Sie sind auf Seite 1von 8

Microelectronics

Journal
Microelectronics Journal 30 (1999) 11551162
www.elsevier.com/locate/mejo

Dynamic non-linear electro-thermal simulation of a thin-film thermal


converter
H. Laiz a, M. Klonz b,*
a

Instituto Nacional de Tecnologa Industrial, CC 157, 1650 San Martn, Argentina


Physikalisch-Technische Bundesanstalt, Bundesalle 100, 38114 Braunschweig, Germany

Abstract
This article describes the electro-thermal simulation of the acdc transfer differences at low frequencies of a thin-film thermal converter.
The dynamic non-linear model includes the temperature dependence of all the material parameters, and the radiation losses. It is used to
optimise the performance of the device at low frequencies, where temperature oscillations are present due to the lack of integration of the
oscillating Joule heat. The results of the simulation are compared with those of the measurements using a digital method. 1999 Elsevier
Science Ltd. All rights reserved.
Keywords: Electro-thermal simulation; Thin-film thermal converter; Low frequency acdc transfer

of the converter with ac and dc input voltages, respectively.


For an ideal thermal converter d is equal to zero.

1. Introduction
In metrology laboratories, achieving the highest accuracy
reference standards for the electrical units are reproduced
and maintained by means of two quantitative experiments,
the Josephson and the Quantum Hall effects [1]. These
experiments were, however, done using dc. To relate the
rms value of an ac signal to these dc units, a transfer method
has to be used. Nowadays the highest accuracy in such
transfers is achieved with thermal converters. In these
devices one or more thermocouples measure the temperature produced by the Joule heat in a resistor by a known dc
voltage (or current) and an unknown ac. Ideally the
temperature should be equal for the two signals of the
same rms value. Thus, the unknown rms ac value can be
related to the equivalent dc. The acdc voltage transfer
difference d of a thermal converter is defined as

Uiac Uidc
Uidc

Uoac Uodc ;

where Uiac is the ac input voltage. Uidc is the dc input


voltage, which when reversed produces the same mean
output voltage as Uiac. Uoac and Uodc are the output voltages

* Corresponding author. PTB Laboratory 2.32, Bundesalle 100, 38116


Braunschweig, Germany. Tel.: 49-531-592-2320; fax: 49-531-5922345.
E-mail addresses: Laiz@inti.gov.ar (H. Laiz), manfred.klonz@ptb.de
(M. Klonz)

1.1. Thin-film multijunction thermal converter


Originally, a thermal converter consisted of a thin
resistive wire of a stable alloy, like Evanohm, with a thermocouple thermally attached to, but electrically insulated
from the resistor by a small glass bead, all inside an evacuated ampoule. The highest accuracy was obtained with
multijunction thermal converters (MJTC) in which up to
120 thermocouples were attached to a bifilar twisted wire.
These devices are difficult to fabricate and thus costly [2].
Modern technologies, like thin-film, photolithographic
techniques and micromechanics have provided the opportunity to redesign the thermal transfer devices. Thin-film
technology allows mass fabrication and thus a dramatic
reduction of the cost [3]. In the thin-film or planar multijunction thermal converter (PMJTC), the bifilar heater and
up to 100 thermocouples are sputtered on a Si3N4/SiO2/Si3N4
sandwich membrane, covering a window which is etched
into a silicon wafer (Fig. 1).
The hot junctions of the thermocouple array are located
along the heater and the cold junctions are arranged symmetrical on the silicon that acts as a heat sink. The thin
dielectric membrane and the thermocouple system offer
low thermal conductance, thus providing a high sensitivity
of the device.
In the audio frequency range, the acdc transfer difference is caused by the Thomson effect in the heater and

0026-2692/99/$ - see front matter 1999 Elsevier Science Ltd. All rights reserved.
PII: S0026-269 2(99)00079-8

1156

H. Laiz, M. Klonz / Microelectronics Journal 30 (1999) 11551162

Fig. 1. PTB planar multijunction thermal converter.

Peltier effect in the connection pads. In the actual design, the


temperature is constant along the heater leading to a reduction of the Thomson heat. Both connection pads of the
heater are placed at a short distance on the silicon frame
electrically insulated by the sandwich membrane, but thermally short-circuited by the high thermal conductance of the
silicon underneath, leading to a reduction of the temperature
difference due to the Peltier effect in the pads.
All types of the thermal converters show an increasing
acdc transfer differences at low frequencies. Some models
were proposed to study the relation between the acdc
transfer difference, lack of integration and non-linearities
like the temperature dependence of material parameters
and radiation losses in single junctions units [4,5]. We introduce in this article a model to study this problem in thin film
units.

2. Electro-thermal model
Modeling of a thermal converter provides a useful tool to
study its behavior and test potential changes in the design. In
the development of the PTB thin-film MJTC, the analog
models and the finite element method (FEM) have been
applied to optimize the converter geometry, with the aim
of improving its sensitivity and temperature profile to minimise the Thomson heat along the heater [6]. The CMOS
thermal converters were modelled by Jaeggi also using the
FEM [7]. In this model, Jaeggi included the Peltier effect
and anisotropic characteristics in the multilayer structures.
These models are, however, not capable of calculating the
acdc transfer difference of the device, because they cannot

Fig. 2. Lumped parameter model of a thermal converter.

Fig. 3. Measured acdc transfer differences d at low frequencies f of the


two PMJTC, with the silicon obelisk underneath the heater (t 1.3 s) and
without obelisk (t 0.032 s).

deal with the ac inputs and do not include non-linearities in


the material parameters, which are crucial for the low
frequency performance.
Measurement of the ac quantities with the thermal
converters implies the transformation of the electrical
energy into thermal energy by means of the Joule heat PJ
dissipated in the heater resistor. The Joule heat flows
through the membrane with the thermal conductance G
(sum of the conduction and radiation losses) and a thermal
capacitance C, producing a temperature rise DT in the hot
junctions of the thermal element. By means of the Seebeck
effect a , this temperature rise is transformed into an output
voltage Uo. Thus, in steady state
Ui ! PJ

Ui2
PJ
! DT
! Uo aTDT:
RH T
GT

This can be depicted in a simple lumped temperaturedependent parameter model (Fig. 2).
The Joule heat oscillates with a frequency that is twice the
frequency of the input voltage. At frequencies well above
the inverse of the thermal time-constant of the converter, i.e.
the time needed by the output voltage to reach 0.63 of its
final value after a step input voltage is applied, the heat is
integrated and the temperature of the device, and thus, the
output voltage are constant, like in the dc.
On the contrary, if the period of the input voltage is large
in comparison to the thermal time-constant, the temperature
of the whole device follows the Joule heat and the output
voltage shows a double-frequency ripple on its dc voltage.
Due to the dependence of the electrical resistance, thermal conductance, and specific heat on temperature and the
high non-linear characteristic of radiation losses, an oscillation in the temperature will result in the mean value of the
heater temperature being different at ac and dc.

H. Laiz, M. Klonz / Microelectronics Journal 30 (1999) 11551162

1157

Fig. 4. Geometry of the model. Due to its symmetry, only the solution in half the membrane is needed.

The temperature rise of the heater is converted into a


voltage by means of the Seebeck effect which in turn is
also temperature-dependant. Consequently, the mean output
voltage will, generally, differ when applying the dc and
low frequency ac inputs, leading to an acdc transfer
difference.
Due to this fact, the PMJTC shows increasing acdc
transfer differences to lower frequencies. The increase of
the time-constant by adding some thermal mass underneath
the heater (the so-called silicon obelisk) has improved the
performance of the device at low frequencies [8]. Fig. 3
shows the acdc transfer differences at low frequencies of
the two PMJTC with different thermal time-constants t . It
can be seen that increasing t shifts the high values of d to
lower frequencies.
As the main aim of this project is to improve the
performance of the device at low frequencies, we
have to find out the relation between the acdc transfer
differences at these frequencies and the temperature
dependence of the material parameters and heat transport mechanisms. To find out this relation, when the
temperature varies with the oscillating input power,
one has to consider the dynamic behavior of the device.
Hence, to determine the temperature distribution in the
PMJTC with time, the dynamic heat diffusion equation
has to be solved under the conditions imposed on its

boundaries
7k7T gx; y; z; t rc

2T
;
2t

where the tensor k is the thermal conductivity of the


medium in W m 1 K 1, gx; y; z; t the heat generation
rate per unit of volume in the medium in W m 3, r the
medium density in kg m 3 and c its specific heat
in J kg 1 K 1. The radiation losses from the surfaces
are evaluated according to the StefanBoltzmann law as
q 1m sAT14 Ta4 ;

where 1 m denotes the emissivity coefficient of the


different materials of the PMJTC, s the StefanBoltzmann constant, A the surface area, T1 the temperature of
the surface and Ta the external temperature. Simulations
were performed in vacuum and in air, where heat losses
by conduction in air are included as a boundary condition.
2.1. Heat generation
In Eq. (3), gx; y; z; t denotes the Joule heat generation
rate per unit of volume in the heater. Due to its particular
bifilar geometry, the electric current density j is not
uniformly distributed. Furthermore, the dependence of the
electrical resistivity r e with the temperature and the

1158

H. Laiz, M. Klonz / Microelectronics Journal 30 (1999) 11551162

expected temperature differences along the heater claims for


the knowledge of the electric field distribution in order to
calculate the generated Joule heat. Consequently, the Joule
heat per unit of volume in (x,y,z) should be calculated as

boundaries of the membrane are


Tx; 0; z; t Ta

10

Tx; y5 ; z; t Ta

PJ x; y; z

E
;
re x; y; z; T

where E is the electric field intensity which is calculated as


the gradient of the electric potential V, which in turn is the
solution of the Laplace equation under appropriate boundary conditions, i.e.


1
7V 0:
6
7
re x; y; z; T

Tx5 ; y; z; t Ta
(b) Neumanns bcs or bcs of the second kind: The plane of
symmetry is
2Tx; y; z
0
2x

x 0:

11

(c) Newton bcs or bcs of the third kind: The lower


surface of the membrane

Hence, the determination of the output voltage of the


PMJTK for a particular time instant t involves the iterative
solution of Eqs. (3) and (6). Afterwards, the output voltage
can be calculated with the Seebeck coefficient a (T) of each
thermocouple for its particular temperature.

2Tx; y; z
hTx; y; z; t Ta
2z

2.2. Calculation of the acdc transfer differences

z z1 on the membrane; z z2 on the heater;

For small voltage excursions, thermal converters have a


potential relationship between the input Ui and output Uo
voltages of the type
Uo kUin ;

where, for ideal devices, n 2. In the case of the PMJTC, n


is between 1.97 and 2.
When we differentiate Eq. (7), and divide it by Uo
dUo
dU
n i
Uo
Ui

and regarding Eq. (1), we can calculate the acdc transfer


difference for small D Uodc Uoac as

Uodc Uoac
nUodc

Uiac Uidc :

The main purpose of the electro-thermal model is to study


the influence of the non-linear material parameters and the
radiation losses in the acdc transfer difference. Consequently, no thermoelectric effects, such as Thomson and
Peltier effects, were included in the model. This means
that the output voltage will be the same with the dc inputs
of both polarities. Thus, to calculate d with Eq. (9), we need
the solution of the thermoelectrically coupled problem with
the ac and dc excitations as electrical boundary conditions in
order to calculate Uoac and Uodc.
2.3. Boundary conditions
Due to the particular symmetry of the device we can solve
the coupled problem only in half of the membrane.
Hence, we have as thermal boundary conditions (see Fig.
4):
(a) Dirichlets bc or bc of the first kind: The external

z 0:

12

The upper surface of the membrane is


2Tx; y; z
hTx; y; z; t Ta
2z

13

z z3 on the thermopile;
where the heat transfer coefficient h accounts for the
conduction through air. No convection exists due to the
small distances between the PMJTC and its cover and
between the PMJTC and its carrier [9]. When we study
the thermal converter in vacuum, h is taken as zero.
The electrical boundary conditions differ when we deal
with the dc or ac input voltages.
2.3.1. Solution with dc input voltage
In this case, the boundary conditions of the first kind for
the electric field problem are (see Fig. 4)
input terminal of the heater
Ux; 0; z Udc =2;

x1 x x2 ;

plane of symmetry U0; y; z 0;

14
0 y y5

15

and a steady-state solution is performed for the electric (Eq.


(6)) and thermal (Eq. (3)) field problems, that is the derivative regarding time is taken as zero. Then, Uo is determined
from the temperature distribution by means of the Seebeck
coefficient aT.
2.3.2. Solution with ac input voltage
In this case the boundary conditions of the first kind for
the electric field problem must be time-dependent. A discretization of the sinusoidal input voltage is performed and a
solution for the coupled electrical and thermal problem is
found in each time step. If the input voltage ui is sinusoidal:
p
16
ui 2U sinvt

H. Laiz, M. Klonz / Microelectronics Journal 30 (1999) 11551162

1159

obtained as a solution for the step k 1.


Initial bc

Fig. 5. Electric potential distribution (a), Joule heat (b), and temperature
distribution (c). Due to its symmetry, only the solution in half the
membrane is needed.

the voltage at time step k is


p
Uik 2U sinkvDt;

17

where Dt is the chosen time step. Consequently the boundary conditions of the first kind for the electrical problem for
the time instant tk kDt, will be
input terminal of the heater
Ux; 0; z Uik =2;

x1 x x2 ;

plane of symmetry

U0; y; z 0;

18
0 y y5 :
19

Although we are dealing with the ac input quantities, a


steady-state solution is performed for the electric field, and
consequently for the Joule heat, in each step. This means
that no capacitive or inductive effects are included, which is
most reasonable when we limit our study to the highest
frequency of 20 Hz. For the solution of the thermal problem,
a transient solution is performed for each step, this means
the time diffusion term in Eq. (3) is included. The thermal
boundary conditions remain constant for all the time steps,
but the internal heat generation of the time step k is the Joule
heat calculated in the same step and the initial temperature
distribution of the time step k is the temperature distribution

Tibc x; y; z; tk Tx; y; z; tk1 :

20

The ANSYS was used to solve the coupled problem. The


first step of this iterative process provides a solution for Eq.
(6). From this, in the second step, the electric field intensity
E is derived and the Joule heat distribution PJ is calculated
using Eq. (5). In the third step Eq. (3) is solved to obtain the
temperature distribution with PJ as the internal heat generation. Fig. 5 depicts the solutions of the three steps. Due to
the temperature dependence of r e (Eq. (6), first step), and of
c, k and the heat transfer coefficient in air h, and the nonlinear characteristic of the radiation lasses (Eq. (3), third
step), the three steps should be iterated until an equilibrium
solution is reached.
As previously shown, the PMJTC under study consists of
a heater and a multilevel thermopile deposited on a dielectric sandwich membrane. We modelled the entire threedimensional structure, but to keep the number of elements
within reasonable limits, the thermopile and the membrane
were modelled as two separate homogeneous layers with an
equivalent thermal conductivity tensor, i.e. direction dependent [7]. Although, due to its small thickness, no significant
temperature variations are expected in the z-directions [6],
the membrane and the thermopile are kept as separate layers
because this simplifies the analysis of the influence of the
material and dimensional changes in the design.
The NewtonRaphson procedure is used to solve the nonlinear set of equations. Convergence is obtained when the
size of the Euclidean norm of the residual is less than 10 3
times a reference value. Each time step usually requires
between two and four iterations to converge. The convergence criterion is based on the electrical current and the heat
flow for the electrical and thermal problems, respectively.
Simulations were carried out with different number of steps
per period. A good compromise between the accuracy and
the speed was found with 20 steps. An increment does not
produce a significant change in the results, but it increases
the time required for a solution.
In Fig. 6(a) the simulated output voltage with an input
frequency of 0.1 Hz and with dc is plotted as a function of
time for a PMJTC with a thermal time-constant t of about
0.2 s. The mean value of the last cycle of the ac voltage is
also plotted. In Fig. 6(b) the same results are plotted with an
input frequency of 10 Hz, in which the effect of the thermal
integration can be observed. With ac input voltages the time
index k is incremented up to the mean value of the output
voltage Uo, for cycle n equals its mean value for cycle n 1.
3. Parameter adjustment and comparison of results with
measurements
As the thermal and electrical parameters of the different
PMJTCs can differ due to tolerance and fabrication procedures, an adjustment of the parameters for a particular

1160

H. Laiz, M. Klonz / Microelectronics Journal 30 (1999) 11551162

Fig. 7. Measured Sm and calculated sensitivities Sc as a function of the


temperature rise DT of the hot junctions at different ambient temperatures.

power coefficient indicates that a change in the input power


is not followed by a proportional change in the output
voltage. As the output voltage is proportional to the input
power, the power coefficient of the sensitivity can be defined
as

wSUo

Fig. 6. Output voltage with the dc and ac input voltages (t 0.2 s). (a) f
0.1 Hz (b) f 10 Hz. The difference D between the output voltage with the
dc and the mean output voltage of the last cycle in the ac is used to calculate
the acdc transfer difference (see Eq. (9)).

1 dS
:
S dUo

23

To adjust the temperature coefficients of the thermal


conductivities and the emissivity coefficients, we performed
steady-state simulations for the sensitivities for different
ambient temperatures and input voltages in vacuum. They
were adjusted in order to match the simulated and measured

thermal converter has to be carried out in order to compare


the results of the simulations with those of the measurements. The sensitivity and the time-constant are used for
the adjusting procedure.
The sensitivity of the device S is defined as the ratio of the
output voltage Uo and its input power PJ.
S

Uo
:
PJ

21

The temperature coefficient of the sensitivity is defined as


[4]

bS

1 dS
:
S dT

22

A low bS is important for obtaining a low dependence of


the output voltage on the ambient temperature. This condition improves both the settling time and the standard
deviation during measurement. It leads to the corresponding
improvement of the standard measurement uncertainty. A

Fig. 8. Measured and calculated ratio of the ac voltage (peak-to-peak) Uoac


superimposed to the dc output voltage Uodc at different frequencies f.

H. Laiz, M. Klonz / Microelectronics Journal 30 (1999) 11551162

1161

Fig. 9. Comparison of the measured and calculated acdc voltage transfer


differences d at frequencies f. (a) PMJTC with t 27 ms (without Si
obelisk in air) (b) PMJTC with t 7.5 s (with Si obelisk in vacuum).

Fig. 10. Influence of the radiation losses on d at low frequencies in air and
in vacuum.

temperature and power coefficients. In Fig. 7 comparisons


of this simulation with measurements are presented for a
PMJTC with a heater resistor of 184 V and a thermal
time-constant of about 7.5 s (in vacuum). These simulations
were performed using dc, this means, with steady-state solutions in which the right term of Eq. (3) is zero. Afterwards,
the heat transfer coefficient in air was adjusted in order to
match the simulated and measured vacuum factor and its
power dependence.
Afterwards the heat capacity rc of the device was
adjusted so as to approximate the amount of the ac superimposed on the dc output voltage for a particular frequency.
Fig. 8 depicts the measured and calculated percentages of
the ac ripple.
Using these adjusted parameters for a particular PMJTC,

we performed comparisons of the calculated and measured


acdc transfer differences for different types of the thermal
converters, in order to test the accuracy of the model. In Fig.
9 the results for the two PMJTC with different timeconstants t are presented.

4. The reasons for the acdc transfer differences at low


frequencies
The developed model is used to investigate the causes and
to search a minimization of d at low frequencies. A good
example for the usefulness of this method is to study the
influence of the radiation losses. The calculations performed
with the model show that in air for a PMJTC with Si obelisk

1162

H. Laiz, M. Klonz / Microelectronics Journal 30 (1999) 11551162

radiation losses represent 3.9% of the total thermal conductance and, consequently, 96.1% corresponds to the conduction through air and the layer materials. In vacuum, the
participation of radiation losses increases to 21.2 and
78.8% corresponds to the conduction through the layer
materials. Moreover, their influence is even greater in the
temperature coefficient b S and power coefficient q SUo of the
device. In air the absolute value of b S is reduced by 16.6%
when radiation is avoided, and in vacuum it can even change
sign. The highly non-linear characteristic of the radiation
losses explains the fact that, although they represent about
4% of the heat flow in air, they have a considerable influence on the b S and q SUo.
Fig. 10 depicts the calculated d at low frequencies with
and without radiation losses for a PMJTC with Si obelisk in
air and in vacuum. In air d decreases about 400 10 6 at
0.01 Hz. The effect is still clearer in vacuum. d is reduced to
very low values and can even reverse the sign, when radiation losses are eliminated. Therefore, an attempt is made to
reduce the radiation losses by means of the deposition of a
reflecting aluminum thin film over the silicon obelisk.
5. Conclusions
We made a dynamic non-linear thermo-electrical model
for a planar MJTC to simulate its inputoutput relation. The
model has the accuracy necessary for the calculation of the
acdc transfer differences of thin-film thermal converters at
low frequencies. It is used to investigate the causes and
search for a reduction of d at low frequencies. This reduction can be achieved by increasing the thermal timeconstant or by compensation of the power dependence of

S. Some possibilities of compensating this power dependence are under investigation:


reduction of radiation losses,
use of thermo-electric materials with appropriate
temperature coefficient of the Seebeck effect,
depositing an extra layer with appropriate temperature
coefficient of the thermal conductivity.

References
[1] T.J. Quinn, Base units of the systeme international dunites, their accuracy, dissemination and international traceability, Metrologia 31
(1995) 515527.
[2] B.D. Inglis, Standards for ACDC transfer, Metrologia 29 (1991) 191
199.
[3] M. Klonz, T. Weimann, Accurate thin-film multijunction thermal
converter on a silicon chip, IEEE Trans. Instrum. Meas. 38 (1989)
335337.
[4] T. Takeishi, Characteristics of vacuo-thermo junctions at ultra-low
frequency, IEEE Trans. Instrum. Meas. 34 (1) (1985) 3441.
[5] F.L. Hermach, Thermal converters as acdc transfer standards for
current and voltage measurements at audio frequencies, J. Res. NBS
48 (1952) 121138.
[6] U. Dillner, E. Kessler, M. Klonz, C. Thomes, Micro-machined planar
acdc multijunction thermal converter: aspects of design and layer
materials, in: H. Reichl, A. Heuberger (Eds.), MICRO SYSTEM Technologies 1994, VDE-Verlag GmbH, Berlin, 1994, pp. 773782.
[7] D. Jaeggi, Thermal converters by CMOS technology, PhD Thesis,
Physical Electronics Laboratory, ETH Zurich.
[8] M. Klonz, T. Weimann, Increasing the time-constant of a thin film
multijunction thermal converter for low frequency application, IEEE
Trans. Instrum. Meas. 40 (2) (1991) 350351.
[9] T. Elbel, Miniaturised thermoelectric radiation sensors covering a wide
range with respect to sensitivity or time constant, Sensor Actuators A
2527 (1991) 653656.