BATCH PROCESSED THERMOELECTRIC ENERGY HARVESTER
CHARATERIZATION AND MODELING
Jiale Su1, Ziyang Wang1, Vladimir Leonov2 and Ruud J.M.Vullers1
1
imec /Holst Centre, Eindhoven, the Netherlands
2
imec, Leuven, Belgium
*Presenting Author: jiale.su@imec-nl.nl
Abstract: A micromachined thermoelectric energy harvester for human body applications has been developed.
This paper reports the device characterization and a corresponding model to validate the performance of the device.
The open circuit voltage is measured in ambient and vacuum. The model, which is an equivalent circuit, is built up
by transferring the measurement setup from thermal domain to electrical domain. We find a good agreement
between the model and the measurement in vacuum. However, the model of ambient does not reflect the
experimental data. We believe the reason to be air convection.
Keywords: Thermoelectric energy harvester, MEMS, Stepper lithography, thermopile, TEG
INTRODUCTION
Temperature differences in/on artificial objects
(machinery, buildings, transport, pipelines) and on
the skin of animals and men can be used to power
autonomous devices. For example, as the first
wearable wireless sensors and medical devices
(Electroencephalograph (EEG) system,
an
Electrocardiography System in a Shirt) fully powered
by thermoelectric generators (TEG) on man have
been demonstrated [1]-[3]. These thermal energy
harvesters have been made by using off-the-shelf
thermopiles. Hence they have ‘high’ fabrication cost
because of the fabrication method for thermopiles.
The fabrication cost of energy harvesters is a key
factor for their acceptance by industry and for
moving them into mass production. Reduction of the
cost can be achieved by using micromachining
technologies and fabricating thousands of devices,
like IC chips, in one run.
In [4], the development of micromachined
thermoelectric harvesters was presented. A matched
load is connected to these devices for
characterization. We were able to obtain the voltage
and power output on the matched load in air,
delivering 1.2V and 0.35µW at a temperature
difference of 30K. In this paper we report additional
measurements, both in vacuum and in air, using a
LabviewTM program: both voltage and temperature of
the hot and cold plates are measured every 0.1sec
automatically. We have also developed a model, in
order to analyze the results.
CHARATERIZATION
Measurement setup
Figure 1 shows a dedicated setup designed for
reliable temperature measurement. A device is placed
between two aluminum blocks, which are separated
by plastic spacers. To have a much larger heat
capacity than the device, the aluminum blocks are
3cm×3cm×2cm, which is around 1800 times larger
than the devices. One of the blocks is heated by a
hotplate, while the other block has large cooling fins
and acts as a cooling plate. The device is glued to
both plates with thermal paste, Thermal Joint
Compound 120[5], which has a thermal conductivity
of 0.735Wm-1K-1. The total thickness of the paste is
about 100 µm. The device is electrically connected
by wire bonding to a small plastic board with large
soldered connecters which can be easily connected to
a voltmeter for the open circuit voltage measurement
or a load for further characterization.
(a)
(b)
Figure 1 (a) Schematic of measurement setup and (b)
Photo of measurement setup
A program is made to read the open circuit
voltage output and the voltage values of
thermocouple I (T1) and thermocouple II (T2) by
three multimeters. The program converts the
thermocouple voltages to the temperature difference
between T1 and T2. The open circuit voltage of two
different devices as a function of temperature
difference is measured in both ambient and vacuum
(pressure of 1mbar) environment, as shown in Figure
2. All plots are linear. We find the voltage output per
unit temperature difference is 256mVK-1in ambient
and 374mVK-1 in vacuum.
Ambient vs. Vacuum
Voltage(V)
7
6
Ambient
5
Vacuum
RAL,B
T2
Repoxy,B
Rsi,B
A
y = 0.3736x + 0.0093
RAIR,d-d’
∆T
Rspacer
RTP
RAIR,b-b’
4
y = 0.2563x + 0.0884
3
T1
2
A’
Repoxy,T
RAL,T
1
Rsi,T
(a)
0
0
5
10
15
20
25
Repoxy,B
RAL,B
T2
Rsi,B
A
∆T (K)
Figure 2 open circuit voltage measurement in
ambient and vacuum
The thermopile behaves differently in ambient
and vacuum. In order to analyze the effect, a model is
built to simulate the open circuit voltage behavior.
This is discussed in the next section.
Modeling
A model is made which describes the whole
measurement setup in analogy to an electric circuit.
The elements between T1 and T2 are transferred
from the thermal domain to the electrical domain,
where temperature difference, heat flow and thermal
resistance correspond to voltage, current and
electrical resistance, respectively. This is done for
both ambient and vacuum. Starting from
thermocouple II (T2), in ambient (see Figure 3(a)),
the heat flow passes through part of the aluminum
block, RAL,B. Then the heat flow divides into three
flows, one through spacer, Rspacer, one through air
between two Al blocks, RAIR,b-b’, and one through
epoxy between bottom Al block and bottom chip,
Repoxy,B, bottom Si chip, Rsi,B, device, RAA’, from A to
A’, top Si chip, Rsi,T, and epoxy between top chip and
top Al block, Repoxy,T. All the heat flows come
together, pass part of top Al block and reach
thermocouple I (T1). But in calculation, RAIR,b-b’ and
Rspacer can be neglected. Because both thermal
resistors, RAIR,b-b’ and Rspacer, are in parallel with a
series of thermal resistors, which are Repoxy,B, Rsi,B,
RAA’, Rsi,T and Repoxy,T, the temperature differences
fall across RAIR,b-b’, Rspacer and the series of thermal
resistors are the same. The model of the device is a
combination of the thermopile and air, so a thermal
resistance of the air between top and bottom die of
device, RAIR,d-d’, and a thermal resistance of
thermopile, RTP in parallel represents the device in
thermal domain, shown between A and A’ in Figure
3(a). Moreover, in ambient, thermal resistance of air
includes air conduction, air convection and radiation.
In vacuum the effects of air conduction and air
convection can be neglected.
Rradiation,d-d’
∆T
Rspacer
RTP
Rradiation,b-b’
T1
A’
Repoxy,T
RAL,T
Rsi,T
(b)
Figure 3 Equivalent circuits of elements between T1
and T2 in (a) ambient and (b) vacuum
Thus only the thermal resistance of radiation is taken
account in the model (shown in Figure 3(b)). The
thermal resistance of each element (except the
thermopile) shown in Figure 3, is listed in Table 1. In
thermal domain, all thermocouples are connected in
parallel. Therefore the thermal resistance of the
thermopile, RTP, can be calculated by dividing the
thermal resistance of one thermocouple by the total
number of the thermocouples. The thermal resistance
of one thermocouple is simulated by using Finite
Element Model (FEM) software, ‘MSC Marc’. A
10µm wide thermocouple with a 1.5µm wide air gap
on the left and right side performs as a basic element
of thermopile, shown in Figure 4(b).
Table 1 Thermal resistance of each element fromT1 to T2
(except thermopile)
Thermal
Conductivity
(Wm-1K-1)
Length
(m)
Width
(m)
Area
(m2)
Thickness
(m)
Thermal
Resistance
(KW-1)
RAL,B
REpoxy,B
RSi,B
RSi,T
REpoxy,T
RAL,T
237
0.7
149
149
0.7
237
0.03
0.0025
0.002
0.005
0.005
0.03
0.03
0.003
0.003
0.003
0.003
0.03
9.0×
10-4
3.0×
10-3
7.5
×10-6
5.0
×10-5
7.5
×10-6
7.5
×10-4
1.5
×10-5
6.8
×10-4
1.5
×10-5
5.0
×10-5
9.0
×10-4
3.0
×10-3
0.01
9.52
0.67
0.30
4.76
0.01
22.15
22.14
22.12
22.11
22.09
22.08
22.06
22.05
22.03
22.01
(a)
22.00
(b)
Figure 5 Temperature difference over one
thermocouple by simulation (a) in ambient and (b)in
vacuum
DISCUSSION
(b)
Figure 4 Finite element model of thermocouple (a)
cross view (b) side view
A relative large difference between theory and
measured data is observed in Figure 6(a), while in
Figure 6(b), a good agreement is found between
theory and measured data. Because the setup is
unchanged, the presence of air is the only difference
between the ambient and vacuum condition. We
believe that the air convection plays a role here, an
effect which needs further investigation in our
devices. In vacuum, where air convection can be
neglected, the model is more accurate.
Simulated vs. Measured
8
22.10
22.09
Measured
6
y = 0.3272x - 0.0029
5
4
3
22.08
2
22.07
1
22.06
Simulated
7
Voltage (V)
In the simulation, thermal conductivity of air is
chosen for the measurement in ambient (0.026Wm1 -1
K ), while in the case of vacuum the conductivity is
2.6×10-5Wm-1K-1. Applying a heat flux of 1mWmm-2
from the bottom face, a temperature difference ∆T of
0.10K is falling over the thermocouple in ambient
(see Figure 5(a)) and 0.15K in vacuum (see Figure
5(b)). The total area of one thermocouple is 5.44×10-4
mm2 ((L) 45µm × (W) 12µm). The heat flow, Q, is
found to be 5.40×10-7W, by multiplying the heat flux
and the area. The thermal resistance of the
thermocouple is 1.85×105KW-1 in ambient and
2.78×105KW-1 in vacuum, which is calculated
according to heat transfer equation, R= ∆T/Q. We
thus find the thermal resistance of the thermopile to
be 104KW-1 in ambient and 156KW-1 in vacuum.
A MATLAB program is written to model the
circuit in Figure 3, where the thermal resistance of
each element (except the thermopile) is calculated in
Table 1. The temperature results measured from T1
and T2 in both ambient and vacuum are imported in
the MATLAB program to calculate the open circuit
voltage. The data are plotted in Figure 6(a) and (b).
y = 0.2563x + 0.0884
0
22.05
0
22.04
5
10
15
∆T (K)
22.03
(a)
22.02
22.01
22.00
(a)
20
25
Furthermore, new poly-SiGe micromachined
thermopiles are being fabricated. The whole
fabrication process will be optimized in several
aspects, such as, avoiding short circuit between
adjacent thermocouples and decreasing specific
contact resistance between Al and n-SiGe, which
were discussed in [4]. The new devices will be
characterized and compared with current devices.
Simulated vs. Measured
7
measured y = 0.3863x + 4E-05
Voltage(V)
6
simulated
5
4
y = 0.3736x + 0.0093
3
2
REFERENCE:
1
0
0
5
10
15
20
∆T(K)
(b)
Figure 6 Simulated data vs. measured data of the
device in both (a) ambient and (b) vacuum
CONCLUSION
In this paper, a dedicated measurement setup is
fabricated to characterize the device. A sophisticated
model is derived by converting the measurement
setup including the device from thermal domain to
electrical domain. The device is measured using the
same setup both in ambient and vacuum. The
theoretical open circuit voltage is calculated
according to the measured temperature differences
through the model. The theoretical value agrees with
the measured results in vacuum but not in ambient.
We believe the reason to be the lack of proper
description of air convection in our model in ambient.
For future work, we will try to adjust our model
in order to have a better agreement between the
model and the measurements.
[1] Leonov, V., Fiorini, P., Sedky, S., Torfs, T., and
Van Hoof, C., “Thermoelectric MEMS Generators as
a Power Supply for a Body Area Network,” Proc.
13th Int. Conf. on Solid-State Sensors, Actuators and
Microsystems (Transducers’05), IEEE, 2005, pp.
291-294.
[2] Van Bavel, M., Leonov, V., Yazicioglu, R. F.,
Torfs, T., Van Hoof, C., Posthuma, N., and Vullers,
R. J. M., “Wearable Battery-Free Wireless 2-Channel
EEG Systems Powered by Energy Scavengers,”
Sensors& Transducers, 2008, Vol. 94, No. 7, pp.
103-115,
http://www.sensorsportal.com/HTML/DIGEST/P_30
0.htm
[3] Vladimir Leonov, Tom Torfs, Chris Van Hoof
and Ruud J. M. Vullers, “Smart Wireless Sensors
Integrated in Clothing: an Electrocardiography
System in a Shirt Powered Using Human Body
Heat”, Sensors & Transducers J., Vol. 107, Issue 8,
August
2009,
pp.
165-176,
http://www.
sensorsportal.com/HTML/DIGEST/P_485.htm
[4] Jiale Su, M Goedbloed, Y Van Andel, M C De
Nooijer, Rene Elfrink, Vladimir Leonov, Z Wang and
Ruud Vullers, ‘Batch process micromachined
thermoelectric energy harvester: fabrication and
characterization’, Journal of Micromachining and
Microengineering, Vol. 20 No. 10, 2010, doi:
10.1088/0960-1317/20/10/104005
[5] http://www.wakefield.com/PDF/accessories.pdf