International Journal of Heat and Mass Transfer 53 (2010) 2001–2011

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

journal homepage: www.elsevier.com/locate/ijhmt

A three-dimensional theoretical model for predicting transient thermal behavior

of thermoelectric coolers
Chin-Hsiang Cheng a,*, Shu-Yu Huang a, Tsung-Chieh Cheng b
a
Department of Aeronautics and Astronautics, National Cheng Kung University, Tainan 70101, Taiwan, ROC
b
Department of Mechanical Engineering, National Kaohsiung University of Applied Sciences, Kaohsiung 807, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Article history: A simulation model is developed and used to predict transient thermal behavior of the thermoelectric
Received 14 August 2009 coolers. The present model amends the previous models, in which the P–N pair is simply treated as a sin-
Received in revised form 3 December 2009 gle bulk material so that the temperature difference between the semiconductor elements was not pos-
Accepted 3 December 2009
sible to evaluate. Based on the present simulation model, the thermoelectric cooler is divided into four
Available online 6 February 2010
major regions, namely, cold end (region 1), hot end (region 2), and the P-type and N-type thermoelectric
elements (regions 3 and 4). Solutions for the three-dimensional temperature fields in the P-type and the
Keywords:
N-type semiconductor elements and transient temperature variations in the cold and the hot ends have
Thermoelectric cooler
Theoretical model
been carried out. The magnitude of the coefficient of performance (COP) of the thermoelectric cooler are
Transient behavior calculated in wide ranges of physical and geometrical parameters. To verify the numerical predictions,
Experiment experiments have been conducted to measure the temperature variations of both the cold and the hot
ends. Close agreement between the numerical and the experimental data of the temperature variations
has been observed.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction eral, a thermoelectric cooler consists of a number of semiconductor

element pairs which are connected electrically in series and ther-
Thermoelectric effects in semiconductors cause electrical cur- mally in parallel, and each pair includes a P-type and an N-type
rents to flow due to temperature gradients and also cause temper- elements. The P-type and the N-type elements are used to ensure
ature gradients when an electrical current is applied. A feature of that the carriers flow in the same direction. Although in theory a
the thermoelectric devices is the absence of moving parts when single piece of semiconductor material could be adopted, the series
they are used to convert heat into electrical energy, and vice versa. connection is used to avoid the high electrical current requirement
Therefore, they can work over a long period of time in high perfor- with the single element.
mance without extra maintenance. This could be a significant An applied current causes charged carriers in the material,
advantage from system simplification and reliability points of whether they are electrons or holes, to diffuse from the cold side
view. Besides, these devices produce no waste matter in the con- to the hot side. These mobile charged carriers migrating to the
version process. Therefore, it is regarded as an attractive method hot side leave behind their oppositely charged and immobile nuclei
for energy conversion. at the cold side, thus giving rise to a temperature difference. Since a
Thermoelectric devices are basically categorized into two oper- temperature difference creates a driving potential for the phonons
ation modes based on the direction of energy conversion: thermo- to drift back to the cold side resulting in heat conduction at equi-
electric cooler (TEC) converting electricity to thermal energy and librium. Improving the performance of a thermoelectric materials
thermoelectric generator (TEG) converting heat into electricity. involves controlling the motion of phonons and electrons/holes.
The thermoelectric cooler is a device in which an electric current The most important performance index of the thermoelectric
is applied to semiconductor devices to produce an appreciable materials is the figure of merit (ZT) which is defined by
temperature difference at the two ends of the semiconductor.
The colder end may then be utilized for cooling purposes. In gen- ZT ¼ a2 T r=k ð1Þ

where a is the Seebeck coefficient; T is the temperature; r is the

electrical conductivity; k is thermal conductivity of the material
* Corresponding author. Address: Department of Aeronautics and Astronautics,
National Cheng Kung University, No.1, Ta Shieh Road, Tainan 70101, Taiwan, ROC.
[1,2]. A higher ZT value in TECs leads to higher COP, and in TEGs
Tel.: +886 6 2757575x63627. it leads to higher thermal efficiency. For the purpose of reaching
E-mail address: [email protected] (C.-H. Cheng). higher figure of merit, one should try to elevate the magnitude of

0017-9310/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2009.12.056
2002 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011

Nomenclature

A cross-sectional area of P-type or N-type element (m2) T temperature (K)

Ah surface area of the hot end (cm2) Tc temperature of the cold end (K)
As total heat transfer surface area of the heat sink (m2) Th temperature of the hot end (K)
C specific heat (J kg1 K1) T1 ambient temperature (K)
h heat transfer coefficient on surface of the heat sink V volume of the thermoelectric elements (m3)
(W m2 K1) x, y, z rectangular coordinates
ha heat transfer coefficient on side surfaces of P-type or
N-type element (W m2 K1) Greek symbols
I electrical current (A) a Seebeck coefficient of thermoelectric elements (V K1)
k thermal conductivity (W m1 K1) e electrical resistivity of thermoelectric elements (X m)
L height of the thermoelectric elements (m) l Thomson coefficient of thermoelectric elements (V K1)
La length of the cross-sectional areas of P-type or N-type q density of thermoelectric elements (kg m3)
element in y-direction (m)
Lb length of the cross-sectional areas of P-type or N-type Subscripts
element in z-direction (m) b cold plate of cold end
M mass (kg) cp ceramic plate of cold end
Q cooling load (W) hp ceramic plate of hot end
q cooling load per unit area (W cm2), Q/Ah n N-type thermoelectric element
Qc heat transfer rate between cold end and P-type or p P-type thermoelectric element
N-type element (W) s heat sink
Qh heat transfer rate between hot end and P-type or N-type
element (W)

the electrical conductivity and reduce the thermal conductivity [3]. number of TEC units required. Chakraborty et al. [14] presented a
Besides, a higher electrical conductivity of the material decreases thermodynamic modeling of a solid state thermoelectric cooling
the electrical resistance for electrical currents and hence lowers device based on the temperature-entropy analysis. Most recently,
the Joule heating, and a lower thermal conductivity of the material Lee and Kim [15] carried out the numerical analysis to evaluate
increases the thermal resistance for heat conduction from the hot the coefficient of performance of the thermoelectric micro-cooler
end to the cold end. with a nearly three-dimensional model which is used to investi-
To the authors’ knowledge, Stilbans and Regel [4] are the earli- gate the effects of the temperature difference and the current.
est researchers who carried out the dynamic behavior of the ther- According to the literature survey [16], it is recognized that
moelectric cooler. They found that the cooling rate is decreased as mostly the previous theoretical models are limited to the one-
the thickness of the thermoelectric elements is increased. Parrot dimensional problems. However, in these existing models, the
[5] extended the study to include the effects of variable heat con- P–N element pair is simply treated as a single bulk material so that
ductivity, which is treated as a function of temperature. Gray [6] the difference in thermal behavior between the two semiconductor
derived a mathematical model for predicting the dynamic behavior elements was not possible to evaluate. Only a few studies were
of TEC; however, the author did not take the hot end’s dynamic conducted for the three-dimensional ones, for example, the model
behavior into consideration. Bywaters and Blum [7] proposed the of Lee and Kim [15].
equivalent stage method (ESM) which was derived by simplifying In practices, the P-type and the N-type single crystal silicon ele-
the governing equations of Gray’s model, and used it to estimate ments are made by doping boron and phosphorous, respectively,
the dynamic behavior of a multi-layer TEC. Huang and Duang into the silicon wafer. According to the existing information, the
[8,9] presented a theoretical model, utilized the numerical solu- properties of the P-type and the N-type silicon are dependent on
tions of the model to control the temperature difference of the the material doped, dopant density and doping time. For example,
TEC device, and introduced the influences of thermal resistance Beadle et al. [17] addressed that the electrical resistivity of N-type
into their model for calculating the efficiency of TEC. Recently, a silicon is dependent on the dopant density. When the N-type sili-
dimensionless mathematical model is constructed by Naji et al. con is doped at a dopant density of 5  1017 cm3, its electrical
[10] to investigate the influences of the dimensionless parameters resistivity can reach 2.3  104 X m. The properties of the two
on the performance of TEC. Authors also found that the cold end semiconductor elements can be rather different under different
temperature was elevated as the electrical resistivity is decreased doping conditions. As a result, it may lead to significant difference
and was reduced as the Thomson coefficient is increased. Xuan in temperature response in the P-type and N-type elements. Under
et al. [11] constructed a general model to study the effects of elec- these circumstances, a three-dimensional model is developed in
trical resistances in the internal and external interface layers on this study that treats the P-type and the N-type semiconductor ele-
the performance of the thermoelectric devices. Huang et al. [12] ments as two separate parts and considers the difference in tem-
addressed the effects of the Thomson effect, the Joule heating, perature between the two elements. The model is also used to
the Fourier’s heat conduction, and the radiation and convection predict transient thermal behavior of the thermoelectric coolers.
heat transfer on the temperature distribution. Authors also found
that the cooling efficiency of the thermoelectric cooler can be im-
proved not only by increasing the figure of merit (ZT) but also by 2. Theoretical model
taking advantage of the Thomson effect. Khire et al. [13] combined
the thermoelectric devices with the solar energy units to design an The schematic of a thermoelectric cooler containing one semi-
active building envelope (ABE) system and mentioned that the conductor element pair is shown in Fig. 1. A thermoelectric cooler
thermal resistance of the heat sink is essential to determine the consists of a number of semiconductor element pairs. Neverthe-
 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011 2003

Ambient lected and compared with the numerical predictions to verify the
Temperature theoretical model.
hAs(Th-T ∞ ) The theoretical model is developed based on the following
T∞
assumptions:

(1) The hot and cold ends of the thermoelectric cooler are
Heat sink assumed to be lumped-capacity systems inside which the
Region 2 temperature is changed with time but is uniformly
(Hot-end, Th) distributed.
Ceramic plate (2) The material of the P-type and N-type elements is homoge-
Qhp Qhn
IαpTh -Iα nTh neous and isotropic. Furthermore, all the thermal and elec-
trical properties of the elements are constant and
P N independent of temperature.
Regions 3 and 4
(3) The heat transfer coefficients on the outer surfaces are con-
Ceramic plate
stant, and the ambient temperature (T1) is fixed at 298 K.
Qcp Qcn
IαpTc IαnTc The governing equations for the present study are expressed in
Region 1 the following statement:

(Cold-end, Tc)
Cold plate 2.1. Cold end (region 1)
Q
As shown in Fig. 1(a), the cold end consists of a ceramic plate
(a) Energy balance for all regions and a cold plate. The cooling load is the heat transfer rate that is
absorbed into the cold plate from the external object that is cooled
by this TEC. By treating the ceramic plate and the cold plate as a
system, the energy equation is expressed as

dT c
P N ðM cp C cp þ M b C b Þ ¼ Q þ Q cp þ Q cn  Iðap  an ÞT c ð2Þ
L L dt
xp zp xn zn where Mcp and Mb denote the masses of the ceramic plate and cold
plate, respectively, and Ccp and Cb the heat capacities; ap and an are
yp Lbp yn Lbn the Seebeck coefficients of the P-type and N-type elements; Q is the
cooling load absorbed from the external object; Qcp and Qcn repre-
Lap Lan
sent the heat conduction rates between the ceramic plate and the
(b) Dimensions of P-type and N-type thermoelectric elements P-type and N-type elements which can be calculated by
Z P
Fig. 1. Schematic of a thermoelectric cooler. @T p ðx; y; z; tÞ
Q cp ¼ kp  dA ð3aÞ
@x x¼0
Z N
@T n ðx; y; z; tÞ
less, the thermal characteristics among the pairs are periodic, and Q cn ¼ kn  dA ð3bÞ
@x
hence, only one pair is considered here. In this study, the solution x¼0

domain is divided into four major regions, namely, cold end (region where kp and kn represent the heat conductivities of the P-type and
1), hot end (region 2), and the P-type and N-type thermoelectric N-type elements, respectively, and A represents the cross-section
elements (regions 3 and 4), as shown in Fig. 1(a). Numerical solu- area of P-type or N-type elements. In Eqs. (3a) and (3b), it is noted
tions for the three-dimensional temperature fields in the P-type that the heat transfer rate is determined in terms of the integration
and the N-type semiconductor elements and transient temperature of the heat flux in direction normal to the plane.
variations in the cold and the hot ends can be carried out. In addi-
tion, based on the obtained temperature data, the magnitude of the 2.2. Hot end (region 2)
coefficient of performance (COP) of the thermoelectric cooler are
calculated in wide ranges of physical and geometrical parameters, As also indicated in Fig. 1(a), the hot end region includes a cera-
including electrical current (I), cooling load per unit area (q), heat mic plate and a heat sink that is used to dissipate heat to the ambi-
transfer coefficients on the surfaces (h and ha) and geometrical ent air. The energy equation for the hot end region is written as
parameters such as total area of heat sink (As) and height of semi-
dT h
conductor elements (L). The P-type and the N-type elements are ðM s C s þ Mhp C hp Þ ¼ Iðap  an ÞT h  Q hp  Q hn  hAs ðT h  T 1 Þ
rectangular columns of equal volume. A heat sink is attached on dt
the hot end of the thermoelectric cooler for heat dissipation. Note ð4Þ
that in this study the temperature distributions within the P-type where Ms, Mhp, Cs, and Chp are the masses and the heat capacities of
and N-type elements are predicted by the three-dimensional mod- the heat sink and the ceramic plate; Qhp and Qhn represent the heat
el. This is the major concern of the present paper. However, the hot conduction rates between the ceramic plate of hot end and the P-
and cold ends of the thermoelectric cooler are minor parts of the type and N-type elements which can be calculated by
modules and their specifications are directly dependent on the
Z p
practical applications. Therefore, they are assumed to be lumped- @T p ðx; y; z; tÞ
capacity bodies to simplify the mathematical model. Q hp ¼ kp  dA ð5aÞ
@x x¼L
For validation of the solution model, experiments are also con- Z N
@T n ðx; y; z; tÞ
ducted for a certain particular cases. Measurement data regarding Q hn ¼ kn
@x  dA ð5bÞ
the temperature variations in the cold and the hot ends are col- x¼L
2004 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011

2.3. P-type and N-type thermoelectric elements (regions 3 and 4) Table 1

Fixed parameters in computation.

An energy balance analysis for the P-type element leads to the Parameter Value Unit Source
energy equation: Ccp 419 1
J kg K 1
[8]
! Cb 400 J kg1 K1 [8]
@T p 2
@ Tp @ Tp @ Tp2 2
lp @T p ep Cs 900 J kg1 K1 [8]
C p qp ¼ kp þ þ  I þ I2 ð6aÞ Chp 419 J kg1 K1 [8]
@t @x2 @y2 @z2 A @x A2 Cp 200 J kg1 K1 [19–23]
Cn 200 J kg1 K1 [19–23]
where Cp, qp, lp, and ep represents the specific heat, the density, the La 1.5  103 m Measured
Thomson coefficient, and the electrical resistivity of P-type element, Lb 1.5  103 m Measured
respectively, and I is the electrical current. Similarly, one may have Mcp 2.95  105 kg Measured
Mb 0 kg No cold plate
the energy equation for the N-type element as
Ms 8.64  105 kg Measured
! Mhp 2.95  105 kg Measured
@T n @2T n @2T n @2T n ln @T n en kp 1.8 W m1 K1 [18]
C n qn ¼ kn þ þ þ I þ I2 ð6bÞ kn 2.2 W m1 K1 [18]
@t @x2 @y2 @z2 A @x A2
Tp 300 K Measured
T1 298 K Measured
Tn 298 K Measured
2.4. Boundary conditions ap 0.00015 V K1 [18]
an 0.00019 V K1 [18]
The boundary conditions on the surfaces of the P-type and the qp 10922.08 kg m3 Measured
N-type elements associated with Eqs. (6a) and (6b) are listed as qn 10922.08 kg m3 Measured
lp 0.00027 V K1 [18]
follows:
ln 0.000156 V K1 [18]
ep 1.2  105 Xm [15]
(1) Surfaces at yp = 0 and yn = 0 en 1  105 Xm [15]

@T p ðx; y; z; tÞ h i
kp ¼ ha T p ðx; y; z; tÞjPy¼0  T 1 ; and ð7aÞ
@y 2.5. Coefficient of performance
@T n ðx; y; z; tÞ h i
kn ¼ ha T n ðx; y; z; tÞjNy¼0  T 1 ð7bÞ The coefficient of performance (COP) is probably the most
@y
important index regarding the performance of the thermoelectric
coolers. In this study, it is defined by
(2) Surfaces at yp = Lap and yn = Lan
QL
COP ¼ ð13Þ
h i W
@T p ðx; y; z; tÞ
 kp ¼ ha T p ðx; y; z; tÞjPy¼Lap  T 1 ; and ð8aÞ where
@y
@T n ðx; y; z; tÞ h i
 kn ¼ ha T n ðx; y; z; tÞjNy¼Lan  T 1 ð8bÞ QL ¼ Q ð14aÞ
@y
 
ep 2 en
W ¼ Iðap  an ÞðT h  T c Þ þ 2
I V p þ 2 I2 V n
(3) Surfaces at zp = 0 and zn = 0 A A
 
lp @T p ln @T n
h i þ  I Vp þ I Vn ð14bÞ
@T p ðx; y; z; tÞ A @x A @x
kp ¼ ha T p ðx; y; z; tÞjPz¼0  T 1 ; and ð9aÞ
@z Note that the net work W is influenced by the Seebeck effect, the
@T n ðx; y; z; tÞ h i
kn ¼ ha T n ðx; y; z; tÞjNz¼0  T 1 ð9bÞ Joule heating, and the Thomson effect. The cooling load per unit
@z area (q), in W/cm2, is calculated with

Q
(4) Surfaces at zp = Lbp and zn = Lbn q¼ ð15Þ
Ah

@T p ðx; y; z; tÞ h i where Q is the cooling load and Ah is the hot end area, which is fixed
 kp ¼ ha T p ðx; y; z; tÞjPz¼Lbp  T 1 ; and ð10aÞ at 0.125 cm2 herein.
@z
@T n ðx; y; z; tÞ h i The magnitude of the coefficient of performance (COP) of the
 kn ¼ ha T n ðx; y; z; tÞjNz¼Lbn  T 1 ð10bÞ thermoelectric cooler is calculated when the steady regime is at-
@z
tained. The transient variation in temperatures and the COP value
at steady state are predicted in wide ranges of physical and geo-
(5) At xp = 0 and xn = 0 metrical parameters. The varied parameters in this study include

T p ðx; y; z; tÞjx¼0 ¼ T c ðtÞ; and ð11aÞ

Table 2
T n ðx; y; z; tÞjx¼0 ¼ T c ðtÞ ð11bÞ Specific heats of thermoelectric materials.

Material C (250 K) C (298 K) C (400 K) Unit Reference

(6) At xp = L and xn = L Bi2Te3 186.587 189.834 194.246 J kg1 K1 [19]
Bi2Te3 154.412 157.588 163.285 J kg1 K1 [20]
Bi2Se3 197.218 203.613 208.774 J kg1 K1 [21]
T p ðx; y; z; tÞjx¼L ¼ T h ðtÞ; and ð12aÞ Sb2Te3 203.257 205.493 209.963 J kg1 K1 [22]
T n ðx; y; z; tÞjx¼L ¼ T h ðtÞ ð12bÞ Bi0.52 Sb1.48Te3 182.860 187.059 194.473 J kg1 K1 [23]
 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011 2005

Mesh with all the cases in computation. Densities of the P-type and N-
11x11x16 type elements (10922.08 kg m3) are determined by direct mea-
330 21x21x31 Th surement in terms of the mass and volume of the element. And,
31x31x46
41x41x61 according to the measurement data, no appreciable difference in
density between these two elements is found. Note that the values
of the specific heats, Cp and Cn, are dependent on the thermoelec-
tric material used; however, it is found that even for the same
315 material, the information is rather scattering. Some of the existing
information of the specific heat taken from Refs. [19–23] are listed
in Table 2. It can be found that the specific heat of the major ther-
T/(K)

Tc moelectric materials is within 154–210 J kg1 K1 and according to

the existing data, no appreciable difference between the P-type
300 and N-type materials is found. Therefore, in this study, the values
of the specific heats are assigned to be Cp = 200 J kg1 K1 and
Cn = 200 J kg1 K1.

3. Numerical methods
285

100 200 300 400 500 600 700 The governing equations are discretized based on the finite-dif-
ference method. These time derivative terms in Eqs. (6a) and (6b)
Time/(s)
are treated by the forward-difference scheme. A 31  31  46 grid
Fig. 2. Grid-independency check by a comparison in hot end and cold end point system is adopted for the solution domain y  z  x after a
temperatures obtained from different grids. careful grid-independence check. Fig. 2 displays the grid-indepen-
dence of the solutions for the case with As = 4.254  105 m2,
h = 20 W m2 K1, ha = 10 W m2 K1, I = 1 A, L = 2.25  103 m,
Table 3
and q = 800 W cm2. Other fixed parameters are equal to those gi-
Base-line case parameters.
ven in Table 3. In this figure, four grid systems, 11  11  16,
Parameter Value Unit 21  21  31, 31  31  46, and 41  41  61, are tested, and the
Ah 1.25  105 m2 comparison in the time histories of the hot end and the cold end
As 4.254  105 m2 temperatures for the case are presented. In addition, the magni-
h 20 W m2 K1
tudes of COP by different grids are also provided in a small table
ha 10 W m2 K1
I 1 A
shown in the figure. It is found that for this case the relative errors
L 2.325  103 m in the temperature solutions and COP between 31  31  46 and
q 800 W m2 41  41  61 grids are all within 0.1%. However, note that the com-
putation time required with the 41  41  61 grids is three times
that with the 31  31  46 grids. Hence, to save computation ef-
the applied current (I), total heat transfer surface area of the heat forts without loss in accuracy, the 31  31  46 grids are adopted.
sink (As), heat transfer coefficient on surface of the heat sink (h),
heat transfer coefficient on side surfaces of thermoelectric ele- 4. Experiments
ments (ha), height of the thermoelectric elements (L), and cooling
load per unit area (q). On the other hand, the parameters are deter- By conducting experiments for the thermal behavior of the
mined either according to direct measurement or from the existing thermoelectric module, the validity of the simulation model is pos-
information [8,15,18–23]. Table 1 displayed the fixed parameters sible to be confirmed in part. Herein a zero-cooling-loading test has

Heat sink

Thermoelectric cooler module

Ceramic fiber blanket

Thermocouples
PC

Data logger
Power supply
Fig. 3. Experimental system.
2006 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011

been performed for validation of the simulation model. The reason (2) Heat sink: A plate-fin heat sink is attached on the top of the
for performing the zero-cooling-load test is to avoid the errors in ceramic plate at hot end for heat dissipation to the ambient.
measuring the magnitude of the cooling load and the heat losses The heat sink is made of aluminum 6063 with a total heat
with it which could lead to uncertainties in the testing conditions. transfer area of As = 4.254  105 m2. Total heat transfer area
The experimental apparatus of the experiments is shown in Fig. 3, of the heat sink was divided by the number of element pairs
which consists of the following components: to yield the heat transfer area per element pair, which is
used in the one pair simulation.
(1) Thermoelectric cooler module: The thermoelectric cooler mod- (3) Ceramic fiber blanket: To fulfill a zero-cooling-loading condi-
ule used is a typical one which comprises 127 pairs of P-type tion, a ceramic fiber blanket (Al2O3 and SiO2) is placed under
and N-type elements connected in series which are clamped the cold end of the thermoelectric model for insulation.
by two ceramic plates. The area of the thermoelectric cooler (4) Power supply: The power supply is used to supply a DC cur-
module is 4 cm4 cm. For this particular case, the dimen- rent pass through the device. The accuracy of the power sup-
sions of the P-type and the N-type elements are ply is ±0.5%.
L = 2.325  103 m, La = 1.5  103 m, and Lb = 1.5  103 m. (5) Thermocouples: Three T-type thermocouples (copper/cop-
In the zero-cooling-load case, q = 0. per-constantan alloy) are installed for temperature

320 360
I = 0.5A I = 1A
Th
Th 350

340
310
T/(K)

330
T/(K)

Tc
320
Tc
300
310

300

290
-Numerical - Numerical
- Experimental 290 - Experimental

280
0 100 200 300 400 500 0 100 200 300 400 500

Time/(s) Time/(s)
(a) I = 0.5 A (b) I = 1 A

425
I = 1.5A Th
400

375
Tc
T/(K)

350

325

300
- Numerical
- Experimental
275
0 100 200 300 400 500

Time/(s)
(c) I = 1.5 A
Fig. 4. Comparison in temperature variation between numerical predictions and experiments at zero-cooling-load per unit area (q = 0). (a) I = 0.5 A, (b) I = 1 A, and (c)
I = 1.5 A.
 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011 2007

measurement, whose error is within ±0.5°. For monitoring geometrical parameters of the simulation model are identical to
the hot and cold ends temperatures, one thermocouple is the specifications of the experimental device.
placed at the interface between the hot end and the heat The comparisons between the numerical predictions and exper-
sink, and one at the interface between the cold end and iments are made at I = 0.5, 1.0, and 1.5 A. All the fixed parameters
the ceramic fiber blanket. The third thermocouple is in the computation are already given in Table 1. Note that the heat
installed for measuring the ambient temperature. transfer coefficients h and ha are assigned to be 20 and
(6) Data logger: Fluke 2620A Data Logger is used for temperature 10 W m2 K1, respectively. The values are corresponding to the
data collection. An RS232 serial interface is used to connect air natural convection situation based on the operating conditions
the data logger to a personal computer for real time data of experiments. Fig. 4 shows the comparison in temperature vari-
acquisition. ation between the numerical predictions and the experiments at
q = 0. It is found by experiments that the hot end temperature in-
The experiment is started as the DC current is applied steadily creases with time monotonically. For the case at I = 0.5 A, the hot
by the power supply. The temperature data of the hot and the cold end temperature reaches a temperature of 313 K in 500 s; how-
sides measured by the thermal couples are transmitted to data ever, in the same period of time it reaches 422 K for the case at
logger and transmitted into computer via RS232 for analysis. The I = 1.5 A. On the other hand, the cold end temperature descends

As [m2]2
I [A] As [m ] Th
380
2 330 8.508×10-5
Th 8.508E-5
4.254×10-5
1.5 4.254E-5-5
1 2.127×10
2.127E-5
360 0.5 Th

Th Th/(K) & Tc/(K) 315

340 Th
Th/(K) & Tc/(K)

Tc
Tc
320 300
Th
Tc Tc
Th
300
Tc
Tc Tc
285
280

100 200 300 400 500 0 100 200 300 400 500
Time/(s) Time/(s)
(a) Th and Tc Fig. 6. Effects of total area of the heat sink on transient temperature variation for
base-line case at steady state.
40

[W/K-m2]
Th
340 h
40
20
5

320 Th
COP

20 Tc
Th
Th/(K) & Tc/(K)

300
Tc

Tc

0 280

0 0.5 1 1.5
I/(A) 0 100 200 300 400 500
(b) COP Time/(s)
Fig. 5. Effects of applied current on transient temperature variation and COP for Fig. 7. Effects of heat transfer coefficient on the surface of the heat sink on transient
base-line case at steady state. (a) Th and Tc; (b) COP. temperature variation for base-line case at steady state.
2008 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011

from start to a minimum value and then goes up to a value higher parameters. In this study, for comparisons a base-line case is cho-
than the ambient temperature. This may be explained by Eq. (2). At sen and analyzed firstly. The physical and geometrical parameters
beginning, the Seebeck effect causes heat output from the cold end with the base-line case are displayed in Table 3. It is noted that the
so that the cold end temperature is decreased. However, after a parameters of the base-line case are identical to the conditions of
period of time, when the temperature of hot end is increased, the the experiments described in the subsequent section. Also note
heat conduction from the hot to the cold end is also increased that the current with the base-line case is assigned to be 1 A
and it results in the increase in the cold end temperature. The dif- according to the specifications of the existing commercial TEC
ference between the hot- and the cold end temperatures is approx- modules. In the parametric study, when a parameter is varied to
imately maintained at 12, 25, and 40 K for the cases at I = 0.5, 1.0, evaluate its effects, other parameters remain the same as with
and 1.5 A, respectively. The numerical predictions by the simula- the base-line case. The following figures are plotted for the base-
tion model closely agree with the experimental data. Hence, the line case; therefore, the conditions for the case are already given
validity of the present simulation model can be ensured. in Tables 1–3.

5. Results and discussion 5.1. Effects of applied current

Next, the simulation model is adopted to perform a parametric Fig. 5 shows the effects of applied current on the transient var-
study to investigate the effects of the physical and geometrical iation in temperatures and the COP value at steady state for the

380
[W/K-m2]
L [mm]
ha
10 Th
15 5
10 Th 360 2.325
320 5

340 Th
Th/(K) & Tc/(K)

Th/(K) & Tc/(K)

320 Th
300
Tc
300
Tc
Tc
Tc
280

280
260
0 100 200 300 400 500 0 100 200 300 400 500
Time/(s) Time/(s)
(a) Th and Tc (a) Th and Tc

0.49
0.5

0.489
0.45

0.488 0.4

0.487 0.35
COP

COP

0.486 0.3

0.25
0.485

0.2
0.484
0.15
0.483
0.1
10 20 30 2 3 4 5 6 7 8 9
ha /(Wm-2K-1) L/(mm)
(b) COP (b) COP
Fig. 8. Effects of heat transfer coefficient on the surfaces of the P-type and N-type Fig. 9. Effects of height of the P-type and N-type elements height of the P-type and
elements on transient temperature variation and COP for base-line case at steady N-type elements on transient temperature variation and COP for base-line case at
state. (a) Th and Tc; (b) COP. steady state. (a) Th and Tc; (b) COP.
 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011 2009

cases with I = 0.5, 1.0. 1.5, and 2.0 A. Note that for this case, the 0.01 m, and it is found that the temperature difference between
cooling load per unit area q is fixed at 800 W cm2. As shown in the two ends increases with L. However, as L is changed from
this figure, the temperature of the hot end and the temperature dif- 0.002325 to 0.01 m, the value of COP is reduced approximately
ference between the hot and cold ends both increase with the ap- from 0.5 to 0.1.
plied current. This is because in Eq. (6a), the heat sink term due to
Thomson effect is proportional to current while the heat source 5.6. Effects of cooling load
term due to Joule heating is proportional to the square of current.
As a result, once the value of current is elevated, the heat source The cooling load could be different in response to various situ-
term gradually dominates, and hence the slope of increase in the ations. Under different cooling loads, the thermal behavior of the
temperatures of the P-type and the hot end are both higher. In a thermoelectric cooler could be different. Effects of cooling load
similar manner, even the cold end temperature is decreased at on thermal behavior are illustrated in Fig. 10 for the base-line case.
the beginning, it is increased to a value exceeding the ambient In the cases shown in Fig. 10(a), the value of q is assigned to be 0,
temperature in the steady state regime. However, if the magnitude 400, 800, and 2400 W cm2. It is observed that both the tempera-
of the current is not high enough (say, 0.5 or 1.0 A), the cold end tures of the hot and the cold ends increase with q. The temperature
temperature will still be maintained at a lower value. The COP is difference (Th  Tc) reaches a maximum value of 24.51 K for the
displayed in Fig. 5(b) as a function of applied current. It is found case at q = 0. Furthermore, as shown in Fig. 10(b), the value of
that the magnitude of COP is rapidly decreased as the current is in- COP linearly increases with q. This complies with Eq. (13) in which
creased. This may be attributed to the increase in the net work (W), the COP appears to be proportional to the cooling load per unit area
which is defined by Eq. (14b), resulting from an increase in current.
2
q [W/cm ]
5.2. Effects of total area of the heat sink 0.24
0.08
Convective heat transfer from the heat sink to the ambient 0.04
340 0
could be influential. Fig. 6 shows the results for the cases with As
varied from 2.127  105 to 8.508  105 m2. According to Eq. (4)
for Th, it is expected that the increasing rate of hot end tempera- Th
ture, dTh/dt, is lowered when the magnitude of hAs(Th  T1) is in-
Th/(K) & Tc/(K)

Th
Th
creased. The curves plotted in Fig. 6 agree with the expectation. 320 Th
The hot end temperature is remarkably reduced by increasing
the total area of the heat sink. Note that the dependence of cold
end temperature on total area of the heat sink is similar to that Tc Tc
of the hot end temperature due to the conduction through the 300
semiconductor elements between the two ends. Tc
Tc

5.3. Effects of heat transfer coefficient on the surface of the heat sink

Fig. 7 shows the effects of heat transfer coefficient on the sur- 280
face of the heat sink on transient temperature variation. Basically, 100 200 300 400 500
to increase heat dissipation by convection one may increases the Time/(s)
value of either As or h, if the temperature difference Th  T1 is kept
constant. The effect of h on the thermal behavior of thermoelectric
(a) Th and Tc
cooler is similar to that of As. When the value of h is assigned to be
5
5, 20, and 40 W m2 K1, the steady state hot end temperature is
340.9 K, 318.8 K, and 309.9 K and the cold end temperature
4.5
313.7 K, 294.3 K, and 286.4 K, respectively.
4
5.4. Effects of heat transfer coefficient on the surfaces of the P-type and
N-type elements 3.5
COP

A heat loss from the surfaces of the P-type and N-type elements 3
causes a reduction in the COP. Fig. 8 shows the effects of ha on tran-
sient temperature variation and COP at steady state. In Fig. 8(b), it 2.5
is observed that as ha is elevated from 5 to 15 W m2 K1, the value
2
of COP is reduced from 0.4896 to 0.483. Meanwhile, an increase in
ha results in a decrease in both Th and Tc; therefore, the tempera- 1.5
ture difference is only slightly influenced by a change in the heat
transfer coefficient. 1

5.5. Effects of height of the P-type and N-type elements 0.5

Effects of the height of the semiconductor elements are exhib- 0

0 0.2 0.4 0.6 0.8
ited in Fig. 9. The influence of an increase in element height is actu- q/(Wcm-2)
ally involved because it increases the thermal resistance against
heat conduction from the hot end to the cold but also increases
(b) COP
electrical resistance as well as the Joule heating effects. In Fig. 10. Effects of cooling load on transient temperature variation and COP for
Fig. 9(a), the value of L is assigned to be 0.002325, 0.005 m, and base-line case at steady state. (a) Th and Tc; (b) COP.
2010 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011

q as long as the magnitude of the net work W is fixed. In this case, sional model particularly when the properties of the two elements
the temperature difference between the hot and the cold ends is are rather different. Plotted in Fig. 11 are the planar temperature
only slightly decreased by increasing the value of q. Therefore, distributions at different z- and x-locations within the P-type and
the magnitude of the net work is not appreciably altered. the N-type elements, of which the electrical resistivities are set
In addition, the Joule heating and the Thomson terms in Eqs. to be 3  105 and 6.5  106 X m, respectively. Note that these
(6a) and (6b) can be regarded as a heat source and a heat sink two values are still within the reasonable range of electrical
terms, respectively. Therefore, in general the Joule heating leads resistivity in practical cases [15]. Other parameters are equal to
to a reduction in the magnitude of COP, whereas the Thomson ef- those of the base-line case. Here, the difference in electrical resis-
fect tends to increase the COP. However, the investigation of the tivity between the two elements results in the discrepancy in tem-
subtle influence of these effects is still open to discussion and perature distribution. The steady state temperature contours on
worth further investigation. the x–y cross-sectional planes at z = 0, Lb/4, and Lb/2, and on the
As stated earlier, the present three-dimensional model treats y–z cross-sectional planes at x = L/4, L/2, and 3L/4, are plotted
the P-type and the N-type semiconductor elements as two separate and the discrepancy in temperature distributions between the
parts. Therefore, one may take advantages of the three-dimen- two elements is clearly seen.

Fig. 11. Steady-state temperature distributions at different locations within the P-type and the N-type elements. (a) At different z-locations. (b) At different x-locations.
 C.-H. Cheng et al. / International Journal of Heat and Mass Transfer 53 (2010) 2001–2011 2011

6. Conclusions 2221-E-006-111-MY3. Authors would also like to thank the Na-

tional Space Organization, Taiwan, ROC, for their partial financial
In this model, the solution domain is divided into four major re- support and help in experiments.
gions, namely, cold end (region 1), hot end (region 2), and the P-
type and N-type thermoelectric elements (regions 3 and 4). The
present model treats the P-type and the N-type semiconductor ele- References
ments as two separate parts so that the difference in thermal
[1] Y. Kraftmakher, Simple experiments with a thermoelectric module, Euro. J.
behavior between the two elements can be evaluated. To verify Phys. 26 (2005) 959–967.
the numerical predictions, experiments have been conducted to [2] G.D. Mahan, Figure of merit for thermoelectrics, J. Appl. Phys. 65 (1989) 1578–
measure the temperature variations in the cold and the hot ends 1583.
[3] R. Venkatasubramanian, E. Siivola, T. Colpitts, B. O’Quinn, Thin-film
for a certain particular cases. Close agreement between the numer- thermoelectric devices with high room-temperature figures of merit, Nature
ical and the experimental data has been observed. The simulation 413 (2001) 597–602.
model is adopted to perform a parametric study to investigate the [4] L.S. Stilbans, A.R. Regel, Thermoelect. Power Generat. 1 (1968) 1341. Soviet
Physics Semiconductors-USSR.
effects of the parameters such as electrical current (I), cooling load [5] J.E. Parrott, A new theory of size effect in electrical conduction, Proc. Phys. Soc.
per unit area (q), heat transfer coefficients on the surfaces (h and Lond. 85 (1965) 1143.
ha) and geometrical parameters such as total area of heat sink [6] P.E. Gray, The Dynamic Behavior of Thermoelectric Devices, Wiley, New York,
1960.
(As) and height of semiconductor elements (L). [7] R.P. Bywaters, H.A. Blum, The transient behavior of cascade thermoelectric
The following conclusions are yielded based on the obtained heat pumps, Energy Convers. 10 (1970) 193–200.
results: [8] B.J. Huang, C.L. Duang, System dynamic model and temperature control of a
thermoelectric cooler, Int. J. Refrigerat. 23 (2000) 197–207.
[9] B.J. Huang, C.L. Duang, A design of thermoelectric cooler, Int. J. Refrigerat. 23
(1) The temperature of the hot end and the temperature differ- (2000) 208–218.
ence between the hot and cold ends both increase with the [10] M. Naji, M. Alata, M.A. Al-Nimr, Transient behavior of a thermoelectric
applied current. It is also found that the magnitude of COP device, Proc. Inst. Mechan. Eng. A: J. Power Energy 217 (2003) 615–
621.
is rapidly decreased as the current is increased. This may [11] X.C. Xuan, K.C. Ng, C. Yap, H.T. Chua, A general model for studying effects of
be attributed to the increase in the net work input resulting interface layers on thermoelectric devices performance, Int. J. Heat Mass
from an increase in current. Transfer 45 (2002) 5159–5170.
[12] M.J. Huang, R.H. Yen, A.B. Wang, The influence of the Thomson effect on the
(2) Convective heat transfer from the heat sink to the ambient performance of a thermoelectric cooler, Int. J. Heat Mass Transfer 48 (2005)
could be influential. The hot end temperature is remarkably 413–418.
reduced by increasing the convective heat transfer of the [13] R.A. Khire, A. Messac, S.V. Dessel, Design of thermoelectric heat pump unit for
active building envelope systems, Int. J. Heat Mass Transfer 48 (2005) 4028–
heat sink. The dependence of cold end temperature on the 4040.
heat rejection from heat sink is similar due to the conduction [14] A. Chakraborty, B.B. Saha, S. Koyama, K.C. Ng, Thermodynamic modeling of a
through the semiconductor elements between the two ends. solid state thermoelectric cooling device: temperature-entropy analysis, Int. J.
Heat Mass Transfer 49 (2006) 3547–3554.
(3) A heat loss from the surfaces of the P-type and N-type ele- [15] K.H. Lee, O.J. Kim, Analysis on the cooling performance of the thermoelectric
ments causes a reduction in the COP; however, the temper- micro-cooler, Int. J. Heat Mass Transfer 50 (2007) 1982–1992.
ature difference is only slightly influenced by a change in the [16] A.J. Minnich, M.S. Dresselhaus, Z.F. Ren, G. Chen, Bulk nanostructured
thermoelectric materials: current research and future prospects, Energy
heat transfer coefficient.
Environ. Sci. 2 (2009) 466–479.
(4) The influence of an increase in element height is actually [17] W.E. Beadle, J.C. Tsai, R.D. Plummer, Quick Reference Manual for Silicon
involved. It is found that the temperature difference Integrated Circuit Technology, John Wiley& Sons, 1985.
between the two ends increases with L. However, as L is [18] D.M. Rowe (Ed.), CRC Handbook of Thermoelectric, CRC Press, USA, 1995, pp.
145–230.
changed from 0.002325 to 0.01 m, the value of COP is [19] D.M. Rowe (Ed.) Thermoelectrics Handbook, CRC Press, USA, 2006, pp.
reduced approximately from 0.5 to 0.1. 27–28.
(5) It is observed that both the temperatures of the hot and the [20] N.P. Gorbachuk, A.S. Bolgar, V.R. Sidorko, L.V. Goncharuk, Heat capacity and
enthalpy of Bi2 Se3 and Bi2 Te3 in the temperature range 58–1012 K, Powder
cold ends increase with cooling load q, and the value of COP Metallur. Metal Ceram. 43 (2004) 284–290.
linearly increases with q. In the cases considered, the tem- [21] A.S. Pashinkin, A.S. Malkova, Heat capacity of solid bismuth selenide (Bi2 Se3),
perature difference between the hot and the cold ends is Zhurnal Fizicheskoi Khimii 79 (2005) 1325–1327.
[22] A.S. Pashinkin, A.S. Malkova, M.S. Mikhailova, Heat capacity of solid antimony
only slightly decreased by increasing the value of q. telluride, Zhurnal Fizicheskoi Khimii 81 (2008) 1–3.
[23] Y.I. Shtern, A.S. Malkova, A.S. Pashinkin, V.A. Fedorov, Heat Capacity of the n-
Acknowledgements Bi2Te2.88Se0.12 and p-Bi0.52Sb1.48Te3 Solid Solutions, Inorgan. Mater. 44 (2008)
1057–1059.

The authors would like to thank the National Science Council,

Taiwan, ROC, for their financial support under Grant: NSC 97-