You are on page 1of 14

Review

Cite This: ACS Appl. Energy Mater. 2020, 3, 2200−2213 www.acsaem.org

Critical Analysis of Single Band Modeling of Thermoelectric


Materials
Harshita Naithani† and Titas Dasgupta*,†
Department of Metallurgical Engineering and Materials Science, Indian Institute of Technology Bombay, Mumbai 400076, India
*
S Supporting Information

ABSTRACT: Study of the electronic band structure of thermoelectric (TE)


materials is fundamental to both its understanding and further development.
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

Currently theoretical models which assume a single-band-based charge


transport are utilized due to their predictive capabilities and ease of
implementation. However, most good thermoelectric materials have complex
Downloaded via NANJING TECH UNIV on October 29, 2020 at 12:54:21 (UTC).

band structures with multiple bands near the band edge. The extent of
applicability of single-band models in such systems is questionable and forms
the objective of this study. To check this, we have chosen five well-known TE
materials and modeled their room temperature properties using the single
parabolic band (SPB) model and the single Kane band (SKB) model. The
room temperature experimental data for these materials were extracted from
literature reports, and the analysis was carried out on a relatively large sample
set (with over 350 data points spread across the various materials). Our
analysis indicates the failure of single-band models in situations where multiple near-degenerate bands are present close to the
band extrema. The associated errors are in the estimated density of states effective mass (mD*), Lorenz number (L), and lattice
thermal conductivity (κL), which in turn result in erroneous predictions of the optimum charge carrier concentration and zT
values. We also find that identifying whether the band edge is parabolic is difficult from a visual comparison of the SPB and SKB
Pisarenko plots since the observed variations are well within the acceptable limits of experimental error. To overcome this
problem, we propose an error analysis technique which can be used to find the best fit model. The error analysis can also be
useful in identifying the dominant charge carrier scattering mechanism as shown from our study. Overall, our work highlights
the need for implementation of multiband modeling while working with materials with complex band structures.
KEYWORDS: single parabolic band model (SPB), single Kane band model (SKB), thermoelectric materials, Seebeck coefficient,
Lorenz number, band mass

■ INTRODUCTION
Thermoelectric materials have the unique ability to directly
be readily obtained from experiments. If the electronic band
structure of a material is known, the transport properties can
interconvert heat and electrical energy.1 Hence they have be predicted. But density functional theory predicts the band
immense application potential as solid state devices where they structure at ground state, which does not reflect the actual state
can be used as generators, refrigerators, or temperature sensors. of the material. Therefore, modeling of thermoelectric
In this world of ever rising demand for consumable energy, materials becomes important. On the basis of the kind of
harvesting waste heat to electricity would not only help in the band which takes part in electrical transport, single parabolic
preservation of fossil fuels but also reduce our carbon footprint. band (SPB) and singe Kane band (SKB) are the most
However, such thermoelectric heat harvesters are not commonly used models. 3,4 They are applicable when
commonplace. This is due to their poor conversion efficiency conduction is limited to just one band, parabolic or
(commercially available thermoelectric devices have around nonparabolic, respectively. One can also go for modeling
5−6%2). Current research in this field is therefore concen- which involves multiple bands; however, solving the equations
trated on improving the material properties. becomes relatively complex. Another parameter to keep in
The transport properties of a material such as the Seebeck mind is the scattering mechanism of charge carriers since these
coefficient (S), electrical conductivity (σ), and hermal models include it as well.5 These models use some
conductivity (κ, the sum of electronic and lattice thermal experimental data as input and return some important material
conductivities) contribute to a dimensionless number called
S 2σ
figure of merit, zT = κ T which in turn determines the Special Issue: Thermoelectrics
efficiency of a thermoelectric material. All of these parameters Received: October 12, 2019
depend upon the charge carrier concentration. Hence this Accepted: November 21, 2019
concentration needs to be optimized.1 These parameters can Published: December 9, 2019

© 2019 American Chemical Society 2200 DOI: 10.1021/acsaem.9b02015


ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Table 1. Single Band Expressions for Charge Carrier Concentration (n), Seebeck Coefficient (S), Hall Prefactor (rH), Lorenz
Number (L), Hall Mobility (μH), and Power Factor (S2σ) for the SPB and SKB Modelsa
property SPB SKB
(2mD*kBT )3/2 * kBT )3/2
(2mD0
n F1/2(η) (0F03/2(η))

ÅÄ ÑÉÑ
3π 2ℏ3 3π 3ℏ3
kB ijj 2F1(η) zy kB ÅÅÅÅ 1F −1 2(η) Ñ
jj − ηzzzz ÅÅ 0 1 − ηÑÑÑÑ
e jk F0(η) { e ÅÅÅÇ F −2(η) ÑÑ
ÑÖ
S

3 F −1/2(η) 3K (K + 2)(0F03/2)(0F −1/2


4 )
rH F1/2(η) 2

Ä É
2Å 2Ñ
4 F0 (η) (2K + 1)2 (0F −1 2)2
Å Ñ
ij kB yz 3F0(η)F2(η) − 4F12(η) ij kB yz ÅÅÅÅ 2F −1 2(η) jij 1F −1 2(η) zyz ÑÑÑÑ
jj zz jj zz ÅÅ 0 1 j z
je z j e z ÅÅ F (η) − jj 0F 1 (η) zz ÑÑÑÑ
2

k { k { ÅÅÇ −2 k −2 { ÑÑÖ
L
F0 2(η)
0 1
F −1/2(η) F −2(η)
μH μ0 rHμ0 0 3/2
2F0(η) F0 (η)

i * y ij 2F (η) yz i * y ij 1F 1 (η) yz
jij 8π zyzμ k 2jjj 2mDkBT zzz F0(η)jjjj 1 − ηzzzz jij 8π zyzμ k 2jjj 2mD0kBT zzz F −2(η)jjjj 0 −1 2 − ηzzzz
j z 0 B jj zz j z 0 B jj zz
3/2 2 3/2 2

k 3e { k h { k F0(η) { k 3e { k h { k F −2(η) {
S2σ 2 2
0 1

a
mD* represents the density of states effective mass of a parabolic band, while mD0
* is the density of states effective mass at the band extremum (E =
0) of a nonparabolic band. μ0 represents the mobility free parameter, and K is the ratio of the longitudinal to transverse effective mass.3,4.

parameters. Lattice thermal conductivity is one such parameter m* the band effective mass; or (b) nonparabolic (Kane

i Ey
Ejjj1 + E zzz = 2m * , where Eg is the band gap of the material.4
which can be obtained. Its knowledge would aid in further type)the energy dispersion relation is of the type
reducing its value to increase the figure of merit.
k g{
ℏ2k 2
The use of the SPB model is widespread for initial analysis of
materials due to its mathematical simplicity. However, it has The quadratic energy term in this equation is to account for
many shortcomings when used to analyze some material the nonparabolic nature of the band. The direct consequence
systems over a wide range of charge carrier concentration. For of nonparabolicity is that the band mass (m*) is a function of
instance, for materials in which electronic transport takes place both energy (E) and band gap (thus temperature), unlike the
through multiple near-degenerate bands, it is likely to fail. In parabolic band where m* is constant. These two models are
such a case multiband modeling would be ideal. Thus, there commonly known as the single parabolic band model and the
are many questions associated with the applicability of present single Kane band model, respectively.
and commonly used models to a particular material. How The solutions of the various transport properties in the SPB
different are SPB and SKB from each other? How accurate is a model3 involve integrals which are represented in terms of the
scattering mechanism in describing the behavior of a material? ∞ ε j dε
“Fermi−Dirac” integral given by Fj(η) = ∫ .5 Here
Which model best fits the experimental data? What can be the 0 1 + exp(ε − η)
error associated with the calculation of lattice thermal ε represents the reduced energy (E/kT) and η the reduced
conductivity? To answer these questions, a comparison of Fermi level (EF/kT). The term “j”, known as the index of the
the available models needs to be made. In this work we have integral, depends on both the transport property and energy
reviewed some common thermoelectric material systems and dependence of the relaxation time of the charge carriers.
analyzed them using the SPB and SKB models. In addition, we Solution of a more complex integral4 is necessary to account
have incorporated the use of a two-band model for transport for band nonparabolicity in the SKB model, where the integral
based on available literature to fit the data, which has provided is given by
∞i
jj− ∂f yzzε n(ε + αε 2)m (1 + 2αε)k dε
j z
more insight into the nature of some materials.

k ∂ε {
Single Band Models (Parabolic and Kane Bands). The n m
F k (η , α) = ∫0
assumption that electrical transport of materials with complex
band structures can be modeled using a single energy (1)
dispersion curve is based on the fact that TE materials are Here the term “α” is known as the band nonparabolicity
heavily doped semiconductors. Thus, only the majority charge parameter and depends on the band gap of the material
carriers and hence the bands associated with them are relevant kT
(α = E ). The terms n, m, and k are indices of the integral
for electrical transport. Further, the majority carrier bands are G

replaced with an equivalent band while working with single- whose values depend on the transport property and the charge
band models. The validity of this assumption is something that carrier scattering mechanism.
will be discussed in the next section. Transport equations for The solutions of both of these integrals require the
the single-band model are based on the Boltzmann transport information on the energy dependence of scattering of the
equation (BTE).5 Depending on the a priori information charge carriers (depends on the scattering mechanism). In the
obtained from the electronic band structure of the material SPB model, the energy dependence of the relaxation time (τ)
system, the single band is assumed to be either (a) parabolic is given by the equation τ = τ0ελ−1/2, where the value of λ
ℏ2k 2 depends on the scattering mechanism. For TE modeling at
the energy dispersion relation is given by E = 2m * , where ℏ is room temperature and above, it is assumed that acoustic
the reduced Planck’s constant, k the Boltzmann constant, and phonon scattering of the charge carriers is the dominant
2201 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

mechanism for which the λ value is zero.5 It is to be noted that are multiple near-degenerate bands close to the band edge or
other scattering mechanisms or multiple mechanisms with the at elevated temperatures where the contribution to transport
same energy exponent can also be implemented in the model. from the minority charge carriers cannot be neglected. For
Energy dependence of relaxation time in the SKB model is such materials the various thermoelectric properties can be
given by the expression4 obtained from9
1 n= ∑ ni
∝ [ε(1 + ε /εg)]−r ε1/2(1 + 2ε /εg)(1 + ε /εg)1/2
τ (2) i (3)

where the parameter r is equal to zero for scattering by σ= ∑ σi


acoustic phonons. i (4)
The input variables to both of these models are the
experimental data of the Seebeck coefficient (S(T)), electrical ∑i Siσi
conductivity (σ(T)), thermal conductivity (κ(T)), and Hall S=
∑i σi (5)
coefficient (RH(T)) at a given temperature (T). The solution
to the SPB model involves a sequential procedure where at
∑i RHiσi 2
each step the number of unknown parameters is restricted to RH = 2
one. An excellent review on the SPB modeling which details (∑i σi) (6)
the solution procedure can be found elsewhere.3 A similar
procedure is adopted for solving the SKB model6 which σσ
i j
κE = ∑ κi + ∑ (Si − Sj)2 T
additionally requires the information on the material band gap σi + σj
i i,j (7)
as an input parameter.
In Table 1 various equations for SPB and SKB that have σ
L=
been used in this work are provided. The corresponding κET (8)
integrals for SPB and SKB are as discussed earlier. It has been
assumed that the charge carriers are scattered by acoustic Here the indices i and j represent contributions from a given
phonons in both models. band. The multiband expression for electronic contribution to
The applicability of the models to a given material system thermal conductivity (κE) has an additional summation term
can be ascertained from the variation of the experimental data (second term in the equation) which represents the
from the theoretically calculated values. Typically a plot of the contribution due to interband transport. For mixed conduction
Seebeck coefficient versus the charge carrier concentration (e.g., one conduction and one valence band), this represents
known as the “Pisarenko plot”3 is drawn to check the band the bipolar thermal transport which can be significant at
edge characteristics. The plot consists of experimental data elevated temperatures. Modeling using multiband expressions
points which represent the variation of the Seebeck coefficient becomes complex as the expressions cannot be reduced to
with charge carrier concentration at a given temperature. For provide a stepwise solution such as the single-band models.
both the SPB and SKB models, an ideal fit would be a Typically this involves simultaneous solution of the transport
theoretical line (with fixed m*D or m*D0 value) that passes equations or least-squares fitting to obtain the desired data.10
through all of the data points. Normally, the validity and hence In this work, we have tried an inverse solution; i.e., from
the applicability of the model are based on a visual judgment of reported data of the various intrinsic parameters, the values for
the Pisarenko plot. Any deviations from the theoretical the transport properties were obtained and compared with
those from the single band models.


Pisarenko line is taken to be proof of band nonrigidity,
presence of multiple bands, or presence of special features in
the electronic band structure, e.g., resonant states.7 Additional REVIEW OF LITERATURE DATA
checks for model validity can be obtained from plots of charge In this section we review the reported experimental data for
carrier mobility (μ) versus charge carrier concentration (n) five state-of-the-art thermoelectric materials. These materials
and zT versus n. It is to be noted that mostly the Pisarenko were chosen as examples of specific differences in the
plot provides the best fit between experiment and theory since electronic structure and are representative of contribution to
the Seebeck coefficient is independent of mobility (depends on electrical transport coming from (i) a single parabolic band
the energy dependence of mobility5) and hence largely (ZnSb), (ii) multiple (two) degenerate bands (Mg2Si0.3Sn0.7),
unaffected by microstructural features. (iii) multiple near-degenerate bands (Mg2Si0.4Sn0.6 and
Solutions of the SPB and SKB equations involve solving the CoSb3), and (iv) a single nonparabolic band (PbTe). Reported
Fermi integrals. This can be easily done numerically using experimental data of S, σ, and RH at 300 K for these
numeric computational packages such as MATLAB or compounds were used as input to the single-band models. The
Octave.8 As an example, defining the Fermi integrals and data set used for the analysis is provided in the Supporting
inverse solving to obtain the reduced Fermi level (η) from the Information. The methodology used for the analysis is
experimental Seebeck coefficient (S) value (which is the first described first followed by the observation and results for
step for both models) using MATLAB is provided in the each material system.
Supporting Information (see Figure S1). Further steps to Methodology. Experimental data were used as input for
obtain the various parameters can be defined from the the SPB and SKB models, and various intrinsic materials
estimated η value. parameters (η, rH, mD*, L, n, and μ0) were calculated for all data
Beyond Single Band Models. To analyze materials where points. From here on, the SKB m*D0 has been referred to as m*D.
charge carriers from multiple bands contribute to electrical Typical plots generated for various materials were (a) variation
transport properties, the single-band models are generally not of Seebeck coefficient (S) with the Hall carrier concentration
valid. Such a scenario can arise in semiconductors where there (nH) (Pisarenko Plot), (b) mD* as a function of nH, (c) power
2202 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Figure 1. Representative plots showing (a) least-squares minimization of Seebeck coefficient error to obtain the mD*av value and (b) least-squares
minimization of μ error to obtain the μav
0 value.

Figure 2. (a) Schematic band structure of the conduction band edge of Mg2Si0.3Sn0.7. (b) Pisarenko plot showing variation of S with Hall carrier
concentration (nH). The lines indicate the theoretical data obtained from the various models. The inset shows a zoomed-in region highlighting the
near overlap of the various theoretical model lines. (c) Variation of mD* values with nH obtained from the SPB and SKB models. The dashed lines
indicate the average m*D values obtained from SPB and SKB models, respectively. (d) Experimental and theoretical data showing the variation of S2σ
with nH. (e) Experimental data plot of the estimated Lorenz number from the SPB model as a function of nH. The theoretical two-band L and the
SPB L (derived from the two-band data) are also plotted. The experimental data for the plots have been taken from refs 11−20.

factor (S2σ) as a function of nH, and (d) Lorenz number as a the Supporting Information. The mD*av value was used to
function of nH. To generate theoretical fits for the data, average generate the theoretical fit in the Pisarenko plot while both
values of m*Dav and μav 0 were calculated using a least-squares m*Dav and μav0 values were needed for generating the S σ plot. In
2

minimization technique. For the m*Dav calculation, the quantity addition, to compare the goodness of fit for the SPB and SKB
∑i(Sexp − Smodel)2 (where Sexp is the experimental data point, model, the root-mean-square error of the Seebeck coefficient
Smodel is the value obtained from the single-band model, and i is data (Srms
error) was used as a metric.
a data point) was minimized as a function of mD*. The mD* value Two parabolic band model equations (provided in the
corresponding to the minimum S least-squares error was taken previous section) were used to explain discrepancies in the
as m*Dav (as shown in Figure 1a). Similarly, the μav 0 value was observed data. The parameters required for solving the
obtained from the minimization of ∑i(μexp − μmodel)2 (μexp is equations were taken from literature or generated from the
experimental data and μmodel is the value obtained from the single-band models (explanation given in individual materials’
model; shown in Figure 1b). Additional details are provided in sections).
2203 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Table 2. Estimated Srms


error Error Values for the Different Materials Obtained Using the SPB and SKB Models

Srms
error

SPB SKB
Mg2Si0.3Sn0.7 11.99 12.37
Mg2Si0.4Sn0.6 52.57 54.20
CoSb3 35.49 37.61
ZnSb 51.99 53.11
PbTe 12.52 (nH full range) 13.25 (nH full range)
19.33 (nH < 1025 m−3) 18.89 (nH < 1025 m−3)
6.85 (I-doped,69 nH full range) 6.67 (I-doped,69 nH full range)
4.79 (I-doped,69 nH > 1025 m−3) 5.20 (I-doped,69 nH > 1025 m−3)

Figure 3. (a) Schematic band structure showing the nearly degenerate conduction bands in Mg2Si0.4Sn0.6. (b) Pisarenko plot showing the variation
of S with nH. For both the SPB and SKB models, the theoretical lines were drawn using multiple mD* values. For better clarity, a zoomed-in image of
the encircled region is provided in the inset. (c) Variation of m*D values with nH obtained from the SPB and SKB models. The solid line represents
the mD* data obtained from SPB using the properties calculated from the two-band model as the input. The dashed lines indicate the mD* values of
the heavy and light bands. (d) Experimental and theoretical data showing the variation of S2σ with nH. (e) Experimental data of the estimated
Lorenz number from SPB model as a function of nH. Theoretical two-band L and SPB L are also plotted. The experimental data for the plots have
been taken from refs 23−31.

Mg2(Si0.3Sn0.7) (n-Type). Mg2Si1−xSnx solid solutions are obtained from different models. For the SKB model, the α
potential n-type thermoelectric materials. The high zT in these parameter was taken to be 0.0637 (on the basis of the reported
materials is due to a combination of enhanced power factor band gap21) and the K value is assumed to be 1. To generate
due to the convergence of conduction bands and reduced the two-band Pisarenko line, information on the density of
lattice thermal conductivity due to alloy scattering.11 The best states mass and mobility of the two individual bands are
TE performance is observed close to x ∼ 0.7 (Mg2(Si0.3Sn0.7)),
required. For the light band (band 1 in Figure 1a), data of m*D1
where the two conduction band minima at the X k-point in the
and μ01 were taken from literature (mD1 * = 1.1me, and μ01 =
reciprocal space become degenerate. The schematic band
structure for this composition is shown in Figure 2a indicating 0.0123 cm2/(V s)).22 These literature values were obtained
the two degenerate bands which contribute to the electrical from SPB modeling in Mg2Si in which the light band alone
transport when this material is n-doped. In Figure 2b, the contributes to the electrical transport. To obtain the
Pisarenko plot is given showing the experimental data points information for the heavy band (mD2 * = 2.19me and μ02 =
(round symbols) along with the theoretical plots (lines) 0.0078 cm2/(V s)), the following equations were used:
2204 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Figure 4. (a) Schematic band structure of the conduction band edge in CoSb3 which consists of two near-degenerate bands. (b) Pisarenko plot
showing the variation of the experimental and theoretically calculated S values as a function of nH. Three different average m*D values were
calculated for the SPB model (over the entire nH range, 2.7me; for nH < 1025 m−3, 1.4me; for nH > 1026 m−3, 3.7me). For the SKB model the average
over the entire nH range was used for the m*D calculation. (c) Variation of the experimental m*D (calculated using the SPB model) with nH. The
values of mD* used for the two-band calculation32 are shown as dashed lines. The solid line represents the mD* data obtained from the SPB model
using the two-band-calculated properties as input. (d) Variation of the experimental and calculated S2σ values as a function of nH. (e) Plot of
experimental data of the estimated Lorenz number from SPB model as a function of nH. The theoretical two-band L and SPB L are also plotted.
The experimental data for the plots have been taken from refs 32−55.

* = ((mD*av )3/2 − (mD1


mD2 * )3/2 )2/3 (9) to an erroneous definition of nH in a multiband scenario). It is
to be noted that the SKB model predicts a lower optimum nH
compared to other models. This is attributed to the higher
(mD*av )3/2 μ0av − (mD1* )3/2 μ
01 value of the effective mass estimated from the SKB model. In
μ02 =
(mD*av )3/2 (10) Figure 2e, the Lorenz number values are shown as a function of
nH. Both band models predict the same values of L in the
(where the m*D and
av
μav
0 values were obtained from the least- entire carrier concentration range.
squares minimization technique using the SPB model). As seen It is to be noted that the close match between the SPB and
from the data, the differences between the different models are two-band data in Figure 2b,d,e is also a result of the way the
less compared to the experimental scatter in the data. This is theoretical plots have been generated. The SPB n value has an
indicative of (a) validity of the SPB model in a degenerate two- r
error due to the assumption of RH = neH which is invalid for a
band scenario (the inset shows minor differences between the
SPB and two-band lines and is attributed to errors in defining two-band system. The associated error depends on the
nH in a multiband scenario) and (b) difficulty in differentiating mobility ratio of the bands (due to the
nμ +n μ
parabolic from a Kane-type nonparabolic band. To find the 1
assumption n = (n 1μ 1+ n 2μ 2)2 ). The error in the SPB n value
better fit, the root-mean-square error of the S data (Srms
error) were
1 1 2 2

calculated as given in Table 2. The data are indicative of the further results in an error in the estimated m*Dav value. Since
band being parabolic. * and μ02 were calculated on the basis of this mD*av value, the
mD2
In Figure 2c, the calculated m*D values from both the SPB two-band data generated from these values also contain this
and SKB values are given. As observed, the SKB model tends error and result in a good match with the SPB data. This
to give a higher absolute value compared to the SPB model. however does not affect the calculated optimum S2σ and zT
The mD* values are also observed to be independent of nH absolute values since the experimental data are taken as input
indicative of the validity of models in the studied range. The and the errors in the intrinsic parameters (m*D and μ0) cancel
S2σ data (shown in Figure 2d) also indicate the validity of the out when used to generate the theoretical values of S2σ and zT.
SPB model in a degenerate two-band scenario (minor The observed effect is thus a translation of the plots along the
mismatch between SPB and the two-band plot is attributed x-axis (n values). In principle, this indicates a failure of the SPB
2205 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Figure 5. (a) Schematic band structure of the valence band edge in ZnSb. Generally only the light band (band 1) is considered while modeling its
TE properties; hence, the heavy band is shown with a question mark. (b) Pisarenko plot showing the variation of the experimental and theoretically
calculated S values as a function of nH. Two different average mD* values were calculated for both SPB and SKB models (over the entire nH range,
0.46me (SPB) and 0.58 me (SKB); for nH > 1025 m−3, 0.72me (SPB) and 0.91me (SKB)). (c) Variation of the experimental m*D (calculated using the
SPB model) with nH. The values of mD* used for the two-band calculation are shown as dashed lines. The solid line represents the mD* data obtained
from the SPB model using the two-band-calculated properties as input. (d) Variation of the experimental and calculated S2σ values as a function of
nH. Data for the plots are taken from refs 56−65.

model in predicting the intrinsic material parameters with the nH dependence of m*D cannot be attributed to band
multiple degenerate bands; however, this in no way affects the nonparabolicity (in which case the mD* calculated from SKB is
accuracy of the predicted optimum extrinsic properties. While expected to be invariant to changes in nH). To probe the
ways to overcome this problem can be the scope of future possible influence of the two near-degenerate bands, two-band
study, in the present work this issue is not discussed any calculations were performed using the same set of m*D and μ0
further. In the Supporting Information, the extent of error in values as that of Mg2Si0.3Sn0.7. The Pisarenko line for the two-
the n values with material parameters of Mg2Si0.3Sn0.7 and band model provides a much better fit to the experimental data
CoSb3 are provided (shown in Figure S2). as compared to the single-band models (shown in the inset of
Overall, the results for Mg2Si0.3Sn0.7 reaffirm this material to Figure 3b). Further, the output from the two band analysis (S
be a “canonical example” of applicability of SPB in a multiband and RH) was taken as the input data for SPB. As evident from
scenario, as mentioned by Uher et al.13 Figure 3c, the m*D values obtained from the two-band data
Mg2(Si0.4Sn0.6) (n-Type). The next material that we have agree with those obtained from experimental data (S and RH),
surveyed is Mg2(Si0.4Sn0.6), which also belongs to the indicating that it is indeed a two-band scenario.
Mg2(Si1−xSnx) family. However, in this material the condition The experimental data of S2σ as a function of nH is shown in
of conduction band degeneracy is lifted (shown in Figure 3a), Figure 3d along with various theoretical calculations. The best
resulting in two nearly degenerate bands at the conduction fit to the data is obtained for the two-band model, which also
band edge with a reported interband separation of 0.061 eV.21 indicates the influence of the two conduction bands on the
In Figure 3b, the Pisarenko plot is shown. As seen from the observed electrical properties. The predictions of S2σ and
figure, unlike Mg2(Si0.3Sn0.7) the single-band lines do not fit optimum nH from the single-band models are seen to be
the data in the entire nH range. To get a better fit, two different dependent on the nH range from which the m*D is estimated,
values of m*Dav were calculated (for both SPB and SKB indicating the failure of single-band models. The other effect of
models), one using the data for the entire carrier concentration using the SPB model is the difference in the estimated Lorenz
range and the other for nH > 1026 m−3. As seen from the plot, number (compared to the two-band model) as seen in Figure
different m*Dav values result in a better fit over specific ranges of 3e. The two-band model predicts a higher value of L in the
nH. This is indicative of the variation of m*D with nH , and as entire nH range. The effect of this on the calculated κL and zT
seen from Figure 3c, there is a systematic increase of m*D with is discussed in the next section.
increasing nH. This trend of increasing mD* is similar for both CoSb3 (n-Type). CoSb3-based materials are also well-
the SPB and SKB models. This observation also indicates that known TE materials. The conduction band edge of this
2206 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Figure 6. (a) Schematic band structure showing the sole conduction band in PbTe. Generally, it is taken to be a Kane-type nonparabolic band, but
our results indicate otherwise. (b) Pisarenko plot showing the variation of S with nH for both SPB (solid line) and SKB (dashed line). A zoomed-in
image of the encircled region is provided in the inset. For lower nH values SKB gives a better fit than SPB, and for higher n values SPB gives a better
fit than SKB. (c) Variation of mD* values with nH obtained from the SPB and SKB models. mD* for SKB deviates more from the average SKB mD* at
higher nH. (d) Experimental and theoretical data showing the variation of S2σ with nH. (e) Experimental data of the estimated Lorenz number from
SPB and SKB models as a function of nH. The experimental data for the plots have been taken from refs 6, 67−76.

material is similar to that of Mg2Si0.4Sn0.6 with two near- impurity band conduction based on studies of low-doped
degenerate bands (shown in Figure 4a). The reported specimens.56 However, the observed increase of mD* even at
interband separation is 0.08 eV and the density of states heavy doping levels and the similarity to previous examples
mass being 0.7me and 4.8me, respectively.32 This material was indicate the possibility of two-band conduction. A recent DFT
studied to check the repeatability of the observed trends for the electronic band structure calculation66 reveals the presence of a
two near-degenerate band scenarios. The plots in Figure 4b−e heavy hole band around 0.8 eV below the valence band
are similar to those for Mg2Si0.4Sn0.6 (compare Figure 3b−e). maximum. To verify the possibility of multiband conduction, a
For the two-band model, necessary parameters were taken two-band model was used to fit the experimental data. Due to a
from the work of Snyder et al.32 (we restrict our model to two lack of literature reports on the properties of the heavy band,
bands (2 CB) as compared to three bands (2 CB + 1 VB) since required data were extracted from the high nH range. It should
our calculation is restricted to 300 K). The obtained overall be noted that our results only highlight the possibility of two-
trends in the Pisarenko plot and variations of m*D, S2σ, and L band conduction in this material and further studies are
with nH are similar to those of Mg2Si0.4Sn0.6. Finer differences, required to get accurate properties of the heavy band. As seen
e.g., the extent of swing in the L data, exist between the two from the data, the two-band model provides a better fit
materials and possible causes for them are discussed in the next (compared to the single-band models) to the S versus nH and
section. Estimation of Srms
error values for this material system is m*D versus nH variations. In Figure 5d, the S2σ data are shown
given in Table 2, which indicates the SPB model has a better fit as a function of nH. Theoretical single-band fits using different
to the data compared to the SKB model. average m*D values highlight the difference in the predicted
ZnSb (p-Type). ZnSb is a well-known TE material, and its optimum S2σ and doping level while working with data at
electrical properties are generally modeled assuming a single different nH ranges. The estimation of Srms error values for this
parabolic band (band 1 in Figure 5a). The Pisarenko plot for material is provided in Table 2, which indicates the SPB model
this material is shown in Figure 5b. As seen from the figure, has a better fit to the data compared to that of the SKB model.
single-band models (both SPB and SKB) do not fit the PbTe (n-Type). Lastly, we have analyzed PbTe. n-type
experimental data over the entire nH range. This is because of PbTe is taken to be the model example of materials in which
the increase of the mD* values with increasing doping level the sole conduction band is nonparabolic and therefore the
(shown in Figure 5c). This has been attributed to the effect of applicability of the SKB model4(schematically represented in
2207 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Figure 6a). Here both SKB and SPB have been applied to the modeled using single-band models. Systematic deviations from
experimental data, and the results were surprising. Throughout the theoretical models are observed in the Pisarenko plot and
this work, a parameter called Srms error (described in the preceding mD* dependence on nH when plotted over a wide carrier
section) has been used to compare SPB and SKB models. A concentration range. Both of these observations are related to
lower value of Srmserror for a model is indicative of that model changes in the interband charge carrier distribution and are
giving a better fit. In Figure 6b, it is difficult to visually indicative of multiband conduction. The extent of the observed
compare SKB and SPB. When the entire range of charge carrier deviation in the Pisarenko plot is also an indicator (of
concentration is taken into consideration, it is found that SPB multiband conduction) because the difference between the
25 −3
gives a better fit (lower Srms
error than SKB). But for nH < 10 m , SPB and SKB lines as observed from our analysis is minimal.
it is found that SKB gives a better fit. To eliminate the errors The only scenario when multiband conduction can be modeled
arising from taking the experimental data with different by single-band models is for multiple degenerate bands with
dopants, test equipment, and synthesis methods, data from a similar effective band masses as observed in Mg2Si0.3Sn0.7.
single research work6 were taken separately and analyzed as From our analysis, the errors associated with modeling a
shown in Figure 7. For the entire range of charge carrier multiband material with single-band models are in the
estimated (a) Lorenz number (which in turn affects the
estimated κL value), (b) optimal doping concentration, and (c)
predictions of maximum S2σ (also zT).
In this section, we aim to provide further insight into these
systematic deviations and errors on the basis of the study of a
two-parabolic-band model. Two possible ratios of the band
effective mass are studied; (i) both the DOS band masses (m1
and m2) are the same and equal to the electron rest mass, and
(ii) a large difference in the DOS band masses with m1 = 0.7me
and m2 = 4.8me (values taken for that of CoSb3). In both of
these cases calculations were performed for interband
separations of 0 and 0.1 eV, i.e., for conditions of degenerate
and near-degenerate bands.
Interpretation of mD* Plot. In Figure 8, the variation of
m*D with n for the two different cases are shown. The m*D values

Figure 7. Experimental and theoretically generated Pisarenko plot for


iodine-doped PbTe, with data taken from ref 6. The Srms error values
calculated in different nH ranges are provided.

concentration, SKB gave a lower value of Srms error, but for the
range above 1025 m−3, SPB Srms error was lower. In any case, for
higher charge carrier concentrations, SPB gives a better fit,
suggesting that on moving away from the band minima the
band becomes parabolic. This is also evident from the m*D
versus nH plot (Figure 6c), which shows a higher deviation of
the effective mass from the average value at higher charge
carrier concentration, indicating a failure of the SKB at higher
doping levels. These findings certainly raise a question
regarding the absolute validity of the Kane band model over
the entire range of charge carrier concentration in PbTe. The
effective mass and Lorenz number calculated from these two
different models are different (Figure 6c,e). Thus, by using the
wrong model, it is possible to end up with wrong values for the
optimum charge carrier concentration and lattice thermal Figure 8. Variation in the calculated mD* with n using the SPB model
conductivity. It is important to stress upon this gradual shift on a two-band material. The upper panel data are for equal band
from the Kane-type nonparabolic to the parabolic nature in the masses, while the lower panel data are that of different band masses.
conduction band because the optimum zT is achieved within a In both cases the constant value (solid line) represents the degenerate
narrow range of n. Thus, identifying the correct model in that scenario.
range (or coming up with a new model altogether) to get the
right concentration is important. p-type PbTe and other were obtained from the SPB model using the generated two-
materials with a known nonparabolic band edge could be band data as input. It is thus similar to studying the measured
studied in a similar way to get more insight into the nature of data of a two-band material using SPB model. In both cases,
the extent of this nonparabolicity. degenerate bands result in constant values of mD* given by mD* =

■ DISCUSSION
The materials analysis in the previous section highlights certain
(Nv1m3/21 + Nv2m3/2
2 )
2/3
(here the band degeneracies Nv1 and
Nv2 are assumed to be unity). For this situation, the SPB model
is valid since the criterion of constant mD* with changing doping
systematic deviations and errors when a multiband material is level (known as rigid band) is satisfied. This also results in an
2208 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

Figure 9. (a) Difference in Lorenz number calculated using the two-band and SPB models as a function of charge carrier concentration. The upper
panel data are for equal band masses, while the lower panel data are those of different band masses. In both cases the interband separation is taken
to be 0.1 eV. (b) Error in the estimated κL (using SPB generated L) plotted as a function of carrier concentration.

accurate prediction of S2σ and optimum n as shown in Figure LσT). In Figure 9b, the estimated errors in κL (ΔκL) if the SPB
2d,e. Thus, the SPB model can be used in multiband scenarios model is applied to a two-band material is given. The ΔκL
where the bands are degenerate and can be expected to values peak at intermediate doping levels (∼1026 m−3) which is
provide accurate information on L, κL, nopt, and zTopt. also where optimum zT values lie (shown in Figure 10). The
In case of near-degenerate bands (in this example 0.1 eV),
the observed mD* variation with n shows a sharp change at
intermediate n values, which plateaus at higher and lower
carrier concentrations. The near-constant m*D values at lower n
are indicative of charge transport dominated by the lower
band. As the doping level increases, there is an increase in
charge transfer to the upper band resulting in a rising mD* value.
At very high doping levels (when all electrons participating in
charge transport can access the upper band) the m*D value
again becomes constant because there is no net change in the
charge distribution with increasing n value. Thus, modeling any
two (or multi)band material using the SPB model would
exhibit this feature if there are multiple doped samples spread
over a wide range of n. In such cases, the SPB model can be
used as an indicator for the presence of multiple bands near the
Fermi level. Even when the two masses are equal, the observed
rise in m*D is ∼50% (and becomes higher if the band masses are
unequal) and hence can be easily observed from experiments.
Errors in Lorenz Number and Its Effect on Lattice
Thermal Conductivity. Lorenz number as a function of
charge carrier concentration was calculated using both the two-
band and SPB models (using the two-band S data as input). In
Figure 9a, the difference between them is plotted as a function
of n. At low doping levels L values from the two-band model
are higher compared to the SPB values. With a rise in doping
level, both the two-band and SPB L initially increase which is
followed by a sharp decrease of the two-band L value until it Figure 10. Variation of zT with n as calculated by the SPB and two-
merges with the SPB line. The downward swing in the two- band models. The upper panel represents the data when the band
band L value is seen to occur at n values where interband mass ratio is unity. In the lower panel the data for the different band
charge transfer is prominent (see Figure 8). masses are shown. Here the SPB calculations were carried out using
The reason for this difference in the L values (calculated three different sets of m*D and μ0 values.
using the two models) is the additional interband charge
transfer component of the electronic thermal conductivity (κine ) estimated peak errors in ΔκL also depend on the band mass
which is accounted for in the two-band model. The κein ratios with deviation from unity leading to larger errors. As
contribution is higher at low doping levels and is expected to seen from our example it can be as high as 15% though further
fall sharply as the upper band gets populated. Thus, using the studies (with varying band mass ratio and interband
SPB model is expected to result in erroneous L values in separation) are necessary to check the possible upper limits.
materials with multiple near-degenerate bands. This in turn The ΔκL values are smaller at low doping levels (even though
influences the calculated lattice thermal conductivity (κL = κ − the difference in L is large) because of the low σ values, while
2209 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

at higher n this is due to the similar L values predicted by both


models. Overall, this study indicates that κL values are
overestimated when using the SPB model on a multiband
near-degenerate system.
Errors in Optimal Charge Carrier Concentration and
zT. The effect of modeling a two-band material with near-
degenerate bands using SPB results in errors in mD*, L, and κL
as discussed previously. The cumulative effect of this can be
seen in the difference in the calculated zT values from the two
models. For the SPB calculations, data generated from the two-
band model was used to derive average m*D and μ0 values using
the least-squares minimization technique. In the case of
identical mass of the two bands (upper panel of Figure 10) the
SPB model predicts a higher value of the zTmax value and a
lower value of the optimum charge carrier concentration. The
differences between the SPB and two-band zTmax become more Figure 11. Pisarenko plots for Mg2Si0.3Sn0.7 with theoretical fits
pronounced when the mass of the two bands are different as generated using different values of λ. The insets show the differences
seen from the lower panel of Figure 10. The origin of this between theoretical fits. Data are taken from refs 11−20.
difference lie in the charge carrier dependence of m*D in two-
band systems. Thus, working with a fixed mD* for zT calculation Table 3. Srms
error Obtained for Various λ Values in Mg2Si0.3Sn0.7
(in th SPB model) results in errors depending on the doping λ mD* (/me) Srms
concentration of the specimen used for m*D extraction. As error

0 (acoustic phonons) 2.68 11.99


shown in the lower panel of Figure 10, if m*D and μ0 values are
0.5 (neutral impurity) 1.79 11.87
extracted from the averaging over the entire n-range, the
1 (polar optical phonons) 1.37 12.00
obtained zTmax (dotted line) is significantly lower than the
2 (ionized impurity) 0.96 12.72
two-band zTmax value. Similarly, using the doping level which
results in the optimum zT from the two-band model results in
a much higher predicted zTmax (dashed−dotted line). A further the basis of experimental data) further support our result that it
zT plot (dashed line) is shown in the figure which represents is possible for this material to have λ = 0.5. A very important
the effect of data extraction from a low-doped sample (n = 1024 point to note is that even though the differences in Srms error are
m−3). These results highlight one of the biggest drawback of marginal, the mD* calculated from these values are very different
using the SPB model in multiband systems. The general from each other. Thus, identifying the correct mechanism is of
practice while using the SPB model has been to use a single utmost importance for calculation of material properties.
composition to extract experimental data for the model. While It is thus evident that single-band models have limitations
this works perfectly in materials with a single band contributing when applied to materials with complex band structures.
to charge transport or, as shown in this work, in materials with However, they can serve as an indicator of the presence of
multiple degenerate bands, it would result in erroneous multiple bands on the basis of the charge carrier variation of
predictions in multiband systems with near-degenerate bands. the calculated properties. Thus, unlike common practice, it is
Identification of Scattering Mechanism. As mentioned advisable to have multiple compositions spread over a few
earlier, a parameter called Srms
error has been used as a statistical orders of magnitude to check for the validity of a single-band
tool to determine how suitable a model is for the experimental approximation. Another observation from this work is the
data. This approach was used to identify the extent of marginal difference between the SPB and SKB Pisarenko plots
applicability of SPB and SKB to a given material. But it can in all of the studied materials. For all of the cases, the
also be used within SPB to identify the scattering mechanism difference is less than the scatter in the experimental data,
which gives the best fit to the experimental data since the SPB making it visually impossible to identify the correct model. We
equations employ the scattering parameter λ. The idea is that have shown that statistical error analysis might be a useful tool
the theoretical Pisarenko plot which results in lowest Srms error is to identify the better model, but further work is necessary for a
the most suitable and so is the corresponding λ. Generalized confirmation. Thus, elimination of other possible sources of
SPB equations with λ as an additional variable were used. For scatter in experimental data (dopant material, synthesis
different values of λ, the theoretical Pisarenko plots were process, and measurement setup) is recommended to check
generated (shown in Figure 11) and Srms error was evaluated. Here, for band parabolicity when studying unknown material
Mg2Si0.3Sn0.7 experimental data were used because, among all systems. Choosing the correct model (parabolic or non-
of the materials surveyed, SPB fits best to this material (Srms error parabolic) becomes relevant due to the significant difference in
data provided in Table 3). It was found that, for λ = 0.5, Srms error the estimated properties (e.g., mD*, L, S2σ, and optimum n)
was the lowest (although marginally). Since the same data are between the two models. The assumption of acoustic phonon
used as the input to SPB with different λ values, it is argued scattering being the dominant mechanism is also something
here that the marginal differences in Srms error should be attributed that is used as a default in SPB modeling. Our results show the
to the accuracy of the model in describing the behavior of the possibility of identifying the suitable λ values on the basis of an
material rather than to experimental errors. This result error analysis of the Pisarenko plot. Further studies are
indicates that neutral scattering mechanism could actually be required on this due to the significant scatter observed in the
the dominant mechanism of electron scattering at room experimental data.
temperature or there could be a mixing of different Overall, our work highlights various shortcomings of single-
mechanisms. Findings of Kim et al.77 for Mg2Si0.1Sn0.9 (on band models and indicates the need for multiband modeling
2210 DOI: 10.1021/acsaem.9b02015
ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

while addressing thermoelectric materials with complex band Notes


structures. Another issue that is highlighted is the significant The authors declare no competing financial interest.


scatter in the reported data which limits the predictive accuracy
of any such band models. Better quality experimental data and ACKNOWLEDGMENTS
thus better measurement practices are therefore required while
working with band structure modeling of TE materials.78 We acknowledge the financial support provided by the


Department of Science and Technology (DST), Government
CONCLUSION of India (Grant SERB/F/I0565/2017-2018). We also acknowl-
We have looked at the applicability of single-band models to edge Dr. J. de Boor (DLR) for providing valuable suggestions
various thermoelectric materials with complex band structures. and discussions.
The studied material systems were chosen to represent
different possible features of the band edge, namely, a single
parabolic band, two degenerate parabolic bands, two near-
■ ABBREVIATIONS
rms
Serror ,
root-mean-square of the error of the Seebeck
degenerate parabolic bands, and a single nonparabolic band. coefficient data
Both SPB and SKB models were used to analyze the data using SKB, single Kane band
standard procedures. To interpret the observed deviations SPB, single parabolic band


from the single-band models, fitting was carried out using two-
parabolic-band equations. SPB works well for materials with
REFERENCES
multiple degenerate bands (especially if the band masses are
similar). In the case of near-degenerate bands, application of (1) Snyder, G. J.; Toberer, E. S. Complex Thermoelectric Materials.
the SPB model results in errors in the calculated Lorenz In Materials for Sustainable Energy; Dusastre, V., Ed.; World Scientific:
number, lattice thermal conductivity, the predicted optimum London, U.K., 2010; p 101, DOI: 10.1142/9789814317665_0016.
charge carrier concentration, and zT values. The magnitude of (2) Champier, D. Thermoelectric generators: A review of
applications. Energy Convers. Manage. 2017, 140, 167−181.
the error depends on the ratio of the band masses and the (3) May, A. F.; Snyder, G. J. Introduction to Modeling Thermo-
interband separation. Differences between SPB and SKB electric Transport at High Temperatures. In Materials, Preparation,
Pisarenko plots are also found to be marginal, making it and Characterization in Thermoelectrics; Rowe, D. M., Ed.; CRC Press:
difficult to predict the nature of the band edge on the basis of Boca Raton, FL, USA, 2012; p 11.
these models. We show the possibility of using statistical error (4) Ravich, Yu. I.; Efimova, B. A.; Smirnov, I. A. In Semiconducting
analysis in such instances for identifying the best fit. Lead Chalcogenides; Stil’bans, L. S., Ed.; Plenum Press: New York,
Interestingly, applying this analysis to PbTe which is known 1970.
to have a nonparabolic band edge indicates a transition from a (5) Fistul, V. I. In Heavily Doped Semiconductors; Tybulewicz, A.,
Kane-type to a parabolic one on increasing the doping level. Ed.; Plenum Press: New York, 1969.
This approach can be used to identify the scattering (6) Pei, Y.; LaLonde, A. D.; Wang, H.; Snyder, G. J. Low effective
mechanism in materials and was attempted in Mg2Si0.3Sn0.7. mass leading to high thermoelectric performance. Energy Environ. Sci.
2012, 5, 7963−7969.
The best fit was obtained for a λ value of 0.5. While further (7) Heremans, J. P.; Wiendlocha, B.; Chamoire, A. M. Resonant
studies are necessary to verify its applicability, this poses the levels in bulk thermoelectric semiconductors. Energy Environ. Sci.
question whether acoustic scattering of charge carriers can be 2012, 5, 5510−5530.
considered as the default mechanism while using these models. (8) Eaton, J. W.; Bateman, D.; Hauberg, S.; Wehbring, R. GNU
Overall, our study indicates the need for multiband modeling Octave version 5.1.0 manual: a high-level interactive language for
while dealing with complex band structures and the single-and numerical computations; GNU, 2019.
models remain effective tools while working with multiple (9) Putley, E. H. The Hall Effect and Semi-Conductor Physics; Dover:
degenerate bands at best. New York, 1968.


(10) Vining, C. B. A model for the high-temperature transport
ASSOCIATED CONTENT properties of heavily doped n-type silicon-germanium alloys. J. Appl.
Phys. 1991, 69, 331.
*
S Supporting Information
(11) Liu, W.; Tan, X.; Yin, K.; Liu, H.; Tang, X.; Shi, J.; Zhang, Q.;
The Supporting Information is available free of charge at Uher, C. Convergence of Conduction Bands as a Means of Enhancing
https://pubs.acs.org/doi/10.1021/acsaem.9b02015. Thermoelectric Performance of n-Type Mg2Si1‑xSnx Solid Solutions.
Initial steps of single band model solutions; experimental Phys. Rev. Lett. 2012, 108, 166601.
(12) You, S. W.; Kim, I. H.; Choi, S. M.; Seo, W. S. Thermoelectric
data used for analysis; various material parameters used
Properties of Bi-doped Mg2Si1−xSnx Prepared by Mechanical Alloying.
for the models; procedure for calculation of mD*av and μav
0 J. Korean Phys. Soc. 2013, 63, 2153−2157.
and their values; comparison of two cases of degenerate (13) Liu, W.; Chi, H.; Sun, H.; Zhang, Q.; Yin, K.; Tang, X.; Zhang,
bands (PDF) Q.; Uher, C. Advanced thermoelectrics governed by a single parabolic


band: Mg2Si0.3Sn0.7, a canonical example. Phys. Chem. Chem. Phys.
AUTHOR INFORMATION 2014, 16, 6893−6897.
(14) Zhang, Q.; Liu, W.; Liu, C.; Yin, K.; Tang, X. F. Thermoelectric
Corresponding Author Properties of Mg2(Si0.3Sn0.7)1‑ySby Solid Solutions Doped with Co as
*E-mail: titas.dasgupta@iitb.ac.in. CoSi Secondary Phase. J. Electron. Mater. 2014, 43, 2188−2195.
ORCID (15) Gao, H.; Zhu, T.; Zhao, X.; Deng, Y. Variations of
Harshita Naithani: 0000-0001-9138-8431 thermoelectric properties of Mg2.2Si1‑ySny−0.013Sb0.013 materials with
different Si/Sn ratios. J. Solid State Chem. 2014, 220, 157−162.
Titas Dasgupta: 0000-0002-8261-9784 (16) Zhang, Q.; Zheng, Y.; Su, X.; Yin, K.; Tang, X.; Uher, C.
Author Contributions Enhanced power factor of Mg2Si0.3Sn0.7 synthesized by a non-

H.N. and T.D. contributed equally to this work. equilibrium rapid solidification method. Scr. Mater. 2015, 96, 1−4.

2211 DOI: 10.1021/acsaem.9b02015


ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

(17) Yin, K.; Zhang, Q.; Zheng, Y.; Su, X.; Tang, X.; Uher, C. (36) Lamberton, G. A.; Bhattacharya, S.; Littleton, R. T.; Kaeser, M.
Thermal stability of Mg2Si0.3Sn0.7 under different heat treatment A.; Tedstrom, R. H.; Tritt, T. M.; Yang, J.; Nolas, G. S. High figure of
conditions. J. Mater. Chem. C 2015, 3, 10381−10387. merit in Eu-filled CoSb3-based skutterudites. Appl. Phys. Lett. 2002,
(18) Yin, K.; Su, X.; Yan, Y.; Uher, C.; Tang, X. In situ nanostructure 80, 598−600.
design leading to a high figure of merit in an eco-friendly stable (37) Puyet, M.; Dauscher, A.; Lenoir, B.; Dehmas, M.; Stiewe, C.;
Mg2Si0.30Sn0.70 solid solution. RSC Adv. 2016, 6, 16824−16831. Müller, E.; Hejtmanek, J. Beneficial effect of Ni substitution on the
(19) Goyal, G. K.; Mukherjee, S.; Mallik, R. C.; Vitta, S.; Samajdar, thermoelectric properties in partially filled CayCo4−xNixSb12 skutter-
I.; Dasgupta, T. High Thermoelectric Performance in Mg2(Si0.3Sn0.7) udites. J. Appl. Phys. 2005, 97, 083712.
by Enhanced Phonon Scattering. ACS Appl. Energy Mater. 2019, 2, (38) Li, X. Y.; Chen, L. D.; Fan, J. F.; Zhang, W. B.; Kawahara, T.;
2129−2137. Hirai, T. Thermoelectric properties of Te-doped CoSb3 by spark
(20) Zhang, Q.; Su, X.; Yan, Y.; Zheng, Y.; Liu, W.; Tang, X. plasma sintering. J. Appl. Phys. 2005, 98, 083702.
Ultrafast and low-cost preparation of Mg2(Si0.3Sn0.7)1−ySby with (39) Zhao, X. Y.; Shi, X.; Chen, L. D.; Zhang, W. Q.; Zhang, W. B.;
superior thermoelectric performance by self-propagating high-temper- Pei, Y. Z. Synthesis and thermoelectric properties of Sr-filled
ature synthesis. Scr. Mater. 2019, 162, 507−511. skutterudite SryCo4Sb12. J. Appl. Phys. 2006, 99, 053711.
(21) Bahk, J. H.; Bian, Z.; Shakouri, A. Electron transport modeling (40) Puyet, M.; Dauscher, A.; Lenoir, B.; Bellouard, C.; Stiewe, C.;
and energy filtering for efficient thermoelectric Mg2Si1‑xSnx solid Müller, E.; Hejtmanek, J.; Tobola, J. Influence of Ni on the
solutions. Phys. Rev. B: Condens. Matter Mater. Phys. 2014, 89, 075204. thermoelectric properties of the partially filled calcium skutterudites
(22) Bux, S. K.; Yeung, M. T.; Toberer, E. S.; Snyder, G. J.; Kaner, R. CayCo4−xNixSb12. Phys. Rev. B: Condens. Matter Mater. Phys. 2007, 75,
B.; Fleurial, J. P. Mechanochemical synthesis and thermoelectric 245110.
properties of high quality magnesium silicide. J. Mater. Chem. 2011, (41) Shi, X.; Kong, H.; Li, C. P.; Uher, C.; Yang, J.; Salvador, J. R.;
21, 12259−12266. Wang, H.; Chen, L.; Zhang, W. Low thermal conductivity and high
(23) Zaitsev, V. K.; Fedorov, M. I.; Burkov, A. T.; Gurieva, E. A.; thermoelectric figure of merit in n-type BaxYbyCo4Sb12 double-filled
Eremin, I. S.; Konstantinov, P. P.; Ordin, S. V.; Sano, S.; Vedernikov, skutterudites. Appl. Phys. Lett. 2008, 92, 182101.
M. V. Some features of the conduction band structure, transport and (42) Anno, H.; Matsubara, K.; Notohara, Y.; Sakakibara, T.; Tashiro,
optical properties of n-type Mg2Si-Mg2Sn alloys. 21st International H. Effects of doping on the transport properties of CoSb3. J. Appl.
Conference on Thermoelectronics 2002, 151−154. Phys. 1999, 86, 3780−3786.
(24) Zaitsev, V. K.; Fedorov, M. I.; Gurieva, E. A.; Eremin, I. S.; (43) Peng, J.; He, J.; Su, Z.; Alboni, P. N.; Zhu, S.; Tritt, T. M. High
Konstantinov, P. P.; Samunin, A.Yu.; Vedernikov, M. V. Thermo- temperature thermoelectric properties of double-filled InxYbyCo4Sb12
electrics of n-type with ZT > 1 Based on Mg2Si-Mg2Sn Solid skutterudites. J. Appl. Phys. 2009, 105, 084907.
(44) Sales, B. C.; Chakoumakos, B. C.; Mandrus, D. Thermoelectric
Solutions. Int. Conf. Thermoelectr. 2005, 189.
properties of thallium-filled skutterudites. Phys. Rev. B: Condens.
(25) Zhang, Q.; He, J.; Zhu, T. J.; Zhang, S. N.; Zhao, X. B.; Tritt, T.
Matter Mater. Phys. 2000, 61, 2475−2481.
M. High figures of merit and natural nanostructures in Mg2Si0.4Sn0.6
(45) Shi, X.; Yang, J.; Salvador, J. R.; Chi, M.; Cho, J. Y.; Wang, H.;
based thermoelectric materials. Appl. Phys. Lett. 2008, 93, 102109.
Bai, S.; Yang, J.; Zhang, W.; Chen, L. Multiple-Filled Skutterudites:
(26) Liu, W.; Tang, X.; Li, H.; Yin, K.; Sharp, J.; Zhou, X.; Uher, C.
High Thermoelectric Figure of Merit through Separately Optimizing
Enhanced thermoelectric properties of n-type Mg2.16(Si0.4Sn0.6)1‑ySby
Electrical and Thermal Transports. J. Am. Chem. Soc. 2011, 133,
due to nano-sized Sn-rich precipitates and an optimized electron
7837−7846.
concentration. J. Mater. Chem. 2012, 22, 13653−13661. (46) Bai, S. Q.; Huang, X. Y.; Chen, L. D.; Zhang, W.; Zhao, X. Y.;
(27) Liu, W.; Zhang, Q.; Yin, K.; Chi, H.; Zhou, X.; Tang, X.; Uher,
Zhou, Y. F. Thermoelectric properties of n-type SrxMyCo4Sb12 (M =
C. High figure of merit and thermoelectric properties of Bi-doped Yb, Ba) double-filled skutterudites. Appl. Phys. A: Mater. Sci. Process.
Mg2Si0.4Sn0.6 solid solutions. J. Solid State Chem. 2013, 203, 333−339. 2010, 100, 1109−1114.
(28) Gao, P.; Berkun, I.; Schmidt, R. D.; Luzenski, M. F.; Lu, X.; (47) Zhang, J.; Xu, B.; Yu, F.; Yu, D.; Liu, Z.; He, J.; Tian, Y.
Bordon Sarac, P.; Case, E. D.; Hogan, T. P. Transport and Mechanical Thermoelectric properties of n-type CoSb3 fabricated with high
Properties of High-ZT Mg2.08Si0.4‑xSn0.6Sbx Thermoelectric Materials. pressure sintering. J. Alloys Compd. 2010, 503, 490−493.
J. Electron. Mater. 2014, 43, 1790−1803. (48) Ballikaya, S.; Uzar, N.; Yildirim, S.; Salvador, J. R.; Uher, C.
(29) Gao, P.; Lu, X.; Berkun, I.; Schmidt, R. D.; Case, E. D.; Hogan, High thermoelectric performance of In,Yb,Ce multiple filled CoSb3
T. P. Reduced lattice thermal conductivity in Bi-doped Mg2Si0.4Sn0.6. based skutterudite compounds. J. Solid State Chem. 2012, 193, 31−35.
Appl. Phys. Lett. 2014, 105, 202104. (49) Qiu, P.; Shi, X.; Qiu, Y.; Huang, X.; Wan, S.; Zhang, W.; Chen,
(30) Jiang, G.; He, J.; Zhu, T.; Fu, C.; Liu, X.; Hu, L.; Zhao, X. High L.; Yang, J. Enhancement of thermoelectric performance in slightly
Performance Mg2(Si,Sn) Solid Solutions: a Point Defect Chemistry charge-compensated CeyCo4Sb12 skutterudites. Appl. Phys. Lett. 2013,
Approach to Enhancing Thermoelectric Properties. Adv. Funct. Mater. 103, 062103.
2014, 24, 3776−3781. (50) Salvador, J. R.; Waldo, R. A.; Wong, C. A.; Tessema, M.;
(31) Zhang, L.; Xiao, P.; Shi, L.; Henkelman, G.; Goodenough, J. B.; Brown, D. N.; Miller, D. J.; Wang, H.; Wereszczak, A. A.; Cai, W.
Zhou, J. Localized Mg-vacancy states in the thermoelectric material Thermoelectric and mechanical properties of melt spun and spark
Mg2−δSi0.4Sn0.6. J. Appl. Phys. 2016, 119, 085104. plasma sintered n-type Yb- and Ba-filled skutterudites. Mater. Sci. Eng.,
(32) Tang, Y.; Gibbs, Z. M.; Agapito, L. A.; Li, G.; Kim, H. S.; B 2013, 178, 1087−1096.
Nardelli, M. B.; Curtarolo, S.; Snyder, G. J. Convergence of multi- (51) Park, K. H.; Seo, W. S.; Shin, D. K.; Kim, I. H. Thermoelectric
valley bands as the electronic origin of high thermoelectric Properties of Yb-filled CoSb3 Skutterudites. J. Korean Phys. Soc. 2014,
performance in CoSb3 skutterudites. Nat. Mater. 2015, 14, 1223− 65, 491−495.
1229. (52) Wang, S.; Yang, J.; Wu, L.; Wei, P.; Zhang, W.; Yang, J. On
(33) Morelli, D. T.; Meisner, G. P.; Chen, B.; Hu, S.; Uher, C. Intensifying Carrier Impurity Scattering to Enhance Thermoelectric
Cerium filling and doping of cobalt triantimonide. Phys. Rev. B: Performance in Cr-Doped CeyCo4Sb12. Adv. Funct. Mater. 2015, 25,
Condens. Matter Mater. Phys. 1997, 56, 7376−7383. 6660−6670.
(34) Nolas, G. S.; Cohn, J. L.; Slack, G. A. Effect of partial void (53) Dong, J.; Yang, K.; Xu, B.; Zhang, L.; Zhang, Q.; Tian, Y.
filling on the lattice thermal conductivity of skutterudites. Phys. Rev. B: Structure and thermoelectric properties of Se- and Se/Te-doped
Condens. Matter Mater. Phys. 1998, 58, 164−170. CoSb3 skutterudites synthesized by high-pressure technique. J. Alloys
(35) Dyck, J. S.; Chen, W.; Uher, C.; Chen, L.; Tang, X.; Hirai, T. Compd. 2015, 647, 295−302.
Thermoelectric properties of the n-type filled skutterudite (54) Li, X.; Zhang, Q.; Kang, Y.; Chen, C.; Zhang, L.; Yu, D.; Tian,
Ba0.3Co4Sb12 doped with Ni. J. Appl. Phys. 2002, 91, 3698−3705. Y.; Xu, B. High pressure synthesized Ca-filled CoSb3 skutterudites

2212 DOI: 10.1021/acsaem.9b02015


ACS Appl. Energy Mater. 2020, 3, 2200−2213
ACS Applied Energy Materials Review

with enhanced thermoelectric properties. J. Alloys Compd. 2016, 677, (76) Yang, L.; Chen, Z. G.; Hong, M.; Wang, L.; Kong, D.; Huang,
61−65. L.; Han, G.; Zou, Y.; Dargusch, M.; Zou, J. n-type Bi-doped PbTe
(55) Chen, C.; Zhang, L.; Li, J.; Yu, F.; Yu, D.; Tian, Y.; Xu, B. Nanocubes with Enhanced Thermoelectric Performance. Nano Energy
Enhanced thermoelectric performance of lanthanum filled CoSb3 2017, 31, 105−112.
synthesized under high pressure. J. Alloys Compd. 2017, 699, 751− (77) Kim, S.; Wiendlocha, B.; Jin, H.; Tobola, J.; Heremans, J. P.
755. Electronic structure and thermoelectric properties of p-type Ag-doped
(56) Boettger, P. H. M.; Pomrehn, G. S.; Snyder, G. J.; Finstad, T. G. Mg2Sn and Mg2Sn1‑xSix (x = 0.05, 0.1). J. Appl. Phys. 2014, 116,
Doping of p-type ZnSb: Single parabolic band model and impurity 153706.
band conduction. Phys. Status Solidi A 2011, 208, 2753−2759. (78) Borup, K. A.; de Boor, J.; Wang, H.; Drymiotis, F.; Gascoin, F.;
(57) Valset, K.; Böttger, P. H. M.; Taftø, J.; Finstad, T. G. Shi, X.; Chen, L.; Fedorov, M. I.; Mueller, E.; Iversen, B. B.; Snyder,
Thermoelectric properties of Cu doped ZnSb containing Zn3P2 G. J. Measuring thermoelectric transport properties of materials.
Energy Environ. Sci. 2015, 8, 423−435.


particles. J. Appl. Phys. 2012, 111, 023703.
(58) Xiong, D. B.; Okamoto, N. L.; Inui, H. Enhanced thermo-
electric figure of merit in p-type Ag-doped ZnSb nanostructured with NOTE ADDED AFTER ASAP PUBLICATION
Ag3Sb. Scr. Mater. 2013, 69, 397−400. This paper was published ASAP on December 9, 2019, under
(59) Prokofieva, L. V.; Konstantinov, P. P.; Shabaldin, A. A.; the incorrect manuscript type. It was changed from a Forum
Pshenai-Severin, D. A.; Burkov, A. T.; Fedorov, M. I. Doping and Article to a Review and reposted ASAP on February 24, 2020.
Defect Formation in Thermoelectric ZnSb Doped with Copper.
Semiconductors 2014, 48, 1571−1580.
(60) Shabaldin, A. A.; Prokof'eva, L. V.; Konstantinov, P. P.; Burkov,
A. T.; Fedorov, M. I. Acceptor impurity of copper in ZnSb
thermoelectric. Mater. Today: Proc. 2015, 2, 699−704.
(61) Valset, K.; Song, X.; Finstad, T. G. A study of transport
properties in Cu and P doped ZnSb. J. Appl. Phys. 2015, 117, 045709.
(62) Guo, Q.; Luo, S. Improved thermoelectric efficiency in p-type
ZnSb through Zn deficiency. Funct. Mater. Lett. 2015, 8, 1550028.
(63) Shabaldin, A. A.; Prokof'eva, L. V.; Snyder, G. J.; Konstantinov,
P. P.; Isachenko, G. N.; Asach, A. V. The Influence of Weak Tin
Doping on the Thermoelectric Properties of Zinc Antimonide. J.
Electron. Mater. 2016, 45, 1871−1874.
(64) Pothin, R.; Ayral, R. M.; Berche, A.; Granier, D.; Rouessac, F.;
Jund, P. Preparation and properties of ZnSb thermoelectric material
through mechanical-alloying and Spark Plasma Sintering. Chem. Eng.
J. 2016, 299, 126−134.
(65) Song, X.; Finstad, T. G. Review of Research on the
Thermoelectric Material ZnSb. In Thermoelectric Power Generation:
A Look at Trends in Technology; Skipidarov, S.; Nikitin, M.; Eds.;
2016; p 119, DOI: 10.5772/62753.
(66) Zhao, G. L.; Gao, F.; Bagayoko, D. Reliable density functional
calculations for the electronic structure of thermoelectric material
ZnSb. AIP Adv. 2018, 8, 105211.
(67) Pei, Y.; May, A. F.; Snyder, G. J. Self-Tuning the Carrier
Concentration of PbTe/Ag 2 Te Composites with Excess Ag for High
Thermoelectric Performance. Adv. Energy Mater. 2011, 1, 291−296.
(68) LaLonde, A. D.; Pei, Y.; Snyder, G. J. Reevaluation of PbTe1‑xIx
as high performance n-type thermoelectric material. Energy Environ.
Sci. 2011, 4, 2090−2096.
(69) Pei, Y.; May, A. F.; Snyder, G. J. Self-Tuning the Carrier
Concentration of PbTe/Ag 2 Te Composites with Excess Ag for High
Thermoelectric Performance. Adv. Energy Mater. 2011, 1, 291−296.
(70) Pei, Y.; Gibbs, Z. M.; Gloskovskii, A.; Balke, B.; Zeier, W. G.;
Snyder, G. J. Optimum Carrier Concentration in n-Type PbTe
Thermoelectrics. Adv. Energy Mater. 2014, 4, 1400486.
(71) Orihashi, M.; Noda, Y.; Chen, L.; Hirai, T. Carrier
Concentration Dependence of Thermal Conductivity of Iodine-
Doped n-Type PbTe. Mater. Trans., JIM 2000, 41, 1282−1286.
(72) Kishimoto, K.; Koyanagi, T. Preparation of sintered degenerate
n-type PbTe with a small grain size and its thermoelectric properties.
J. Appl. Phys. 2002, 92, 2544−2549.
(73) Jaworski, C. M.; Heremans, J. P. Thermoelectric transport
properties of the n-type impurity Al in PbTe. Phys. Rev. B: Condens.
Matter Mater. Phys. 2012, 85, 033204.
(74) Nielsen, M. D.; Levin, E. M.; Jaworski, C. M.; Schmidt-Rohr,
K.; Heremans, J. P. Chromium as resonant donor impurity in PbTe.
Phys. Rev. B: Condens. Matter Mater. Phys. 2012, 85, 045210.
(75) Cohen, I.; Kaller, M.; Komisarchik, G.; Fuks, D.; Gelbstein, Y.
Enhancement of the thermoelectric properties of n-type PbTe by Na
and Cl co-doping. J. Mater. Chem. C 2015, 3, 9559−9564.

2213 DOI: 10.1021/acsaem.9b02015


ACS Appl. Energy Mater. 2020, 3, 2200−2213

You might also like