Daniel
Souchay
a,
Stefan
Schwarzmüller
a,
Hanka
Becker
b,
Stefan
Kante
b,
G. Jeffrey
Snyder
c,
Andreas
Leineweber
b and
Oliver
Oeckler
*a
aInstitute for Mineralogy, Crystallography and Materials Science, Faculty of Chemistry and Mineralogy, Leipzig University, Scharnhorststr. 20, 04275 Leipzig, Germany. E-mail: oliver.oeckler@gmx.de
bInstitute of Materials Science, TU Bergakademie Freiberg, Gustav-Zeuner-Str. 5, 09599 Freiberg, Germany
cDepartment of Materials Science and Engineering, Northwestern University, Evanston, IL 60208, USA
First published on 19th August 2019
Different established synthesis methods such as melt-casting followed by annealing as well as melt-spinning or ball-milling followed by hot-pressing or spark plasma sintering may significantly influence the properties of thermoelectric materials. The comparison of microstructures obtained by different synthesis routes (water-quenching and melt-spinning followed by spark plasma sintering) reveals the indirect nature of the beneficial influence of cobalt germanide precipitates on the thermoelectric properties of germanium telluride and germanium antimony tellurides (GST materials). Cobalt germanide precipitates significantly influence the thermoelectric properties of GST materials: the thermoelectric figure of merit zT of (GeTe)17Sb2Te3 obtained by quenching melts in water increases from 1.6 to 1.9 (at 450 °C) by introducing cobalt germanide precipitates. They drastically reduce the grain and sub-grain sizes of the GST matrix. Melt-spinning followed by spark plasma sintering leads to nanoscopic cobalt germanide precipitates, whose effect on the thermoelectric properties, especially the phononic thermal conductivity, surprisingly seems to be marginal. This is due to the already significantly reduced (sub-)grain sizes in such polycrystalline GST samples as revealed by orientation (“channeling”) contrast in backscattered-electron micrographs. Melt-spun and spark-plasma-sintered GST material exhibits similar thermoelectric figures of merit zT values as water-quenched samples with cobalt germanide precipitates. In addition to providing easier access to samples with small (sub-)grains by simple quenching, the precipitates may stabilize the GST or GeTe microstructure by Zener pinning as cobalt is insoluble in the matrix material. This is very favorable concerning cyclability during thermoelectric measurements and potential applications. The heterostructured materials and their thermoelectric properties are long-term stable. Hall effect measurements of heterostructured GST (melt-spun and spark-plasma-sintered) indicate that the charge carrier concentration is near optimum.
Thermoelectric generators that convert waste heat to electrical energy may contribute to establish a multitude of sensors such as thermostatic radiator valves, transmitters, light sensors or aircraft monitoring,9,10 which require only a few hundred micro- or milliwatts to operate. Thus, only small temperature differences are necessary to make them autonomous as no cables or battery changes are necessary if they have access to waste heat.11 In addition, thermoelectric generators can provide independent energy supply in places where no other form of electricity is available, e.g. using thermoelectric generators in ventilators for stoves12 or in space missions.11 Thermoelectric materials may thus contribute to sustainable and renewable energy management, e.g. combining thermoelectric generators with salinity gradient solar pond, photovoltaic systems or by harvesting the waste heat of hot spring thermal energy.13,14
Many thermoelectric materials are being explored for such power generation applications, such as chalcogenides,15e.g. germanium telluride co-doped with BiTe and Cu with a zT value (see below) up to 1.55 in the temperature range of 325 to 450 °C,16 or PbTe co-doped by Na and Cl with Na on cation sites providing additional electrons to the conduction band, resulting in a zT value of 1.27 at 375 °C.17 Other recent examples include PbxSn1−xTe alloys, where Bi2Te3 doping increases zT in the low-temperature range (210 to 340 °C) compared to pristine PbxSn1−xTe samples,18 as well as silicides15 such as higher manganese silicides exhibiting textured behavior leading to ∼10% zT enhancement parallel to the pressing direction. Such anisotropic behavior in non-cubic polycrystalline thermoelectric materials should be carefully considered.19 Concerning the rather isotropic half-Heusler compounds, e.g. doping Ti0.3Zr0.35Hf0.35NiSn with Sb and/or Bi increases in zT value up to 0.58 at 550 °C.20
However, research for optimized thermoelectric materials faces several problems, especially their thermal stability and their rather low efficiency. The latter depends on the dimensionless figure of merit zT = S2σT/(κe + κph), where κe and κph are the electronic and phononic thermal conductivities, respectively, S the Seebeck coefficient, σ the electrical conductivity and T the temperature. In spite of the seemingly simple equation, the individual transport properties contributing to zT are highly correlated: σ depends on the charge carrier density n and the mobility μ according to σ = neμ (Drude model).21 High S would be beneficial for high zT values and requires a high effective mass m* according to S = (8π2kB2m*T) (π/3n)2/3 (3eh2)−1, assuming a parabolic band model with energy independent scattering approximation.22 However, a high m* would mean low μ according to μ = eτ/m* (τ = scattering relaxation time) which results in a low σ. High σ, on the other hand, results in high total thermal conductivity κ according to the Wiedemann–Franz law κeσ−1 = LT (L = Lorenz number).23 A decrease in L is correlated with an increase in thermopower, i.e. the absolute value of Seebeck coefficient according to L = 1.5 + e(−|S|/116) (where L is in 10−8 W Ω K−2 and S in μV K−1).24 This equation is a good approximation when a single parabolic band is applicable and when acoustic phonon scattering dominates. κph is one parameter which can be influenced by real-structure effects such as domain boundaries or nanoscale precipitates. However, the energy range of long-wavelength acoustic phonon modes has to be in the same range as the presumed phonon scattering centers as shown for e.g. skutterudites.25 The so-called phonon-glass electron-crystal (PGEC) concept26 focuses on disorder and “rattling” atoms that may enhance phonon scattering by reducing the mean free path of phonons, whereas σ ideally remains almost unaffected if electron or hole conduction is associated with the rigid framework.
The free path distribution range of phonons and thus the thermal conductivity may further be reduced by micro- or nanoscaled side phases in heterostructured systems.27 For example, such secondary phases in GST materials such as elemental germanium28 or antimony29 have been shown to increase zT significantly. Endotaxially intergrown nanoinclusions of SrTe in PbTe lead to peak zT values up to ∼2.2 at 640 °C.30 LAST (Pb–Sb–Ag–Te) thermoelectric materials obtained by quenching melts exhibit Ag–Sb-rich nanoscale inclusions with coherent or semicoherent interfaces in a PbTe-rich matrix; and further nanoscale inclusions related to compositional fluctuations of Ag, Sb and Pb/Sn have been described in the system Ag(Pb1−ySny)mSbTe2+m.31,32 Such decrease in lattice thermal conductivity has, however, also been explained by solid solution effects, e.g. for PbTe alloyed with MgTe33 and strain effects in PbTe with nanoinclusions of SrTe.34
Instead of nano- and micro-inclusions of second phase, pronounced real structure effects in chemically homogeneous GST materials are also beneficial concerning their thermoelectric properties.6,35–39 Compounds (GeTe)nSb2Te3 with n ≥ 3 exhibit a disordered rocksalt-type high-temperature modification. Quenching this phase affords a metastable pseudocubic modification with short-range vacancy ordering and herringbone-like twinned nanostructures. The planar defects present reduce thermal conductivity drastically, most likely by scattering long-wavelength phonons.38,40 For n < ∼12, the stable room temperature (RT) phases correspond to long-periodically ordered layered structures that consist of rocksalt-type building blocks separated by van der Waals gaps. Increasing the GeTe content to n ≥ 12 leads to materials that can be considered as doped variants of GeTe. In the GST class of materials, many elements are suitable for substitution, e.g. Mn,41 Sn,42 Cd,43 Li,44,45 Ag,46 or Cr47 replacing Ge; As48 or In49,50 replacing Sb, and Se35 replacing Te. These substitutions change the electronic structure and charge carrier density and thus influence the thermoelectric performance.
Comparable to GST composite materials (see above), some heterostructured variants of GeTe exhibit high zT values. Ge0.87Pb0.13Te51 with zT = 1.9 at 500 °C and Ge0.87Pb0.13Te with 3 mol% Bi2Te352 (zT = 1.9 at 500 °C) both contain PbTe-rich domains as a side phase. The heterostructured material with the nominal composition Ge0.9Sb0.1Te0.9Se0.05S0.0553 represents an Sb-doped GeTe1−xSex matrix containing GeS1−xSex precipitates. By substituting Te in GeTe with Se and S, κph was reduced by phonon scattering due to mass fluctuations. As an effect of spark plasma sintering, κph was further reduced by alloying with Sb, which leads to additional point defects and grain boundaries. However, these heterostructured samples combine substitution of the matrix material with additional precipitates, which makes it difficult to trace individual effects.
Furthermore, the microhardness of such precipitate-containing samples is increased compared to GeTe.53 In general, heterostructuring affects mechanical properties by maintaining the microstructure, which is well-known e.g. for δ-ferrite precipitates in steel.54 Grain-boundary migration can be impeded by small particles, an effect known as Zener pinning.55,56 Such particles generate a pinning pressure as it is energetically unfavorable to move past the particle because new boundaries must be created. This counteracts the driving force pushing the boundaries during annealing (mainly the energy stored by grain boundary curvature, especially at high-angle grain boundaries). This aspect is highly relevant for thermoelectric materials with respect to maintaining the microstructure and thus ensuring long-term performance.
In order to study the influence of precipitates in GST materials, a composition close to the optimal charge carrier concentration is most relevant. As precipitates should not influence the GST material by substitution, cobalt germanides are a particularly good choice as cobalt is almost insoluble in GeTe57 and GST with cobalt-rich precipitates.58,59 Thus, this is an optimal model system to analyze the effect of precipitates on thermoelectric properties. Comparing (GeTe)nSb2Te3 with different values of n with and without cobalt germanide precipitates is an intriguing way to elucidate the influence of precipitates on thermoelectric properties. Since good thermoelectric properties for GST with 1 wt% of CoGe2 precipitates had been demonstrated, these compositions served as basis for this study.58 Different synthesis routes such as quenching in water compared with melt-spinning (MS) followed by spark plasma sintering (SPS) allow for comparing more or less coarse microstructures.
Selected-area electron diffraction (SAED), high resolution transmission electron microscopy (HRTEM) and further EDX measurements were executed on a Philips CM-200 STEM (LaB6 cathode, 200 kV, super-twin lens, point resolution 0.23 nm) equipped with an RTEM 136-5 EDX detector (EDAX, Genesis64 software). A double-tilt low-background sample holder (Gatan) was used. For TEM investigations, the bulk material was manually cut, mechanically polished using a dimple grinder (Gatan) and then Ar-ion-thinned (Duo-Mill, GATAN). Evaluation of SAED data was done using the Analysis65 software. For HRTEM and SAED simulations, jEMS66 was used.
Samples with GeTe as the matrix material exhibit the α-GeTe structure type at RT, both after the initial synthesis and after additional thermal treatment during thermoelectric measurements. This holds for water-quenched as well as MS/SPS samples (Fig. S3 and S4, ESI†). Compounds (GeTe)nSb2Te3 (n = 12, 17 and 19), both pristine and as matrix materials, exhibit average structures that correspond to the rhombohedral α-GeTe type (Fig. S5, ESI†). The pseudocubic structure described for n = 1269 is not detected after MS preparation. This may be due to the small grain size in contrast to that in compact ingots, which enables relaxation to rhombohedral metrics in MS flakes. After densification by SPS, including rapid cooling from 450 °C, the structure of the (GeTe)12Sb2Te3 matrix corresponds to the metastable pseudocubic state (Fig. S6, ESI†), which is in good agreement with literature.38 GST samples with n = 17 and n = 19 retain the average rhombohedral α-GeTe-type structure. After thermal treatment during thermoelectric measurements, PXRD patterns of all samples investigated show this structure type, as indicated by broadened and/or split reflections (Fig. S6 and S7, ESI†). The likely presence of vacancy ordering, which eventually may lead to van der Waals gaps, cannot be evaluated on the basis of PXRD data, since the corresponding diffraction patterns are very similar (pseudo-homometry).70 Note that in the case of extended van der Waals gaps, the α-GeTe type only approximates the exact average structure of heavily disordered materials.
SEM-EDX measurements confirm that the composition of the different GST and GeTe matrices are in good agreement with the initial weights (Tables S3–S5, ESI†) and corresponding SEM images show heterostructured materials (example in Fig. 1). As it is typical for GeTe,71 small amounts of Ge precipitates were observed in some SEM images of some Sb-free samples. As weight fraction of precipitates is low, their structure could not be elucidated from PXRD patterns. TEM-EDX point measurements of the cobalt germanide precipitates, however, confirm compositions of approximately Co5Ge7 and CoGe2 (Fig. S8 and Table S6, ESI†). In addition to previously observed58 precipitates with Co5Ge7-type structure (space group I4mm, a ≈ 7.6 Å, c ≈ 5.8 Å),72 the structure and compositions of precipitates with the EDX analysis result Co30.5(1)Ge66.3(4)Sb0.3(1)Te3(2) correspond to CoGe2 (space group Cmce, a = 10.82 Å, b = 5.68 Å and c = 5.68 Å)73 according to electron diffraction. The corresponding SAED patterns and tilt angles between them are given in Fig. S9 and S10, ESI.† The pseudo-tetragonal metrics often leads to twinning, which was typically observed in SAED patterns. As both Co5Ge7 and CoGe2 precipitates are present in the heterostructured materials, the idealized stoichiometry of the precipitates slightly deviates from the nominal composition. The cobalt germanide precipitates may also contain very small amounts of Te, which most likely has no pronounced effect on their weight fraction. As GST materials exhibit a certain homogeneity range,36 this deviation is negligible as no other types of precipitates were detected. In the following sections, all heterostructured samples are represented by their formal nominal compositions in order to avoid the repetition of this discussion.
Quenched (GeTe)12Sb2Te3, however, exhibits a pseudocubic structure at RT, which transforms to a rocksalt-type high temperature structure (T1; bottom right in Fig. 3). Upon slow cooling, this sample also transforms to a distorted variant of rhombohedral α-GeTe type as average structure (T2; Fig. 3), which is in good agreement with literature.38,69
Phase transitions are visible because reflections of the rocksalt-type structure type split in a characteristic way. These phase transitions correlate with the hystereses in the measurements of thermoelectric properties (e.g. electrical conductivity, Fig. 4 and Table S7, ESI†). As the phase transition from GeTe type to rocksalt-type itself is a displacive phase transformation, these structure models represent average structures and the hystereses (Fig. 4) are caused by vacancy diffusion and element ordering during heating and cooling as reported in the literature.75–77
Fig. 4 Thermoelectric properties of (GeTe)12Sb2Te3vs. (CoGe2)0.15(GeTe)12Sb2Te15 (left column), (GeTe)17Sb2Te3vs. (CoGe2)0.2(GeTe)17Sb2Te3 (middle column), and (GeTe)19Sb2Te3vs. (CoGe2)0.22(GeTe)19Sb2Te3 (right column): Seebeck coefficients (first row), electrical conductivities (second row), power factor (third row), thermal conductivities and corresponding phononic part (L = 1.5 + e(−|S|/116) where L is in 10−8 W Ω K−2 and S in μV K−1, fourth row), and zT values (fifth row). All data points of the MS/SPS synthesis route were merged from 3 cycles between 50 °C and 500 °C, not taking into account the first heating curve which is affected by relaxation of metastable states; the individual cycles of the MS/SPS samples with the best average zT values (GeTe)19Sb2Te3 and (CoGe2)0.22(GeTe)19Sb2Te3 are presented in Fig, S13 und S14, ESI;† note that red dots and black triangles often overlap (especially at n = 17), as they exhibit similar values. |
The thermal and electrical conductivities of MS/SPS samples are lower than those of water-quenched ones as the (sub-)grain sizes of the GST and GeTe matrix is drastically reduced because of the rapid solidification conditions of the MS process, which might also induce strain. As the small (sub-)grain sizes of MS samples outweighs the effect of the precipitates on the GST matrix material, their influence on the thermoelectric properties of GST is much less pronounced for MS/SPS samples than for water-quenched ones. Thus, the precipitates influence the thermoelectric properties in a rather indirect way. Their mere presence does not significantly contribute to phonon scattering and thus does not significantly reduce κph even though their size is drastically reduced in MS/SPS samples. However, they influence the microstructure of the GST matrix, which explains their pronounced effect in water-quenched samples, whereas their effect cannot further enhance the beneficial effect of MS in terms of reducing the grain and/or subgrain size in the GST matrix. The cobalt germanides improve the long-term stability of the materials by stabilizing the microstructure by means of Zener pinning. Thus, the precipitates are also very beneficial in MS/SPS samples even if they do not enhance zT. In (CoGe2)0.22(GeTe)19Sb2Te3 (MS/SPS), zT values remain stable in several consecutive cycles up to 500 °C (Fig. S16, ESI†), whereas in pristine (GeTe)19Sb2Te3 (MS/SPS), κ increases and therefore zT gradually decreases upon thermal cycling (Fig. S17, ESI†). This can be explained by growing grains in pristine GST where the grain size is not stabilized by Zener pinning. Upon thermal cycling, increasing electrical and thermal conductivities and decreasing Seebeck coefficients result in slowly decreasing zT values.
However, the preparation of GeTe and its heterostructured variants by MS/SPS seems generally favorable in terms of zT value compared to corresponding water-quenched samples (Fig. 5). CoGe2 and Co5Ge7 precipitates (and a very small fraction of Ge precipitates in GeTe samples) were detected as mentioned in Section Phase composition and crystal structure. Formally adding 1 wt% of Co5Ge7 to GeTe (nominal composition) turns out not to significantly influence the thermoelectric properties compared to (CoGe2)0.01GeTe or GeTe. It turned out that heterostructuring GeTe with cobalt-containing precipitates is not very promising as discussed below.
The evaluation of charge carrier concentration and mobility by temperature-dependent Hall measurements gives further insights into optimal doping levels and scattering mechanisms. The application of an effective mass model (see formulae on the last page of the ESI†) reveals the theoretically possible maximum zT value for an optimal charge carrier concentration and thus the potential for further improvement via doping. Altogether, the approach of heterostructuring GeTe with cobalt germanides is less promising compared to GST hetero-structures, as the average carrier concentration of (CoGe2)0.01GeTe is much higher than optimal. Therefore, doping GeTe would have more impact than heterostructuring. In this respect, GST can be viewed as an optimally doped variant of GeTe. The Hall carrier mobility (Fig. 6) decreases with temperature, following a relationship between T−3/2 or T−1, which would correspond to phonon scattering in non-degenerate and degenerate semiconductors, respectively. As the curves exhibit small changes in slope, the exact exponent cannot be determined reliably and one may assume an intermediate type of behavior, which is quite typical in complex alloys. Both relationships are compatible with the fact that scattering of phonons, probably both acoustic and optical, limits the charge carrier mobility.78,79 Other forms of scattering, e.g. impurity scattering, would have a different temperature dependence. Therefore, the single parabolic band (SPB) model with the acoustic phonon scattering approximation was applied in the temperature range with an approximately linear increase of the Seebeck coefficient and the electrical resistivity, i.e. up to ca. 250 °C. With increasing Sb content, the Hall carrier mobility decreases. The effective charge carrier mass m* according to effective mass modelling of all three compositions changes only very little in the temperature range from 50 °C to 250 °C and no significant changes in charge carrier concentration and mobility were detected (cf.Fig. 6). This confirms that the precipitates do not lead to significant changes of the matrix by mutual doping. (CoGe2)0.15(GeTe)12Sb2Te3 and (CoGe2)0.22(GeTe)19Sb2Te3 exhibit almost optimal charge carrier concentrations. Therefore, (CoGe2)0.22(GeTe)19Sb2Te3 has the better dopant-independent intrinsic properties compared to (CoGe2)0.15(GeTe)12Sb2Te3 and thus the better average zT value in the temperature range from 50 °C to 250 °C. The best performing water-quenched samples with the composition (CoGe2)0.20(GeTe)17Sb2Te3 has the same average zT value of 1.17 as (CoGe2)0.22(GeTe)19Sb2Te3 (MS/SPS) – but is accessible in a more straightforward way.
Fig. 6 Slope of the Hall carrier mobility (top left), effective mass (top middle) and the Hall carrier concentration (top right) of MS/SPS samples of (CoGe2)0.01GeTe (red triangles), (CoGe2)0.15(GeTe)12Sb2Te3 (blue triangles) and (CoGe2)0.22(GeTe)19Sb2Te3 (green squares). For these samples, the calculated zT values as a function of pH at 100 °C (black line), 150 °C (red line), and 200 °C (blue line) and 250 °C (dark green) are shown (assuming SPB behavior, middle row), measured values depicted as squares; Seebeck coefficient as a function of log(pH) (Pisarenko plot, bottom row). Details of the calculations are described at the end of the ESI† and can be found in ref. 78 and 79. |
SEM backscattered-electron orientation (“channeling”) contrast imaging with respect to the (sub-)grain size of the GST matrix material shows a significant difference in case of water-quenched samples. Introducing cobalt germanide precipitates into the GST matrix leads to an enhancement concerning thermoelectric performance of samples with microstructures that contain small grains and/or subgrains. Therefore, precipitates can be beneficial for the products of melt-casting syntheses by significantly reducing the grain, subgrain and domain sizes and would be an interesting approach for other materials to enhance their thermoelectric properties. This is especially true if the precipitates contain an element that is insoluble in the matrix, which prevents them from coarsening during annealing. Thus, Co-containing precipitates are an ideal way of heterostructuring GST materials. In MS/SPS samples, on the other hand, the grain size of the matrix material is mainly influenced by the rapid solidification conditions of MS, which leads to similar zT values of GST or GeTe and the corresponding heterostructured materials. MS/SPS samples exhibit a lower thermal conductivity, lower electrical conductivity combined with an increased Seebeck coefficient compared to water-quenched samples due to the reduced (sub-)grain size of the matrix material. In addition, heterostructuring may contribute to higher hardness (compare ref. 53) combined with an increased cyclability during thermoelectric measurements. We attribute the latter to the Zener pinning effect, which helps maintaining the (sub-)grain size of the matrix material and could be an interesting approach for future investigations in order to boost long-term stability of thermoelectric materials. As both synthesis routes lead to good thermoelectric materials, it has yet to be evaluated, which of the presented syntheses are better suited for upscaling, once these materials are considered for possible applications.
Footnote |
† Electronic supplementary information (ESI) available: Details of syntheses and sintering, density and Dulong-Petit Cp values, powder X-ray diffraction patterns, scanning and transmission electron microscopy (including EDX and SAED), and thermoelectric measurements. See DOI: 10.1039/c9tc03410b |
This journal is © The Royal Society of Chemistry 2019 |