AbstractThis paper presents a theoretical and experimental

investigation of MEMS thermal biosensors in flow-through

measurements of metabolites. A model is developed to consider
convective heat transfer effects and enzyme kinetics using the
Michaelis-Menton theory. Numerical solutions from the model
are compared with experimental data obtained from a
prototype MEMS thermal biosensor. Satisfactory agreement is
found between the theoretical and experimental results,
indicating that the model is useful for designing MEMS thermal
biosensors with optimized performance.
I. INTRODUCTION
MEMS thermal biosensors, which measure the
concentrations of metabolites (e.g., glucose) by the
heat released from enthalpy changes of associated enzymatic
reactions, have important applications in metabolite detection
and monitoring [1-3]. However, the behavior of such sensors
has not been adequately analyzed, hindering optimal sensor
design and exploration of new applications. While heat
transfer effects in MEMS thermal biosensors have been
modeled [4, 5], this is not sufficient to determine the overall
sensor performance because of the omission of the kinetics of
enzymatic reactions. We previously reported a model for
MEMS thermal biosensors operating in flow-injection as well
as flow-through operation modes [6]. However, convective
heat transfer effects in the device were not considered
adequately. In addition, the binary enzyme kinetics model
used is not valid when the substrate concentration is around
saturated. Consequently, while this earlier model generally
agrees with the experimental data, significant errors exist. To
address these issues, we pursue a systematical study of the
device by considering convective heat transfer effects, which
occur in the porous media of enzyme-attached beads packed
inside the device, and the more accurate Michaelis-Menton
enzyme kinetics model. Numerical simulations have been
performed with FEMLAB and a prototype MEMS thermal
biosensor has been fabricated and characterized for glucose
detection. We have found that simulation results from the
current model quantitatively agree with the experimental data,
indicating that convective heat transfer effects significantly
influences the behavior of the device. Based on the model,
optimal designs of MEMS thermal biosensors can be
efficiently carried out.
L. Wang is with the Mechanical Engineering Department, Carnegie
Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213 USA (phone:
412-2682514; e-mail: [email protected]).
Q. Lin is with the Mechanical Engineering Department, Carnegie Mellon
University, 5000 Forbes Ave., Pittsburgh, PA 15213 USA (e-mail:
[email protected]).
II. MEMS THERMAL BIOSENSORS
Fig. 1 illustrates a MEMS thermal biosensor for flow
-through measurement. The device (Fig. 1a) primarily
consists of two identical microfluidic chambers (i.e. sample
and reference chambers) by bonding a PDMS sheet with
microfluidic features onto the chip surface. The chambers are
connected to their respective inlets and outlets by
microchannels and contain beads coated with enzymes (Fig.
1b). The beads are trapped in the chambers by a narrow gap
formed by a weir structure and chip surface (Fig. 1b). Each
chamber contains a freestanding thin polymeric membrane as
a mechanical support and sensing part. A thermopile is
integrated on-chip with its hot and cold junctions, formed
between chromium and nickel, respectively on the
membranes in the sample and reference chambers to measure
the temperature difference between them. Fabricated from
SU-8 photoresist, the thin membranes have a low thermal
conductivity, which results in excellent thermal isolation as
well as minimized thermal mass of the biosensor structure.
Fig. 2 shows a microscopic image of the sensor chip and an
image of the packaged device.
(a)
(b)
Fig. 1. MEMS thermal biosensor schematic: (a) top view, and (b)
cross-sectional view.
Glucose oxidase and catalase enzymes are immobilized
(using streptavidin-biotin chemistry) on polystyrene beads of
9.77 m nominal diameter, which are packed on the sensor
membrane for measurements of glucose. The thermopile
output voltage is measured to detect the enzymatic reaction of
glucose. During measurements, the glucose solutions are
Theory and Experiments of MEMS Thermal Biosensors
Li Wang and Qiao Lin
M
Reference
chamber
On chip
heater
Metal
thermopile
Block weir
Inlet
Sample
chamber
Access
channels
Outlet
A
B C
PDMS
cover
Bonding pad
Metal thermopile On-chip heater
Inlet Outlet
Chamber Membrane Channel
Membrane
Beads Weir
PDMS
cover
Bonding pad
Metal thermopile
Inlet Outlet
Chamber Membrane Channel
Membrane
Beads Weir
Proceedings of the 2005 IEEE
Engineering in Medicine and Biology 27th Annual Conference
Shanghai, China, September 1-4, 2005
0-7803-8740-6/05/$20.00 2005 IEEE.
1301
pumped through the sample chamber and the buffers for
reference chamber. The temperature difference between the
sample and reference chambers due to the heat difference
released from the enzymatic reactions reflects the
concentration difference between the sample and buffer
solutions. The temperature difference is measured by the
thermopile and used to obtain the concentrations of the
sample substrate. To obtain experimental data to verify the
MEMS thermal biosensor model, we have performed the
experiments by fixing the glucose concentration and varying
flow rates at which samples and buffers are introduced.
Polymer
membrane
On chip heaters
100 m
Metal
thermopile
Tubing
Chamber
Fig. 2. Images of a (a) microscopic view of bare and (b) packaged MEMS
thermal biosensor.
III. MEMS THERAL BIOSENSOR MODELING
Our MEMS thermal biosensor model is in Fig. 3, which
shows the cross-section schematic of the sensor chamber
along the fluid flow. In this 2D problem, substrate and
reference solutions flow through the chambers packed with
enzyme-functionalized beads. The kinetics of enzymatic
reactions of the substrate, forced mass and heat convection in
the fluid and heat conduction in the membranes are
considered. The membrane center is the location of hot
junctions of the thermopile temperature sensor whose voltage
gives the temperature change due to the enzymatic reactions.
The thermopile thickness is ignored, as it is far less than the
polymer membrane thickness. However, the conduction heat
loss via the thermopile lines is included in terms of the
average thermal conductivity of the membranes. With the low
thermal conductivity and large thickness of the PDMS cover
compared with the thin membrane, the upper boundary of the
channel is assumed to be thermally insulated. Since the
device inlet and outlet are located far away from the chambers,
the fluid flow in the chambers can be regarded as being fully
developed through a porous media formed by the packed
beads. Then in the Darcy flow regime [7] the longitudinal
volume-average velocity u is uniform over the channel cross
section, which implies that u=-Q/2a, where Q is the flow rate
per unit channel width.
Forced convection of heat and substrate mass in the fluid is
governed by [7]
2 2
2 2
( )
( )
p
p
f
q
T T T T
u
t x x y c
o

c c c c
+ = + +
c c c c

(1)
2 2
2 2
( )
s s s s s
p
C C C C dC
u D
t x x y dt
c c c c
+ = + +
c c c c
(2)
Substrate
L
t
2a
y
x
Beads
Q
T

Polymeric membrane Temperature sensor

Substrate
L
t
2a
y
x
Beads
Q
T

Polymeric membrane Temperature sensor

Fig. 3. Model of MEMS thermal biosensor in flow-through mode.
Where
/( )
p p f
c o =
is thermal diffusivity of the porous
media, p f
q q =
is the average density of power generated in
the chamber, T is the chamber temperature, f
q
is thermal
power density in the substrate solution, is porosity of the
porous media,
( )
f
c
is the heat capacity of the solution,
s
C is the substrate concentration, p
D
is the substrate
diffusivity and p

is the average thermal conductivity of

porous media. Heat conduction in the membrane is governed
by
2
2
| 0
m
m p y a
T T
t
x y

=
c c
+ =
c c
(3)
|
m y a
T T
=
=
(4)
where
m
T and
m
are the average temperature over the
membrane thickness t (t<<L) and average thermal
conductivity of the membrane, respectively.
Heat is released from the enzymatic reactions, whose
kinetics is governed by the Michaelis-Menten equation [8]
2 s e s
M s
dC k C C
dt K C
=
+
(5)
2 s e s
f
M s
dC Hk C C
q H
dt K C
A
= A =
+

(6)
where
2
k is the turnover number of the enzyme,
M
K is the
Michaelis constant,
e
C
is the enzyme concentration and H A
is the reaction enthalpy change. Introducing dimensionless
parameters
*
0
~ / ,
s s
C C C
*
~ / ,
amb
T T T
*
~ / , x x L
*
~ / 2 , y y a
* 2
~ /( / ),
p
t t L D
*
0
~ /
M M
K K C
and
*
~ /
d d amb
T T T where C
o
and T
amb
are initial substrate concentration and ambient
temperature, and combining Eqs. (1,2,3,5,6), we can then
formulate the problem in dimensionless form
* * 2 * 2 * *
2
2 1 * * *2 *2 * *
( )
s s s s s
a
M s
C C C C C
r L B
t x x y K C
c c c c
+ = +
c c c c +
(7)
* * * 2 * 2 *
2 0 2 * * *2 *2 * *
( )
( )
s
M s
C T T T T
r r B
t x x y K C
c c c c
+ = + +
c c c c +
(8)
*
2 * 2 *
1 *2 *
2
| 0
2
p
m
y
m
T L T
x at y

=
c c
+ =
c c
(9)
where the dimensionless parameters
2
/
p
r uL D = is mass
Peclet number showing relative importance of the mass
convection rate vs. diffusion rate,
0
/
p p
r D o =
shows the
1302
relative importance of thermal conduction vs. mass diffusion
effects,
2
1 2 0
/
e p
B L k C D C =
is the ratio of the substrate
consumption rate vs. mass diffusion rate,
2
2 2 0
/ ( )
e p f
B HL k C D T c = A
is the ratio of heat production rate
from reaction vs. heat conduction rate, and
a L L
a
2 / =
is the
ratio of the length and height of the chamber. The problem is
closed by the following boundary and initial conditions.
* * *
*
* *
*
*
* *
*
*
* *
*
(0, , ) 1
(1, , ) 0
1
( , , ) 0
2
1
( , , ) 0
2
s
s
s
s
C y t
C
y t
x
C
x t
y
C
x t
y
=
c
=
c
c
=
c
c
=
c
* * *
*
* *
*
*
* *
*
* *
* *
(0, , ) 1
(1, , ) 0
1
( , , ) 0
2
(0, ) 1
(1, ) 1
m
m
T y t
T
y t
x
T
x t
y
T t
T t
=
c
=
c
c
=
c
=
=
* * *
* * *
( , , 0) 1
( , , 0) 1
s
T x y
C x y
=
=
We then carry out the numerical simulations with the
parameters in Table 1 for flow-through measurements with
different substrate concentrations and flow rates.
TABLE 1. TYPICAL PARAMETERS OF MEMS THERMAL BIOSENSORS FOR
SIMULATION OF DIFFERENT SUBSTRATE CONCENTRATIONS AND FLOW RATES
D
p
(m
2
/s) (c)
f
(J/ul.K) C
e
(M) Width (um)
9.2E-10 0.0042 1.925E-5 2000

p
(W/K.m) 2a (um) L (um)
0.4764 0.4 200 2000
H (KJ/mol) K
M
(mM)
m
(W/K.m) k
2
(s
-1
)
180 1 0.2 1000
IV. RESULTS AND DISCUSSION
To evaluate the significance of heat convection compared to
heat conduction, we can calculate dimensionless thermal
Peclet number Pe=2au/
p
, which characterizes the ratio of
heat convection and conduction effects. With a flow rate of
0.5 ml/h and using the parameters listed in Table 1, Pe=0.7,
which shows that convection heat transfer effects are of the
same order as conduction effects, and therefore should be
considered for our problem. We have performed 2D
numerical simulation using the FEMLAB finite element
analysis package [9]. Figs 4 and 5 show typical calculated
fluid temperature and glucose concentration distributions
inside the chamber. It is observed that the reaction starts from
the chamber entrance (located on the left) and the substrate
concentration drops from there, resulting in the release of heat
and rise of temperature along the chamber. We then take the
simulated temperatures at membrane center corresponding to
the thermopile junctions, and calculate the thermopile voltage
values by multiplying the thermoelectric coefficient (1.85
mV/K). By plotting the thermopile output voltage versus flow
rates or pumped substrate concentrations in Figs. 6-7, we
compare the results from current model with the experimental
data and simulation results from the previous model [6].
Fig. 4. Temperature map in the chamber from FEMLAB simulation (initial
substrate concentration: 31.25 mM)
Fig. 5. Concentration map in the chamber from FEMLAB simulation (initial
substrate concentration: 31.25mM)
0
2
4
6
8
0 1 2 3 4
Flow Rate (ml/h)
D
e
v
i
c
e

O
u
t
p
u
t

(

V
)
Ref. [6]
Model
Experiment
Fig. 6. Device output at different flow rates (glucose concentration: 31.25
mM)
In Fig. 6, we can observe that results from the current
model are in much improved agreement with the
experimental data in the entire flow rate range. When flow
rates are smaller than 0.05 ml/h, the current model is close to
the experimental as well as the previous model. This is
explained by the small Pe value (<0.07), which implies that
convective heat transfer effects are almost negligible. At
larger flow rates, the heat loss via convective heat transfer is
more pronounced as Pe becomes larger. The current model,
which considers these effects, gives lower thermopile voltage
Chamber
Membrane
Membrane Chamber
1303
predictions compared with the previous model [6] and is
more accurate.
In flow-through mode, part of the heat generated by the
enzymatic reaction is taken away by the flowing substrate
solution via forced convection, which is more pronounced at
larger flow rates. On the other hand, a larger flow rate
supplies more substrate to the reaction chamber, generating
larger thermal power. These are two competitive effects, and
therefore, an optimal flow rate exists that maximizes the
thermopile output. In the experiment, an optimal flow rate of
0.5 ml/h is observed, which is same as predicted by the
current model. In contrast, the previous model that ignores
convection effects is in error by 40% in the predicted optimal
flow rate.
Fig. 7 shows the simulated and experimental results of
device output when glucose solutions of different
concentrations are pumped through the chamber with a fixed
0.5 ml/h flow rate. We see that the simulated and
experimental thermopile output both increase when the
glucose concentration becomes larger. This can be attributed
to the small flow rate (Q=0.5 ml/h) we used in the
measurements. Flow rates can influence the extent of
completion of the enzymatic reaction inside the chamber.
When Q is small, the substrate has sufficient residence time in
the chamber to interact with the enzymes and therefore is
completely consumed. When Q becomes larger, the substrate
in the chamber is replenished before it is completely reacted.
The critical flow rate Q
c
=k
2
C
e
V/C
s
can be calculated by
comparing the residence time t
r
=V/Q and consumption time
t
c
= C
s
/k
2
C
e
of substrate solution in the reaction chamber. Here
we used a fixed flow rate Q=0.5 ml/h. This is smaller than the
critical flow rate Q
c
=1.77 ml/h when the glucose
concentration is 31.25 mM. The critical flow rate is even
larger for smaller glucose concentrations. Therefore, the
glucose solutions used in our experiment are all completely
consumed in the chamber and the reactions are limited by
substrate supply. This results in the increase of thermopile
output when the concentrations of supplied glucose solution
become larger.
The current model predicts that the device output is
nonlinear at higher glucose concentrations (Fig. 7). This
nonlinearity, which is not predicted by the previous model,
can be explained by the non-uniformity of substrate
consumption in the chamber. When the concentration is small,
the glucose solution is completely consumed near the
entrance to the chamber, and there is no reaction towards the
chamber exit. This results in non-uniform distributions of
reaction power density inside the chamber and significant
temperature non-uniformities, with the thermopile junctions
at a large distance from locations of maximal temperatures.
As the concentration increases, the substrate will be available
for reaction in a larger portion of chamber, resulting in
improved power and temperature uniformities. In this case,
the thermopile junctions will become closer to locations of
maximal temperature, and the thermopile signal will therefore
increase faster when the substrate concentration becomes
higher.
0
2
4
6
8
0 10 20 30 40
Glucose Concentration (mM)
D
e
v
i
c
e

O
u
t
p
u
t

(

V
)
Ref. [6]
Model
Experiment
Fig. 7. Device output at different glucose concentrations (flow rate: 0.5 ml/h)
V. CONCLUSION
We have conducted theoretical and experimental analysis of
MEMS thermal biosensors. Convective heat transfer effects
and Michaelis-Menton enzyme kinetics are included in the
model. Numerical solutions of the model have been
performed and compared with experimental data as well as
results from a previous model that does not include
convection effects. Satisfactory agreement between the
current model and experimental data has been observed.
Nonlinear behavior at larger rates in flow-through glucose
detection is also predicted.
REFERENCES
[1] A. W. Vanherwaarden, P. M. Sarro, J. W. Gardner, and P. Bataillard,
"Liquid and Gas Micro-Calorimeters for (Bio)Chemical Measurements,"
Sensors and Actuators A-Physical, vol. 43, pp. 24-30, 1994.
[2] B. Xie, M. Mecklenburg, B. Danielsson, O. Ohman, and F. Winquist,
"Microbiosensor Based on an Integrated Thermopile," Analytica Chimica
Acta, vol. 299, pp. 165-170, 1994.
[3] J. M. Kohler and M. Zieren, "Chip Reactor for Microfluid Calorimetry,"
Thermochimica Acta, vol. 310, pp. 25-35, 1998.
[4] A. Wolf, A. Weber, R. Huttl, J. Lerchner, and G. Wolf, "Sequential Flow
Injection Analysis Based on Calorimetric Detection," Thermochimica Acta,
vol. 337, pp. 27-38, 1999.
[5] C. Auguet, F. Martorell, F. Moll, and V. Torra, "Identification of
Micro-scale Calorimetric Devices II - Heat Transfer Models from Two- or
Three-dimensional Analysis," Journal of Thermal Analysis and Calorimetry,
vol. 70, pp. 277-290, 2002.
[6] L. Wang, D. M. Sipe, Q. Lin, Modeling and Characterization of MEMS
Thermal Biosensors, ASME International Mechanical Engineering
Congress & Exposition, IMECE-61769, Nov. 13-19, 2004, Anaheim, CA
[7] D. A. Nield and A. Bejan, Convection in Porous Media, 2nd ed. New
York: Springer, 1999.
[8] D. Voet and J. G. Voet, Biochemistry, 3rd ed. New York: J. Wiley & Sons,
2004.
[9] FEMLAB-Multiphysics Modeling, www.comsol.com/
1304