Download as pdf or txt
Download as pdf or txt
You are on page 1of 33

Version of Record: https://www.sciencedirect.

com/science/article/pii/S246860691730045X
Manuscript_e4418f1b8f297c16f52986197bd0e386

Recent progress in p-type thermoelectric magnesium silicide based solid solutions

J. de Boor1+, T. Dasgupta2, U. Saparamadu3, E. Müller1,4, Zhifeng Ren3

1: Institute of Materials Research, German Aerospace Center (DLR),D- 51147 Koeln, Germany

2: Department of Metallurgical Engineering and Materials Science Indian, Institute of


Technology Bombay Mumbai, Maharashtra 400076, India

3: Department of Physics and Texas Center for Superconductivity, University of Houston,


Houston, TX 77204, USA

4: Institute of Inorganic and Analytical Chemistry, Justus Liebig University Giessen,D- 35392
Giessen, Germany

Thermoelectric materials can convert waste or process heat directly into usable electrical
energy. They are therefore a fascinating opportunity to increase the energy efficiency of
various industrial processes and thus reduce fossil fuel consumption. For the temperature
range of 500 K to 800 K – where a large fraction of the reusable heat is available –
magnesium silicide based solid solutions are among the most promising thermoelectric
materials. They combine very good thermoelectric properties with a high material
availability, low cost of raw materials and environmental compatibility making them suitable
for large scale applications. The thermoelectric properties of the n-type material have been
found superior to the p-type, however, for a practical application of magnesium silicide
based thermoelectric generators a p-type material with good thermoelectric properties is
indispensable. Due to progress in synthesis techniques and the identification of efficient
dopants there has been significant progress in the optimization of the p-type, resulting in
several reports with a thermoelectric figure of merit exceeding 0.5.

We analyze experimental data and theoretical results and rationalize the recent progress. In
particular we investigate the interplay between thermoelectric properties and composition,
electronic band structure, and the effect of doping by various elements. Overall we find that
the optimization of p-type Mg2X requires consideration or more parameters than the n-type
material. Achieving the optimum carrier concentration is experimentally challenging and
depends on the dopant species as well as the Si:Ge:Sn ratio in the compounds. Although
debated frequently, the overwhelming majority of the data indicates a rigid electronic band
structure for the most popular dopants Ag, Si, Li, and Na. On the other hand, the doping
efficiency differs between the dopants as well as the corresponding hole mobilities. Further
progress in the material development can be expected if the various defect densities can be
controlled and the bipolar effect can be decreased further.

1
© 2017 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
1. Introduction

1.1 Magnesium silicide as high potential thermoelectric material

Thermoelectricity deals with the direct conversion of electrical energy into heat and vice versa.
Thermoelectric devices can use electrical power for heating and cooling and are therefore used for
temperature regulation, e.g., in car seats, semiconductor lasers, and the biochemical industry [1, 2].
On the other hand they can also make use of waste heat or process heat immanently to system
operation flowing from a hot source to a cold reservoir as, e.g., in heating systems of buildings, and
convert it into usable electrical energy. As waste heat is ubiquitous and not exploited efficiently in a
multitude of energy intensive processes, the potential for thermoelectric waste heat recovery is
tremendous and includes applications like the exhaust heat from cars and process heat of steel
casting and forming processes [3, 4]. A large fraction of the potentially available waste heat can be
found in the mid-temperature region between 500 K and 800 K, which is too hot for the
commercially available thermoelectric generators based on (Bi,Sb)2(Te,Se)3 materials. Heat to
electricity conversion with a high efficiency requires materials with good thermoelectric properties,
i.e., with a large thermoelectric figure of merit = , where is the electrical conductivity,
the Seebeck coefficient, the thermal conductivity, and the absolute temperature. The
development of efficient materials with high for this temperature regime has therefore been a
focus of the thermoelectric materials research in the last decades, resulting in attractive
thermoelectric properties for several material classes like nanostructured PbTe [5-7], CoSb3-based
Skutterudites [8-10], half-Heusler compounds [11-13], SnSe [14], Cu2Se-based materials [15, 16], and
Mg2Si-based solid solutions [17-24]. The constituent elements of Mg2Si are among the most
abundant in the earth’s crust: Si and Mg are the 2nd and 8th most abundant elements, respectively,
giving this class a crucial advantage in view of potential large scale applications and economical
attractiveness [25, 26]. A further material class-specific advantage is their low mass density , which
is by a factor of 2-3 lower than that of Skutterudites or PbTe, resulting in a higher specific figure of
merit / [27]. This weight advantage is higher for the binary Mg2Si or Si-rich compositions [28-30]
but gives magnesium silicides (of any composition) an advantage where weight is crucial, i.e., in
airborne or mobile applications.

Furthermore, magnesium silicides are a technologically relatively advanced material class where
some systematic contacting studies are reported [31-36] as well as fabrication and analysis of lab
prototypes for thermoelectric modules [37-42]; Mg2Si-based thermoelectric generators are currently
entering the status of industrial commercialization. The excellent thermoelectric properties of Mg2Si
based solid solutions are caused by a reasonably low thermal conductivity due to alloying with Sn
and Ge, a high carrier mobility and a high power factor due to a convergence of conduction
bands for compositions around and between Mg2Si0.35Sn0.65 and Mg2Ge0.25Sn0.75 [17, 18, 43].
However, attractive thermoelectric properties with 1.1 < < 1.5 have been reported only for
the n-type material, with the p-type being clearly inferior [44, 45]. Nevertheless, thermoelectric
generators require both n- and p-type materials to work efficiently, ideally with similar
thermochemical and thermomechanical properties. As good p-type Mg2X (X = Si, Ge, Sn) has not
been available for a long time, n-type Mg2X has often been combined with higher manganese silicide
(HMS) to build thermoelectric generators [46]. However, the thermomechanical properties such as
coefficient of thermal expansion matches relatively poorly between these materials leading to the
2
long-term stability as an issue of the generators [38, 41]. Consequently, an efficient p-type Mg2X is
highly sought.

Exemplary experimental data for n- and p-type Mg2X is compared in Figure 1a) and shows the
significantly lower over the whole temperature range.

Figure 1. a) The comparison of the experimental figure of merit for n- and p- type Mg2X [23, 47] shows inferior
properties of the p-type over the whole temperature range. b) Calculated vs. carrier concentration and ,
respectively, at 700 K [48]. While the agreement between theory and experiment is good for the n-type, there is further
optimization potential for the p-type.

On the other hand, theoretical calculations predict a ≈ 0.8 at 700 K for the p-type (Figure 1b),
worse than the n-type material due to the missing band convergence but significantly better than
what has been achieved experimentally. It can also be seen that the optimized properties for the p-
type are expected at a much higher charge carrier concentration level, which poses a significant
challenge in the experimental realization. This discrepancy between experimental and theoretical
results for the p-type material has stimulated experimental research in p-type Mg2Si yielding notable
improvements.

The review is dedicated to p-type Mg2X as the n-type has been thoroughly covered in the literature
[44, 45, 49, 50]. In this review we will summarize experimental and theoretical findings on p-type
Mg2X and try to rationalize the recent progress in the material development. We will specifically
address the differences between the possible dopants for Mg2X, the influence of defects on the
thermoelectric properties and the applicability of simple electronic band structure models. In the
outlook section we will summarize the main challenges for further material development and discuss
strategies for future research.

1.2 Physical properties of Mg2X

3
The iso-structural compounds Mg2X (X = Si, Ge, Sn) have the cubic anti-fluorite crystal structure
(Space Group: Fm3m) with the Mg atoms occupying the tetrahedral positions and the X atoms at the
corners and face centers, see Figure 2 [44]. Deviation from the 2:1 composition has been reported in
both theoretical [51, 52] and experimental publications in Mg2Si [53] with the presence of Mg at the
dodecahedral interstitial site, indicated as gridded sphere in Figure 2. The Mg2X compounds form
solid solutions amongst each other with varying extent of solubility. The Mg2Si-Mg2Sn system is the
most intensively studied due to high reported figures of merit (for n-doped material) and low
cost. In this system, a miscibility gap exists though the composition of the solid solution end-
members is debated [54, 55]. The lattice parameters in the solid solution vary according to Vegard’s
law with values of 6.354 Å (Mg2Si) and 6.764 Å (Mg2Sn) for the end members [56]. An additional
feature of these solid solutions is their low density that varies between 2 g/cm3 (Mg2Si) and 3.6 g/cm3
(Mg2Sn).

Figure 2. Crystal structure of Mg2X (X = Si, Ge, Sn). The center position is potentially occupied by interstitial atoms.
Reprinted with permission from [Liu, X., et al., Advanced Electronic Materials, 2, p. 1500284, 2016] Copyright (2017) by
Wiley" [52].

1.3 Electronic band structure


The Mg2X compounds are indirect band gap semiconductors with similar electronic band structure.
The band gap decreases with increased atomic weight of X from approximately 0.78 eV for Mg2Si to
0.35 eV for Mg2Sn [17, 57, 58]. The solid solutions show intermediate values for the band-gap and
are usually assumed to follow Vegard’s law [17], this is, however, not totally agreed upon [18]. In
Figure 3a) the electronic band structure of Mg2Si is shown [59].

4
Figure 3. (a) The electronic band structure of Mg2Si. (b-d) Band-structure of Mg2Si, Mg2Ge, and Mg2Sn showing the band
splitting at the valence band edge [59]. Reprinted with permission from [Kutorasinski, K., et al., Physical Review B, 89, p.
115205-1-8, 2014] Copyright (2017) by the American Physical Society."

The conduction band minima (CBM) are located at the X symmetry point in the 1st Brillouin zone
while the valence band maximum (VBM) is positioned at point in the zone center. Considering the
symmetry of the lattice, this results in valley degeneracy (NV) of 3 for the CBM and NV of 1 for the
VBM. The special feature in these compounds is the presence of two almost degenerate bands at the
conduction band extremum. The orbital contributions of these bands are different, thus mixing Mg2Si
with Mg2Sn or Mg2Ge with Mg2Sn can result in solid solutions with two degenerate bands at the
CBM. This feature is used to engineer compositions with enhanced power factor and results in high
in n-doped solid solutions [17-23, 43, 60]. Best values ( ≈ 1.4) in the Mg2Si-Mg2Sn solid
solutions are observed for compositions M2Si1-xSnx with 0.6 < " < 0.7, in the range of the
theoretically predicted band convergence [17, 18, 47]. For the Mg2Ge-Mg2Sn system optimum
properties are found for " ≈ 0.75 [43, 60, 61], while no band convergence has been observed for the
Mg2Si-Mg2Ge system due to the similarity of the band structure in both compounds.

The Mg2X compounds share similar VB edge features with two degenerate bands and two different
masses termed as the heavy hole and light hole bands at the VBM along with a third band with lower
energy (split-off band) (see Figure 3 (b-d)). This splitting is attributed to spin-orbital interactions and
an increase in band separation is observed with weight of the X element. The calculated values [59]
of the inter-band separation are 0.036 eV for Mg2Si, 0.2 eV for Mg2Ge and 0.525 eV for Mg2Sn and
are similar to those found experimentally [62]. The spin-orbit interaction also influences the band
mass with a decrease observed in Mg2Ge and Mg2Sn. Both these effects of reduced band mass and
splitting of the bands result in a lower value of density of states mass ($%∗ ) in Mg2Sn compared to
Mg2Si. The influence of solid solution formation on the band splitting of the VB has not been
reported and provides scope for further study.

5
2. P-type Doping of Mg2X

2.1 Overview

In this section we want to give an overview over the elements that have been successfully employed
to p-dope Mg2X. The optimization of the carrier concentration is a fundamental prerequisite for good
thermoelectric properties and one of the fundamental optimization parameters for any
thermoelectric material [3]. For p-type Mg2X the issue is of particular importance: first, the
experimental realization of samples with sufficient number of charge carriers has been proven to be
difficult, often leading to samples with too low ' and hence unoptimized thermoelectric properties
[63-67]. Secondly, the choice of the dopant might influence the material properties more than for
the n-type material. For the n-type material the most popular dopants Sb and Bi seem to obey the
rigid band picture [68], i.e., addition of dopants shifts the chemical potential of the electrons and
enhances the number of charge carriers but the band structure itself remains unaffected.
Disregarding small differences in carrier mobility and lattice thermal conductivity the choice of the
dopant does not significantly influence the thermoelectric properties. This has been confirmed
experimentally for Sb and Bi for n-type Mg2X where > 1.2 has been found for both dopants, see
e.g., discussion in [49, 61, 68] or compare [18, 22, 47]. For the p-type it has been argued repeatedly
that the rigid band model is not applicable for all dopants. The conclusion has been reached based on
band structure calculations [65, 68] or the analysis of experimental transport data [69]. In this case
the choice of dopant is naturally of utmost importance for the optimization of the thermoelectric
material. We will therefore discuss the relevant dopants for p-type Mg2X, investigate the differences
between these and discuss the interplay between dopant, carrier concentration and composition,
i.e., the Si:Ge:Sn ratio. The applicability of the rigid band picture will be discussed in Section 3.3.

Table I lists all (to the best of our knowledge) reported p-type dopants for Mg2X.

Table I: List of dopants that have been successfully studied to synthesize p-type Mg2X. The substituted atom species (Mg
or X = Si, Ge, Sn) is also given, together with the maximum reported carrier concentration. It is expected that ≥ +. , ×
.+/+ 0123 is required for optimum thermoelectric properties. Note that the data from [70] for Ag doped Mg2Sn is
omitted due to the large amount of metallic secondary phases.

Dopant Substitutes pmax /10 4


cm27 Composition Reference
Ag Mg 0.6 Mg2Si0.25Sn0.75 [71], [57, 58, 72-
77] [64, 73]
Li Mg 4.9 Mg2Si0.3Sn0.7 [69], [63, 64, 66,
74, 78-82]
Na Mg 0.59 Mg2Ge0.25Sn0.75 [82],[67, 69, 80,
83]
Ga X 3.2 Mg2Si0.3Sn0.7 [84],[65, 69, 82]
K Mg [80]
In X [83]

Theoretical and experimental studies indicate that Ag substitutes Mg [73, 74, 76, 85] and is thus able
to provide holes. In particular Kim et al., reported the electronic structure and the thermoelectric
properties of Ag-doped Mg2Sn and Mg2Si1-xSnx (x = 0.95, 0.9) [73]. According to their calculations, Ag
is likely to form a resonant like state if it substitutes to the Sn site; this was also found by Bourgeois
et al. [68]. If Ag goes to the Mg site, a rigid band like behavior is predicted by the calculations. The
6
experimental results reveal no indications for a resonant state behavior which suggests that Ag
occupies the Mg site. This is in agreement with all other experimental reports, where Ag has been
used for p-type doping of single crystals in earlier work [57, 58] and also more recently [70, 86-89].
Prytuliak et al. investigated the substitution of various lattice sites in the Mg2Si structure by dopants
using synchrotron and neutron diffraction. Surprisingly they found that Ag occupies the Si position
[90], not the Mg position as reported by most papers. This apparent contradiction points out that the
preferential substitution site can depend on the synthesis conditions: according to the calculation
from Tani et al. the AgSi substitution is preferable for Mg-rich synthesis conditions while the AgMg
substitution is preferable for Si-rich synthesis conditions [74]. It should also be noted that Prytuliak et
al. found 10 wt% of secondary phases (MgAg, MgO, Si) in their material, possibly influencing the
substitution site. Promising thermoelectric properties have been reported by Kim et al. who were
able to obtain a peak power factor of more than 2 × 1027 W m29 K 2 at 300 K [73], by Tang et al.
who obtained a peak figure of merit 0.45 at 690 K which has been achieved for 5% Ag doping at a
carrier concentration of ≈ 6 × 109; cm27 for melt-spun Mg2 Si0.4Sn0.6 [71] and by Jiang et al., who
obtained = 0.38 for Mg2Ge0.4Sn0.6 [77].

DFT calculations for Mg2Si and Mg2Ge from several groups indicate that Li preferentially substitutes
Mg and that this results in a shift of the Fermi level towards the valence band, resulting in p-type
conduction [68, 74, 91]. In particular, Nieroda et al. have compared the formation energies for LiMg,
LiSi and Lii (Li at interstitial position) and found that LiMg is the most stable configuration [66, 92].
Several of the recent experimental papers focus on Li as dopant [69, 78-81] and confirm Li as an
effective p-type dopant. Zhang et al., demonstrated an improved figure of merit of ≈ 0.5 and a
power factor of 1.5 × 1027 W m29 K 2 at 750 K by Li doping to the Mg-site of Mg2Si0.3Sn0.7 [69].
Later reports by Gao et al., Tang et al., and Isachenko et al. confirm the reported good thermoelectric
properties for the same or similar compositions [78-80]. Gao et al. used Li2CO3 instead of using Li
metal as a dopant for Mg2Sn0.6Si0.4 due to the highly reactive nature of Li [93]. Here, the reaction of
Mg with Li2CO3 will give some impurity phases which can be seen in the XRD pattern. Nevertheless
the reported carrier concentration and the hole mobility show that the transport properties are
comparable with the best reported data. De Boor and Saparamadu et al. showed that Li is also an
effective dopant for Mg2Ge0.4Sn0.6 and that this material has a comparable figure of merit [81].

As a group I element, Na is expected to behave similarly to Li and act as acceptor substituting Mg [68,
91]. Tada et al. used metallic Na and CH3COONa as dopants for Mg2Si0.25Sn0.75 [67]. The highest carrier
concentration is shown for CH3COONa doping and better thermoelectric performance is shown
compared to (metallic) Na doping. Zhang et al. and Gao et al. both directly compared Li doping with
Na doping and both found a much lower carrier concentration and hence poorer thermoelectric
performance for Na at comparable nominal dopant concentration [69, 80]. In a recent work
Saparamadu and de Boor et al. could show experimentally that the carrier concentration in Na-doped
Mg2Ge0.25Sn0.75 can be enhanced to ' ≈ 6 × 109; cm27 using high energy ball milling, resulting in
comparable performance of Li and Na-doped samples [82].

As a group 13 (3A) element Ga could in principle act as n- and p-type dopant. However, several DFT
calculations show that Ga preferentially substitutes the X atoms [68, 74, 91], which is confirmed
experimentally by the observed p-type conduction [65, 84, 94]. Liu et al. systematically varied the Ga
content in Mg .9 >Si4.7 Sn4.B C92D GaD and were also able to achieve high hole concentrations ' ≈

7
3 × 10 4 cm27 and carrier mobilities of ≈ 20 cm /Vs. With these high carrier concentrations they
were able to obtain a peak figure of merit of 0.35 at 650 K.

Al has also been considered as a p-type dopant [68, 83], however, mostly n-type is observed
experimentally [74, 95-99]. Other experimentally investigated dopants include In [83] and K.
However, for the latter one the obtained thermoelectric properties were poor, presumably due to a
low carrier concentration [69, 80]. In and K have also been both theoretically predicted to work as p-
type dopants [68, 85]. Further predicted dopants include Cu and B [68, 91]; however, B has been
reported as n-type dopant [100], while the experimental confirmation for as Cu p-type dopant is yet
missing [99]. The reason for that lies at least partially in the low solubility of the dopants due to
relatively high formation energies of the respective lattice defects [74, 91].

A preliminary comparison of the dopants is attempted in Figure 4, where the reported are
listed for the four main dopants Ag, Li, Na, and Ga. It can be seen that the best properties and most
publications with > 0.4 used Li as dopant. On the other hand it is clear that for all of these
four dopants ≈ 0.4 has been achieved. The possible reasons for the observed differences
(dopant solubility and efficiency, changes to the band structure, etc.) will be discussed in the
following.

Figure 4. a) Maximum figure of merit for p-type Mg2X; the best results are usually obtained for Mg2A1-xSnx (with A = Si, Ge
and 0.20 < x < 0.4). Data for Ga taken from [65, 82, 84], for Ag from [64, 70-72, 75, 77], for Na from [67, 82] and for Li
from[64, 69, 78-80, 101]. b) Maximum reported carrier concentration vs. Si content of the composition. Higher (thus
better) values for have been obtained for heavier compositions containing little Si.

8
A further crucial parameter for the carrier concentration is the sample composition, i.e., the Si:Ge:Sn
ratio. This is visualized in Figure 4b, where a summary of literature results of ' is given as a
function of Si content. The data obviously shows some scatter as it ignores the differences between
the dopants, the chosen synthesis approach, the differences between Ge and Sn and the dopant
content I. It should also be noted that for some of the highest ' secondary phases have been
observed [70, 84]. It can nevertheless be concluded that reaching ' > 5 × 109; cm-3 becomes more
facile with lower Si content. This is in line with the findings from several authors, e.g., Liu et al. [84],
Ihou-Mouku et al. who were able to double p at the same Ga content going from Mg2Si to
Mg2Si0.4Ge0.6 [65], and Kim et al. who found that roughly four times the Ag content is required to
achieve the same p for Mg2Si0.1Sn0.9 compared to Mg2Sn [73]. Comparing Mg2Si0.4Sn0.6 and
Mg2Ge0.4Sn0.6 de Boor and Saparamadu et al. found that the Ge-containing samples show a higher '
at the same Li content (and same synthesis conditions) [81]. It can be concluded that a high carrier
concentration is more feasible for heavier compositions and that the thermoelectric optimization of
lighter compositions with a high Si content is particularly challenging as 'JKL > 5 × 109; cm-3 can be
expected [48, 102].

We will now attempt to explain the observed differences between the different dopants as well as
the influence of the composition on the carrier concentration ' and the thermoelectric properties.
Possible differences between the dopants are: their solubility in the matrix, the ionization of the
dopants, and the interplay with intrinsic defects in Mg2X. These might also play a role in the observed
dependence of ' on composition. It should furthermore be noted that the synthesis route and
the detailed processing conditions have an influence on the carrier concentration (for a given
composition). This is not specific to p-type Mg2X but can be observed for almost all thermoelectric
materials.

2.2 Dopant solubility

A natural explanation for the observed differences of the reported thermoelectric properties and the
achievable carrier concentration is the solubility of the dopants, which can differ between the
dopants due to different ionic radii and electronegativity [74, 91].

The formation of Ag containing impurity phases has been reported in several papers, e.g., in the work
on Mg2Sn by Chen and Savvides et al. [70, 86-89]. They found MgAg as a side phase even for
relatively small dopant concentrations in Mg2SnAgy, with I ≈ 0.0045. For Mg2Ge0.4Sn0.6, Jiang et al.
deduced an Ag solubility limit of I ≈ 0.02 from the observed shift of the lattice parameter [77]. On
the other hand, Tang et al. studied the thermoelectric properties of Ag-doped Mg2Si0.4Sn0.6 and
deduced a solubility limit of I ≈ 0.05 due to the observed change of p with y [71]. The observed
differences emphasize the importance of the synthesis technique for the dopant solubility: Jiang et
al. used a melt synthesis in an ampoule route while Tang et al. used melt-spinning which provides
very high cooling rates [103], possibly retaining the Ag in a metastable state in the material. Several
papers report peculiar transport data for Ag-doped samples [72, 99, 104]. In those, the Seebeck
coefficient first rises with temperature as expected for a degenerate semiconductor, then decreases
due to the onset of intrinsic conduction, partially with an observed sign change. For even higher
temperatures a partial recovery of the Seebeck coefficient is often observed, which cannot be

9
understood in a simple two-band model with a fixed number of extrinsic charge carriers [99].
Prytuliak et al. studied Ag-doped Mg2Si using neutron and synchrotron diffraction and deduced
migration of Ag in and out of the matrix upon thermal cycling the sample [90]. A temperature
dependent number of extrinsic charge carriers can qualitatively explain the peculiar transport data
but the exact mechanism remains unclear. Furthermore it should also be noted that the investigated
samples contained a quite large fraction (≈ 10 MN%) of secondary phases (MgAg, MgO, Si), which is
not necessarily representative for the other reports. Overall solubility of Ag in the matrix appears to
be at least one of the factors limiting maximum carrier concentration, especially for compounds with
high Si content.

The situation appears to be different for Li: Nieroda et al. studied Mg2-yLiySi and refined the Li content
using X-ray and neutron diffraction. They obtained I = 0.26 but did not observe any Li containing
impurities using XRD, EDX, and neutron diffraction [66]. This is in line with studies of Li-doped ternary
compounds where also quite high Li contents have been employed (I > 0.1C, without observation of
precipitated Li [69, 78, 80]. This indicates a high solubility of Li in Mg2X. Chemical analysis of the
samples by inductively coupled plasma mass spectrometry (ICP) from Gao et al. also shows a
reasonably good correlation between nominal and actual content, excluding that a large fraction of Li
is lost during the synthesis process [80].

For Na the experimental data is sparse. Tada et al. reported on the synthesis of Mg2-yNaySi0.25Sn0.75
and did not observe Na containing secondary phases up to I = 0.03. Similarly Saparamadu did not
find Na containing side phases for Mg2-yNayGe0.25Sn0.75 for I = 0.03 [82]. Tani et al. calculated the
formation energy for both LiMg and NaMg in Mg2Ge and found a lower formation energy for LiMg,
indicating a lower solubility for Na [91].

For Ga doping Liu et al. found Ga-containing impurity phases in Mg2(Si0.3Sn0.7)1-yGay for I ≥ 0.05,
indicating finite solubility of Ga in the material [84]. Using electron probe micro analysis (EPMA) they
also showed that the actual Ga concentration is around one order of magnitude lower than the
nominal concentration, independent of y. This is in quantitative agreement with the findings from
Ihou-Mouku et al. who also found significantly lower actually Ga content than expected for both
Mg2Si and Mg2Si0.4Ge0.6 [65]. It should also be noted that Ga and Mg can form a number of phases
which can act as Ga sinks and thus limit the solubility of Ga in Mg2X [105].

Little is known on the influence of the Mg2X composition on the solubility of the dopants. Comparing
the theoretical results from Tani et al. on Mg2Si and Mg2Ge, smaller formation energy is observed for
Ga in Mg2Ge, indicating higher solubility. This is in agreement with the ionic radii of Ga being closer
to that of Ge than Si. However, further research is necessary to estimate the solubility limits of the
individual dopants in Mg2X. It should also be noted that XRD is not very sensitive to secondary phases
so that solubility limits obtained from powder diffraction should be regarded as an upper limit.

2.3 Dopant ionization

A charge carrier concentration lower than expected can also be due to partial thermal activation of
the dopants. This is the case if the dopant creates a sub-band with a relatively large energy difference
to the valence band as has been observed e.g., for transition metals in Si [106]. However,
temperature dependent carrier concentration measurements of Ag-doped Mg2X show no significant
change above 100 K [70, 73] which would be expected as carriers would be excited from such a sub-
10
band in this temperature range. Furthermore in the vast majority of the papers and for all relevant
acceptors electrical conductivity and Hall mobility follow the behavior expected for a constant carrier
concentration above 300 K [69, 80, 81] up to the onset of the thermal activation of the intrinsic
charge carriers. This indicates that thermal activation of extrinsic charge carriers is complete at the
temperatures of interest and thus incomplete ionization cannot be an explanation for the observed
peculiarities.

2.4 Interplay between defects and dopants and their effect on the thermoelectric
properties

A carrier concentration that is lower than expected from the nominal dopant concentration can also
result from compensation effects of electrically doping intrinsic lattice defects. In Mg2X, defects are
ubiquitous and have a strong influence on the thermoelectric properties of the material [84, 107-
109]. Kato et al. calculated the formation energies in Mg2Si for different defects and identified Mg at
the interstitial position (Mgi) as the most relevant one [51]. Their calculations also predicted, that
irrespective of the experimental synthesis conditions, Mg2Si will always be n-type. Kubouchi et al.
studied the site occupancy in Mg2-xSi and found a relatively constant occupancy of 0.5% for the
interstitial position, confirming the presence of Mgi experimentally [53]. Mgi is positively charged and
provides electrons to the crystal, rendering the intrinsic material n-type. This is in agreement with
the experimental work on not intentional doped (“undoped”) Mg2Si which is exclusively found to be
n-type conducting [57, 72, 107]. Liu et al emphasized the importance of point defects in a recent
work and calculated formation energies and concentrations for point defects in Mg2Si, Mg2Ge, and
Mg2Sn [52]. They found Mgi and Mg vacancies VMg to be the most relevant defects, followed by
antisite defects on the Mg position, see Figure 5a).

Figure 5. a) Formation energies of point defects in Mg2Si under Si-rich conditions as a function of the Fermi energy of the
electrons. Note that a Fermi energy of 0 corresponds to the valence band maximum and that the band gap has a width
here of only 0.21 eV due to the method of calculation. The defects with the lowest formation energy (and hence highest
concentration) are Mg vacancies and Mg interstitials. b) Defect concentration and resulting net carrier concentration for
intrinsic Mg2Si, Mg2Ge, and Mg2Sn at 800 K; the chemical potential of Mg can be influenced experimentally by the
synthesis conditions. Note that while Mg2Si is found to be always n-type, Mg2Ge and Mg2Sn can be obtained n or p-type

11
according to the calculations from [52]. Reprinted with permission from [Liu, X., et al., Advanced Electronic Materials, 2,
p. 1500284, 2016] Copyright (2017) by Wiley".

While Mgi is positively charged and acts as an electron donor (PQR → PQRT + 2V 2 ), VMg is negatively
charged and acts as an acceptor (WXY + 2V 2 → WXY 2
). The dominant charge carrier type for intrinsic
Mg2X thus depends on which defect is dominant. Liu et al. found that for Mg2Si Mgi is always the
dominant defect, in agreement with the work from Kato and the experimental findings. For Mg2Ge
and Mg2Sn the dominant defect type and carrier type depend on the actual synthesis conditions and
can be controlled to some extent by synthesizing in Mg-rich (=> n-type) or Mg-poor conditions (=> p-
type). Experimentally, undoped Mg2Ge was found both n- and p-type [58, 110], while Mg2Sn is often
p-type below room temperature [70, 89]. Furthermore Isoda et al. could tune the carrier type in
Mg2Si0.25Sn0.75 from n-type to p-type if the material was synthesized Mg deficient [64], in qualitative
agreement with the calculations from Liu et al. According to their calculations the processing window
for p-Mg2Sn is smaller than that of Mg2Ge, experimental verification of that feature remains to be
done.

It should also be noted that the formation energy of charged defects depends on the Fermi level and
the defect concentration in doped Mg2Si will thus depend on the dopant concentration. With respect
to p-type Mg2X it is particularly important that the formation energy for the electron donating Mgi
decreases to lower Fermi energy (i.e., towards increasing hole concentration) while the formation
energy of the oppositely charged VMg increases. Both effects counteract the original increase of the
hole concentration. Summarizing experimental and theoretical findings, the following conclusions
can be drawn:

1. Defects in Mg2X can provide a significant number of charge carriers and can have a significant
influence on the thermoelectric properties.
2. The concentration ratio of defects acting as electron donors (mainly Mgi) and acceptors (VMg)
decreases going from Mg2Si over Mg2Sn to Mg2Ge. Extrapolating into the quasi-binary solid
solutions the donor/acceptor ratio decreases from Si-rich to Si-poor compositions of Mg2X.
3. The concentration of Mgi increases with decreasing Fermi energy of the electrons.
Presumably, this trend is valid for impurity-doped Mg2X as well, indicating that electrons are
donated to the crystal due to a change of intrinsic defect concentration upon intended p-
type doping, reducing the net hole concentration and counteracting the desired increase of
holes.

These hypotheses need further experimental proof as they extrapolate from the binary solutions to
the quasi-ternary solutions and transfer the results from undoped Mg2X to doped Mg2X, however,
they are in line with the experimental results. The second conclusion nicely explains the experimental
finding that achieving a high hole concentration is much easier in Si-poor solid solutions than in Si-
rich ones, see Figure 4. Furthermore, Kim et al. studied intrinsic Mg2Si1-xSnx with 0.9 < " < 1 and
found a decrease of the (positive) low temperature Seebeck coefficient with increasing Si content as
well as a sign change at lower temperatures, indicating a decrease of the (net) hole concentration
with increasing Si content [73]. They also noted that roughly four times the amount of Ag is required
to obtain comparable carrier concentration in Mg2Si0.1Sn0.9 compared to Mg2Sn.

12
The third conclusion essentially says that dopant atom concentration and net hole concentration
cannot be expected to be equivalent to each other, but the hole concentration will be less than
expected due to compensation effects of the increasing Mgi concentration. Even more, a
proportionality cannot be expected since the defect energy, and with it the degree of compensation
itself, will vary with the Fermi energy.

To further check the validity of the hypotheses, the analysis of Mg2X doped with Li is particularly
relevant as the Li solubility was found to be high and is thus not the limiting factor for the achievable
hole concentration. It has been shown furthermore that Li is not lost during synthesis [66, 69, 80].
The occupation of the interstitial position 4b by Li was suggested by Gao et al. as a possible reason
for the observed low dopant efficiency (p/Li atom density) [80]. However, Nieroda et al. studied the
occupation of the 4b position and found a more negative formation energy for LiMg than Lii and also a
negative occupancy trying to refine Li on the 4b position using neutron and synchrotron data [66].
This shows that Lii is unfavorable in Mg2Si. It does not rule out completely the possibility of Lii for the
general Mg2X system but indicates that Lii is not the explanation for the observed poor dopant
efficiency. That leaves a defect induced compensation mechanism as a plausible explanation. A
summary of experimental data on the doping efficiency of Li in Sn-rich Mg2X solid solutions is shown
in Figure 6.

Figure 6. a) Li content [ vs. experimentally determined carrier concentration for samples with a composition of Mg2XySn1-
y (X = Si, Ge; 0.25 < y < 0.4). The dashed line ( = \]^ C is obtained under the assumption that each Li atom provides one
free hole and for a lattice constant of _ = `. ` × .+2.+ 1. Although Li doping allows for high carrier concentrations, the
experimentally observed values remain far below the limiting value, especially for higher [ values. b) The ratio of
experimental and theoretical value indicates that the dopant efficiency increases with increasing Sn content of the

13
samples and that replacing Si by Ge leads to a higher dopant efficiency. Figure adapted from [81], data from
[81],[82],[81],[80],[78],[69] legend from top to bottom.

It can be seen that the dopant efficiency is far below unity and that it does not increase with
increasing Li-content (or p) as would be expected for a constant number of compensating electrons
due to defects. Rather, it can be concluded from Figure 6 that the number of compensating electrons
increases drastically upon increasing doping, in line with the calculations shown in Figure 5 and our
hypotheses. The differences between different reports (for similar compositions) emphasize the
influence of the synthesis conditions. However, comparing Mg2Si0.4Sn0.6 with Mg2Ge0.4Sn0.6 from the
same synthesis route, de Boor and Saparamadu found a better dopant efficiency for the Ge
containing samples [81] (blue ‘+’ vs. yellow square). Moreover, comparing Mg2Ge0.4Sn0.6 with
Mg2Ge0.25Sn0.75 (blue ‘+’ vs. red ‘x’) better dopant efficiency was obtained for the Sn-richer samples,
all findings in agreement with the above stated hypotheses.

For Ag, Ga, and Na a decrease in dopant efficiency with dopant content has also been observed [67,
71, 82], but with these elements other effects are also likely to play a role. It should also be noted
that more complex defects can also influence the electrical properties in Mg2X. This was highlighted
in a recent theoretical work from Han and Shao [76]. They considered Ag-doped Mg2Si and showed
that p-type conductivity is achieved through linear clustering of Ag with Mg interstitials (Ag-Mgi-Ag).
The positive charge of substitutional Ag is compensated to a large extend by the negative charge of
Mgi, resulting in relatively low carrier concentrations. Experimental verification of the relevance of
complex defects for Ag or p-doped samples is highly desirable.

2.5 Dopant summary

A survey of the literature results shows that achieving a high hole concentration is challenging for
Mg2X, in particular for binary Mg2Si. As this is an indispensable prerequisite for p-type Mg2X with
good thermoelectric properties we have tried to rationalize the reasons for the observations. It has
been found that the obtained results differ between the possible p-type dopants. From theoretical
and experimental findings it can be deduced that the solubility limit for Ag and Ga can be one of the
carrier concentration limiting factors, especially for Si-rich Mg2X. On the other hand, no solubility
issues for Li have been reported. Instead the formation of defects in Mg2X (in particular Mgi and VMg)
appears to be of major influence. An analysis of the literature results indicates that the formation of
defects compensates the desired p-doping, resulting in the observed low carrier densities.
Furthermore, the number of n-doping defects increases with increasing dopant concentration but
decreases with X changing from Si to Sn and to Ge in Mg2X. This indicates that the optimization of p-
type Mg2X is more involved than the optimization of the n-type as the choice of X influences not only
carrier mobility, band structure, and lattice thermal conductivity but also the obtainable carrier
concentration.

3. Thermoelectric Transport

3.1 Introduction

14
In the following section we want to investigate the thermoelectric transport in p-type Mg2X. We will
particularly address the question whether the rigid band model is applicable for p-Mg2X and how
good the predictive power of the simple single parabolic band model (SPB) is. After discussing the
differences between the dopant species in the previous section we will also try to give a conclusive
advice on the best dopant.

The basic assumptions of the SPB are that the electronic transport in a material is governed by a
single parabolic band (i.e., carrier effective mass is independent of carrier concentration: $∗ a
$∗ >bC) and that the band structure of the material does not change upon dopant
addition/substitution; the lattice thermal conductivity c L enters as a parameter. The SPB model has
been employed extensively for the n- and p-type material [18, 30, 47, 69] and has been presented in
detail previosuly [111-113]. The basic equations are as follows:

ef >2 + λCijT9 >kC


=d g d kl,
V >1 + λCij >kC
(3.1)

7/
2$%∗ ef
b = 4n g l i9/ >kC. (3.2)
o
b 1.5i4.r >kC>0.5 + 2sCi t24.r >kC
bp = , qp =
qp >1 + λC ij >kC
(3.3)

where iu >kC is the Fermi integral of order v, k the reduced (electro)chemical potential, ef the
Boltzmann’s constant, V the elementary charge, and $%∗ the density of states effective mass. The
true carrier concentration b differs from the experimentally accessible Hall carrier density bp by the
Hall factor qp . Note that b refers to the carrier concentration in a general or specifically the electron
concentration in n-type materials while we use ' when discussing specifically the hole concentration
in p-type materials. The scattering factor s describes the energy dependence of relaxation time
according to the scattering mechanism: w = w4 x t24.r.

z{ >9TtC>7TtC}~•€ }~• 2> TtC }~••


From k, the Lorenz number y = >9TjC }‚
can be calculated, which allows for
|
the determination of the lattice thermal conductivity c L + ƒ„ = d … + ƒ„ = d y + ƒ„ ,
provided that the bipolar contribution ƒ„ is negligible, i.e., at sufficiently low temperature.
The thermoelectric figure of merit can be written as [111]

= (3.4)
y + >†‡C29

ˆ‰| ‹€ zŒ 9.r ‹∗ 9.r


with † = Š Ž >1 + sCit and ‡ = μ4 Š ‹• Ž .r
/ c L. ‡ and μ4 are material parameters
7 • €
independent of carrier concentration, μ4 is related to the measured Hall mobility by μ4 = ‘’ ∗ >1 +
sCit />0.5 + sCi t 24.r .
For the analysis of the experimental data we have employed s = 0, corresponding to the energy
dependence of carrier scattering with acoustic phonons (AP) and potential fluctuations due to the
solid solution formation (alloy scattering) [114, 115]. The assumption that these mechanisms are
dominant is reasonable for Mg2X at room temperature and above. The incorporation of dopants will
initiate some ionized impurity scattering, but several calculations show that this is very small
compared to AP scattering for Mg2Si at the usual doping levels [48, 102].
15
The SPB does not account for the contribution of the intrinsic charge carriers that are thermally
excited and have a strong influence on the transport properties for low extrinsic carrier
concentrations and/or high temperatures. Thus, for accurate modelling of the high temperature
behavior of doped Mg2X multiband modelling is required, see e.g., [48, 116].
As the analysis is easier if data from different sources can be compared directly with each other, we
have extracted the experimental results ( , , , bp C graphically and employed the equations given
above to calculate , c L , $%∗ , ‘p , and ‘4 . For the calculation of ‘p , a temperature independent
Hall carrier concentration bp was assumed. Sometimes it was necessary to back-calculate the
experimental results, e.g., if ”•– was given or and the power factor. As everything except , , is
thus calculated and not measured data, the markers in the following plots do not represent
experimental data but are means for better distinguishing of the curves.

3.2 Thermal transport

Solid solution formation has been identified as one general possibility to enhance the thermoelectric
properties of a material class [114] and has been studied intensively for n-type Mg2X [44, 117, 118].
As can be seen from Figure 7a) this has also been realized for p-type Mg2X with the lowest lattice
thermal conductivity for the Mg2(Si,Sn) system with similar fractions of Si and Sn. Differences
between reports on the same composition can (besides measurement uncertainties) be due to
different synthesis techniques which influence the thermal conductivity via the micro- and
nanostructure of the material. One difference between n- and p-type Mg2X is the importance of the
bipolar contribution to the thermal transport, as shown by the comparison for n- and p-type data for
Mg2Si0.4Sn0.6 in Figure 7a). At > 650 K a much more pronounced increase for the p-type material
can be observed. This is partially due to the higher carrier concentration that is easier accessible in
the n-type material but also due to the higher mobility of the electrons compared to that of the holes
and thus higher (partial) conductivity of the electrons as minority carriers [119] in the p-type
material. The choice of the dopant species is sometimes considered as a possibility to decrease c L
further as the dopants also introduce mass fluctuation and strain fields into the material, albeit to a
smaller degree as the fraction is usually < 5 at%. Direct studies on that are sparse for the p-type
material and the comparison between different reports is non-conclusive if different synthesis
techniques have been employed. As no conclusive study for all dopants is available, we compare the
results on Mg2-xYxSi0.4Sn0.6 (with Y = Li, Ag) from Tang et al (synthesized by melt spinning) in b) [71,
78], a study from Saparamadu et al. on Mg2Ge0.25Sn0.75 comparing Li, Ga, and Na in c) [82], and results
from the Tang group on Mg2Si0.3Sn0.7 comparing Li and Ga [69, 84].

16
Figure 7. a) Lattice thermal conductivity for (doped) Mg2Si and its solid solutions. Solid solution formation drastically
decreases ˜™_š , the more the higher the mass difference between the constituents is. For p-type material the bipolar
contribution at higher temperatures is significantly larger, as can be seen by comparing data for n- and p-Mg2Si0.4Sn0.6.
Data from: Mg2Si [66], Mg2Si0.4Ge0.6 [65], Mg2Si0.3Sn0.7 [69], p-Mg2Si0.4Sn0.6 [78, 81]; n- Mg2Si0.4Sn0.6 [22]; from each
publication the data of the “best” sample is shown. b-d): Lattice thermal conductivity of Mg2X for different dopants.
Always, data are compared from the same synthesis technique and (partially) the same authors. While the results from
Tang indicate a significantly lower ˜™_š + ˜›^ for Ag-doped samples compared to Li-doped ones [71, 78], the comparative
study from Saparamadu reveals only slightly different ˜™_š + ˜›^ for Ga compared to Na and Li [82] and no significant
differences are found comparing the Li- and Ga-doped samples from [69, 84]. Note that the differences at higher T are
due to the bipolar contribution which is different between the samples due to different extrinsic carrier concentration
levels.

The results from Tang et al. show a remarkably reduced c L >≈ 20%) for the Ag doped samples
compared with the Li doped samples at lower temperatures. Different behavior at higher
temperatures is due to different doping levels of the samples where a higher concentration of
majority carriers lowers bipolar effects. The study by Saparamadu et al. shows the largest c L for the
Ga doped samples and the smallest for the Li doped ones, however, the differences are relatively
small. The data from Liu et al. and Zhang et al. on Ga and Li doped samples reveals no clear trend.
Some influence of Ag on the thermal conductivity is qualitatively plausible as the mass contrast
between Mg and Ag (PXY ≈ 24 g mol29 , PžY ≈ 108 g mol29 ) is larger than between the other
substitutions (PŸR ≈ 7 g mol29 , P • ≈ 23 g mol29 P R ≈ 28 g mol29 , P¡• ≈ 70 g mol29 , P ¢ ≈
119 g mol29 C. However, it can not be excluded that some synthesis parameter has been varied
between the two papers so a truly comparative study is necessary to confirm this hypothesis. For the
other dopants it can be concluded that the differences are not larger than the experimental errors.

3.3 Electrical transport

17
This subsection is dedicated to the analysis of the carrier mobility ‘ and the carrier (density of states)
effective mass $%∗ . Both parameters govern the electrical transport in thermoelectric materials and
are thus of fundamental importance for the optimization of p-Mg2X. We will pay particular attention
to the interplay between composition, dopant, and mobility as well as to the applicability of the SPB
model to the material system.

For a given material (band structure, temperature, carrier density, etc.) a high mobility of the carriers
is desired as the drift mobility ‘% is proportional to the electrical conductivity = bV‘% and is thus

beneficial for thermoelectric conversion. ‘% = ‹ , where the relaxation time is usually modelled
¤

using a power law w = w4 V t24.r, and $R is the inertia effective mass, which is assumed to be identical
/7
to the single valley effective mass $∗ ¥¦ ; it furthermore holds $%∗ = §¦ $¥¦ ∗
, where §¦ is the
multiplicity of valleys of a certain band according to crystal symmetry. If several scattering
mechanisms are present, the specific relaxation times are assumed to add inversely w 29 = ∑wR29
(Matthiessen’s rule).

It has been found that for all binary and ternary compositions the hole mobility is lower than that of
the electrons. This is supported by the results from Zaitsev et al., who state ‘¢ = >405 / 530 /
320C cm V 29 s29 and ‘© = >65, 110, 260C cm V 29 s29 for Mg2Si, Mg2Ge, and Mg2Sn,
respectively [44, 83, 94]. The difference between ‘¢ and ‘© is partially due to the different curvature
of the valence and conduction bands, resulting in different $¥¦∗
, see Figure 3. Kutorasinski et al.
showed in particular that the density of states effective mass $%∗ close to the VB maximum is less
than half in Mg2Sn compared to that in Mg2Si, explaining the observed trend [59].

One might expect a weighted average for the quasi-binary solutions superimposed with a decrease
due to additional alloy scattering of the charge carriers [114]. When comparing experimental data it
also needs to be considered that for increasing carrier concentration ' the carrier (drift) mobility ‘%
and the measured ‘p usually decrease due to the increasing carrier-carrier scattering. We have
therefore summarized in Figure 8a) the data of room temperature ‘p for the Mg2Si1-xSnx system
considering data from samples with ' > 109; cm27 only.

18
Figure 8. a) Hall mobility ª«,¬ at ≈ 3++ - for doped Mg2Si1-xSnx. As ª« = ª« > C data from samples with > , ×
.+.® 0123 is shown as thick symbols, while data with .+.® 0123 < < , × .+.® 0123 is shown in thin markers and
lower doped samples are discarded completely. b-d): Hall mobility ª«,¬ at ≈ 3++ - for different compositions comparing
directly different p-type dopants. Data taken from [64, 65, 67, 69, 71, 73, 78-81, 84] (a), [71, 78] (b), [82] (c), [69, 84] (d).

It can be seen that the obtained mobilities are roughly one order of magnitude lower for Mg2Si than
that for Mg2Sn, in qualitative agreement with the finding from Zaitsev et al. Furthermore a clear and
monotonous trend to smaller mobilities for higher Si content can be discerned, emphasizing the
challenge in optimizing the Si rich compositions. Mg2(Ge,Sn) samples have been omitted in Figure 8a)
for the sake of simplicity but it has been observed that replacing Si by Ge enhances carrier mobility as
would be expected due to less pronounced alloy and or AP scattering of the charge carriers, see [81]
or compare [82] with [64]. As the concentration of single point defects presumably decreases with Si-
content (see Figure 5b) these are unlikely to be the cause for the observed trend. On the other hand,
calculations on Ag-doped Mg2Si from Han and Shao indicated linear clusters of Ag- and Mgi to be the
reason for the mobility reduction.

The influence of the dopant species is highlighted in Figure 8b-d) where data from the same Mg2X
compositions but different dopants are compared. As ‘ is heavily influenced by the employed
synthesis procedure, comparison is most instructive if only data from the same synthesis approach
(ideally: same laboratory) are compared. All three examples show a decrease of ‘’ with increasing
carrier density. This is partially due to increased carrier-carrier scattering, but also due to the
increased $%∗ with increasing ', see [48] and discussion below Figure 9. No large difference can be
concluded from the comparison of Ag and Li from the Tang et al. data [71, 78] and between Na and
Ga from Saparamadu [82] as the overlapping window in ' is too small. On the other hand,
comparison between Ga and Li in Figure 8c) indicates higher mobility for Li, which is in agreement
with the results from Liu et al. and Zhang et al. [69, 84] compared in Figure 8d).

Overall, the mobility data shows a strong influence of the composition with heavier compositions
showing significantly enhanced mobility. With respect to the dopant species Ga seems to lead to
lower mobilities. If complex defects were crucial for the hole mobility in Mg2X (as indicated by Han
19
and Shao for Ag doped Mg2Si) a clear dependence on dopant element would be expected due to
difference in ionic radii and substituted position. Further systematic studies are required to elucidate
this point. Furthermore, as the highest mobilities in the Mg2X system have been reported for Ag-
doped Mg2Sn a direct comparison including Ag would be highly interesting.

The effective mass is a fundamental material parameter and part of the quantity ‡ =

>‹• /‹€ C•.¯ °€ .r
that is often used to assess the potential of a thermoelectric material (Eq. (3.4)) [60,
±²³
69, 120]. The dependence of the thermoelectric properties on the effective mass is complex as ‡ ∝
>$%∗ C9.r , but ‘ = ‘>$∗ C. Indeed it can be shown that for acoustic phonon scattering as dominant
scattering mechanism ‘žµ ∝ >$¥¦
∗ C2 .r
, and thus ‡ ∝ ¶
∗ , where the difference between $¥¦

and
‹·¶
the inertial mass $¸∗ has been neglected [120]. This indicates a low effective mass to be beneficial
even if a large $¥¦

corresponds to a large Seebeck coefficient.

Calculations consistently indicate flatter bands and larger (single valley) effective masses for the
valence band compared to the conduction band, see Figure 3. Fitting experimental data for
Mg2(Si,Sn) to multiband solutions of the Boltzmann transport equation, Bahk et al. obtained $¹’ ∗
=
1.0 $4 and $’’ = 1.5 $4 but $º9 ≈ 0.4 d 0.5 $4 and $º7 ≈ 0.38 $4, where $4 is the free
∗ ∗ ∗

electron rest mass and X1 and X3 denominate the two lowest conduction bands in Mg2X [48].
Similarly Satyala and Vashaee used $¹’ ∗
= 1.0 $4 and $’’∗
= 2.0 $4 and $º9 ∗
= $º7

≈ 0.26 $4
(directionally averaged) for their calculations of binary Mg2Si [102]. The higher effective mass leads
to a significantly lowered mobility: Employing ab initio calculations for the band structure Han and
Shao calculated a mobility ratio of roughly 1:8 between holes and electrons in binary Mg2Si, in
reasonable agreement with experimental data [76].

With respect to the influence of the group 4 element X Kutorasinski et al. found a decrease in $¥¦∗

going to heavier compositions [59]. This is in agreement with findings for the n-type where a change
of effective mass with composition has been predicted and experimentally observed [18, 121]. The
observed decrease of the hole mobility with increasing Si content (Figure 8a)) is in agreement with
this prediction, however as mobility is not only controlled by $¥¦

, direct experimental evidence for a
significant dependence of $¥¦ is still missing.

The applicability of the rigid band and SPB model is controversially discussed in the literature. Most
authors assume the applicability a priori and check the validity by comparing experimental and
theoretical results for >bC, the so-called Pisarenko plot. This usually gives good agreement, see e.g.,
the discussion on Ag-doped Mg2Si1-xSnx by Kim et al. or that on Li-doped Mg2Si0.4Sn0.6 and
Mg2Ge0.4Sn0.6 by de Boor et al. [81]. On the other hand DFT calculations by Ihou-Mouku et al. show a
strong modification of the DOS in the vicinity of the Fermi energy due to the substitution of Si by Ga
and the experimental results indicate an increase in $%∗ with increasing ', both points are in
disagreement with the parabolic rigid band picture [65]. Furthermore Zhang et al. found deviations
from the Pisarenko plot for both Li and Ga-doped Mg2Si0.3Sn0.7 as well as a higher $% for the Ga-
doped samples [69].

For a complete picture and to observe general trends we have compiled the available experimental
data in Figure 9. The Pisarenko plot in Figure 9a) shows that almost all effective masses fall between
0.8 $4 < $%∗ < 2 $4 for the whole compositional range of Mg2X and all dopant species. The only
20
exception is the data from Ihou-Mouku which shows exceptionally high $%∗ values [65]. With respect
to the dopant species no difference can be observed, all dopants show a slight increase in DOS
effective mass with increasing '. On average $%∗ increase from $%∗ >109; cm27 C ≈ 1 $4 to
$%∗ >3 × 10 4 cm27 C ≈ 1.7 $4 as can be discerned from Figure 9b). With respect to composition the
heavier compositions have on average a slightly lower $%∗ ,. This is in agreement with the trend
expected from the theoretical results of Kutorasinski et al. , who obtained density of states effective
mass $%,XY

R ≈ 1.5 $4 , $%,XY ¡| ≈ 0.8 $4 , $%,XY ¢ ≈ 0.5 $4 for the holes close to the
∗ ∗

valence band edge[59]. They also observe an increase of $%,



>'C also in agreement with the trend of
the experimental data.

Figure 9. Seebeck coefficient at room temperature »> C (Pisarenko plot) (a) and DOS-effective mass ¼∗½ as a function of
carrier concentration (b). In a) the dopants are highlighted while in b) the composition is indicated. While the data shows
some scatter, a slow increase of ¼∗½ > C can be discerned but no remarkable influence of the dopant species. All data has
been calculated assuming SPB and AP scattering for ≈ 3++ ¾. Data taken from [64, 65, 67, 69, 71, 73, 78-82, 84].

For the calculation of the theoretical curves and the effective masses SPB with s = 0 (dominant
scattering mechanism: AP and alloy scattering) was employed. Kim et al. measured the scattering
parameter for compositions close to Mg2Sn and obtained d0.2 < s < 1.2 (note the different
definition of s in the original work) corresponding to a mixture of acoustic phonon, optical phonon
and alloy scattering; for the modelling they therefore used s = 0.5 [73]. Employing s = 0.5 in the
above analysis results in slightly lower $%∗ , but it does not significantly enhance the quality of the fit
nor does it change the fundamental observations.

From the analysis of the available experimental data the following can be concluded:

• The rigid band model seems to be well applicable and no particular influence of any dopant
species on $%∗ is visible.
• The SPB is not strictly valid as $%∗ = $%∗ >'C. On the other hand, within a finite range for ',
the change is not dramatic and the use of an average value like $%∗ = 1.3 $4 (corresponding
to the average ' between 3 × 109; cm27 and 3 × 10 4 cm27 ) should still yield to
acceptable agreement between modelling and experimental data.
• A trend towards lower $%∗ is observed for heavier compositions, shifting the 'JKL to lower
values and overall favoring Si-poor compositions with respect to thermoelectric
performance.

21
As fundamental reasons for the deviation from the SPB model several things are possible: First, the
assumption of parabolic bands is generally an approximation that is expected to hold close to the
band extremum, i.e., for low carrier concentrations. Furthermore, band structure calculations from
Zhang et al. showed a non-degeneracy of the heavy and light valence band [69]. In this case a change
in $%∗ with changing ' (and hence chemical potential of the carriers) is expected as the relative
contribution of each band varies. However, this has not been verified by other calculations and
further theoretical and experimental work is necessary to elucidate this point.

The power factor (PF) of a material is directly linked to the power output of a TEG and
determines the electrical power that can be produced from a given heat source. In cases where the
input heat is “freely” available it is even more relevant than the thermoelectric figure of merit which
determines the conversion efficiency. The highest power factor values > 2 × 1027 W m29 K 2
have been reported for Mg2Sn [73, 89], while the ternary compositions with the higher
thermoelectric figure of merit can be optimized to > 1.5 × 1027 W m29 K 2 [69, 78, 80]. de
Boor and Saparamadu et al. showed in particular, that replacing Si by Ge leads to an enhancement of
the PF , similar to the results in the n-type material [43, 60]. However, is a local quantity with
respect to temperature while for most application scenarios the integrated properties over a
relatively large temperature interval are important. While a high (peak) power factor at a certain
Å
[ÃÅ Æ >ÄC ÈÄ]
temperature is thus beneficial, the engineering power factor >¿iC…ÀÁ = Ç
Å is a more
ÃÅ Æ Ê>ÄC ÈÄ
Ç
suitable quantity to compare and evaluate the potential of thermoelectric materials [122]. It should
be noted that >¿iC…ÀÁ is more a device quantity than a material property which is proportional to Δ
for constant material properties, the values for ¿i and >¿iC…ÀÁ are thus not comparable anymore.
On the other hand >¿iC…ÀÁ is directly related to the output power density of an ideal device from
that material and hence a suitable quantity for a comparison of materials from a practical point of
view.

Figure 10a) shows the > C results for (the best) samples from Zhang et al. and Isoda et al. as well
as >¿iC…ÀÁ >Δ C [64, 69]. It can be seen that the >¿iC…ÀÁ are quite similar, although the sample from
Zhang et al. has a higher maximum value and a different temperature dependence. We have used
p = 700 K, corresponding to Δ = 400 K as up to this temperature data is available for most of the
samples. The compositional analysis in Figure 10b) shows higher >¿iC…ÀÁ for Sn-rich compositions as
can be expected due to higher mobility and higher (available) carrier concentration. Again, the
difference between Mg2(Si,Sn) and Mg2(Ge,Sn) solid solutions is not small with >¿iC…ÀÁ roughly 10%
larger for the Mg2GeSn compositions.

22
Figure 10. a) Power factor Ì»/ and engineering power factor >ÍÎCÏÐÑ for two samples from Zhang et al. and Isoda et al.
[64, 69]. >ÍÎCÒ Ó increases with increasing temperature difference corresponding to a larger possible output power. b)
>ÍÎCÏÐÑ vs. Si content in the Mg2(SiSn) system; furthermore available data from Mg2(Ge,Sn) and Mg2(Si,Ge) is also
indicated. We have chosen Ô = 3++ ¾ and Õ = Ö++ ¾ for the calculation. From each study/composition only the data
for the samples with the best >ÍÎCÏÐÑ are shown. References are as in Figure 9 with the color of the labelling indicating
the dopant type.

3.4 Thermoelectric Figure of Merit

For (doped) Mg2Si the highest maximum figure of merit was reported from Mars et al. who found
= 0.17 at 900 K [75]. The reason for the poor properties is mainly the low hole concentration,
see Figure 4. For Mg2Ge no systematic doping studies are available, while for Mg2Sn higher hole
concentrations and better thermoelectric properties were reported, e.g., = 0.3 at 500 K from
Chen et al [70]. The best properties with × > 0.5 have been observed for Sn-rich solid solutions.
This is partially due to the reduction of the lattice thermal conductivity caused by increased phonon
scattering and is a general feature of solid solutions [114].

One fundamental parameter for material optimization is the optimum carrier concentration 'JKL .
Within the SPB a single value for 'JKL can be identified that gives the for a certain
temperature, varying only weakly with temperature. In practice it has been shown that for n-type
Mg2X an interval of 1.5 × 10 4 cm27 < bJKL < 3 × 10 4 cm27 can be identified. The finite interval

23
size can be explained by the influence of synthesis on microstructure and hence on c L and ‘ that is
not explicitly accounted for in the SPB. Nevertheless this fact plus the often found validity of the
Pisarenko plot indicate good predictive power of the SPB for n-type Mg2X [47, 49]. The
thermoelectric figure of merit of p-type Mg2X at 700 K is plotted in Figure 11a). At low carrier
concentrations ' < 3 × 109; cm27 is small due to non-optimized electronic properties (as
expected in the SPB) and the influence of bipolar transport. On the other hand the figure shows a
very broad maximum with ≥ 0.4 for 0.8 × 10 4 cm27 < ' < 6 × 10 4 cm27, which is in contrast
to the observations on the n-type material and the predictions from the SPB model. Furthermore,
the 'JKL obtained in individual studies differ significantly for similar compositions and the same
dopant, compare, e.g., the results in ref [69, 80]. The reason for that is currently unclear and requires
further investigations. Note that all data with > 0.4 has been obtained for compositions with
Mg2(Si,Ge)1-xSnx with 0.6 ≤ " ≤ 0.8, i.e., in a relatively narrow composition range. The broad
maximum for 'JKL is therefore not caused by its dependence on the composition.

The thermoelectric figure of merit is linked to the conversion efficiency of a TEG and therefore
usually employed to rate thermoelectric materials [3]. However, alike the power factor, it is a local
quantity with respect to temperature, which is not ideal to judge the potential of a material with
respect to its benefit in a thermodynamic device operated over a temperature difference.
Å
>ÃÅ Æ >ÄC ÈÄC
The engineering figure of merit >× C…ÀÁ = Å
Ç
Å > pd ÙC is a temperature-
ÃÅ Æ >ÄC ÈÄ ÃÅ Æ Ê>ÄC ÈÄ
Ç Ç
integrated quantity and shows (in contrast to ) an almost linear relationship with device efficiency
[122]. The relation between >× C…ÀÁ and more traditional (material) averaging quantities is currently
discussed [123], we use it in Figure 11b) to illustrate the progress in the material development over
the past decade. The >× C…ÀÁ for several Sn-rich p-type solid solutions are compiled with respect to
dopant species in Figure 11b). Li doped samples show the best thermoelectric properties (with
respect to >× C…ÀÁ between 300 K and 700 K) in agreement with the data for × from Figure 4.
The results for the other dopants Ag, Na, and Ga are lower, but of comparable magnitude. It can also
be seen that the intensifying research on p-type Mg2X has led to significant progress over the recent
years. Nevertheless, it should also be noted that there is still a significant gap to the n-type material
which can exhibit >× C…ÀÁ of 0.6-0.7 for the same temperature interval [60, 61], corresponding to
conversion efficiency above 10%.

24
Figure 11. a) Thermoelectric figure of merit at 700 K of all samples from the same publications as in Figure 8 and Figure 9.
High values have been obtained for between Ú × .+.® 0123 ÛÐÜ ` × .+/+ 0123 , showing a very broad
distribution with respect to carrier concentration. b) >Ý CÏÐÑ between 300 K and 700 K (Na data from [82] only up to 674
K), only the result for the highest >Ý CÏÐÑ from each publication is shown, the respective dopant species is also
indicated. The best results have been obtained for Li-doped samples. Note also that the differences in >Ý CÏÐÑ can be
small even if the value for ZT at 700 K is quite different (e.g., data from [65, 69, 78]). Additional references in b): [72, 75,
77].

With respect to composition the existing data indicates that the replacement of Si by Ge in
Mg2(Si,Ge)1-xSnx does apparently not lead to enhanced >× C…ÀÁ and hence conversion efficiency.
However, the amount of available data is not yet large so that this conclusion remains preliminary.
Furthermore, from a device point of view, similar conductivities for n- and p-leg are desirable and as
the Mg2(Ge,Sn) solid solutions have larger electrical conductivities, they might be matched easier
with the optimized n-type material.

It should be noted that while z at 700 K favors ' > 0.8 × 10 4 cm27 and Sn-rich solutions the
difference in >× C…ÀÁ between samples with high and low carrier concentration is not tremendously
large, see e.g., the result for Ga-doped Mg2Si0.4Ge0.6 from Ihou-Mouku et al. with >× C…ÀÁ ≈ 0.2 and
' ≈ 3 × 109; cm27 [65]. Similarly, it is also worth pointing out that the >× C…ÀÁ for the best samples
from Isoda and Zhang are comparable >ZTC…ÀÁ ≈ 0.2 even if the maximum figure of merits are quite
different: = 0.32 and = 0.5 [64, 69]. The sample from Isoda has a much lower carrier
concentration and hence shows a steeper increase in at lower but a decrease beyond 610 K,
while the from Zhang increases over the whole temperature range showing better at higher .
The temperature averaging inherent in >ZTC…ÀÁ evens that out.

4. Conclusion

The optimum carrier concentration is of crucial importance to any thermoelectric material and often
the first parameter to be optimized for a certain material class [3]. Theoretical calculations predict
'à©– ≈ 4 × 10 4 cm27 for Mg2Si [102, 124]. This is significantly higher for the n-type due to the
larger effective mass of the holes compared to that of the electrons. A different calculation from
Bahk et al. including also solid solutions from Mg2Si and Mg2Sn predicts a similar range with 'JKL
increasing with increasing Sn- content [48]. However, Bahk et al. have employed a constant hole
effective mass, while detailed band structure calculations indicate a higher effective mass for Si-rich
compositions, indicating even higher 'JKL increasing with decreasing Sn- content [59, 112].

We have compiled the available experimental data and found a broad range for the optimum carrier
concentration (0.8 × 10 4 cm27 < ' < 6 × 10 4 cm27, see Figure 11)) but not the clear optimum of
>'C predicted by the SPB or more complex multiband models. While some scatter in the data is
due to variations in synthesis technique and composition the fundamental reason is not clear yet.

The Mg2X system offers a further degree of freedom for optimization due to the almost complete
miscibility of Mg2Si, Mg2Ge, and Mg2Sn. We have summarized in Figure 12 the impact of the Si
content of the composition on several parameters that are crucial for material optimization. Mg2Si

25
and Si-rich compositions have low densities which are favorable for most applications and
particularly important for mobile applications [37, 125, 126]. Si-rich compositions are also known to
be more stable, increasing the operating temperature range [127]. Dedicated investigations on p-
type Mg2X are yet to come, but it can be expected that the trend observed from the n-type holds for
the p-type as well.

Figure 12. Optimization of p-type Mg2X with respect to Si:Ge:Sn ratio is complex and involves several, partially
contradicting conditions. The arrows indicate the compositional side that is favorable, while the vertical component
indicates the change of the respective parameter with composition.

Bipolar transport usually limits the thermoelectric properties of the materials at high temperatures.
The number of thermally excited charge carriers is controlled by the band gap, which decreases from
á¡ ≈ 0.78 eV for Mg2Si to á¡ ≈ 0.35 eV for Mg2Sn, again favoring Si-rich compositions. On the other
hand the experimentally accessible hole concentration and the hole mobility clearly favor Si-poor
compositions. While the latter may partially be due to subtle changes of the valence band structure
and a decrease of the effective mass with decreasing Si-content, the achievable hole concentration
appears to be limited by the dopant solubility for some of the dopants. Solubility increases towards
the heavier compositions for these dopants, favoring Si-poor solid solutions. A summary of the
experimental and theoretical works shows that defects have a strong impact on the thermoelectric
properties of Mg2X, influencing carrier concentration and mobility. As the influence of defects
favoring p-type conduction increases towards Si-poor solid solutions achieving higher hole
concentrations is more facile in those, even if dopant solubility is not an issue. Finally, the lattice
thermal conductivity minimizes at intermediate compositions.

Synthesis of p-type Mg2X is more feasible under Mg-poor conditions. While for n-type material
usually some excess Mg is used to compensate for Mg loss during the synthesis, good p-type is
obtained for weighing in of nominal composition, again emphasizing the importance of defects in the
material.

So far, materials with the best thermoelectric properties have been found for compositions around
Mg2(Si,Ge)0.3Sn0.7, however the experimental data is not overwhelmingly numerous and the best
composition is not yet clear due to the number of contradicting optimization conditions. These are
partially a special feature specific to p-type Mg2X, making the material optimization more challenging
but leaving room for further optimization. For Si-rich compositions the expected optimum carrier
26
concentration is higher than that of Si-poor compositions due to change in effective mass. The
experimentally achieved values of the carrier concentration are too low by one order of magnitude
and significant improvement can be expected for Si-rich compositions if higher carrier concentration
levels can be achieved.

With respect to the dopant species we can conclude that Ag, Li, Ga, and Na have been utilized to
synthesize material with decent thermoelectric properties. From analysis of the experimental data
we conclude that all dopants show rigid band behavior, although there is some debate about that.
Examination of the available mobility data indicate higher mobilities for Ag and Li doped samples
compared to Ga-doped ones. For Ag finite solubility appears to be a problem in some reports;
consequently the best properties have been reported for Li-doped samples.

Overall there has been significant progress in the development of p-type Mg2X in the last decade.
Due to employment of different dopants and new synthesis techniques the carrier density has been
enhanced significantly. The employment of Li as dopant in particular has led to high carrier
concentrations, enhancing the power factor and thus the practically achievable output power. In
combination with solid solution formation this has led to maximum values of ≈ 0.6 for the
thermoelectric figure of merit. P-type is thus comparable to higher manganese silicide [128-130] and
the development of Mg2X-based thermoelectric generators is underway with detailed results
expected in the near future.

5. Strategies for further improvements

While there has been considerable progress in the development of efficient p-type Mg2X a further
optimization is highly desirable and would enhance the practical potential for Mg2X-based TEGs.
While speculative, based on the detailed analysis presented in the previous sections, following points
are suggested:

(a) Optimizing via stoichiometry

The competing effects of low achievable doping concentration (for the Si rich compositions due to
presence of defects with a donor nature) and low operation temperature and (for the Sn rich
compositions due to the reduced band gap) indicate a trade-off. Taking into consideration the
general requirement of a band gap of ≈ 5 d 10 ef (where T is the operation temperature) in TE
materials, suitable base compositions would be Mg2X with X having a silicon to tin ratio between
80:20 and 60:40. The choice of dopant is restricted to the elements Ag, Li, and Ga. As discussed in
the previous sections, Li doping results in the best properties. However, the effects of multiple
doping on the desired carrier concentration have not been sufficiently studied and requires further
probe. Combination of dopants (Ag+Ga and Li+Ga) which target substitution in both Mg and Si sites
would be of specific interest due to the possibility of higher charge carrier concentration and
increased alloy scattering.

The synthesis techniques also require further investigations and offer possibilities for an
enhancement of . The solubility of dopants can depend on the synthesis method with non-
equilibrium techniques resulting in higher solubility [6]. Furthermore, the formation of point defects

27
in Mg2X has been theoretically shown to depend on the synthesis conditions (Si rich or Mg rich). Thus
finding the optimal processing route which minimizes formation of the Mg interstitial defect would
result in enhancement. Studies comparing equilibrium processing techniques (e.g., solid state
synthesis) with non-equilibrium techniques (e.g., melting, ball milling and hot pressing) can thus lead
to a fundamental understanding on the defect chemistry in this material system and also help in
improving the material TE properties.

(b) Suppression of Bipolar Thermal Conduction

High bipolar thermal transport is one of the current limitations of the p-type thermoelectric
performance [119]. This is more severe due to the lower mobility of holes compared to electrons and
requires heavy doping (achieving which in itself is a challenge for p-type Mg2X) for suppression. Due
to its direct relation to the material band gap, increasing the band gap is one of the techniques for
increasing the onset temperature for noticeable bipolar effects. This has been successfully
demonstrated in n-type Mg2(Si,Sn) solid solution with Ge substitution [116]. Similar studies (Ge
substitution) in the p-type counterpart is thus expected to result in suppression of bipolar thermal
transport. Other methods to suppress bipolar thermal transport rely on successfully trapping the
minority charge carriers using microstructural (grain boundary modification) or chemical (trap
impurity levels) modifications [124, 131]. Experimental investigation of the solubility and stability of
different impurities and their effect on the thermoelectric transport properties is therefore required.
The choice of the impurity element is speculative however analogies can be drawn from the
extensively studied elemental Si and Ge systems and different transition metal elements could be
attempted. The other possibility involving grain boundary modification is also promising and further
investigation on bulk nanostructuring and its effect on bipolar thermal transport is required.

(c) TE property enhancement using embedded nanoparticles

The so called “nanoparticle-in-alloy” approach is a generic method for improving the performance in
TE materials [132]. Embedded nanoparticles in a thermoelectric matrix act as additional phonon
scattering centers due to their interaction with the mid to long wavelength phonons. Additionally
induced strain at the TE matrix-nanoparticle interface can result in further decrease in the lattice
thermal conductivity. This approach has been successfully demonstrated in the lead chalcogenide
system with significant increase in [5, 6]. The other benefit of embedded nanoparticles is the
possibility of filtering low energy electrons which contribute negatively to the Seebeck coefficient.
For this to occur, special combination of the matrix/nanoparticle with near match in the absolute
energy levels (band alignment) is required. Theoretical estimates in n-type Mg2(Si,Sn) indicate
notable improvement in z due to energy filtering [48]. While not reported, an even higher effect is
expected for p-type Mg2X due to a more favorable electron/phonon scattering length ratio [79, 124,
133].

Acknowledgements

J. de Boor would like to thank the Helmholtz Association for initiating this collaboration. He would
also like to thank the DLR space research for funding of the current work, N. Farahi for good and

28
thorough feedback as well as S. d. L. for support with the figures. Z.Ren would like to acknowledge
the funding from DOE under grant DE-SC0010831.

[1] Bell LE. Science 2008;321:1457.


[2] He W, Zhang G, Zhang X, Ji J, Li G, Zhao X. Applied Energy 2015;143:1.
[3] Snyder GJ, Toberer ES. Nat Mater 2008;7:105.
[4] Ebling DG, Krumm A, Pfeiffelmann B, Gottschald J, Bruchmann J, Benim AC, Adam M, Labs R,
Herbertz RR, Stunz A. J. Electron. Mater. 2016;45:3433.
[5] Biswas K, He JQ, Blum ID, Wu CI, Hogan TP, Seidman DN, Dravid VP, Kanatzidis MG. Nature
2012;489:414.
[6] Tan G, Shi F, Hao S, Zhao L-D, Chi H, Zhang X, Uher C, Wolverton C, Dravid VP, Kanatzidis MG.
Nature Communications 2016;7:12167.
[7] Tan GJ, Zhao LD, Shi FY, Doak JW, Lo SH, Sun H, Wolverton C, Dravid VP, Uher C, Kanatzidis
MG. J. Am. Chem. Soc. 2014;136:7006.
[8] Rogl G, Grytsiv A, Rogl P, Peranio N, Bauer E, Zehetbauer M, Eibl O. Acta Mater. 2014;63:30.
[9] Tan GJ, Liu W, Wang SY, Yan YG, Li H, Tang XF, Uher C. Journal of Materials Chemistry A
2013;1:12657.
[10] Qiu YT, Xi LL, Shi X, Qiu PF, Zhang WQ, Chen LD, Salvador JR, Cho JY, Yang JH, Chien YC, Chen
SW, Tang YL, Snyder GJ. Adv. Funct. Mater. 2013;23:3194.
[11] Sakurada S, Shutoh N. Appl. Phys. Lett. 2005;86:082105.
[12] Li JF, Liu WS, Zhao LD, Zhou M. NPG Asia Mater. 2010;2:152.
[13] Yan XA, Joshi G, Liu WS, Lan YC, Wang H, Lee S, Simonson JW, Poon SJ, Tritt TM, Chen G, Ren
ZF. Nano Lett. 2011;11:556.
[14] Zhao LD, Lo SH, Zhang YS, Sun H, Tan GJ, Uher C, Wolverton C, Dravid VP, Kanatzidis MG.
Nature 2014;508:373.
[15] Brown DR, Day T, Caillat T, Snyder GJ. J. Electron. Mater. 2013;42:2014.
[16] Liu HL, Shi X, Xu FF, Zhang LL, Zhang WQ, Chen LD, Li Q, Uher C, Day T, Snyder GJ. Nat. Mater.
2012;11:422.
[17] Zaitsev VK, Fedorov MI, Gurieva EA, Eremin IS, Konstantinov PP, Samunin AY, Vedernikov MV.
Phys. Rev. B 2006;74:045207.
[18] Liu W, Tan XJ, Yin K, Liu HJ, Tang XF, Shi J, Zhang QJ, Uher C. Phys. Rev. Lett.
2012;108:166601.
[19] Liu XH, Zhu TJ, Wang H, Hu LP, Xie HH, Jiang GY, Snyder GJ, Zhao XB. Adv. Energy Mater.
2013;3:1238.
[20] Liu W, Tang X, Sharp J. J. Phys. D: Appl. Phys. 2010;43:085406.
[21] Khan AU, Vlachos NV, Hatzikraniotis E, Polymeris GS, Lioutas CB, Stefanaki EC,
Paraskevopoulos KM, Giapintzakis I, Kyratsi T. Acta Mater. 2014;77:43.
[22] Gao P, Berkun I, Schmidt R, Luzenski M, Lu X, Bordon Sarac P, Case E, Hogan T. J. Electron.
Mater. 2013;43:1790.
[23] Dasgupta T, Stiewe C, de Boor J, Müller E. physica status solidi (a) 2014;211:1250.
[24] Farahi N, Prabhudev S, Botton GA, Salvador JR, Kleinke H. ACS Appl. Mater. Interfaces
2016;8:34431.
[25] LeBlanc S, Yee SK, Scullin ML, Dames C, Goodson KE. Renew. Sust. Energ. Rev. 2014;32:313.
[26] Gaultois MW, Sparks TD, Borg CKH, Seshadri R, Bonificio WD, Clarke DR. Chem. Mater.
2013;25:2911.
[27] Samunin AY, Zaitsev VK, Konstantinov PP, Fedorov MI, Isachenko GN, Burkov AT, Novikov SV,
Gurieva EA. J. Electron. Mater. 2013;42:1676.
[28] Bux SK, Yeung MT, Toberer ES, Snyder GJ, Kaner RB, Fleurial JP. J. Mater. Chem.
2011;21:12259.

29
[29] Zhao J, Liu Z, Reid J, Takarabe K, Iida T, Wang B, Yoshiya U, Tse JS. J. Mater. Chem. A
2015;3:19774.
[30] de Boor J, Gupta S, Kolb H, Dasgupta t, Mueller E. J. Mater. Chem. C 2015;3:10467.
[31] de Boor J, Gloanec C, Kolb H, Sottong R, Ziolkowski P, Müller E. J. Alloys Compd.
2015;632:348.
[32] Ferrario A, Battiston S, Boldrini S, Sakamoto T, Miorin E, Famengo A, Miozzo A, Fiameni S, Iida
T, Fabrizio M. Mater. Today-Proc. 2015;2:573.
[33] Sakamoto T, Taguchi Y, Kutsuwa T, Ichimi K, Kasatani S, Inada M. J. Electron. Mater.
2015;45:1321.
[34] Sakamoto T, Iida T, Honda Y, Tada M, Sekiguchi T, Nishio K, Kogo Y, Takanashi Y. J. Electron.
Mater. 2012;41:1805.
[35] Zhang B, Zheng T, Wang Q, Zhu Y, Alshareef HN, Kim MJ, Gnade BE. J. Alloys Compd.
2017;699:1134.
[36] de Boor J, Droste D, Schneider C, Janek J, Mueller E. J. Electron. Mater. 2016;45:5313.
[37] Kim HS, Kikuchi K, Itoh T, Iida T, Taya M. Mater. Sci. Eng. B-Adv. Funct. Solid-State Mater.
2014;185:45.
[38] Nakamura T, Hatakeyama K, Minowa M, Mito Y, Arai K, Iida T, Nishio K. J. Electron. Mater.
2015;44:3592.
[39] Nemoto T, Iida T, Sato J, Suda H, Takanashi Y. J. Electron. Mater. 2014;43:1890.
[40] Sakamoto T, Iida T, Ohno Y, Ishikawa M, Kogo Y, Hirayama N, Arai K, Nakamura T, Nishio K,
Takanashi Y. J. Electron. Mater. 2014;43:1620.
[41] Skomedal G, Holmgren L, Middleton H, Eremin IS, Isachenko GN, Jaegle M, Tarantik K,
Vlachos N, Manoli M, Kyratsi T, Berthebaud D, Dao Truong NY, Gascoin F. Energy Convers. Manage.
2016;110:13.
[42] Tarantik KR, König JD, Jägle M, Heuer J, Horzella J, Mahlke A, Vergez M, Bartholomé K.
Materials Today: Proceedings 2015;2:588.
[43] Liu W, Kim HS, Chen S, Jie Q, Lv B, Yao M, Ren Z, Opeil CP, Wilson S, Chu CW, Ren Z. Proc Natl
Acad Sci U S A 2015;112:3269.
[44] Zaitsev VK, Federov MI, Eremin IS, Gurieva EA. Thermoelectrics on the Base of Solid Solutions
of Mg2B IVCompounds. In: Rowe DM, editor. Thermoelectrics Handbook: Macro to Nano. Boca
Raton, USA: CRC, 2005.
[45] Fedorov MI, Zaitsev VK. Silicide Thermoelectrics: State of the Art and Prospects. In: Rowe
DM, editor. Thermoelectrics and its Energy Harvesting: Modules, Systems, and Applications in
Thermoelectrics. Boca Raton, FL: CRC Press,Taylor & Francis Group, 2012.
[46] Kaibe H, Aoyama I, Mukoujima M, Kanda T, Fujimoto S, Kurosawa T, Ishimabushi H, Ishida K,
Rauscher L, Hata Y, Sano S. Development of thermoelectric generating stacked modules aiming for
15% of conversion efficiency. ICT: 2005 24th International Conference on Thermoelectrics. New York:
IEEE, 2005. p.227.
[47] Liu W, Chi H, Sun H, Zhang Q, Yin K, Tang X, Zhang Q, Uher C. Phys Chem Chem Phys
2014;16:6893.
[48] Bahk JH, Bian ZX, Shakouri A. Phys. Rev. B 2014;89:075204.
[49] de Boor J, Dasgupta T, Mueller E. Thermoelectric Properties of Magnesium Silicide Based
Solid Solutions and Higher Manganese Silicides. In: Uher C, editor. Materials Aspect of
Thermoelectricty. Taylor & Francis, 2016.
[50] Zaitsev VK, Isachenko GN, Burkov AT. Efficient Thermoelectric Materials Based on Solid
Solutions of Mg2X Compounds (X = Si, Ge, Sn), 2016.
[51] Kato A, Yagi T, Fukusako N. J Phys Condens Matter 2009;21:205801.
[52] Liu X, Xi L, Qiu W, Yang J, Zhu T, Zhao X, Zhang W. Adv. Electron. Mater. 2016;2:1500284.
[53] Kubouchi M, Hayashi K, Miyazaki Y. J. Alloys Compd. 2014;617:389.
[54] Kozlov A, Grobner J, Schmid-Fetzer R. J. Alloys Compd. 2011;509:3326.
[55] Jung IH, Kang DH, Park WJ, Kim NJ, Ahn S. Calphad 2007;31:192.

30
[56] Sondergaard M, Christensen M, Borup KA, Yin H, Iversen BB. J. Electron. Mater.
2013;42:1417.
[57] Morris RG, Redin RD, Danielson GC. Physical Review 1958;109:1909.
[58] Redin RD, Morris RG, Danielson GC. Physical Review 1958;109:1916.
[59] Kutorasinski K, Wiendlocha B, Tobola J, Kaprzyk S. Phys. Rev. B 2014;89:115205.
[60] Mao J, Kim HS, Shuai J, Liu Z, He R, Saparamadu U, Tian F, Liu W, Ren Z. Acta Mater.
2016;103:633.
[61] Saparamadu U, Mao J, Dahal K, Zhang H, Tian F, Song S, Liu W, Ren Z. Acta Mater.
2017;124:528.
[62] Vazquez F, Forman RA, Cardona M. Physical Review 1968;176:905.
[63] Isoda Y, Tada S, Nagai T, Fujiu H, Shinohara Y. Mater. Trans. 2010;51:868.
[64] Isoda Y, Tada S, Nagai T, Fujiu H, Shinohara Y. J. Electron. Mater. 2010;39:1531.
[65] Ihou-Mouko H, Mercier C, Tobola J, Pont G, Scherrer H. J. Alloys Compd. 2011;509:6503.
[66] Nieroda P, Kolezynski A, Oszajca M, Milczarek J, Wojciechowski KT. J. Electron. Mater.
2016;45:3418.
[67] Tada S, Isoda Y, Udono H, Fujiu H, Kumagai S, Shinohara Y. J. Electron. Mater. 2014;43:1580.
[68] Bourgeois J, Tobola J, Wiendlocha B, Chaput L, Zwolenski P, Berthebaud D, Recour Q, Gascoin
F, Scherrer H. Functional Materials Letters 2013;06:1340005.
[69] Zhang Q, Cheng L, Liu W, Zheng Y, Su X, Chi H, Liu H, Yan Y, Tang X, Uher C. PCCP
2014;16:23576.
[70] Chen HY, Savvides N, Dasgupta T, Stiewe C, Mueller E. Phys. Status Solidi A-Appl. Mat.
2010;207:2523.
[71] Tang X, Zhang Y, Zheng Y, Peng K, Huang T, Lu X, Wang G, Wang S, Zhou X. Appl. Therm. Eng.
2017;111:1396.
[72] Akasaka M, Iida T, Matsumoto A, Yamanaka K, Takanashi Y, Imai T, Hamada N. J. Appl. Phys.
2008;104:013703.
[73] Kim S, Wiendlocha B, Jin H, Tobola J, Heremans JP. J. Appl. Phys. 2014;116:153706.
[74] Tani JI, Kido H. Intermetallics 2008;16:418.
[75] Mars K, Ihou-Mouko H, Pont G, Tobola J, Scherrer H. J. Electron. Mater. 2009;38:1360.
[76] Han XP, Shao GS. J. Mater. Chem. C 2015;3:530.
[77] Jiang G, Chen L, He J, Gao H, Du Z, Zhao X, Tritt TM, Zhu T. Intermetallics 2013;32:312.
[78] Tang X, Wang G, Zheng Y, Zhang Y, Peng K, Guo L, Wang S, Zeng M, Dai J, Wang G, Zhou X.
Scripta Mater. 2016;115:52.
[79] Isachenko GN, Samunin AY, Gurieva EA, Fedorov MI, Pshenay-Severin DA, Konstantinov PP,
Kamolova MD. J. Electron. Mater. 2016;45:1982.
[80] Gao P, Davis JD, Poltavets VV, Hogan TP. J. Mater. Chem. C 2016;4:929.
[81] de Boor J, Saparamadu U, Mao J, Dahal K, Müller E, Ren Z. Acta Mater. 2016;120:273.
[82] Saparamadu U, De Boor J, Mao J, Song SW, Tian F, Liu W, Ren ZF. in preparation.
[83] Fedorov MI, Zaĭtsev VK, Eremin IS, Gurieva EA, Burkov AT, Konstantinov PP, Vedernikov MV,
Samunin AY, Isachenko GN, Shabaldin AA. Physics of the Solid State 2006;48:1486.
[84] Liu W, Yin K, Su X, Li H, Yan Y, Tang X, Uher C. INTERMETALLICS 2013;32:352.
[85] Zwolenski P, Tobola J, Kaprzyk S. J. Electron. Mater. 2011;40:889.
[86] Chen HY, Savvides N. J. Electron. Mater. 2010;39:1792.
[87] Savvides N, Chen HY. J. Electron. Mater. 2010;39:2136.
[88] Chen HY, Savvides N. J. Cryst. Growth 2010;312:2328.
[89] Chen HY, Savvides N. J. Electron. Mater. 2009;38:1056.
[90] Prytuliak A, Godlewska E, Mars K, Berthebaud D. J. Electron. Mater. 2014;43:3746.
[91] Tani J-i, Takahashi M, Kido H. J. Alloys Compd. 2009;485:764.
[92] Kolezynski A, Nieroda P, Wojciechowski KT. Computational Materials Science 2015;100:84.
[93] Gao P, Davis JD, Poltavets VV, Hogan TP. Journal of Materials Chemistry C 2016;4:929.
[94] Isachenko GN, Zaĭtsev VK, Fedorov MI, Burkov AT, Gurieva EA, Konstantinov PP, Vedernikov
MV. Physics of the Solid State 2009;51:1796.

31
[95] Tani J-i, Kido H. Intermetallics 2013;32:72.
[96] Tani J-i, Kido H. J. Alloys Compd. 2008;466:335.
[97] Battiston S, Fiameni S, Saleemi M, Boldrini S, Famengo A, Agresti F, Stingaciu M, Toprak MS,
Fabrizio M, Barison S. J. Electron. Mater. 2013;42:1956.
[98] Hu XK, Barnett MR, Yamamoto A. J. Alloys Compd. 2015;649:1060.
[99] Sakamoto T, Iida T, Matsumoto A, Honda Y, Nemoto T, Sato J, Nakajima T, Taguchi H,
Takanashi Y. J. Electron. Mater. 2010;39:1708.
[100] Kubouchi M, Hayashi K, Miyazaki Y. Scripta Mater. 2016;123:59.
[101] De Boor J, Saparamadu U, Mao J, Dahal K, Mueller E, Ren ZF. Acta Mater. 2016;in press.
[102] Satyala N, Vashaee D. J. Electron. Mater. 2012;41:1785.
[103] Tang XF, Xie WJ, Li H, Du BL, Zhang QJ, Tritt TM, Uher C. High-Performance Nanostructured
Thermoelectric Materials Prepared by Melt Spinning and Spark Plasma Sintering. In: Rowe DM,
editor. THERMOELECTRICS AND ITS ENERGY HARVESTINGMATERIALS: MATERIALS, PREPARATION,
AND CHARACTERIZATION IN THERMOELECTRICS. Boca Raton: CRC Press, 2012.
[104] Berthebaud D, Gascoin F. J. Solid State Chem. 2013;202:61.
[105] MTDATA. vol. 2017: National Physical Laboratory.
[106] Tavendale AJ, Pearton SJ. Journal of Physics C: Solid State Physics 1983;16:1665.
[107] Dasgupta T, Stiewe C, Hassdorf R, Zhou AJ, Boettcher L, Mueller E. Phys. Rev. B
2011;83:235207.
[108] Nolas GS, Wang D, Beekman M. Phys. Rev. B 2007;76:235204.
[109] Du ZL, Zhu TJ, Chen Y, He J, Gao HL, Jiang GY, Tritt TM, Zhao XB. J. Mater. Chem.
2012;22:6838.
[110] Shanks HR. J. Cryst. Growth 1974;23:190.
[111] May AF, Snyder GJ. Introduction to Modeling Thermoelectric Transport at High
Temperatures. In: Rowe DM, editor. Thermoelectrics and its Energy Harvesting: Materials,
Preparation, and Characterization in Thermoelectrics. CRC Press, 2012.
[112] Flage-Larsen E, Lovvik OM. Band Structure Guidelines for Higher Figure-of-Merit: Analytic
Band Generation and Energy Filtering. In: Rowe DM, editor. Thermoelectrics and its Energy
Harvesting: Materials, Preparation, and Characterization in Thermoelectrics

CRC Press, 2012.


[113] Fistul VI. Heavily Doped Semiconductors: Springer, 1969.
[114] Wang H, LaLonde AD, Pei YZ, Snyder GJ. Adv. Funct. Mater. 2013;23:1586.
[115] Harrison JW, Hauser JR. Phys. Rev. B 1976;13:5347.
[116] Zhang L, Xiao P, Shi L, Henkelman G, Goodenough JB, Zhou J. J. Appl. Phys. 2015;117:155103.
[117] Bellanger P, Gorsse S, Bernard-Granger G, Navone C, Redjaimia A, Vives S. Acta Mater.
2015;95:102.
[118] Zaitsev VK, Tkalenke EN, Nikitin EN. Sov. Phys. Solid State 1969;11.
[119] Wang S, Yang J, Toll T, Yang J, Zhang W, Tang X. Sci Rep 2015;5:10136.
[120] Wood C. Rep. Prog. Phys. 1988;51:459.
[121] Tan XJ, Liu W, Liu HJ, Shi J, Tang XF, Uher C. Phys. Rev. B 2012;85.
[122] Kim HS, Liu W, Chen G, Chu C-W, Ren Z. PNAS 2015;112:8205.
[123] Armstrong H, Boese M, Carmichael C, Dimich H, Seay D, Sheppard N, Beekman M. J. Electron.
Mater. 2017;46:6.
[124] Satyala N, Vashaee D. J. Appl. Phys. 2012;112.
[125] Kousksou T, Bédécarrats J-P, Champier D, Pignolet P, Brillet C. J. Power Sources
2011;196:4026.
[126] Samson D, Kluge M, Fuss T, Schmid U, Becker T. J. Electron. Mater. 2012;41:1134.
[127] Skomedal G, Burkov A, Samunin A, Haugsrud R, Middleton H. Corros. Sci. 2016;111:325.
[128] Girard SN, Chen X, Meng F, Pokhrel A, Zhou JS, Shi L, Jin S. Chem. Mater. 2014;26:5097.
[129] Chen X, Girard SN, Meng F, Lara-Curzio E, Jin S, Goodenough JB, Zhou J, Shi L. Adv. Energy
Mater. 2014;4:1400452.

32
[130] Bernard-Granger G, Soulier M, Ihou-Mouko H, Navone C, Boidot M, Leforestier J, Simon J. J.
Alloys Compd. 2015;618:403.
[131] Zhao LD, Dravid VP, Kanatzidis MG. Energ Environ Sci 2014;7:251.
[132] Mingo N, Hauser D, Kobayashi NP, Plissonnier M, Shakouri A. Nano Lett. 2009;9:711.
[133] Pshenai-Severin DA, Fedorov MI, Samunin AY. J. Electron. Mater. 2013;42:1707.

33

You might also like