows the motional resistance of RCDR without
shown in Fig. 6. any insulating layer (i.e., d2 ¼ 0; d0 ¼ d1), which is used by
The motional resistance of disk resonator can be calculated Clark et al. [11].
using the following equation [28]: As shown in Fig. 7, the motional resistance with 2 lm air gap is
over 500 M-Ohm, for which the output signal is not measurable.
Rx ¼ ceq =ðg1 g2 Þ (3) Making the small gap sizes leads to low level of motional resist-
ance (solid line). However, the motional resistance of the reduced
where ceq is a damping element of the disk identified based on gap (dashed line) is still in the acceptable range while avoiding
Ref. [11]. The Q-factor for calculating ceq can be estimated using the expensive and difficult fabrication processes.

Journal of Mechanical Design APRIL 2012, Vol. 134 / 041005-5

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Fig. 9 A simulation chain of performance analysis

The motional resistance of the reduced gap resonator decreases


incredibly in comparison with that of the air gap resonator. There-
Fig. 7 Motional resistance of the reduced gap resonator ver- fore, using reduced-gap design helps the designer to arrange a fab-
sus the air gap resonator design (Clark et al. [11]) for different rication process flow to get a much smaller gap size and also
gap size (d1) achieve acceptable motional resistances.
In this section, the mass sensitivity of the RCDR biosensor is
For the purpose of validation, a simulation model for motional modeled with FEM models. Then, a theoretical model is devel-
resistance is developed using COMSOL MultiphysicsV. In the
R
oped to obtain the motional resistance of the biosensor consider-
simulation, slide and squeeze film damping of the surrounding air ing an added Si3N4 layer in the gap space. Afterwards, the
and structural damping of the resonator are modeled to take the obtained model is verified with a full simulation model. In the
energy dissipation (due to vibration) into account. The simulated next section, we discuss design of the RCDR biosensor based on
motional resistance for air gap and reduced gap resonators are the simulation and analysis models described in this section.
plotted versus bias voltages in Fig. 8. The motional resistance
from the theoretical equation obtained by Clark et al. [28] are 4 Simulation Based Robust Design of RCDR
plotted with solid line, and those of the reduced gap based on our
derivation, Eq. (10), with dashed line. Corresponding results of
Biosensors
our simulation models of air and reduced gaps are illustrated in As mentioned in Sec. 2.3, our design approach taken in this
the figure. study is simulation-based robust design. For the purpose of com-
The result trends obtained from the simulation generally agree parison, we consider two design scenarios, deterministic optimiza-
with the theoretical results from Eq. (10). The difference between tion and robust optimization. The simulation and analysis models
the simulated and theoretical results is caused by the difference in explained in Sec. 3 are employed in the optimization process of
the system damping. In our simulation model, only slide and these scenarios and fabrication uncertainties mentioned in Sec.
squeeze film damping due to air viscosity and the structural damp- 2.2 are considered in this robust design. As shown in Fig. 9,
ing are considered. All other unknown damping factors in the real D; t; and SD are input parameters of the mass sensitivity simula-
system further decrease the Q-factor in practice, which has been tion model and D; t; d0; d2; Vp; and Vac are those of the motional
experimentally measured and used in calculating Eq. (10). There- resistance analysis model. Among these parameters, D and t are
fore, the simulated motional resistances are lower than the calcu- shared input parameters of the both models. In the robust design
lated motional resistance. optimization, input parameters are classified into control and noise
factors. D, t, SD, d2, Vp, and Vac are control factors, and d0 is the
noise factor.
Considering fabrication limits, the initial gap size (d0) is set as
one of the noise factors since the minimum achievable d0 is 2 lm
with normal lithography process [32] and the tolerance of the gap
size is 10%. Si3N4 layer thickness is controllable and its tolerance
is negligible. Vp and Vac from the impedance analyzer are also
controllable very accurately. Therefore, in this study, d2, Vp and
Vac are control factors without noise. However, D, t, and SD may
include fabrication uncertainty up to 10% of the feature size and
those are considered as control factors with noise. The ranges of
all control and noise factors and those tolerances considered in
this design are listed in Table 2. These tolerances are identified
based on the specifications of traditional fabrication equipment
[33], a voltage supplier (MIC POWERSUP/2807), and a network
analyzer (AGILENT 4294A).
Section 4 consists of design constraints formulation in Sec. 4.1,
a metamodeling approach for efficient robust design optimization
in Sec. 4.2, problem formulation for the two design scenarios in
Sec. 4.3, and design results and discussion in Sec. 4.4.

Fig. 8 The motional resistance of the RCDR versus bias volt- 4.1 Constraints for Design Exploration. In addition to the
age at Vac50.1 V; from the top, simulation for air gap, theoreti- two main performance of the RCDR, additional design constraints
cal equation for air gap by Clark et al., Eq. (10) for reduced gap are identified. In our fabrication process, lithography is one of the
and simulation for reduced gap main processes applied, which can achieve trenches and holes

041005-6 / Vol. 134, APRIL 2012 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Table 2 Ranges and tolerance of control and noise factors sinusoidal signal). Therefore, the breakdown voltage constraints
of the two capacitors are
Parameters Ranges Tolerance (3r)
Z01 ðI2  I1 Þ  Es d1  0 and Z02 ðI4  I5 Þ  Es d1  0
Disk diameter (D) 100 lm  D  1000 lm 610% of feature size
Disk thickness (t) 2 lm  t  60 lm 610% of feature size
Stem diameter (SD) 2 lm  SD  60 lm 610% of feature size All aforementioned design ranges, constraints conditions and
Bias voltage (Vp) 1 (V)  Vp  40 (V) 60.001% of voltage design goals are summarized in optimization formulation in Sec.
AC voltage (Vac) 0.01 (V)  Vac  1 (V) 60.001% of voltage 4.3.
Si3N4 thickness (d2) 0.5 lm  d2  1.5 lm 60.0001 lm
Initial gap space (d0) d0 ¼ 2 lm 60.2 lm
4.2 Metamodel for Mass Sensitivity. Mass sensitivity is
estimated based on two different computationally intensive mod-
els as mentioned in Sec. 3. In robust design, this time consuming
with aspect ratio of 30. This ratio should be considered between
models should be replaced by a surrogate model (or metamodel)
disk thickness (t) and gap space (d0) in this case. Therefore, first
with an acceptable level of precision for efficient design explora-
constraint condition is
tion. In this study, we employ Kriging to replace the computer
t=d0  30 simulation and manage nonlinear the system response. Kriging is
originated from Geostatistics but widely used in engineering
design to replace time-consuming deterministic computer simula-
Second constraint is a geometrical constraint of disk resonator. As tions. Moreover, it is more flexible than other surrogate models,
mentioned in Sec. 3, the main assumption is that the disk resona- such as Response Surface Model, for us to represent highly non-
tor vibrates in radial mode. This assumption is true when the disk linear behavior of systems. Kriging model is expressed as
is not too thin or thick in comparison with its radius. In other
yðxÞ ¼ f ðxÞT b þ Z ðxÞ, where x denotes input parameters, f ðxÞT b
word, if the thickness of disk is very small in comparison with its is a vector of regression models that globally approximates the
radius, it may vibrate in undesired natural modes (tangential or space. The Gaussian stochastic process, ZðxÞ, is the amount of
vertical) rather than the radial mode. The same consideration must localized deviations from the regression model in order to interpo-
be taken into account for the ratio of radius and stem diameter. late observation points. Please refer the work of Jones and his co-
Therefore, the second constraint condition for satisfying dynamic
authors [36] for more details of Kriging.
stability of the RCDR is With Latin Hypercube sampling, 180 experimental points are
obtained based on the unloaded and loaded resonance frequency
5  D=t  25 and 5  D=SD  25
simulation models (refer Fig. 10). In Fig. 10, Kriging model for
the mass sensitivity is plotted at SD ¼15 lm. In this model, first-
However, these constraints are uncertain due to the limited knowl- order global regression function best suits to our requirements
edge about dynamic stability of the RCDR, which is model struc- (simplicity and accuracy) and the stochastic process are modeled
ture uncertainty (or epistemic uncertainty) as mentioned in Sec. 2.2. by Gaussian correlation function. From the model, we can observe
This fact should be considered while designing RCDR biosensors. that the achieved mass sensitivities of the RCDR biosensor are
We will further discuss our approach for this issue in Sec. 4.4. very high, comparing with those of other mass sensitive biosen-
Lastly, the voltage of both capacitors (C01 and C02 in Fig. 6) sors listed in Table 1. The main reason of this high mass sensitiv-
should not exceed its breakdown voltage. The breakdown voltage ity of the square resonator in comparison with state-of-the-art
of a capacitor with air in its gap is Vbreakdown ¼ Es  d1 where Es , mass sensors [37,38] is its relatively thin resonator structure
the dielectric strength of air, is 107 V=m [34,35]. Voltage of the (2 lm).
capacitors are VC01 ¼ Z01 ðI2  I1 Þ and VC02 ¼ Z02 ðI4  I5 Þ, where
I1 ; I2 ; I4 , and I5 are the currents calculated from the circuit shown
in Fig. 6 and Z01 and Z02 are capacitor impedances (Z0i ¼ 1ðjCi xÞÞ, 4.3 Design Exploration Formulation. In this study, we for-
where j is the imaginary unit and x is the angular velocity of the mulate two design scenarios, which are multiobjective

Fig. 10 Kriging model of mass sensitivity based on sampling points obtained from simulation

Journal of Mechanical Design APRIL 2012, Vol. 134 / 041005-7

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Table 3 Deterministic optimization formulation 2.3, we need to formulate the robust design problem based on
Type III robust design. In this work, we employ error margin indi-
Given ces (EMIs) to implement Type III robust design problem since the
FSm: Kriging model for mass sensitivity developed in Sec. 4.2. dynamic stability constraints in Sec. 4.1 are still immature at this
FRx: Model for motional resistance developed in Sec. 3.2
time. As shown in Fig. 11, an EMI is calculated based on the
Find mean of a performance (ly) and upper/lower deviations (Dyupper
D, t, SD, d2, Vp, and Vac and Dylower) from specified performance limitations (upper
Satisfy requirement limit (URL) or lower requirement limit (LRL)) or
n Design space: constraint boundaries.
D ¼ ½100; 1000; t ¼ ½2; 40; SD ¼ ½2:40 In the case where ‘Smaller is Better’, EMI¼(URL-ly)/ Dyupper,
d2 ¼ ½0:5; 1:5; Vp ¼ ½1; 40; and Vac ¼ ½0:01; 1 which means that the EMI becomes larger when the location of ly
is farther from a upper requirement limit (URL) and/or Dyupper
n Fabrication limitation: gets the smaller. In all cases depicted in Fig. 11, the EMI becomes
g1ðt; d0 Þ ¼ t  30  d0  0; where d0 ¼ 2 lm larger as decisions become more reliable. Specifically, the EMI is
n Dynamic stability constraints:
g2ðD; tÞ ¼ 5t  D  0; g3ðD; tÞ ¼ D  25t  0 larger than zero when the mean of a performance is better than
g4ðD; SDÞ ¼ 5SD  D  0; and g5ðD; SDÞ ¼ D  25SD  0 URL or/and LRL. The EMI is larger than one when performance
n Breakdown voltage constraints: deviation limit is better than URL or/and LRL. The EMIs, formu-
g6ðD; t; d2; Vp; VacÞ ¼ Z01  ðI2  I1 Þ  Es  d1  0 and lated in this manner, are leveraged in a solution search algorithm
g7ðD; t; d2; Vp; VacÞ ¼ Z02  ðI4  I5 Þ  Es  d1  0 to find solution sets which are robust to uncertainty in constraints
as well as input variables. EMI is used in solution search algo-
where d1 ¼ d0  d2; and Es ¼ 107 V=m
rithms to find the robust optimum point of the sensor design.
Minimize More detail descriptions about Error Margin Indices (EMIs) are
 þ
z ¼ w1  dSm þ w2  dRx , found in Ref. [40]. Based on the construct of EMI, a robust design

where dSm ¼ 1  FSm ðD; t; SDÞ=T Sm and optimization problem of the RCDR biosensor is formulated as
þ
dRx ¼ FRx ðD; t; d2; d0 ; Vp; VacÞ=T Rx  1
shown in Table 4.
In this cDSP formulation, we search for the value (or mean
value) of the control factors, assuming the uncertainty in the fac-
tors (i.e., D, t, SD, and d0) are normally distributed. The devia-
deterministic optimization problem and robust design problem, tions of uncertain parameters are set as three time of those
for the purpose of comparison. In Scenario 1, a traditional deter- standard deviations, as listed in Table 2. The lower limitation of
ministic optimization problem, our device is designed to maxi- mass sensitivity is set as 60 lm2 =ng the highest mass sensitivity
mize the mass sensitivity (Sm) as well as minimize the motional available from the current mass detection based biosensors. The
resistance (Rx), satisfying the fabrication limit, dynamic stability, upper limitation of motional resistance is set as 500 M-ohm from
and breakdown voltage constraints. This multiobjective design the network analyzer, AGILENT 4294A, specifications. Seven
problem is formulated based on compromise decision support liner and nonlinear constraints shown in Table 3 are converted
problem (cDSP) [39] as shown in Table 3. into EMI constraints (i.e., EMIgi) considering the variations of the
In this cDSP formulation (Given, Find, Satisfy, Minimize), we uncertainty parameters. The values of the EMI constraints must be
find six design parameters with given Kriging model for mass sen- greater than or equal to unity. EMIs of the RCDR performances
sitivity and motional resistance model. The minimum achievable (i.e., EMISm and EMIRx) must also be greater than equal to unity
gap of 2 lm is set as constant parameter here. Five linear and two to satisfy the performance requirements. Finally, we set a multiob-
nonlinear constraints are formulated as the fabrication, dynamic jective function based on the goal programming formulation. The
stability, and breakdown voltage constraints. The multiobjective performance goals (i.e., EMISm_goal and EMIRx_goal) are set as the
function is an Archimedean formulation [39] with the goal pro- EMIs obtained from two separate single objective optimization
gramming approach. The target values, TSm and TRx, are the opti- problems. Calculating all EMIs, we set the performance devia-
mum results of two separate single objective optimization tions as three times of estimated standard deviations. The standard
problems. deviations are estimated based on the second moment method and
Scenario 2 is multiobjective robust design problem to maximize the partial derivatives are numerically calculated. However, the
the mean of Sm and minimize the mean of Rx, minimizing the motional resistances best fit a lognormal distribution rather than a
deviations of the two RCDR performances. As mentioned in Sec. normal distribution; therefore, upper deviation (DRxupper of the

Fig. 11 Mathematical construct of error margin indices [40]

041005-8 / Vol. 134, APRIL 2012 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Table 4 EMI-based robust design formulation

Given
x ¼ fD; t; SDgð3r~x ¼ 10% of mean sizeÞ
Uncertain design variables (control factors with noise): ~
Deterministic design variables (control factors without noise): x¼fVp, Vac, d2g
Uncertain fixed parameters (noise factor): d0 (ld0 ¼ 2, 3rd0 ¼ 0:2)
FSm: Kriging model for mass sensitivity developed in Sec. 4.2
LNRx: Model for log motional resistance developed in Sec. 3.2
Find
l~x and x
Satisfy
(1) Design space:
     
lD ¼ 100 þ 3rD; 1000  3rD ; lt ¼ 2 þ 3rt; 40  3rt ; lSD ¼ 2 þ 3rSD; 40  3rsD ;
d2 ¼ ½0:5; 1:5; Vp ¼ ½1; 40; and Vac ¼ ½0:01; 1
(2) Constraints for fabrication limit, dynamic stability, and breakdown voltage
URL  lgi
EMIgi ¼  1; where i ¼ 1; …; 7 ðnumber of constraintsÞ
3rgi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

 2  
P3 @gi @gi 2 2
URL ¼ 0; lgi ¼ giðl~x ; x; ld0 Þ and rgi ¼  r2x~j þ  rd0
j¼1 @~ xj @d0
(3) EMI formulation for the RCDR performances
EM ISm ¼ ðlSm  LRLSm Þ=3rSm  1; where LRLSm ¼ 60; lSm ¼ FSm ðl~x Þ, and
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uX  
u 3 @FSm 2 2
rSm t rx~j
j¼1
@~ xj
   
EM IRx ¼ URLRx  exp lRx þ 0:5r2Rx =DRxupper  1; where URLRx ¼ 500  106 lRx
¼ LNRx ðlD ; lt ; ld0 ; xÞ; DRxupper ¼ expðlRx Þ  expðlRx þ 3rRx Þ and
s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
     ffi
@LNRx 2 2 @LNRx 2 2 @LNRx 2 2
rRx ¼ rD þ rt þ rd0
@D @t @d0
Minimize
 
z ¼ w1  dEM I Sm þ w2  dEM I Rx
 
where dEM I Sm ¼ 1  EM ISm =EM ISm goal and dEM I Rx ¼ 1  EM IRx =EM IRx goal

lognormal distribution are calculated in a numerical manner. In located behind of that of deterministic optimization, and the per-
our case, the numerical second moment method is a reasonable formances gap between the two design results is much larger at
approach for uncertainty analysis because our system performan- higher mass sensitivity. It is because the two RCDR performances
ces are not highly nonlinear and the standard deviations of the are much more sensitive to the fabrication uncertainty at the
design ables are small enough. This assumption will be validated smaller scale.
in the next section. From the EMI robust design exploration, combinations of the
two performances among the Pareto sets are list in Table 5. The
motional resistances achieved in the design are lower than the
4.4 Design Results and Discussion. For the purpose of com- URL (i.e., 500 M-Ohm). Therefore, it is better to select the higher
parison, we consider two scenarios, the RCDR biosensors w/o mass sensitivity design among the solution sets, which is bold
Si3N4. First, the design exploration results with Si3N4 are illus- faced in the table. The mean of achieved mass sensitivity of the
trated in Fig. 12. As expected, the results of the EMI based robust RCDR biosensors reaches up to 261.2 lm2/ng, which is about five
design (Table 4) are more conservative than those of deterministic times higher than those of existing BAW/SAW sensors. In addi-
optimization (Table 3). The Pareto frontier of the robust design is tion, with 99.9% confidence, it is designed to have the minimum
sensitivity of 253.6 lm2/ng even if fabrication tolerances are the
worst. In case of using a measurement device with lower accu-
racy, the design for lower motional resistance would be the better
selection. The mean of motional resistance may decrease down to
6.484 M-Ohm and the upper bound 14.74 M-Ohm with 99.9%
confidence.
We have discussed the uncertain constraints (from g2 to g5)
due to the limited knowledge in RCDR dynamic stability condi-
tion in Sec. 4.1. We believe that it should be a safer choice to find
a solution with the higher EMI for the uncertain constraints while
selecting the best robust design. The minimum EMIs (i.e., mini-
mum[EMIg2-5]) among the four EMI constraint evaluations at the
Pareto optimum sets are also listed in Table 5. Among the Pareto
optimum sets, minimum[EMIg2-5] at the selected (i.e., bold faced)
RCDR design are the largest; therefore, our selection would be a
safer choice under the potential errors in the dynamic stability
constraints.
In the EMI-based robust design formulation in Table 4, the
uncertainty analyses for the constraints and performances are per-
Fig. 12 Feasible region of deterministic and EMI based robust formed in a numerical manner for efficient design exploration. For
design with Si3N4 layer the nonlinear models (i.e., breakdown voltage constraint, motional

Journal of Mechanical Design APRIL 2012, Vol. 134 / 041005-9

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


Table 5 Pareto optimum sets in the EMI based robust design

Mean D Mean t Mean SD d2 Vp Vac Minimum Sm Lower Rx Upper


(lm) (lm) (lm) (lm) (V) (V) EMISm EMIRx [EMIg2-5] (lm2=ng) Sm (M-Ohm) Rx

680 32 38 1.30 40 0.01 1.006 59.77 1.765 72.70 60.08 6.484 14.74
680 32 39 1.30 40 0.01 1.007 59.77 1.765 73.29 60.09 6.484 14.74
700 31 36 1.30 40 0.01 1.502 59.60 1.072 74.38 64.81 6.502 14.78
… … … … … … … … … … … … …
110 6 8 1.30 40 0.01 26.68 1.053 3.636 261.2 253.6 221.9 486.0

Fig. 14 Feasible region of deterministic and EMI based robust


design without Si3N4 layer

resistance, and mass sensitivity models), accuracy of the numerical


uncertainty analysis in Table 4 is checked at the selected design
point, which is located in the relatively high nonlinear area of the
models. With 10,000 latin hypercube (LHC) sampling points for
the four multivariate random variables (i.e., D, t, SD, and d0), three
probability distributions of the results are shown in Fig. 13.
Normal probability distributions certainly fit the samples of
breakdown voltage constraint and mass sensitivity models, and a
lognormal probability distribution also does the samples of
motional resistance model. The results of the numerical uncer-
tainty estimation and LHC sampling are compared in Table 6 and
the level of accuracy is shown to be high enough.
Reduced gap with Si3N4 layer is found to be beneficial for
developing high performance and cost effective RCDR biosensor.
As shown in Fig. 14, the feasible regions of both deterministic
optimization and EMI-based robust design formulations are much
reduced for the RCDR without the Si3N4 layer. The maximum
achievable mass sensitivity is about 100 lm2/ng with 250 M-ohm
for the motional resistance. Compared to the mass sensitivities of
conventional mass detection based immunosensors, it is not high
enough.

Fig. 13 Probability distributions of LHC sampling results


based on the nonlinear models; (a) breakdown voltage con-
5 Conclusion
straint (normal distribution), (b) motional resistance(lognormal In this work, we design a biological mass sensor based on a ra-
distribution), and (c) mass sensitivity (normal distribution) dial contour-mode disk resonator. While designing the sensors,
Table 6 Comparison of uncertainty analysis results between LHC sampling and numerical estimation

10000 LHC sampling Numerical estimation

Mean Lower limit (99.9%) Upper limit (99.9%) Mean Lower limit (99.9%) Upper limit (99.9%)

Breakdown voltage constraint 2.868 1.088 2.859 1.087


Motional resistance (M-ohm) 220.3 485.4 221.9 486.0
Mass sensitivity (lm2=ng) 261.2 253.6 261.2 253.6

041005-10 / Vol. 134, APRIL 2012 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms


we focus on the challenges related to modeling and analysis of the [14] Heitmann, V., Reib, B., and Wegener, J., 2007, “The Quartz Crystal Microba-
sensor performances, low cost fabrication processes, and robust lance in Cell Biology: Basics and Applications,” Piezoelectric Sensors, 5, pp.
303–338.
design of the biosensor under fabrication uncertainties to develop [15] Ward, M. D., and Delawski, E. J., 1991, “Radial Mass Sensitivity of the Quartz
a high performance and cost effective biosensors. Crystal Microbalance in Liquid Media,” Anal. Chem., 63(9), pp. 886–890.
We developed and validated a series of simulation models for [16] Xu, T., Wang, Z., Miao, J., Yu, L., and Li, C. M., 2008, “Micro-Machined Pie-
estimating mass sensitivities of the sensors. Modal and harmonic zoelectric Membrane-Based Immunosensor Array,” Biosens. Bioelectron.,
24(4), pp. 638–643.
analysis models are developed simulating immobilization process [17] Zuniga, C., Rinaldi, M., Khamis, S. M., T. S., Jones, T. S., Johnson, A. T., and
of immunosensors. From a set of modal analyses, we identified Piazza, G., 2009, “DNA-Decorated Carbon Nanotubes as Sensitive Layer for
resonant frequencies of an empty disk and a loaded disk with bio- AlN Contour-Mode Resonant-MEMS Gravimetric Sensor,” Proceedings of
entities, from which the mass sensitivity of the biosensors is esti- 22nd IEEE International Conference on Micro Electro Mechanical Systems
(MEMS), Sorrento, Italy, pp. 320–323.
mated. From the harmonic analyses, we checked the dynamic sta- [18] Shavezipur, M., Ponnambalam, K., Khajepour, A., and Hashemi, S. M., 2008,
bility of the RCDR. As expected, the dynamic behavior of the “Fabrication Uncertainties and Yield Optimization in MEMS Tunable Capaci-
RCDR was very stable in radial excitation mode. Depositing a tors,” Sens. Acutators, A, 147(2), pp. 613–622.
dielectric material (such as Si3N4) layer within the gap between [19] Dewey, A., Ren, H., and Zhang, T., 2000, “Behavioral Modeling of Microelec-
tromechanical Systems (MEMS) With Statistical Performance-Variability
disk and electrode, we could achieve a device with lower motional Reduction and Sensitivity Analysis,” IEEE Trans. Circuits Syst., II: Analog
resistance while maintaining fabrication cost low. The circuit Digital Signal Process, 47(2), pp. 105–113.
analysis model for devices with the dielectric material layer has [20] Fan, Z., Wang, J., and Goodman, E., 2005, “An Evolutionary Approach for Robust
been developed and validated by the results of a multiphysics sim- Layout Synthesis of MEMS,” Proceedings of IEEE/ASME International Confer-
ence on Advanced Intelligent Mechatronics, Monterey, California, pp. 1186–1191.
ulation model. [21] Olivieri, M., Scotti, G., and Trifiletti, A., 2005, “A Novel Yield Optimization
In this paper, we highlight the issue of uncertainties originated Technique for Digital CMOS Circuits Design by Means of Process Parameters
from MEMS fabrication processes. A disk resonator based biosen- Run-Time Estimation and Body Bias Active Control,” IEEE Trans. Very Lagre
sor is designed to be insensitive to the fabrication uncertainty to Scale Integr. (VLSI) Syst., 13(5), pp. 630–638.
[22] Shavezipur, M., Ponnambalam, K., Hashemi, S. M., and Khajepour, A., 2008,
realize low cost mass product of the biosensors. Our approach for “A Probabilistic Design Optimization for MEMS Tunable Capacitors,” Micro-
managing the uncertainty is simulation-based robust design. For electron. J., 39(12), pp. 1528–1533.
the efficient robust design exploration, the computationally inten- [23] Sigmund, O., 2009, “Manufacturing Tolerant Topology Optimization,” Acta
sive simulation model has been replaced with a Kriging model. Mech. Solida Sinica, 25(2), pp. 227–239.
[24] Allen, J. K., Seepersad, C., Choi, H. J., and Mistree, F., 2006, “Robust Design
An Error Margin Index based robust design problem is formulated for Multiscale and Multidisciplinary Applications,” ASME J. Mech. Des.,
to manage various types of uncertainties. From the design results, 128(4), pp. 832–843.
we realized that highly sensitive (up to 250 lm2/ng) mass detec- [25] Janshoff, A., Galla, H.-J., and Steinem, C., 2000, “Piezoelectric Mass-Sensing
tion based biosensors may be mass-produced with low cost fabri- Devices as Biosensors—An Alternative to Optical Biosensors?,” Angew.
Chem., 39(22), pp. 4004–4032.
cation processes (i.e., traditional MEMS fabrication methods) [26] Huang, S., Yang, H., Lakshmanan, R. S., Johnson, M. L., Wan, J., Chen, I. H.,
based on our approach and design. Wikle Iii, H. C., Petrenko, V. A., Barbaree, J. M., and Chin, B. A., 2009,
“Sequential Detection of Salmonella Typhimurium and Bacillus Anthracis
Acknowledgment Spores Using Magnetoelastic Biosensors,” Biosens. Bioelectron., 24(6), pp.
1730–1736.
This research was supported by Basic Science Research Pro- [27] Ivanovska, I. L., de Pablo, P. J., Ibarra, B., Sgalari, G., MacKintosh, F. C., Car-
gram through the National Research Foundation of Korea (NRF) rascosa, J. L., Schmidt, C. F., and Wuite, G. J. L., 2004, “Bacteriophage Cap-
sids: Tough Nanoshells With Complex Elastic Properties,” Proc. Natl. Acad.
funded by the Ministry of Education, Science and Technology Sci. U.S.A., 101(20), pp. 7600–7605.
(Grant No. 2011-0026468). [28] Jing, W., Ren, Z., and Nguyen, C. T. C., 2004, “1.156-GHz Self-Aligned
Vibrating Micromechanical Disk Resonator,” IEEE Trans. Ultrason. Eng.,
51(12), pp. 1607–1628.
References [29] Weinstein, D., and Bhave, S. A., 2007, “Internal Dielectric Transduction of a
[1] Grieshaber, D., MacKenzie, R., Vörös, J., and Reimhult, E., 2008, 4.5 GHz Silicon Bar Resonator,” IEEE International Electron Devices Meeting,
“Electrochemical Biosensors-Sensor Principles and Architectures,” Sensors, IEEE, New York, Vols. 1 and 2, pp. 415–418.
8(3), pp. 1400–1458. [30] Pourkamali, S., Ho, G. K., and Ayazi, F., 2007, “Low-Impedance VHF and
[2] Trojanowicz, M., 2009, “Main Concepts of Chemical and Biological Sensing,” UHF Capacitive Silicon Bulk Acoustic-Wave Resonators; Part II: Measurement
Combinatorial Methods for Chemical and Biological Sensors, pp. 25–60. and Characterization,” IEEE Trans. Electron Devices, 54(8), pp. 2024–2030.
[3] Janata, J., 2009, Principles of Chemical Sensors, Springer, New York. [31] Khine, L., and Palaniapan, M., 2009, “High- Q Bulk-Mode SOI Square Resonators
[4] Hoummady, M., Campitelli, A., and Wlodarski, W., 1997, “Acoustic Wave with Straight-Beam Anchors,” J. Micromech. Microeng., 19(1), p. 015017.
Sensors: Design, Sensing Mechanisms and Applications,” Smart Mater. Struct. [32] Koester, D., Cowen, A., Mahadevan, R., Stonefield, M., and Hardy, B., 2003,
6(6), pp. 647–657. MUMPs Design Handbook, Cronos Integrated Microsystems, Research Trian-
[5] Koller-Lucae, S., 1999, “Resonating Viscosity Sensors: Micro Versus Macro gle Park, NC.
Approach,” Ph.D. thesis, ETH, Zurich. [33] Bangtao, C., and Miao, J., 2007, “Influence of Deep RIE Tolerances on Comb-
[6] Nirschl, M., Blüher, A., Erler, C., Katzschner, B., Vikholm-Lundin, I., Auer, S., Drive Actuator Performance,” J. Phys. D, 40(4), pp. 970–976.
Vörös, J., Pompe, W., Schreiter, M., and Mertig, M., 2009, “Film Bulk Acoustic [34] Howatson, A. H., 1976, An Introduction to Gas Discharges, Pergamom, New York.
Resonators for DNA and Protein Detection and Investigation of In Vitro Bacte- [35] Pai, S. T., and Zhang, Q., 1995, Introduction to High Power Pulse Technology,
rial S-Layer Formation,” Sens. Actuators, A, 156(1), pp. 180–184. World Scientific, Singapore.
[7] Ogi, H., Naga, H., Fukunishi, Y., Hirao, M., and Nishiyama, M., 2009, “170- [36] Jones, D. R., Schonlau, M., and Welch, W. J., 1998, “Efficient Global Optimi-
MHz Electrodeless Quartz Crystal Microbalance Biosensor: Capability and Limi- zation of Expensive Black-Box Functions,” J. Global Optim., 13(4), pp.
tation of Higher Frequency Measurement,” Anal. Chem., 81(19), pp. 8068–8073. 455–492.
[8] Tigli, O., and Zaghloul, M. E., 2010, “Surface Acoustic Wave (SAW) Bio- [37] Marx, K. A., 2003, “Quartz Crystal Microbalance: A Useful Tool for Studying
sensors,” Proceedings of the 53rd IEEE International Midwest Symposium on Thin Polymer Films and Complex Biomolecular Systems at the Solution-
Circuits and Systems (MWSCAS), Seattle, WA, pp. 77–80. Surface Interface,” Biomacromolecules, 4(5), pp. 1099–1120.
[9] Lazcka, O., Campo, F. J. D., and Muñoz, F. X., 2007, “Pathogen Detection: A [38] Shen, W., Mathison, L. C., Petrenko, V. A., and Chin, B. A., 2010, “Design and
Perspective of Traditional Methods and Biosensors,” Biosens. Bioelectron., Characterization of a Magnetoelastic Sensor for the Detection of Biological
22(7), pp. 1205–1217. Agents,” J. Phys. D, 43(1), pp. 015004–015013.
[10] Johnson, R. A., 1983, Mechanical Filters in Electronics, Wiley, New York. [39] Bras, B., and Mistree, F., 1995, “A Compromise Decision Support Problem for
[11] Clark, J. R., Hsu, W. T., Abdelmoneum, M. A., and Nguyen, C. T. C., 2005, Axiomatic and Robust Design,” ASME J. Mech. Des., 117(1), pp. 10–19.
“High-Q UHF Micromechanical Radial-Contour Mode Disk Resonators,” [40] Choi, H.-J., Austin, R., Allen, J. K., McDowell, D. L., Mistree, F., and Benson,
J. Microelectromech. Syst., 14(6), pp. 1298–1310. D. J., 2005, “An Approach for Robust Design of Reactive Powder Metal Mix-
[12] Clark, J. R., Abdelmoneum, M. A., and Nguyen, C. T. C., 2003, “UHF High- tures Based on Non-deterministic Micro-Scale Shock Simulation,” J. Comput.-
Order Radial-Contour-Mode Disk Resonators,” Proceedings of the IEEE Inter- Aided Mater. Des., 12(1), pp. 57–85.
national Frequency Control Symposium and PDA Exhibition Jointly With the [41] Zimmermann, C., Rebiere, D., Dejous, C., Pistre, J., and Planade, R., 2002,
17th European Frequency and Time Forum, Tampa, Florida, pp. 802–809. “Love-Waves to Improve Chemical Sensors Sensitivity: Theoretical and Exper-
[13] Clark, J. R., Wan-Thai, H., and Nguyen, C. T. C., 2000, “High-Q VHF Micro- imental Comparison of Acoustic Modes,” Proceedings of the IEEE Interna-
mechanical Contour-Mode Disk Resonators,” Tech. Dig. – Int. Electron. Devi- tional Frequency Control Symposium & PDA Exhibition, New Orleans,
ces Meet., San Francisco, California, pp. 493–496. Louisiana, pp. 281–288.

Journal of Mechanical Design APRIL 2012, Vol. 134 / 041005-11

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 05/02/2013 Terms of Use: http://asme.org/terms

You might also like