This page is the companion explanation to thermoelectric-simulator-en.html. It derives, without skipping intermediate steps, where the results you saw in the simulator come from: “as carriers are added, |S| decreases while σ and κe increase,” “the ZT peak appears at 1019–1021 cm−3 (1019–1020 cm−3 for many materials),” “the height of the peak is determined by the quality factor B alone,” and “the efficiency is determined by the average ZT.” The main tools are the Fermi–Dirac distribution, the density of states, integration by parts and Taylor expansion. The Boltzmann equation and the relaxation-time approximation (§3), the gamma function (§5) and simple differential equations (§10) also appear (the values of the Fermi–Dirac integrals and the maximization of ZT are left to numerical computation). This page goes beyond Section 5-3, “Thermoelectric Conversion” (pp. 96–98), of the textbook and Lecture 12 of “Inorganic Materials Science,” and is written with second- and third-year undergraduates in mind. A gentle explanation of why a temperature difference produces a voltage is given in §1.1; how to read the screen of each simulator tab, what the controls mean, and the approximations used in the calculation are collected in Appendix A.
When a temperature difference is applied across a material, a voltage appears (the Seebeck effect). Conversely, when a current is passed through the junction of two different materials, heat is absorbed or released at the junction (the Peltier effect). Using the former, electricity can be taken directly from heat (thermoelectric power generation); using the latter, a refrigerator with no moving parts (Peltier cooling) can be made. Thermoelectric cooling with semiconductors was demonstrated by Goldsmid and Douglas in 1954, and the subsequent course of materials development is summarized in the review by Snyder and Toberer. The performance of both is determined by the same set of material constants, and the dimensionless figure of merit combines them into one.
S is the Seebeck coefficient, σ the electrical conductivity, and κ the thermal conductivity, which is the sum of κe carried by the carriers and κL carried by lattice vibrations (phonons). The numerator S2σ is called the power factor. Looking at Eq. (1) alone, it seems that one should simply make S and σ large and κ small, but the three quantities are not independent. Increasing the carrier concentration raises σ, but it lowers |S| and raises κe. Solving this tug of war quantitatively is the purpose of this page.
Before going into what is inside S in Eq. (1), let us first grasp, with almost no equations, why a temperature difference produces a voltage. Suppose one end of a rod of n-type semiconductor is heated and the other end is cooled. In a nondegenerate electron gas the average kinetic energy of an electron is (3/2)kBT, so electrons on the hot side move around more vigorously than those on the cold side. The more vigorously particles move, the farther they spread, so the electrons as a whole tend to diffuse from the hot side to the cold side. As a result, electrons accumulate on the cold side, which becomes negatively charged, while the positive charge of the donor ions that released the electrons is left behind on the hot side. The electric field created by this charge imbalance acts to push the electrons heading toward the cold side back toward the hot side, and when it balances the drive of diffusion, the charge imbalance stops growing. In this way a potential difference arises between the two ends. This is the origin of the Seebeck voltage (the magnitude a voltmeter reads also includes a contribution from the temperature dependence of the Fermi level; see the box below and the explanation that follows). In a p-type material the diffusing particles are positively charged holes, so the cold side becomes positive and the hot side, left with the negative charge of the acceptor ions, becomes negative; the direction of the voltage is reversed. This is why measuring the sign of the voltage tells you the carrier type. You can watch this in the animation of simulator tab ① (Appendix A.2).
Let us fix the sign convention. As in Eq. (2) of §2, this page takes the temperature difference to be ΔT = Th − Tc > 0 and defines (potential of the cold side Vc) − (potential of the hot side Vh) = SΔT. The relation S = −ΔV/ΔT in the lecture slides is read as follows: if the potential difference ΔV is measured in the same direction as the temperature difference, as “hot side − cold side” (ΔV = Vh − Vc), then −ΔV/ΔT = (Vc − Vh)/ΔT, which is the same as the convention of this page. In an n-type material the cold side is negative (Vc < Vh), so S < 0; in a p-type material the cold side is positive, so S > 0.
Consider nondegenerate electrons whose relaxation time does not depend on energy (r = 0). In an open circuit (zero current) the interior of the sample remains almost electrically neutral (the charge imbalance needed to create the electric field is tiny compared with the number of carriers), so the electron concentration n equals the donor concentration and is almost constant, independent of position.
(a) Electrostatic part: the difference in electrostatic potential (balance between diffusion and the electric field). The pressure of the electron gas, nkBT, varies from place to place with the temperature, and it pushes the electrons toward the cold side with a force per unit volume of −d(nkBT)/dz = −nkBdT/dz. Per electron this is −kBdT/dz, which, for dT/dz < 0 (with z taken from the hot side toward the cold side), points in the positive z direction, that is, toward the cold side. When τ is constant, multiplying both sides of Eq. (9) by the momentum pz and summing over all states turns the right-hand side into a frictional force −nm*⟨vz⟩/τ proportional to the mean velocity (that is, to the current), and the left-hand side into the sum of the pressure gradient d(nkBT)/dz and neℰ (neℰ is the electrostatic force per unit volume acting on the electrons, −neℰ, with its sign reversed; that is, the sum of the force from the pressure gradient, −d(nkBT)/dz, and the electrostatic force, −neℰ, balances the frictional force) (for a nondegenerate parabolic band, m*⟨vz2⟩ = kBT). At zero current the frictional force vanishes, so the forces acting on a single electron balance. Writing the electron charge as q = −e, we have
(the second term is the push due to heat, and the electric force qℰ = kBdT/dz cancels it). In terms of the electrostatic potential φ (ℰ = −dφ/dz), dφ/dz = (kB/e)dT/dz, so integrating from the hot end to the cold end gives φc − φh = (kB/e)(Tc − Th) = −(kB/e)ΔT. The electrostatic potential on the cold side is lower by (kB/e)ΔT. What the animation of simulator tab ① shows as particle motion is this part (a).
(b) Chemical-potential part: the difference in chemical potential (the temperature dependence of the Fermi level). However, what a voltmeter measures is not the electrostatic potential itself but the difference in the electrochemical potential of the electrons (the Fermi level) (see the explanation of ℰ* in §2.1). The height of the Fermi level measured from the conduction band bottom at that position, EF − Ec = ηkBT, follows from n = Nceη in Eq. (26) as
This shows that it varies with temperature even when n is constant (we used d ln Nc/dT = 3/(2T)). If the temperature difference is small (so that η is almost the same at both ends), the difference between the cold end and the hot end is (EF − Ec)c − (EF − Ec)h = kB(η − 3/2)(Tc − Th) = (3/2 − η)kBΔT. The voltmeter potential is V = φ − (EF − Ec)/e (this splits the electrochemical potential of the electrons, −eV, into the electrostatic energy −eφ and the chemical-potential part EF − Ec; differentiating gives ℰ* = ℰ + (1/e)dEF/dz of §2.1). Therefore
we obtain S = −(kB/e)(5/2 − η). This coincides with Eq. (25) with τ constant (λ = 1/2, r = 0). Note that in the nondegenerate case η < 0, so the two parts add in the same direction, and that (b) is larger than (a). For example, at η = −2, (a) is 1 and (b) is 3.5 (both in units of (kB/e)ΔT), so |S| = 4.5 × 86.17 μV/K = 388 μV/K. The visible part, “hot particles are pushed out,” is only a portion of what the voltmeter reads. When τ depends on energy (r ≠ 0), splitting ℰ in the same way from ℰ* with J = 0 in Eq. (14), part (a) becomes (1 + r)(kB/e)ΔT (for r > 0, faster electrons are less likely to be scattered and travel farther, so the push toward the cold side becomes stronger), while part (b) is unchanged, so the total is r + 5/2 − η.
Why |S| is small in metals. In a metal the electrons are degenerate, and only the electrons within about kBT of EF take part in transport (Eq. (11) of §3.1). On the hot side, the number of electrons occupying states above EF increases, while the number of empty states below EF also increases. Electrons with energies above EF tend to flow from the hot side to the cold side, but at energies below EF there are fewer electrons on the hot side, so they tend to flow in the opposite direction, from the cold side to the hot side. As a result, the contributions above and below EF almost cancel, only the slight asymmetry of σ(E) around EF remains, and |S| stays at a few μV/K. The quantitative expression of this is Mott’s formula, Eq. (20) of §4.
Consider a sample whose temperature is distributed along its length as T(z). When no current flows (open circuit), the Seebeck coefficient is defined by the potential difference that appears between the two ends. As in the simulator, the sign is chosen so that (potential of the cold side) − (potential of the hot side) = SΔT. Written locally, this reads
Here ℰ* is the gradient of the potential V measured by the voltmeter, with the sign reversed. Because the Fermi level also varies from place to place inside the sample, ℰ* is the electrostatic field ℰ plus the gradient of EF, that is, the electrochemical field ℰ* = ℰ + (1/e)dEF/dz (here EF means the height measured from the conduction band bottom Ec(z) at that position. The part by which Ec(z) itself moves up and down with the electrostatic potential corresponds to ℰ in the first term. The reference from which the energy E is measured in §3 is likewise Ec(z) at that position). In an n-type material, electrons on the hot side diffuse toward the cold side and the cold side becomes negatively charged, so S < 0. When the temperature is uniform, the ratio of the current density J to the heat flux density JQ that it carries is the Peltier coefficient.
When a current is passed through the junction of two materials A and B, Π differs on the two sides, so the difference between the heat carried in and the heat carried out, (ΠA − ΠB)J, is absorbed or released at the junction. This is the Peltier heat. The third effect, the Thomson effect, is treated in §2.4.
When both the temperature gradient and the electric field are small, the flows J and JQ can be written as linear expressions in the “forces.” If we choose the forces so that the entropy production rate is the sum of “flow × force,” Jℰ*/T + JQd(1/T)/dz, the relations take the form
Here Lij are coefficients determined by the material. From the fact that the microscopic equations of motion are symmetric under time reversal, Onsager showed that, with the forces chosen in this way, the off-diagonal coefficients are equal (the reciprocity theorem).
Noting that d(1/T)/dz = −(1/T2)dT/dz, we read off the four coefficients from Eq. (4).
Substituting the reciprocity theorem, Eq. (5), into the expression for the Peltier coefficient gives
Let us rewrite Eq. (4) in a practical form. Solving the first equation for ℰ* and writing the result in terms of σ and S gives ℰ* = J/σ + SdT/dz. Substituting this into the second equation and using σT = L11 and S = L12/(TL11), we obtain
The heat flux is the sum of “the Peltier heat carried by the current” and “heat conduction due to the temperature gradient.” The device calculation in §10 starts from this equation.
Suppose that, in the steady state, a uniform current J flows through a sample that has a temperature gradient. The energy flow is the sum of the heat flux JQ and VJ, carried by the charge at potential V, and this is constant, independent of position: d(JQ + VJ)/dz = 0. Substituting here dV/dz = −ℰ* = −J/σ − SdT/dz and JQ = STJ − κdT/dz (Eq. (7) with Π = ST put in), and using d(ST)/dz = SdT/dz + T(dS/dT)dT/dz, we obtain
(the second line is obtained from the first after the SJdT/dz terms cancel, by multiplying the whole by −1). The first term is the heat flowing in by heat conduction, the second term is the Joule heat (positive regardless of the direction of the current), and the third term is the reversible Thomson heat, whose sign changes with the direction of the current. Its coefficient μTh = TdS/dT is called the first Kelvin relation (using Π = ST, it can also be written μTh = dΠ/dT − S). Although this is the reverse of the order of derivation, it is common to call this the first relation and Π = ST of §2.3 the second relation, and this page follows that convention. If S does not depend on temperature, the Thomson heat vanishes. The constant-property model of §10 uses this approximation.
We calculate the coefficients Lij of §2 from the motion of the carriers. The tools are the Boltzmann equation and the relaxation-time approximation.
The electron distribution function f is slightly displaced from the equilibrium distribution f0 by the electric field and the temperature gradient, and scattering tries to bring it back to f0. In the steady state the two balance, and if the rate of return is written 1/τ (the relaxation-time approximation), we have
Here −eℰ is the force acting on an electron and pz is the momentum. Replacing f on the left-hand side by f0 (keeping terms to first order in the deviation) gives δf = f − f0 = −τ[vz∂f0/∂z − eℰ∂f0/∂pz]. Since f0 = 1/(ey + 1), with y = (E − EF)/kBT, depends on position through EF(z) and T(z), we use df0/dy = kBT∂f0/∂E and the chain rule to obtain
Substituting these into δf and collecting terms with eℰ + dEF/dz = eℰ*, we obtain
Since δf is proportional to −∂f0/∂E, the only states whose distribution is displaced and that take part in transport are those at energies where −∂f0/∂E is non-negligible.
Let g(E) be the number of states per unit volume and per unit energy (the density of states, including spin). The current density is the sum over states of the product of the charge −e and the velocity vz of each state, weighted by δf; the heat flux density is the same with the heat E − EF carried in place of the charge. In an isotropic system, averaging over the direction of the velocity gives vz2 → v2/3, so inserting Eq. (11) we obtain
Here we define the energy-resolved conductivity σ(E) and its weighted integrals 𝒦k.
Using g v2τ/3 = σ(E)/e2, Eq. (12) becomes more compact:
The procedure is the same as in §2.3. Setting J = 0 gives ℰ* = −[𝒦1/(eT𝒦0)]dT/dz. Comparing this with Eq. (2) gives S; substituting ℰ* into the second equation gives JQ = −[(𝒦2 − 𝒦12/𝒦0)/(e2T)]dT/dz, from which κe follows. Setting dT/dz = 0 gives J = 𝒦0ℰ* and JQ = −(𝒦1/e𝒦0)J, from which σ and Π follow.
Π = ST holds automatically. In fact, rewriting Eq. (14) in the form of Eq. (4) gives L11 = T𝒦0 and L12 = L21 = −T𝒦1/e, so the microscopic calculation satisfies the reciprocity theorem.
Regard σ(E)(−∂f0/∂E) as a weight that expresses “how much current is carried by carriers of which energy,” and write the average with this weight as ⟨A⟩ = ∫A σ(E)(−∂f0/∂E)dE/𝒦0. Since 𝒦1/𝒦0 = ⟨E − EF⟩ and 𝒦2/𝒦0 = ⟨(E − EF)2⟩, Eq. (15) becomes
Fig. 1. The transport window (acoustic-phonon scattering, σ(E) ∝ E − Ec). Left: nondegenerate (η = −2); right: degenerate (η = 4). The horizontal axis is the energy x measured from the conduction band bottom (in units of kBT). The gray dashed line is the density of states g, orange is g(−∂f0/∂E), the blue fill is the transport window σ(E)(−∂f0/∂E), and red is the window multiplied by (x − η)/6 (the factor 1/6 is there to fit the vertical axis; dividing the net area of the red by the area of the blue and multiplying by 6 gives the window average quoted below). Orange and blue are scaled so that each has maximum 1, and the gray g is scaled to be 1 at the right edge, so the heights of curves of different colors cannot be compared with one another. The window average ⟨x − η⟩ = |S|/(kB/e) is 4.06 on the left (|S| = 350 μV/K) and 0.79 on the right (68 μV/K).
In a metal, EF lies deep inside the conduction band (EF − Ec ≫ kBT), and σ(E) varies smoothly near EF. So we expand, treating the window width kBT as small.
−∂f0/∂E is an even function of E − EF, and its total integral is ∫(−∂f0/∂E)dE = f0(−∞) − f0(∞) = 1. Expanding a smooth function H(E) in a Taylor series around EF and integrating, the first-order term vanishes because it is an odd function, and we obtain
In the second line we used the value of the second moment. Putting y = (E − EF)/kBT and f(y) = 1/(ey + 1), we have (−∂f0/∂E)dE = (−df/dy)dy, and using the evenness and integration by parts we get
The last integral is evaluated by expanding 1/(ey + 1) = Σk≥1(−1)k+1e−ky and using ∫0∞ye−kydy = 1/k2, which gives Σ(−1)k+1/k2 = (1 − 2/22)Σ1/k2 = (1/2)(π2/6) = π2/12 (the even-numbered terms are first included and then subtracted twice).
We apply Eq. (17) to 𝒦0 and 𝒦1. For 𝒦0, H = σ(E), and to lowest order 𝒦0 ≈ σ(EF). For 𝒦1, H = σ(E)(E − EF); since H(EF) = 0 and H′ = σ′(E − EF) + σ,
Substituting this into S = −𝒦1/(eT𝒦0) of Eq. (15) gives Mott’s formula (this form, written with σ(E), is due to Cutler and Mott).
For 𝒦2, H = σ(E)(E − EF)2, and H″ = σ″(E − EF)2 + 4σ′(E − EF) + 2σ, so H″(EF) = 2σ(EF), and therefore 𝒦2 ≈ (π2/3)(kBT)2σ(EF). On the other hand, 𝒦12/𝒦0 is of order (kBT)4 and can be neglected, so from Eq. (15), κe ≈ 𝒦2/(e2T) = (π2/3)(kB/e)2σT. That is,
This is a universal value that does not depend on the material, and it is the Wiedemann–Franz law of metals (§8).
We approximate the region near the conduction band bottom by a parabolic band of effective mass m*, E − Ec = p2/2m*. The density of states and the square of the velocity are then
Writing the energy dependence of the relaxation time with an exponent λ that represents the scattering mechanism, τ ∝ (E − Ec)λ−1/2, Eq. (13) gives
For acoustic-phonon scattering τ ∝ E−1/2, so λ = 0 (σ(E) ∝ E − Ec); for ionized-impurity scattering τ ∝ E3/2, so λ = 2. The simulator uses λ = 0. In the convention in which τ ∝ Er is written, r = λ − 1/2.
When EF lies below the conduction band bottom by several kBT or more (η ≪ −1), the Fermi–Dirac distribution can be approximated by the Boltzmann distribution f0 ≈ eη−x, so that −∂f0/∂E = eη−x/kBT. Using E − EF = kBT(x − η) and dE = kBTdx, the proportionality constant of σ(E) and kBT cancel between numerator and denominator, and the window average is
(Γ(s) = ∫0∞xs−1e−xdx is the gamma function, and Γ(s + 1) = sΓ(s). eη also cancels.) Substituting into Eq. (16), we obtain
Here 2 + λ is “the average energy of the carriers, weighted by the window and measured from the conduction band bottom,” and −η is “how far above EF the conduction band bottom lies”; both are in units of kBT. Since all the carriers lie above EF, the window is on only one side of EF and |S| is large (Fig. 1, left).
Using g(E) of Eq. (22) and writing (E − Ec)1/2dE = (kBT)3/2x1/2dx, the carrier concentration in the nondegenerate case is
In the first line we used Γ(3/2) = ∫0∞x1/2e−xdx = √π/2. In the third line we put in ħ = h/2π, that is, 1/ħ3 = (2π)3/h3, and combined the numerical factors as (1/2π2)·(2π)3·(√π/2) = 8π3√π/4π2 = 2π3/2, and finally wrote π3/2(2m*kBT)3/2 = (2πm*kBT)3/2. This Nc is called the effective density of states. Nc(300 K, m* = 1) = 2.51×1019 cm−3. Substituting η = ln(n/Nc) into Eq. (25) and writing ln(Nc/n) = ln Nc − (ln 10)log10n, we obtain
§4 and §5 were approximations at the two extremes. The single parabolic band (SPB) model treats the intermediate region (η ~ 0) exactly. Keeping the assumptions of a parabolic band and τ ∝ (E − Ec)λ−1/2, we integrate the Fermi–Dirac distribution without approximating it.
Only F0 can be written with elementary functions (integrate d[−ln(1 + eη−x)]/dx = 1/(1 + ex−η) from 0 to ∞). The others require numerical integration. The limits are Fj → Γ(j + 1)eη for η ≪ −1 and Fj → ηj+1/(j + 1) for η ≫ 1 (this page and the simulator use the definition without the factor 1/Γ(j + 1)). The forms that appear in transport integrals can be reduced to Fj by integration by parts (r > 0).
Insert σ(E) = σE0x1+λ (σE0 is a constant), E − EF = kBT(x − η) and (−∂f0/∂E)dE = (−∂f0/∂x)dx into 𝒦k of Eq. (13), expand (x − η)k, and apply Eq. (29) term by term; this gives
Using the abbreviations A2 = (2 + λ)F1+λ/[(1 + λ)Fλ] and A3 = (3 + λ)F2+λ/[(1 + λ)Fλ], Eq. (30) gives 𝒦1/𝒦0 = kBT(A2 − η) and 𝒦2/𝒦0 = (kBT)2(A3 − 2ηA2 + η2). Substituting into Eqs. (15) and (16) gives
In L the terms containing η cancel (as it should, since a variance does not depend on where the origin is placed). For acoustic-phonon scattering in the simulator (λ = 0), A2 = 2F1/F0 and A3 = 3F2/F0, so the reduced quantities s = |S|/(kB/e) and l = L/(kB/e)2 are
If the carrier concentration is integrated with the Fermi–Dirac distribution as it is, without making the Boltzmann approximation in the first line of Eq. (26), we obtain
(the coefficient follows from the fact that in Eq. (26) we had (1/2π2)(2m*kBT/ħ2)3/2·(√π/2) = Nc). For λ = 0 the electrical conductivity is σ = 𝒦0 = σE0F0(η), and the mobility is μ = σ/ne. For η ≪ −1, F0 → eη and F1/2 → (√π/2)eη, so μ approaches the value μ0 = σE0/(eNc), which does not depend on η. Putting σE0 = eNcμ0 back, we obtain
| Quantity | Limit for η ≪ −1 | η = −10 | Limit for η ≫ 1 | η = 50 |
|---|---|---|---|---|
| s = |S|/(kB/e) | 2 − η (= 12) | 12.00002 | π2/(3η) (= 0.065797) | 0.065797 |
| l = L/(kB/e)2 | 2 | 2.00001 | π2/3 − s2 (= 3.28554) | 3.28554 |
| μ/μ0 | 1 | 0.99999 | (3√π/4)η−1/2 (= 0.18800) | 0.18790 |
For the degenerate side, l follows from F1 ≈ η2/2 + π2/6 and F2 ≈ η3/3 + (π2/3)η, which give A2 ≈ η + π2/(3η) and A3 ≈ η2 + π2; then Eq. (31) gives π2/3 − s2. The degenerate limit of μ follows from F0 → η and F1/2 → (2/3)η3/2.
From Eqs. (32) and (34), the power factor is
so the material parameters are gathered in the leading coefficient and the η dependence is only s2F0. Maximizing this numerically gives η = 0.668, |S| = 167 μV/K, n/Nc = 1.26, independent of the material.
In the denominator of Eq. (1), put κe = LσT (the definition of the Lorenz number, Eq. (16)), write S = s(kB/e) and L = l(kB/e)2, and then divide the numerator and denominator by (kB/e)2σT.
In the remaining κL/[(kB/e)2σT], insert σ = Nceμ0F0 from Eq. (34), which gives
Therefore
Here, substituting Nc = 2(2πm*kBT/h2)3/2 (Eq. (26)) into B gives the quality factor displayed in tab ③ of the simulator.
(Eq. (39) is evaluated in SI units. The m* in Table 3 is a multiple of me = 9.109×10−31 kg, so multiply by me to convert it to kg; μ0 is in cm2/Vs, so multiply by 10−4 to convert it to m2/Vs, and then insert the values.)
The result of numerically maximizing Eq. (38) with respect to η is given in Table 2 and Fig. 2.
| B | Maximum of ZT | ηopt | |S| at the top (μV/K) | nopt/Nc | L/(kB/e)2 at the top |
|---|---|---|---|---|---|
| 0.05 | 0.183 | 0.30 | 187 | 0.967 | 2.212 |
| 0.1 | 0.340 | 0.08 | 200 | 0.815 | 2.182 |
| 0.2 | 0.604 | −0.21 | 218 | 0.645 | 2.147 |
| 0.3 | 0.827 | −0.41 | 231 | 0.547 | 2.127 |
| 0.5 | 1.200 | −0.69 | 250 | 0.431 | 2.101 |
| 1 | 1.900 | −1.10 | 280 | 0.299 | 2.072 |
| 2 | 2.865 | −1.55 | 314 | 0.199 | 2.048 |
Fig. 2. The maximum of ZT (green, left axis) and the optimal reduced Fermi level ηopt (blue dashed line, right axis) as functions of the quality factor B. Computed by maximizing Eq. (38) with respect to η using the same Fermi–Dirac integral code as the simulator. The black dots are the three presets (B is calculated with Eq. (39)).
The 1019–1020 cm−3 in the heading is a rough guide for typical values of m* and T near room temperature (for materials with large m* used at high temperatures, it is higher, as in the latter part of the list below and the oxide in Table 3). As Table 2 shows, even when B is varied from 0.05 to 2, nopt/Nc stays in the range from 0.967 to 0.199, so the optimum is always where n is 0.2–1 times Nc, that is, where EF lies within 2kBT of the conduction band bottom. Since Nc(300 K, m* = 1) = 2.51×1019 cm−3, for example for B = 0.5 (nopt/Nc = 0.431), nopt takes the following values:
“A heavily doped semiconductor, neither an insulator nor a metal,” is optimal because, when EF is placed near the conduction band bottom, the balance among the bias of the window (|S|), the number of carriers in the window (σ) and the width of the window (κe) is at its best.
| Preset | T (K) | m* | μ0 (cm2/Vs) | κL (W/mK) | B | Maximum of ZT | nopt (cm−3) | |S| at the top (μV/K) |
|---|---|---|---|---|---|---|---|---|
| Room temperature (Bi2Te3-type magnitudes) | 300 | 1.2 | 300 | 0.8 | 0.442 | 1.098 | 1.58×1019 | 242 |
| Mid temperature (PbTe-type magnitudes) | 700 | 0.9 | 90 | 1.0 | 0.572 | 1.317 | 3.22×1019 | 252 |
| Oxide (low mobility) | 900 | 4.0 | 2 | 2.5 | 0.089 | 0.308 | 8.92×1020 | 196 |
The maximum ZT of the three presets is exactly the value read off the curve of Fig. 2 from the value of B. The oxide has a small B because its μ0 is small and its κL is large; and because its m* is large and hence its Nc is large, nopt rises to nearly 1021 cm−3.
From Eq. (39), there are four ways to increase B.
Using the Lorenz number of Eq. (16), the thermal conductivity carried by the carriers can be written as
Here L varies with η, and its two limits are
The nondegenerate value is obtained by putting Fj → Γ(j + 1)eη in Eq. (31), which gives A2 → (2 + λ)Γ(2 + λ)/[(1 + λ)Γ(1 + λ)] = 2 + λ and A3 → (3 + λ)(2 + λ); then A3 − A22 = (2 + λ)(3 + λ) − (2 + λ)2 = 2 + λ gives the result. The degenerate value is Eq. (21). For λ = 0, the nondegenerate value is 2(kB/e)2 = 1.485×10−8 W Ω K−2 and the degenerate value is (π2/3)(kB/e)2 = 2.443×10−8 W Ω K−2, and their ratio is π2/6 = 1.645.
In experiments S can be measured but η cannot be known directly, so Kim et al. proposed an expression that approximates L of the SPB (acoustic-phonon scattering) from |S| alone.
Compared with L of the same SPB (λ = 0) as in the simulator, when η is varied from −10 to 50 the difference stays within the range from −3.6 % (near |S| = 36 μV/K) to +3.7 % (near |S| = 217 μV/K).
As Eq. (39) shows, B ∝ μ0/κL. If only κL can be lowered without lowering the mobility, the ZT peak becomes higher. We derive the idea behind this from the kinetic theory of phonons.
Regard phonons as a “gas of particles carrying heat,” and let C be the heat capacity per unit volume, v the speed of sound and Λ the mean free path. A phonon crossing the plane z = z0 in a direction making an angle θ with the normal brings an energy appropriate to the temperature at the position of its last scattering, z0 − Λ cos θ. The energy density there is lower than at z0 by CΛ cos θ dT/dz (the uniform part vanishes on averaging over directions), so multiplying by the z component of the velocity, v cos θ, and averaging over directions gives
(Averaging over the sphere gives ⟨cos2 θ⟩ = 1/3.) C and v are largely fixed by the material, so what can be varied by design is the mean free path Λ.
When several scattering processes occur independently, the scattering frequencies v/Λ add up, so
Here U denotes Umklapp scattering between phonons, pd denotes point defects such as the mass fluctuation in a solid solution, and b denotes grain boundaries. In a polycrystal of grain size d, Λb ≈ d. Tab ④ of the simulator lumps everything other than the grain boundaries into a single value Λph (the gray approximation) and takes 1/Λ = 1/Λph + 1/d. The electron mobility also satisfies μ ∝ τ ∝ Λ, so the same form holds with the electron mean free path Λe, and
The simulator adds to this a multiplication factor on κL due to solid-solution alloying and a lower limit κmin for κL (Eq. (62) of §11). At the carrier concentrations of 1019–1020 cm−3 used in thermoelectric materials, the phonon mean free path is often longer than that of electrons, so by choosing a grain size with Λe ≪ d ≲ Λph one can selectively block just the heat.
| Grain size d | κL (W/mK) | μ0 (cm2/Vs) | B | Maximum of ZT |
|---|---|---|---|---|
| 1 mm | 0.800 | 300.0 | 0.442 | 1.098 |
| 10 μm | 0.797 | 299.9 | 0.443 | 1.101 |
| 1 μm | 0.771 | 298.5 | 0.456 | 1.123 |
| 100 nm | 0.600 | 285.7 | 0.561 | 1.299 |
| 50 nm | 0.500 | 272.7 | 0.642 | 1.425 |
| 20 nm | 0.371 | 240.0 | 0.761 | 1.594 |
| 10 nm | 0.300 | 200.0 | 0.785 | 1.627 |
| 5 nm | 0.255 | 150.0 | 0.694 | 1.501 |
When the grain size decreases to about Λph, κL decreases and B increases. If it is made smaller still, κL approaches the lower limit κmin and no longer decreases, while only μ0 drops as the grain size approaches Λe, so B begins to decrease. In this model the maximum of ZT is highest, 1.63, at about 12 nm (simulator tab ④ scans the grain size over 91 points and shows the best point, so its readout is “11 nm (ZT 1.63)”).
Slack described the ideal thermoelectric material as “a glass for phonons and a crystal for electrons” (Phonon Glass, Electron Crystal). Typical examples are filled skutterudites, in which guest atoms (Ba, Yb, etc.) placed inside the cage-like crystal structure rattle, and clathrates, which work by the same mechanism (rattling does not occur if the cages are left empty). However, no matter how much disorder is introduced, the mean free path cannot become shorter than the interatomic distance. Cahill et al. estimated the thermal conductivity in the limit where the lifetime of each vibration is shortened to half a period, and found that at high temperatures the minimum thermal conductivity is
(na: number density of atoms; vl and vt: speeds of sound of the longitudinal and transverse waves). For example, for na = 5×1028 m−3, vl = 3000 m/s and vt = 1800 m/s, κmin ≈ 0.50 W/mK. The κmin of the simulator is a parameter that represents this lower limit.
Once the ZT of a material is known, we derive how much efficiency is obtained when it is made into a thermoelectric device that combines p-type and n-type legs. This analysis based on the constant-property model has been the standard method since Ioffe’s textbook. Note that only in this section and in Problem 5 of §13, η denotes the conversion efficiency (not the reduced Fermi level).
As in Fig. 3, the p-type and n-type legs are joined at one end by an electrode on the hot side (Th), and a load resistance is connected across the ends on the cold side (Tc). We assume that S, σ and κ do not depend on temperature (the constant-property model; the Thomson heat in Eq. (8) is then 0). The whole device is described by the following three quantities.
The figure of merit of the device is defined as follows.
If the p and n legs have properties (|Sp|, σ, κ) of the same magnitude and differing only in the sign of S, and the same leg length ℓ and cross-sectional area A, then S = 2|Sp|, R = 2ℓ/(σA) and K = 2κA/ℓ, so Z = 4Sp2/(4κ/σ) = Sp2σ/κ, which coincides with the figure of merit of the material. Let the load resistance be mR (m: load ratio) and ΔT = Th − Tc.
Fig. 3. A thermoelectric power-generation device (p–n pair) and the flow of heat. Holes (p-type) and electrons (n-type) both flow from the hot side to the cold side, so the current flows downward in the p-type leg and upward in the n-type leg, and goes once around through the load. The heat Qh entering from the hot side is the sum of the Peltier heat SThI absorbed at the hot junction and the heat conduction KΔT, minus half of the Joule heat, I2R/2, that is generated inside the legs and returns to the hot side (Eq. (51)).
The current flowing in the circuit is the electromotive force SΔT divided by the sum of the internal and load resistances, and the power taken out at the load is P = I2mR.
To find the heat Qh entering the device from the hot side, consider the temperature distribution inside one leg of length ℓ and cross-sectional area A. Joule heat ρJ2 per unit volume is generated uniformly inside the leg (ρ = 1/σ, J = I/A), so the steady-state heat conduction equation (Eq. (8) with the Thomson heat set to 0) and its solution are
(Differentiating twice gives −ρJ2/κ, and the boundary conditions are satisfied at z = 0 and ℓ.) The conducted heat flowing into the leg from the hot end z = 0 is
(ρJ2Aℓ = I2ρℓ/A.) The first term on the right is the thermal conductance of the leg × ΔT, and the second term is half of the Joule heat generated in the leg’s resistance ρℓ/A. In other words, exactly half of the Joule heat generated inside the leg returns to the hot side, and the heat that must be drawn from the hot side is reduced by that amount (the other half flows to the cold side). Summing over the two legs and adding the Peltier heat absorbed at the hot junction (ΠI = SThI, from Eq. (6)) gives
We put Eq. (48) into η = P/Qh. Here Qh = S2ThΔT/[R(1 + m)] − S2ΔT2/[2R(1 + m)2] + KΔT. Dividing both the numerator P and the denominator Qh by S2ΔTTh/[R(1 + m)2] and using RK/S2 = 1/Z, we have
Writing the denominator as D(m) = (1 + m) − ΔT/2Th + (1 + m)2/ZTh, we have η ∝ m/D, and the condition is dη/dm ∝ (D − mD′)/D2 = 0. Using D′ = 1 + 2(1 + m)/ZTh, we obtain
(The first line uses (1 + m)2 − 2m(1 + m) = (1 + m)(1 − m); the second line is obtained by multiplying the whole by ZTh and rearranging.) Therefore
At the optimum D = mD′, so η = (ΔT/Th)·M/D = (ΔT/Th)/D′(M). Using Z = (M2 − 1)/T̄, we have 2(1 + M)/ZTh = 2(1 + M)T̄/[(M2 − 1)Th] = (Th + Tc)/[(M − 1)Th], so
| Typical application | Th/Tc (K) | ZT̄ | Carnot efficiency | ηmax | Fraction of Carnot |
|---|---|---|---|---|---|
| Low-temperature waste heat | 500/300 | 1.0 | 40.0 % | 8.23 % | 20.6 % |
| Mid-temperature waste heat | 800/300 | 1.0 | 62.5 % | 14.47 % | 23.2 % |
| Spacecraft RTG (simulator button) | 1275/575 | 0.8 | 54.9 % | 10.46 % | 19.1 % |
This is the case where a current I is passed by a power supply and heat is pumped up from the cold side. The cooling capacity Qc is the Peltier heat STcI absorbed at the cold junction, minus half of the returning Joule heat and the conducted heat leaking in from the hot side. The electric power W supplied from outside is the sum of the work done against the Seebeck electromotive force and the Joule heat.
Multiplying the numerator and denominator of the coefficient of performance φ = Qc/W by R/S2, and using ξ = IR/S (which has the dimension of temperature) and 1/Z = RK/S2, we have
The condition dφ/dξ = 0 is N′W̃ − NW̃′ = 0, where N is the numerator and W̃ the denominator. When the two products are expanded and subtracted, the terms in ξ3 and TcΔT ξ cancel each other.
(The third line is obtained by multiplying by −Z and using Tc + ΔT/2 = T̄. The fourth line is the positive solution of the quadratic equation, with √(4ΔT2 + 4ZT̄ΔT2) = 2ΔT M.) We put this ξ back into Eq. (57). Using 1/Z = T̄/(M2 − 1), the numerator becomes TcΔT/(M − 1) − ΔT2/2(M − 1)2 − ΔT T̄/(M2 − 1), and the denominator becomes ΔT2/(M − 1) + ΔT2/(M − 1)2 = ΔT2M/(M − 1)2. Taking the ratio, canceling (M − 1)2/ΔT, and multiplying the numerator and denominator by M + 1, we obtain
(The numerator in the second line is a rearrangement of Tc(M2 − 1 − M + 1) − (ΔT/2)·2M, and the last step uses Tc + ΔT = Th.) The first factor is the Carnot coefficient of performance Tc/(Th − Tc). For ZT̄ = 1, cooling from 300 K down to 270 K gives φmax = 1.130 (Carnot 9.00), and cooling down to 250 K gives 0.444 (Carnot 5.00), so widening the temperature difference lowers it sharply. Maximizing Qc/W by varying the current numerically gives the same value.
For φmax > 0 we need M > Th/Tc. Squaring both sides and substituting M2 = 1 + Z(Th + Tc)/2, we obtain
The same result can also be obtained directly by setting Qc = 0. From Eq. (56), KΔT = STcI − I2R/2, and the right-hand side is maximal where d/dI = STc − IR = 0, that is, at I = STc/R, so
For example, for ZTc = 1 and Tc = 300 K, ΔTmax = 150 K. Conversely, to cool from the 300 K heat-rejection side down to 250 K (ΔT = 50 K), M > Th/Tc requires ZT̄ > (300/250)2 − 1 = 0.44.
thermoelectric-simulator-en.html implements the equations of this page as follows.
κL and μ0 on the right-hand side are the values chosen with the sliders (the values for sufficiently large grain size). The default values are Λph = 50 nm, Λe = 5 nm, κmin = 0.2 W/mK and a = 1. For d ≫ Λph, κL approaches aκL, and for d → 0 it approaches κ1. The second line is an assumption for teaching purposes: “if solid-solution alloying lowers κL by a fraction 1 − a, the mobility is lowered by one quarter of that fraction.”
| Topic | Where to look |
|---|---|
| The Seebeck effect: diffusion of hot charged particles and charge imbalance (§1, §2, §3) | thermoelectric-simulator-en.html ① (Appendix A.2) |
| S, σ, κ and the ZT peak (§6, §8) | same, tab ② (Appendix A.3) |
| Height and position of the peak, the quality factor B (§7) | same, tab ③ (Appendix A.4) |
| Lowering κL, PGEC (§9) | same, tab ④ (Appendix A.5) |
| Conversion efficiency and the cooling coefficient of performance (§10) | same, tab ⑤ (Appendix A.6) |
| Operating temperature ranges and typical ZT of real materials | same, tab ⑥ (Appendix A.7) |
| How atomic levels become bands (the starting point of the density of states in Eq. (22)) | band-simulator-en.html / Sections 5-1 and 5-2 of the textbook (Lecture 11) |
| Phonons and heat capacity (C in Eq. (43)) | heatcapacity-simulator-en.html / Section 4-1 of the textbook (Lecture 10) |
| The other main topic of Lecture 12: superconductivity | superconductivity-simulator-en.html |
| Making nanostructured bulk materials: sintering and suppression of grain growth (grain size in §9) | sintering-derivation-en.html / Section 3-4 of the textbook (Lecture 9) |
| Standard textbooks and reviews on thermoelectric conversion | H. J. Goldsmid, Introduction to Thermoelectricity; D. M. Rowe (ed.), CRC Handbook of Thermoelectrics; the review by Snyder and Toberer (see the reference list) |
Solution. From Eq. (27), |S| = (kB/e)[2 + λ + ln(Nc/n)]. Doubling n decreases ln(Nc/n) by ln 2, so |S| decreases by (kB/e) ln 2 = 86.17 × 0.6931 = 59.7 μV/K, and S = −240 μV/K. This change depends on neither λ nor Nc. (For λ = 0 the initial η is 2 − 300/86.17 = −1.48, which, as cautioned in §5, differs by a few percent from the value of the exact SPB.)
Solution. The Seebeck coefficient of one pair is Sp − Sn = 400 μV/K, so its electromotive force is 400 μV/K × 50 K = 20 mV. For 100 pairs in series it is 2.0 V. The basic configuration of a thermoelectric module is to connect the pairs thermally in parallel (all pairs receive the same ΔT) and electrically in series.
Solution. From Eq. (21), κe = L0σT = 2.443×10−8 × 5.96×107 × 300 ≈ 437 W/mK. The thermal conductivity of copper at room temperature is known to be about 400 W/mK, which shows that in a metal, most of the heat is carried by electrons.
Solution. Converting to σ = 105 S/m, S2σ = (2×10−4)2 × 105 = 4×10−3 W/mK2 (= 40 μW/cmK2). From Eq. (1), ZT = 4×10−3 × 300/1.5 = 0.80.
Solution. From Eq. (54), M = √2 = 1.4142, so the load resistance is set to 1.41 times the internal resistance. From Eq. (55), ηmax = (200/500) × (1.4142 − 1)/(1.4142 + 300/500) = 0.4 × 0.4142/2.0142 = 8.23 %. This is 20.6 % of the Carnot efficiency of 40 %.
Solution. By Eq. (39), B ∝ μ0/κL, so both (a) and (b) double B, from 0.442 to 0.883. By Eq. (38), ZT(η) is determined by B alone, so both the maximum of ZT and ηopt are the same in (a) and (b). Furthermore, since m* and T are the same, Nc is also the same, and nopt = Nc(2/√π)F1/2(ηopt) agrees as well. Calculating with the same code as the simulator gives, in both cases, a maximum ZT of 1.756 and nopt = 1.08×1019 cm−3 (the reference values are 1.098 and 1.58×1019 cm−3). In tab ② of the simulator, κL = 0.4 W/mK for (a) can be set exactly. The μ0 slider, however, has a logarithmic scale and cannot be set to exactly 600; at the nearest value, 603 cm2/Vs, the maximum of ZT is displayed as 1.761 (nopt is the same, 1.08×1019 cm−3). Since ηopt decreases as B increases (Table 2), the optimal carrier concentration moves to the lower-concentration side.
Solution. Since B is the same, the maximum of ZT and ηopt do not change. In nopt = Nc(2/√π)F1/2(ηopt), Nc ∝ m*3/2, so nopt becomes 23/2 = 2.83 times larger. If you choose “Vary m* and keep B constant” in tab ③ of the simulator, the “ZT max” values in the table agree in all four rows, and you can confirm that nopt moves in proportion to m*3/2.
Solution. From Eq. (22), g ∝ E1/2 and v2 ∝ E, and since τ is constant, Eq. (13) gives σ(E) ∝ E3/2 (this corresponds to λ = 1/2 in Eq. (23)). Putting d ln σ/dE = 3/(2EF) into Eq. (20) gives S = −(π2/2)(kB/e)(kBT/EF) = −4.935 × 86.17 μV/K × (0.02585/5) ≈ −2.2 μV/K. Even when the integral of Eq. (15) is evaluated numerically without expanding it, the result is −2.2 μV/K, so Mott’s formula is a good approximation. This is two orders of magnitude smaller than the several hundred μV/K of semiconductors.
The origin of the Seebeck effect, in which a temperature difference produces a voltage, lies in the tendency of charged particles with large kinetic energy on the hot side to diffuse toward the cold side (§1.1). The simulator first shows this in an animation in ①, and then moves on to the quantities that determine the performance of thermoelectric materials. ② is the core of this teaching material: with the single parabolic band model it calculates S, σ and κe on the spot as functions of carrier concentration, and you confirm that, because the three quantities get in each other’s way, the figure of merit ZT forms a peak, whose top appears at 1019–1021 cm−3 (1019–1020 cm−3 for many materials). In ③ you see that the height of the peak is determined by the quality factor B alone and that its position moves with the effective mass, in ④ the idea of lowering the lattice thermal conductivity (PGEC), in ⑤ the conversion efficiency and the cooling coefficient of performance, and in ⑥ where real materials are used.
In Lecture 12 (thermoelectric conversion and superconductivity) of “Inorganic Materials Science” at Tokyo University of Science, ①, ② and ⑤ are treated as the basic part, and they correspond to Section 5-3, “Thermoelectric Conversion” (pp. 96–98), of the designated textbook, Understanding Inorganic Materials Chemistry (in Japanese) (Reference 17). ③, ④ and ⑥ are advanced material. However, the idea of PGEC in ④ is also covered in the slides of Lecture 12. Open these tabs when you become interested in materials design and research.
| Tab | What you see on the screen | Level | Corresponding parts of the main text |
|---|---|---|---|
| ① Seebeck effect | The difference in thermal fluctuations of charged particles on the hot and cold sides, diffusion toward the cold side and the resulting charge imbalance, and the current that flows when the circuit is connected | Basic | §1.1, §2.1, A.2 |
| ② S, σ, κ and ZT peak | |S|, σ, the power factor, the fraction of κe and the ZT peak as functions of carrier concentration | Basic | §6, §8, A.3 |
| ③ Peak height and position — Quality factor B | The ZT peaks when one parameter is varied in four ways, and a table comparing their tops | Advanced | §7, A.4 |
| ④ Reducing κ (PGEC) | The maximum of ZT as a function of grain size, and how κL and μ0 decrease | Advanced (the idea of PGEC is also in the slides of Lecture 12) | §9, §11, A.5 |
| ⑤ Conversion efficiency and cooling | The maximum power-generation efficiency and the maximum cooling coefficient of performance as functions of average ZT, and the Carnot limits | Basic | §10, A.6 |
| ⑥ Applications of real materials | The operating temperature range and typical ZT of each material system, with applications, advantages and disadvantages | Advanced | §9.3, A.7 |
Every tab has a figure (canvas) on the left; below it, depending on the tab, come a legend, a “message” box, and a readout or a comparison table,
and the controls panel is on the right (below the figure on smartphones).
The “message” is a short notice that is rewritten to match the current state every time you operate a control.
Because the simulator is embedded in lecture screens, the explanations on the screen are kept short, and the “More details” link at the bottom of the controls panel takes you to the corresponding subsection of this appendix (A.2–A.7).
If you append #seebeck, #curve, #peak, #kappa, #eff or #mat
to the address, the page is displayed with that tab open (the links in Table 6 of the main text use this).
Adding ?embed=1 gives an embedded view that hides the heading and the introduction (the row of tabs remains), and
writing something like ?only=curve displays only that tab, with the row of tabs also hidden.
This tab lets you see the origin of the Seebeck effect, in which a temperature difference produces a voltage, as the motion of charged particles. The physics was explained in §1.1. The key points are the following three.
The order in which to watch the animation. At first it is n-type, with Tc = 300 K, ΔT = 300 K (hot side 600 K), |S| = 200 μV/K and the switch OFF, and it starts moving immediately. The electrons move around in the rod while being scattered. The electrons on the hot side (left) have a darker color (large kinetic energy), move fast and leave long trails, while those on the cold side (right) have a lighter color and move slowly. The vibration of the immobile ions is also drawn larger toward the hot side. Within a few seconds a charge imbalance appears, positive at the end on the hot side and negative at the end on the cold side (the columns of red ⊕ and blue ⊖ just inside the electrodes, and the light red and blue shading of the rod), and the terminal voltage V settles around SΔT = −60 mV. At this point the number of electrons in the rod has hardly changed; only about 0.5 % of the total charge, in very thin layers at the two ends, is imbalanced (see “Model” below). When you turn the switch ON and connect the + side and the − side with an external wire, some of the electrons that have arrived at the electrode on the cold side enter the wire, go around the external circuit and return to the rod from the electrode on the hot side. Since the charge imbalance decreases and the electric field weakens, diffusion from the hot side to the cold side proceeds again in the rod, and the current keeps flowing as long as the temperature difference is maintained. This is thermoelectric power generation: the force with which heat pushes the charged particles out plays the role of the electromotive force of a battery. The current in the external circuit (in the direction of the flow of positive charge) goes from the + side to the − side; in n-type the hot side is +, and in p-type the cold side is + (this is why the current directions in the p-type leg and the n-type leg are opposite in Fig. 3 of the main text). When you turn the switch back OFF, charge accumulates again and the open-circuit state is restored. The ratio of the root-mean-square speeds, √(Th/Tc), is 1.41 by default. To see the difference more clearly, set Tc = 250 K and ΔT = 400 K (hot side 650 K), which gives 1.61; conversely, if ΔT is lowered to 100 K with Tc kept at 300 K, it becomes 1.15, and the difference is hardly noticeable.
The upper figure (animation).
The lower figure. The one on the left, “Charge imbalance (bars) and potential (line)”, has the position along the rod on the horizontal axis (hot side at left, cold side at right) and shows the charge imbalance in each of the 40 intervals (time-averaged; red is positive, blue is negative; the left scale only gives the direction +, 0, −), and the potential φ(x) with the hot-side end taken as 0 (black line, right scale; the unit is mV, with the same conversion as in “Converting to the on-screen voltage” below). Notice that the charge imbalance appears only in thin layers at the two ends, the interior is almost neutral, and the potential changes almost linearly. The one on the right, “Terminal voltage V and current I (switch ON)”, has real time from 24 s in the past up to “now” on the horizontal axis, and V and I are drawn in separate upper and lower frames so that they are not mistaken for each other. In the wide upper frame, the black line is the terminal voltage V (mV) and the black dashed line is the open-circuit value SΔT. The vertical axis extends up to 1.3|SΔT| on the side of SΔT, and on the opposite side leaves room only for the swing due to thermal noise (|S| × 25 K). In the narrow lower frame, the green line is the relative value of the current I, drawn only while the switch is ON (the upper limit of the scale is 1.5 × ΔT/100 K, but at least 0.5). On screens narrower than 560 px, the two graphs are placed one above the other.
The legend below the figure. There are ten items: carriers (the darker the color and trail, the larger the kinetic energy), ions (immobile; the size of the vibration is exaggerated in this schematic), the background color of the rod (temperature), the ⊕ ⊖ columns (a schematic of the excess charge accumulated at the two ends), the terminal +/−, the electrons in the wire, the positive charge imbalance, the negative charge imbalance, the line of the terminal voltage V, and the line of the current I.
Buttons for narrow screens. On screens 760 px wide or less, the controls panel moves below the figures and becomes far away, so right under the upper figure the “Play” / “Pause” and “Switch ON” / “Switch OFF” buttons and the sentence “Tapping the switch in the figure also toggles it.” are shown. They work the same as the corresponding buttons in the controls panel.
Controls panel.
The default values are n-type, Tc = 300 K, ΔT = 300 K (hot side 600 K), |S| = 200 μV/K, switch OFF, light bulb as the load, and playing (paused if the OS is set to reduce motion).
Readouts. These are the carrier type, the Seebeck coefficient S (with sign), the temperature difference ΔT (with the temperatures of the hot and cold sides), the open-circuit electromotive force SΔT (mV), the present terminal voltage V (mV), the state of the switch (“OFF (open circuit)”, or ON with the type of load), and the current I (relative). The following section, “Carrier motion (the figure is a 2D schematic; the energies are 3D values)”, displays the mean kinetic energy (3/2)kBT of the hot side and of the cold side (in meV; since (3/2)kB/e = 0.1293 meV/K, 77.6 meV at 600 K and 38.8 meV at 300 K), the theoretical value √(Th/Tc) of the ratio of the root-mean-square speeds vrms(hot side)/vrms(cold side) (1.414 by default), and the value measured from the particles. The measured value is obtained by fitting v2 of all the carriers with a linear function of position, a + bx/L, by least squares (time-averaged) and taking the ratio of the values at the two ends, √(a/(a + b)); it agrees with the theoretical value to within 1 %. If you instead compare the particles in the hot-side third and the cold-side third of the rod, their mean temperatures are Th − ΔT/6 and Tc + ΔT/6, so the ratio comes out smaller than √(Th/Tc) (√(550/350) = 1.25 by default).
The message box. It shows one of the following, depending on the state (“(paused)” is appended while the animation is paused). For about 10 seconds just after the ΔT slider is lowered to 0 (model time 4): “Because the temperature difference has disappeared, the charge that had built up is spreading out and returning to its original state.” If ΔT remains 0 after that: “the fluctuations are equally large at both ends, the charge is not imbalanced, and the voltage just fluctuates around 0.” Just after a reset or a change of temperature or switch, while the open-circuit voltage is still growing: “the electrons (holes) that move vigorously on the hot side diffuse toward the cold side, and charge is beginning to build up unevenly, negative on the cold side and positive on the hot side (the reverse for p-type).” Once balanced in open circuit: “the electric field created by the accumulated carriers and the remaining ions pushes the diffusion back, and the two are balanced at V ≈ SΔT.” With the switch ON, for n-type it says “the electrons accumulated on the cold side return to the hot side through the wire, and a current flows through the external circuit from the hot side (+) to the cold side (−),” and for p-type “a current flows through the external circuit from the cold side (+) to the hot side (−) (the electrons in the wire go from the hot side to the cold side)”; in both cases it also says that “the terminal voltage falls below SΔT, and the current keeps flowing as long as there is a temperature difference.”
Model — units and temperature distribution. From here on we describe what the animation actually calculates (in this A.2 only, x is not the reduced energy but the position along the rod, corresponding to z in §1.1; likewise λ is not the scattering exponent but the screening length, L is not the Lorenz number but the length of the rod, and ℓ is not the leg length of §10 but the mean free path). The calculation is done in units in which the length L of the rod is the unit of length, T0 = 300 K is the unit of temperature, and the carrier mass m, kB and the magnitude of the charge |q| are all set to 1. The unit of time is L/√(kBT0/m), and the unit of potential is kBT0/|q|. The charge of a carrier is q = −|q| for n-type and q = +|q| for p-type. x is measured from the electrode on the hot side (x = 0 is the hot side and x = L is the cold side), and the temperature is assumed to vary linearly along the rod. This is the steady state of heat conduction, and changes of temperature due to the heat carried by the carriers, the Peltier heat and the Joule heat are not considered (passing a current does not change the temperature distribution).
Model — motion and scattering. The carriers are two-dimensional classical particles and move according to the equations of motion above between scatterings (time step Δt = 0.002). They are reflected like a mirror at the upper and lower walls and at the electrodes at the two ends (the aspect ratio of the rod is only a matter of appearance adjusted to the screen and does not affect the results of the calculation). At each step a carrier is scattered with probability Δt/τ = 0.1 (relaxation time τ = 0.02), and when it is scattered its two velocity components are redrawn from the Maxwell distribution at the local temperature T(x) (a normal distribution with mean 0 and variance kBT(x)/m for each component). This represents the exchange of heat with the lattice (phonons): a carrier scattered on the hot side becomes fast, and one scattered on the cold side becomes slow. As a result ⟨v2⟩ ≈ 2kBT(x)/m (two dimensions) at each position in the rod, and vrms ∝ √T. Because the relaxation time does not depend on energy, this corresponds to the case r = 0 of §1.1. The mean free path is ℓ = √(πkBT/2m)·τ ≈ 0.025L (300 K; 0.046L at 1000 K), and the carriers diffuse through the rod while being scattered many times. Because the system is two-dimensional, the mean kinetic energy per particle is kBT, but the readouts display the actual three-dimensional value (3/2)kBT (the speed ratio √(Th/Tc) does not depend on the dimension).
Model — ions and the electric field. The ions do not move and are treated as a background of charge distributed uniformly in the rod in the same number N as the carriers, so the whole is always electrically neutral. The rod is divided into 40 intervals, the carriers are distributed to the centers of the two neighboring intervals according to distance (the cloud-in-cell method), and the charge of interval j is ρj = q(nj − N/40). The electric field is obtained from the one-dimensional Gauss’s law and linearly interpolated to the positions of the particles to calculate the force (the field is not time-averaged; the particles are pushed by its instantaneous value).
The field is ℰ = 0 at both ends because the whole is neutral and there is no electric field outside the rod. Vsim is the potential on the cold side minus the potential on the hot side, in the same direction as the sign convention of this page. The permittivity ε is chosen so that the screening length (Debye length) λ for the carrier density n0 = N/L is λ0 = 0.07L at 300 K, namely ε = λ02N (127.0 in the units above), so that λ(T) = 0.07L√(T/300 K) (0.064L at 250 K, 0.099L at 600 K, and 0.128L at 1000 K). The time for the charge imbalance to be canceled by the electric field (the dielectric relaxation time) is τd = λ2m/(kBTτ) = λ02/τ ≈ 0.25 (independent of temperature; about 0.6 s in real time), and the product of the plasma frequency and the relaxation time is ωpτ = τ/λ0 ≈ 0.29, so the charge imbalance settles without oscillating (overdamping).
Model — balance in open circuit and the electromotive force. When the switch is OFF and enough time has passed, the current becomes 0. Inside the rod the carrier density n is almost constant, so, as with the electrostatic part (a) of §1.1, the thermal pushing force on one carrier, −kBdT/dx, and the electric force qℰ balance, so that
holds. For n-type (q = −e) the potential on the cold side is lower by (kB/e)ΔT, and for p-type it is higher; the sign agrees with the sign of S. In both two and three dimensions the pressure in one direction per particle is m⟨vx2⟩ = kBT, so this result does not depend on the dimension. The charge that creates this electric field gathers, by Gauss’s law, in layers of thickness about λ at the two ends. In n-type, electrons are depleted at the hot end, leaving the positive charge of the donors, and electrons are in excess at the cold end (the reverse for p-type). The interior is almost neutral. The amount of imbalanced charge is about λ02ΔT/T0 of all the carriers, which for the default ΔT = 300 K is only 0.49 % (the equivalent of about 127 out of 25920). A difference of this size cannot be seen in the numbers of the 216 dots that are drawn, so it is shown by the columns of ⊕ and ⊖ inside the electrodes (schematic), the shading of the rod, and the bar graph in the lower figure.
However, ℰ = 0 at the electrodes, so inside the screening layers at the two ends the field is weaker than the interior value ℰb (in the Debye approximation, ℰ ≈ ℰb(1 − e−x/λh) from the hot-side end), and the potential difference between the terminals is c(kB/q)ΔT, that is, multiplied by a coefficient c ≈ 1 − (λh + λc)/L smaller than 1 (λh and λc are the screening lengths on the hot and cold sides). The simulator uses the following coefficient, fitted to numerical experiments, which also includes the kinetic effects near the ends and the coarseness of the intervals.
By default (600 K and 300 K), c = 0.771 (0.831 from the Debye-approximation formula). In numerical experiments Vsim/[(kB/q)ΔT] is 0.71–0.78 depending on the conditions, and dividing by c gives 0.97–1.00. The screening length of a real thermoelectric semiconductor is of the order of nm (about 1 nm when estimated with the nondegenerate formula for a carrier concentration of 1019 cm−3 and a relative permittivity of 10), which is negligible compared with a sample of mm size, and c can be regarded as 1. In this model the screening length is taken as large as 7 % of the length of the rod, so this correction is needed.
Converting to the on-screen voltage. As the balance above shows, the potential difference in this model is always (kB/e)ΔT × c regardless of the material, that is, c × 86 μV/K in terms of |S|. The real |S| depends on the material through the chemical-potential part (b) of §1.1 and the way the carriers are scattered, and in thermoelectric semiconductors it is much larger than this. So on the screen, the temperature difference ΔTeff calculated back from the potential difference of the particles is multiplied by S, which is the |S| in the controls panel with the sign of the type attached, to give the terminal voltage.
V̄sim is the exponential moving average of Vsim with a time constant of 1 (2.5 s in real time). In the steady state of the open circuit, ΔTeff ≈ ΔT, so V ≈ SΔT (0.97–1.00 times SΔT in the numerical experiments); when the switch is ON and the charge imbalance decreases, ΔTeff becomes smaller and V goes down. The |S| slider only changes the mV scale on the screen; it does not change the motion of the particles, the charge imbalance or the relative value of the current. The potential φ(x) in the lower figure is also converted to mV with the same scaling (multiplied by |S| × T0/c). The fluctuation of the voltage becomes smaller as the number of simulated carriers N is increased (the thermal noise of the potential between the terminals is of the order of √(kBT/N)/λ in the units above (L = |q| = 1), or (kBT/|q|)√(L/N)/(λ/L) when the dimensions are restored. In numerical experiments with ΔT = 0, the standard deviation of the instantaneous Vsim is about 0.07–0.09 kBT0/|q|). If only the 216 drawn carriers were simulated, the instantaneous fluctuation would become larger than the default signal, so N = 25920 carriers are simulated and every 120th of them, 216 in all, is drawn (every 240th, 108 in all, on screens narrower than 560 px). Even so, the displayed voltage fluctuates by about 11–13 K in standard deviation when converted to ΔT (about 4 % of the default ΔT = 300 K).
The circuit rule. When the switch is ON, for a carrier reflected at an electrode we calculate the drop in potential energy when it moves through the wire to the opposite electrode, and the probability that it enters the wire:
(The potentials are averaged with a time constant of 0.02.) Only when u > 0 is the carrier let into the wire, with this probability. The same rule is applied at both ends, and the direction of the current is not set by hand but is determined by the potential created by the particles. Since the number passing through the wire is proportional to the potential difference, this corresponds to Ohm’s law, and G plays the role of the ease with which the load conducts (the conductance). The light bulb has G = 0.08, and “wire (nearly a short circuit)” has G = 3. A carrier that has entered the wire returns to the rod from the electrode on the opposite side at that very instant (a metal wire is full of electrons, so it is treated as if the amount that enters at one end comes out at the other). The velocity on return is generated at the temperature of that electrode: the component pointing away from the electrode is drawn from a flux-weighted half-Maxwell distribution, and the transverse component from a Maxwell distribution. The number of carriers staying in the wire is always 0, and the total number of carriers in the rod, N, does not change.
The direction and magnitude of the current. The current I is the net number of carriers that moved through the rod from the hot side to the cold side (per unit time, averaged with a time constant of 1), divided by 150 to give a relative value. 150 is a unit chosen so that the value comes out close to 1 when the rod is short-circuited at ΔT = 100 K (the theoretical value of the short-circuit current obtained from the balance of the carrier flux is NτkBΔT/(mT0L2) = 173, and the numerical experiments give 153–159). The directions are as follows.
In the numerical experiments (Tc = 300 K), connecting the light bulb lowers the terminal voltage to about 0.43 times the open-circuit value, and the relative value of the current is 0.60, 1.19, 1.83 (default) and 2.42 for ΔT = 100, 200, 300 and 400 K (proportional to ΔT). With “wire”, the terminal voltage is about 0.02 times the open-circuit value, that is, nearly 0, and the current is 1.02, 2.09, 3.08 and 4.13 for ΔT = 100, 200, 300 and 400 K. The 0.43 for the light bulb and 0.60/1.02 ≈ 0.59 add up to almost 1 because the rod behaves as “a battery with electromotive force SΔT and an internal resistance”, so the terminal voltage falls linearly as SΔT − IRint. The brightness of the light bulb is made proportional to the power |IV|. When the switch is turned back OFF, the terminal voltage returns to the open-circuit value to within 1 %.
How time advances. Time 1 in the model corresponds to 2.5 s of real time (it advances by 0.4 per second). On a screen at 60 frames per second, about 3.3 steps are calculated per frame, with a maximum of 40 steps per frame. The relaxation time τ is 0.05 s in real time, and the dielectric relaxation time is about 0.6 s. The averaging time constants are 1 (2.5 s) for the terminal voltage, the current and the potential in the lower figure, and 3 (7.5 s) for the shading of the rod and the bar graph of charge in the lower figure. The animation runs only while tab ① is displayed and at least one of the two figures is visible on the screen; it stops when you switch to another tab, hide the browser tab or scroll the figures off the screen. If the OS is set to “reduce motion”, it starts in the paused state.
| Quantity | Value | Remarks |
|---|---|---|
| Number of simulated carriers N | 25920 | 216 are drawn (108 on screens narrower than 560 px). 56 ions (27) are drawn separately on a lattice |
| Number of intervals for finding the electric field | 40 | Carriers are distributed with the cloud-in-cell method |
| Time step Δt | 0.002 | 5 ms in real time |
| Relaxation time τ | 0.02 | Probability of being scattered in one step: 0.1 |
| Mean free path ℓ (300 K) | 0.025 | √(πkBT/2m)·τ |
| Screening length λ0 (300 K) | 0.07 | λ ∝ √T. ε = λ02N = 127.0 |
| Dielectric relaxation time τd | 0.245 | λ02/τ, independent of temperature |
| Correction coefficient for the potential difference c | 0.771 | The value for the default 600 K / 300 K. 1 − 0.052 − 1.046λ0(√θh + √θc), θ = T/T0 |
| Conductance of the load G | 0.08 / 3 | Light bulb / wire (nearly a short circuit) |
| Unit of the relative value of the current | 150 | Number of carriers per unit time. Almost 1 for a short circuit at ΔT = 100 K |
| Correspondence of time | 1 = 2.5 s | Advances by 0.4 per second |
| Random numbers | mulberry32 | Seed 20260917. A reset gives the same initial configuration |
Rough values of |S|. Metals have |S| ≈ a few μV/K, good thermoelectric semiconductors have 150–250 μV/K, and semiconductors close to insulators have several hundred μV/K or more. For example, a material with |S| = 200 μV/K and ΔT = 300 K (the default of ①) gives an electromotive force of 60 mV, whereas a metal (|S| = 5 μV/K) gives only 1.5 mV for the same ΔT. The reason why |S| of a metal is small is given at the end of §1.1 and in Mott’s formula in §4, and how |S| of a semiconductor decreases with the carrier concentration is explained by the Pisarenko relation in §5.
Thermocouples. A thermocouple is a thermometer that joins two materials A and B and measures the electromotive force V = (SA − SB)ΔT. If a circuit were made of only one material, the same temperature difference would also act on the wires leading to the voltmeter and the electromotive forces would cancel, so two materials with different S are combined. Most temperature measurements in industrial furnaces rely on this principle. The p–n pair of Problem 2 in §13 is also the same addition (Sp − Sn).
Viewing it as the difference between Fermi–Dirac distributions. The same thing can also be seen as the difference between the Fermi–Dirac distributions on the hot and cold sides (Fig. A1). For easy comparison, the figure puts the Fermi levels EF of the two sides together at the origin of the horizontal axis (in a real sample, EF − Ec itself also changes with temperature, as in the (b) chemical-potential part of §1.1).
Fig. A1. The Fermi–Dirac distribution f(E) = 1/[e(E − EF)/kBT + 1] on the hot side (red, Th = 400 K, kBT = 34.5 meV) and the cold side (blue, Tc = 300 K, kBT = 25.9 meV). The horizontal axis is the energy measured from the Fermi level (meV). The red fill is the part above EF that is more occupied on the hot side, and the blue fill is the part below EF that is more empty on the hot side; the two areas are equal.
This is the tab at the heart of this teaching material. With the single parabolic band (SPB) model and acoustic-phonon scattering (relaxation time τ ∝ E−1/2, λ = 0), it calculates on the spot the Seebeck coefficient S, the electrical conductivity σ, the electronic thermal conductivity κe and ZT as functions of the carrier concentration n (the equations are in §6–§8, the implementation in §11). When carriers are added, |S| decreases while σ and κe increase — the three quantities get in each other’s way, so ZT = S2σT/κ forms a peak.
The upper figure, “The Seebeck coefficient falls while the electrical conductivity rises”. The horizontal axis is the carrier concentration n (log scale), drawn from 1017 cm−3 up to the smaller of 1022 cm−3 and the concentration corresponding to η = 50 (because for small m* and low temperature, η = 50 does not reach 1022 cm−3). Blue is |S| (left axis, μV/K; the upper limit of the scale is the maximum of |S| in the displayed range rounded up to a multiple of 100 μV/K, and it is capped at 1000 μV/K), and red is σ (right axis, S/cm, log scale from 10−2 to 106). The background bands are 1017–1018 cm−3, “Lightly doped”; 1019–1020 cm−3, “Heavily doped” (green); and 1021–1022 cm−3, “Metal-like”; a band whose width inside the displayed range is less than 0.6 decades is left unnamed. The gray dotted line marks the position of the n selected in the controls panel.
The lower figure, “That is why the figure of merit forms a peak”. The thick green line is ZT (left axis), orange is the power factor S2σ normalized by its maximum (right axis), and the gray dashed line is κe/(κe + κL), that is, the fraction of the heat carried by electrons (right axis). The green dot is the top of the ZT peak (“ZT max” and its value), the orange dot is the maximum of the power factor (“PF max”), and the green band is 1019–1020 cm−3.
Controls panel.
The equation box of the controls panel contains Eq. (1) and κe = LσT (the Wiedemann–Franz law, Eq. (40)). L is the Lorenz number, which changes with the degree of degeneracy; for acoustic-phonon scattering it is 1.49×10−8 W Ω K−2 in the nondegenerate case and 2.44×10−8 W Ω K−2 in the degenerate case. The metallic constant L0 = 2.44×10−8 W Ω K−2 corresponds to its degenerate limit (Eq. (41); “Misconception 4” of §8).
Readouts. Under “Values at the selected n” it shows the carrier concentration n, the reduced Fermi level η, S (with sign), the mobility μ (Eq. (34)), σ, the power factor (μW/cmK2), the Lorenz number L, κe and κL, and ZT; under “Top of the peak” it shows the optimal carrier concentration nopt, the maximum of ZT, S at that point, the n at which the power factor is maximal (with a note to that effect if it lies on the higher-concentration side of the ZT peak), and the quality factor B (Eq. (39)). For example, with the room-temperature preset as it is (p-type), at the selected n = 3.15×1019 cm−3 we have η = 0.29, S = +187.6 μV/K and ZT = 0.970, and at the top of the peak nopt = 1.58×1019 cm−3, the maximum of ZT is 1.098, S = +242 μV/K, the maximum of the power factor is at 4.29×1019 cm−3, and B = 0.442.
The message box, and what to read off from the figures. The “message” briefly tells whether the n at the top of the ZT peak lies within the range 1019–1020 cm−3, or on the higher or lower side of it, and where the peak of the power factor is (whether it is on the higher-concentration side of the ZT peak). What follows is the background that you should read off from the figures. In a material with large m* and small B the top rises to near 1021 cm−3, and in a material with small m* or large B it moves to the lower-concentration side (§7.3). On the left side of the figure (few carriers), S is large but there is hardly any σ; on the right side (close to a metal), σ is large but S vanishes, and in addition κe grows and heat leaks. That the “heavily doped semiconductor”, neither an insulator nor a metal, is the best is the result of this tug of war (§7.3). Also, the peak of the power factor S2σ (orange) is shifted to the higher-concentration side of the peak of ZT (green). This is because, as κe grows with n, the peak of ZT is pushed back toward the lower-concentration side (“Misconception 3” of §6.5).
This tab takes one chosen material parameter, varies it in four ways, and draws the resulting ZT peaks on top of one another to show how they move. The claim in the background is that of §7: “the height of the ZT peak and its position measured in η are determined by the quality factor B alone.”
The figure and the table. The figure “How does the ZT peak move?” has the carrier concentration on the horizontal axis (1017–1022 cm−3, log scale) and ZT on the vertical axis, and draws four curves, each with a dot at its top (the green band is 1019–1020 cm−3). The table “Comparison of the peak tops” below the figure lists, for each condition, B, the maximum of ZT, nopt and |S| at the top. The equation box of the controls panel is the quality factor of Eq. (39) (which goes back to the material factor of Chasmar and Stratton).
Controls panel. Under “Parameter to vary” choose one of the following five; the material parameters below it (the same presets, types and sliders as in ②; there is no slider for n) serve as the reference. The four curves are the reference values multiplied by the factors shown in the table.
What to read off from the figure for each option. The “message” shows only the one bold sentence of each of the following items and a short supplement.
This tab draws how the maximum of ZT changes as the grains of a polycrystal are made smaller. The idea is in §9 and the equation used for the calculation is Eq. (62) of §11.
The figure “Smaller grains block the heat first”. The horizontal axis is the grain size d (log scale, 5 nm to 1 mm). The thick green line is the maximum of ZT (left axis) when the optimal carrier concentration is taken at each grain size, drawn by scanning the grain size at 91 points. The teal dashed line is κL divided by its original value (the κL in the controls panel), and the red dotted line is μ0 divided by its original value (both on the right axis, “Ratio to bulk”). The gray dotted line and the green dot are the grain size chosen in the controls panel and the maximum of ZT there.
Controls panel.
Readouts. At the selected grain size it shows κL (what percentage of the original), μ0 (what percentage of the original) and the maximum of ZT; as the reference for comparison, the maximum of ZT at a grain size of 1 mm (including only the effect of alloying), the ratio of the two, the “improvement factor”, and the grain size at which ZT is maximal in this model (the maximum among the 91 points) together with its ZT. With the default values, the grain size that maximizes ZT is displayed as “11 nm (ZT 1.63)” (Table 4, §9.2).
The message box, and what to read off from the figure. When the grain size is reduced to about Λph, phonons are scattered at grain boundaries and κL decreases. The “message” briefly tells which of the following four cases the settings fall under. The meaning of each case is as follows.
This tab draws what efficiency a device made from a material with a known average ZT can achieve. The equations are Eq. (55) (power generation) and Eq. (59) (cooling) of §10, which give ideal upper limits from the constant-property model. The equation box of the controls panel is Eq. (55); the first factor is the Carnot efficiency and the second factor is the fraction determined by the material. This η is the conversion efficiency and is a different quantity from the reduced Fermi level η of ② (see the “Notation” box). The “average ZT” on the horizontal axis corresponds to ZT̄ = Z(Th + Tc)/2 of §10.
The upper figure, “Power generation”. The horizontal axis is the average ZT (0–4) and the vertical axis is the maximum power-generation efficiency (%); the blue curve is Eq. (55) and the red dashed line is the Carnot efficiency (Th − Tc)/Th. The blue dot is the efficiency at the ZT chosen in the controls panel.
The lower figure, “Cooling (Peltier)”. This is the maximum coefficient of performance COP when heat is pumped up with the cold side kept at Tc and the hot (heat-rejecting) side kept at Th; the teal curve is Eq. (59) and the red dashed line is the Carnot coefficient of performance Tc/(Th − Tc).
The same Th and Tc are used for power generation and cooling.
Since this temperature difference cannot be produced when √(1 + ZT) ≤ Th/Tc (§10.6), the curve is drawn only to the right of that ZT, and if it cannot be produced with the present ZT, the message “This ΔT cannot be produced with the current ZT = … (required ZT = …)” is displayed in place of the dot.
When the required ZT = (Th/Tc)2 − 1 exceeds the horizontal-axis range 4, the curve itself is not drawn and a message to that effect is displayed.
Controls panel.
Hot side Th (320–1300 K, default 800 K), cold side Tc (200–700 K, default 300 K; kept at least 10 K below Th), and the average ZT of the material (0–4, default 1).
The three buttons are examples of applications: “Waste-heat power” is 800 K / 300 K with ZT = 1.0, “Peltier cooling” is 300 K / 270 K with ZT = 1.0, and “Spacecraft RTG” is 1275 K / 575 K with ZT = 0.8.
Readouts. For power generation it shows the Carnot efficiency, the maximum efficiency of thermoelectric generation and its fraction of the Carnot efficiency; for cooling, the Carnot COP, the maximum COP of thermoelectric cooling (with a note if the temperature difference cannot be produced) and the minimum ZT needed to produce this ΔT.
For example, with the default values (800 K / 300 K, ZT = 1) the maximum efficiency is 14.47 % (23.2 % of the Carnot 62.5 %, Table 5); with “Peltier cooling” the COP is 1.130 (Carnot 9.00) and the minimum required ZT is 0.23; and with “Spacecraft RTG” the efficiency is 10.46 % (19.1 % of the Carnot efficiency).
The message box, and what to read off from the figures. The “message” tells the upper limit of the power-generation efficiency at the present ZT and its fraction of the Carnot efficiency, and the maximum COP of cooling (or, if the temperature difference cannot be produced, says so). Even with ZT = 1 the power-generation efficiency stays at about 20 % of the Carnot efficiency (it depends on the temperatures), and it reaches the Carnot efficiency only as ZT → ∞. This is why a practical rule of thumb is ZT ≳ 1, and why it is said that ZT ≳ 2–3 is needed to compete in efficiency with existing heat engines and refrigerators. Moreover, this is an ideal upper limit, and real modules have lower values because of the contact resistance of the electrodes and heat leaks (“Misconception 7” of §10). Thermoelectric conversion is nevertheless used because it has no moving parts, is quiet and is hard to break. Examples are planetary probes that keep running for decades on nuclear batteries (RTGs), and the temperature control of lasers, which dislike vibration. In cooling, the larger the temperature difference ΔT, the lower the COP, and when √(1 + ZT) ≤ Th/Tc that temperature difference cannot be produced in principle (§10.6).
This tab lines up, as horizontal bars, the operating temperature range and a typical ZT for representative families of thermoelectric materials. The horizontal axis of the figure “Operating temperature range and typical ZT by material system (textbook level)” is temperature (200–1300 K), and the bands at the top indicate “Cooling, room temp.” (250–400 K), “Waste heat (mid-temp.)” (500–850 K) and “High temp., space” (900–1300 K). The bar of the system chosen under “Material system of interest” in the controls panel is drawn dark, the corresponding row of the table “Comparison of material systems” below the figure is highlighted, and the “message” shows an explanation of that system (initially oxides). The contents of the table are given in Table A3.
| Material system | Temperature range | Typical ZT | Main uses | Advantages | Disadvantages |
|---|---|---|---|---|---|
| Bi2Te3-based | 200–450 K | ≈ 1 | Peltier cooling; small power generation near room temperature | Best track record at room temperature | Te is scarce and expensive |
| Mg2Si- and Mg3Sb2-based | 450–800 K | ≈ 1–1.5 | Mid-temperature waste-heat power generation | Abundant resources; lightweight | Oxidizes easily |
| PbTe-based | 500–850 K | ≈ 1–2 | Mid-temperature waste-heat power generation; spacecraft | High ZT at mid temperatures | Pb is toxic and Te is scarce |
| Skutterudites (CoSb3-based) | 500–850 K | ≈ 1–1.5 | Mid-temperature waste-heat power generation | Low κL from the rattling of guest atoms packed in cages | Toxicity and limited supply of Sb |
| SiGe | 900–1300 K | ≈ 1 | Nuclear batteries (RTG) | Stable at high temperatures | Ge is expensive |
| Oxides (Ca3Co4O9, SrTiO3-based, etc.) | 700–1100 K | ≈ 0.2–0.5 | High-temperature waste heat in air | Stable in air, nontoxic, abundant resources | Low ZT (low mobility) |
Explanations by material system. The “message” shows only the key point, which is the first sentence of each item.
All the figures and numbers of the simulator are calculated on the spot by the JavaScript of the page; they are not curves prepared in advance. However, this model is “for understanding trends” and does not quantitatively predict the ZT of a specific substance. The details of the implementation (the scan in η, the Fermi–Dirac integrals, the equations for the transport coefficients, how the top of the peak is found, the temperature options of ③, Eq. (62) of ④, and the minimum ZT of ⑤) are collected in §11, so here only the main points and what is not written in §11 are given.
The numbers on this page were obtained by calculating the equations in the main text with both the same code as the simulator (Simpson’s rule for the Fermi–Dirac integrals) and an independently written calculation (Fermi–Dirac integrals via the polylogarithm, and numerical optimization), and checking that they agree. The material parameters of the presets are educational values that give the order of magnitude, not measured values of specific substances.
Companion simulator: Thermoelectric Conversion — Why Is There an Optimal Carrier Concentration?