Thermoelectric Seebeck
Thermoelectric Seebeck
Thermoelectric Seebeck
Lord Kelvin
(1824-1907; William Thomson)
From Lecture to the
Institution of Civil Engineers, 3 May 1883
Seebeck effect: A temperature difference between two points in a conductor or semiconductor results in a
voltage difference between these two points. Stated differently, a temperature gradient in a conductor or a
semiconductor gives rise to a built-in electric field. This phenomenon is called the Seebeck effect or the
thermoelectric effect. The Seebeck coefficient gauges the magnitude of this effect. The thermoelectric
voltage developed per unit temperature difference in a conductor is called the Seebeck coefficient. Only the
net Seebeck voltage difference between different metals can be measured. The principle of the
thermocouple is based on the Seebeck effect.
1. The Seebeck Effect and Normal Metals
Consider an aluminum rod that is heated at one end and cooled at the other end as depicted in Figure 1. The
electrons in the hot region are more energetic and therefore have greater velocities than those in the cold
region1. Consequently there is a net diffusion of electrons from the hot end toward the cold end which
leaves behind exposed positive metal ions in the hot region and accumulates electrons in the cold region.
This situation prevails until the electric field developed between the positive ions in the hot region and the
excess electrons in the cold region prevents further electron motion from the hot to cold end. A voltage is
therefore developed between the hot and cold ends with the hot end at positive potential. The potential
1
The conduction electrons around the Fermi energy have a mean speed that only has a small temperature dependence. This
small change in the mean speed with temperature is, nonetheless, important in understanding the thermoelectric effect.
difference V across a piece of metal due to a temperature difference T is called the Seebeck effect2. To
gauge the magnitude of this effect we introduce a special coefficient which is defined as the potential
difference developed per unit temperature difference, i.e.
S=
dV
dT
Seebeck coefficient
(1)
E
Hot
Temperature, T
Cold
EF
EF
f(E)
+
+
Hot +
+
+
Conductor
Voltage V
f(E)
Cold
V =
SdT
(2)
To
The voltage difference in Equation (2) above is for the cold end with respect to hot as in the
convention for S.
The average energy Eav per electron in a metal in which the density of states g(E) E1/2 is given by
(see, for example, Ch. 4 in Principles of Electronic Materials and Devices, McGraw-Hill),
Thomas Seebeck observed the thermoelectric effect in 1821 using two different metals as in the thermocouple which is the
only way to observe the phenomenon. It was Thompson (Lord Kelvin) who explained the observed effect.
5 2 kT 2
3
Eav (T ) = EFO 1 +
(3)
5
12 EFO
where EFO is the Fermi energy at 0 K. It is clear from Equation (3) that the Fermi-Dirac distribution actually
extends to much higher energies when the temperature is raised, as depicted in Figure 1, so that the average
energy per electron, as determined by Equation (3), is actually greater in the hot end. Consequently the
more energetic electrons in the hot end diffuse toward the cold region until a potential difference V is built
up which prevents further diffusion. We should also note that the average energy per electron as
determined by Equation (3) also depends on the material by virtue of EFO.
T+ T T
T
Hot
Cold
Figure 2
Table 1
Seebeck coefficients of selected metals (from various sources).
S at 0 C
S at 27 C
EF
( V K )
( V K )
(eV)
Na
3.1
2.2
-12.5
2.0
3.8
1.8
11.6
2.78
7.1
1.38
4.7
9.7
Metal
-1
-1
Al
1.6
Mg
1.3
Pb
1.15
1.3
Pd
9.00
9.99
Pt
4.45
5.28
Mo
+4.71
+5.57
Li
+14
Cu
+1.70
+1.84
7.0
1.79
Ag
+1.38
+1.51
5.5
1.14
Au
+1.79
+1.94
5.5
1.48
Zn
9.4
Suppose that a small temperature difference of T results in a voltage difference V between the
accumulated electrons and exposed positive metal ions as depicted in Figure 2. Suppose that one electron
manages to diffuse from the hot region to the cold region. It has to do work against the potential difference
V which is eV. This work done against V decreases the average energy of the electron by Eav from
Eav(hot) to Eav(cold):
eV = Eav(T + T) Eav(T)
Substituting for Eav(T) from Equation (3) and expanding (T + T) and neglecting T2 term we
obtain,
eV
2 k 2TT
2 E FO
2 k 2T
2eE FO
(4)
For example, for Al, EFO = 11.6 eV so that at T = 300 K (27 C), Equation (4) predicts 0.94 V Kwhich is of the order of the experimentally inferred value of about 1.8 V K-1. Table 1 summarizes some
typical experimental values for the Seebeck coefficient of a selection of metals where it is apparent that the
values are in the microvolt per Kelvin range. It is also startlingly apparent that there are metals with positive
Seebeck coefficients such as copper. This mean that electrons migrate from cold to hot end of a copper bar.
It should be emphasized that the above explanation is based on assuming that the conduction
electrons in the metal behave as if they were free. This means that the density of states g(E) E1/2 up to
and beyond the Fermi energy. It also means that the electron energy E = KE = 1/2m e*v2 and that the
effective electron mass m e* is constant; energy independent. Further, electrons with higher energy have
greater mean speeds and longer mean free paths so that they diffuse from the hot to cold region. These
assumptions only apply to what are called normal metals (e.g. Na, K, Al etc.).
1
x
Energy
Figure 3
Consider two neighboring relatively hot and cold regions, H and C, in a conductor at one instant as
shown in Figure 3. The electron concentrations in H and C are n and n. The width of the H region is , the
mean free path (MFP) along x in H and that of C is , the MFP along x in C. Electrons in H moving
towards the interface and, within a distance , cross the interface into C. Only half of these would be
moving towards C so that the number of electrons that cross into C is 1/2(n); assume that the cross
sectional area is unity. If is the mean scattering time then the electron flux (number of electrons flowing
per unit area per unit time) from H to C is 1/2(n)/. There is a similar electron flux from C to H so that the
net flux from H to C is
n n
(5)
2 2
We can write n n + (dn/dx)x, + (d/dx)x, and + (d/dx)x. Taking x = ( +
)/2 , then Equation (5) becomes,
=
or
2 n n n2
+
2 x 2 x 2 2 x
(6a)
2 n n n2 ln
+
2 x 2 x 2 x
(6b)
It is clear that the net electron migration, whether from hot to cold, or cold to hot, is determined by
the energy dependence of the electron concentration n, MFP and the mean scattering time . For example,
if the MFP increases strongly with energy, /E and hence /x will be negative. This will make in
Equation (6b) positive (+x direction) so that electron diffusion will be from hot to cold and the
thermoelectric power will be negative. In those metals in which decreases strongly with the energy,
electrons migrate from cold to hot and the thermoelectric power is positive. These conclusions apply
primarily to metals.
By including the energy dependence of the scattering processes, Mott and Jones have derived the
following expression for the Seebeck coefficient,
2 k 2T
x
3eE FO
(7)
where x is a numerical constant that depends on the energy dependences of various charge transport
parameters; a few examples for x are given in Table 1. These x values allow Equation (7) to agree with the
experimental Seebeck coefficients at 273 K given EFO.
Equation (7) does not apply to metals in which electrons can be scattered from one transport band
to another transport band as in transition metals. In transitions metals (e.g. Ni) s and d bands overlap and
indeed the d-band may even be substantially full. The electrons in these two bands have different effective
masses and different mean free paths with different energy dependences.
3. The Thermocouple
Consider an aluminum rod heated at one end and cooled at the other end. Suppose that we try to measure
the voltage difference V across the aluminum rod by using aluminum connecting wires to a voltmeter as
indicated in Figure 4. The same temperature difference, however, now also exists across the aluminum
connecting wires and therefore an identical voltage also develops across the connecting wires, opposing that
across the aluminum rod. Consequently no net voltage will be registered by the voltmeter. It is, however,
possible to read a net voltage difference, if the connecting wires are of different material, i.e. have a different
Seebeck coefficient than that of aluminum, so that across this material the thermoelectric voltage is different
than that across the aluminum rod as in Figure 5.
The Seebeck effect is fruitfully utilized in the thermocouple (TC), shown in Figure 5, which uses
two different metals with one junction maintained at a reference temperature T0 and the other used to sense
the temperature T. The voltage across each metal element depends on its Seebeck coefficient so that the
potential difference between the two wires will depend on SA SB. The emf between the two wires, VAB =
VA VB, by virtue of Equation (2), is then given by
VAB =
(S
SB ) dT =
T0
AB
dT
(8)
T0
where SAB = SA SB is defined as the thermoelectric power for the thermocouple pair A-B. For the chromelalumel (K-type) TC, for example, SAB 40 V K-1 at 300 K.
Al
Hot
0 C
100 C
Cold
0
Al
Al
Figure 4
Al
Hot
0 C
100 C
Cold
0
Ni
Ni
The output voltage from a TC pair obviously depends on the two metals used. Instead of tabulating
the emf from all possible pairs of materials in the world, which is an impossible task, engineers have
tabulated the emfs available when a given material is used with a reference metal which is chosen to be
platinum. The reference junction is kept at 0 C (273.16 K) which corresponds to a mixture of ice and
water. Some typical materials are listed in Table 2 to compare their emfs.
By using the expression for the Seebeck coefficient, Equation (7), in Equation (8) we can readily
show, through simple mathematics, that the integration leads to the familiar thermocouple equation,
VAB = aT + b( T ) 2
Thermocouple Equation
(9)
where a and b are the thermocouple coefficients and T = T T 0 is the temperature with respect to the
reference temperature, T0 (273.16 K). The inference to engineers from Equation (9) is that the emf output
from the thermocouple wire does not depend linearly on the temperature difference, T, and consequently
we have to use a look-up table either ourselves or on the computer memory to convert the emf to the
temperature difference.
Figure 5 shows the emf output vs. temperature for various thermocouples where it should be
immediately obvious that the voltages are small, typically few tens of a microvolt per degree temperature
difference. At 0 C, by definition, the TC emf is zero. The K-type thermocouple, the chromel-alumel pair, is
a widely employed general purpose thermocouple sensor up to 1200 C.
Table 2
Thermoelectric emf for metals at 100 and 200 C with respect to Pt and the reference junction at 0 C.
MATERIAL
EMF, mV
EMF, mV
Elements
At 100 C
At 200 C
Copper, Cu
0.76
1.83
Gold, Au
0.78
1.84
Aluminum, Al
0.42
1.06
Molybdenum
1.45
3.19
Nickel, Ni
1.48
3.10
Palladium, Pd
0.57
1.23
Platinum, Pt
Silver, Ag
0.74
1.77
Tungsten, W
1.12
2.62
Thermocouple Materials
Alumel
1.29
2.17
Chromel
2.81
5.96
Constantan
3.51
7.45
Copper, Cu
0.76
1.83
Iron, Fe
1.89
3.54
90%Pt-10%Rh
(Platinum-Rhodium)
0.643
1.44
Hot
0 C
100 C
Cold
0
Cu
Cu
Solution
We essentially have the arrangement shown in Figure 6. For each metal there will be a voltage across it
given by integrating the Seebeck coefficient. From the Mott-Jones equation,
V =
T0
x 2 k 2T
x 2 k 2 2
SdT =
dT =
(T T02 )
3eE FO
6eE FO
T0
T
The emf (VAB) available is the difference in V for the two metals so that
VAB = VA VB =
x
2k 2 xA
B (T 2 T02 )
6e E FAO E FBO
(10)
189 V
Cu
Hot
Cold
Meter
201 V
Figure 7
Thermocouple EMF calculations that closely represent experimental observations require
thermocouple voltages for various metals listed against some reference metal. The reference is usually Pt
with the reference junction at 0 C. From Table 2 we can read Al-Pt and Cu-Pt emfs as VAl-Pt = 0.42 mV and
VCu-Pt = 0.76 mV at 100 C with the experimental error being around 0.01mV, so that for the Al-Cu pair,
VAl-Cu = VAl-Pt VCu-Pt = 0.42 mV 0.76 mV = 0.34 mV or 340 V
There is a reasonable agreement with the calculation using the Mott-Jones equation.
3.2. Example 2: The thermocouple equation
We know that we can only measure differences between thermoelectric powers of materials as in the
thermocouple since the thermally induced voltages cancel when both metals are the same. When two
different metals, A and B, are connected to make a thermocouple as in Figure 6, then the net emf is the
voltage difference between the two elements from Equation (10).
VAB = VA VB =
i.e.
x
2k 2 xA
B (T 2 T02 )
6e E FAO E FBO
where C is a constant that is independent of T but dependent on the material properties (x, EFO for the
metals).
10
We can now expand VAB about T0 by using Taylor's expansion for a function f(T),
f(T) f(T0) + T(df/dT)0 + 1/2(T)2(d2f/dT2)0
where F = VAB and T = T T 0 and the derivatives are evaluated at T0. The result is the thermocouple
equation:
VAB(T) = a(T) + b(T)2
where the coefficients a and b are 2CT0 and C respectively.
NOTATION
C
e
E
Eav
EF
EFO
EMF
F(T)
g(E)
H
k
me
m e*
MFP
n
S
SA B
cold
electronic charge (magnitude only)
energy of an electron
average electron energy (depends on the density of states)
Fermi energy
Fermi energy at 0 K
emf, electromotive force (open circuit voltage)
function of temperature
density of states
hot
Boltzmanns constant
mass of the electron (in free space)
effective mass of the electron in a crystal
mean free path
concentration of conduction electrons (number of conduction electrons per unit volume)
Seebeck coefficient; thermoelectric power
SA SB
T
TC
u
V
VAB
v
x
USEFUL DEFINITIONS
11
Fermi energy (EF) or level may be defined in several equivalent ways. Fermi level is the energy level corresponding to the
energy required to remove an electron from the semiconductor; there need not be any actual electrons at this energy
level. The energy needed to remove an electron defines the work function . We can define the Fermi level to be
below the vacuum level. EF can also be defined as that energy value below which all states are full and above which
all states are empty at absolute zero of temperature. E F can also be defined through a difference. A difference in the
Fermi energy, E F , in a system is the external electrical work done per electron either on the system or by the
system just as electrical work done when a charge e moves through a electrostatic potential energy (PE) difference is
eV. It can be viewed as a fundamental material property. In more advanced texts it is referred to as the chemical
potential of the semiconductor.
Mean free path is the mean distance traversed by an electron between scattering events. If is the mean free time between
scattering events, and v is the mean speed of the electron, then the mean free path, = v.
Mean free time is the average time it takes to scatter a conduction electron. If t i is the free time between collisions
(between scattering events) for an electron labeled as i, then = ti averaged over all the electrons. The drift mobility
is related to the mean free time by d = e / me. The reciprocal of the mean free time is the mean probability per unit
time that a conduction electron will be scattered, or, put differently, the mean frequency of scattering events.
Semiconductor is a nonmetallic element (e.g. Si or Ge) that contains both electrons and holes as charge carriers in contrast
to an enormous number of electrons only as in metals. A hole is essentially a "half-broken" covalent bond which has
a missing electron and therefore behaves effectively as if positively charged. Under the action of an applied field, the
hole can move by accepting an electron from a neighboring bond, thereby passing on the "hole". Electron and hole
concentrations in a semiconductor are generally many orders of magnitude less than those in metals, thus leading to
much smaller conductivities.
Sir Nevill Mott (1905-1996)
At a personal level, I always found
Sir Nevill to possess the typical
helpful and pleasant personality
that people with his outstanding
intellect so often exhibit.
Professor Joe Marshall
University of Wales at Swansea,
September 1996