Single-step preparation and consolidation of reduced early-transition-metal oxide/metal n-type thermoelectric composites

Reduced early transition metal oxides/metal composites have been identiﬁed here as interesting thermoelectric materials. Numerous compositions in the Nb-rich portion of the WO 3 –Nb 2 O 5 system have been studied, in composite formulations with elemental W. Spark plasma sintering (SPS) has been employed to achieve rapid preparation and consolidation of composite materials containing W metal precipitates with characteristic length scales that range from under 20 nm to a few microns, that exhibit thermal conductivities that are constant from 300 K to 1000 K, approximately 2.5 W m − 1 K − 1 . Thermoelectric properties of these n-type materials were measured, and the highest-performing compositions were found to reach ﬁgure of merit zT values close to 0.1 at 950 K. The measurements point to higher zT values at yet-higher temperatures. 14 W/Nb composition. good intrinsic

FIG. 1. WO 3 -Nb 2 O 5 phases tend to form complex tetragonal tungsten bronze structures, which feature 3-dimensional slabs of corner-sharing NbO 6 octahedra connected by WO 4 tetrahedra. W 3 Nb 14 O 44 is one such example of many possible homologues; the size of the slabs can be changed with W/Nb composition. The high polyhedral connectivity is important for good electrical transport, while the intrinsic structural defects (such as the crystallographic shear planes and W/Nb site substitution) are important for low thermal transport.

I. INTRODUCTION
The promise of solid-state energy conversion and the recovery of energy from waste heat in devices that have no moving parts has led to vigorous research on thermoelectric materials. However, widespread deployment of thermoelectric modules has been limited by some combination of low efficiency, low crustal abundance of the key elements in routinely-studied thermoelectrics, and poor thermal stability of materials. Oxide materials potentially help address some of these issues, but few high-performance n-type materials have been discovered. The performance of thermoelectric materials is usually measured by the figure of merit zT given by zT = (S 2 T )/(ρκ); a function of the electrical resistivity ρ, the Seebeck coefficient S, and the thermal conductivity κ of the material at temperature T . A recent datamining study carried out by some of us, 1 had pointed out that empirically, high performing thermoelectric materials operating at (typically) 1000 K have Seebeck coefficients in the range of 200 µV K −1 to 300 µV K −1 , and electrical resistivities close to 1×10 −3 Ω cm. In many oxide materials, attaining high Seebeck coefficients is frequently not as challenging as achieving these relatively metallic resistivity values. 2 In addition, small-unit-cell oxides such as spinels, perovskites, and rutiles, are usually highly connected, and therefore have relatively high lattice thermal conductvities, which is undesirable. In this study, early transition-metal oxides with relatively large unit cells, prepared as composites with metals, have been rationally selected as potentially being able to overcome the pitfalls of high resistivity coupled with low thermal conductivity that many bulk oxides display.
The WO 3 -Nb 2 O 5 phases selected for study here tend to form complex tetragonal tungsten bronze structures, which feature 3-dimensional slabs of corner-sharing NbO 6 octahedra connected by WO 4 tetrahedra. 3 Nb 2 O 5 has many polymorphs, but the dominant 2 room-temperature phase is a layered structure with distorted and irregular Nb polyhedra. 4 This poor structural connectivity is not ideal for electrical transport and electronic doping. However, WO 3 -Nb 2 O 5 phases can be doped to have low electrical resistivity, in part because of the highly connected 3D structures that derive from the perovskite aristotype.
Highly connected structures also tend to have high thermal conductivity, but this is mitigated in the tungsten bronze structure-types because of the complex, large-unit cell structures. In the study here, which W/Nb substitution, oxygen deficiency, and crystallographic shear ( Figure 1) also potentially decrease thermal conductivity. Indeed, Winter and Clarke have shown the thermal conductivity of stoichiometric d 0 analogues to the materials studied here to be about 2 W m −1 K −1 from 300 K to 1300 K. 5 In this contribution, the thermoelectric properties of composite samples from the Nbrich portion of the WO 3 -Nb 2 O 5 system are reported. The samples were prepared in single step of short duration that combined preparation from starting materials (elemental metals and oxides) with consolidation. The technique employed for the one-step preparation and consolidation was spark-plasma sintering or SPS, perhaps more accurately referred to as current-assisted pressure-activated densification. 6 This is a rapid preparatory route to react materials, with typical reaction taking minutes, instead of hours or days, and yielding dense, consolidated solids that can be directly employed for physical property measurements. The preparation of phase-pure single-cation WO 3−δ 7 and multi-cation Cr 2 WO 6 8 have been previously demonstrated using SPS, proving that ion mobility is sufficient to allow diffusion and phase equilibrium in WO 3 systems at these reaction temperatures and time-scales. Conventional solid-state preparation of the materials W-Nb-O oxide materials studied here requires extended heat treatments at high temperatures (e.g., 1350 • C in sealed platinum crucibles). 9 In addition to being time-consuming, extended heat treatments at high temperatures also result in difficulties with stoichiometry control, because of volatilization of precursor oxides; a problem often encountered in the preparation of oxides of W. 9 Furthermore, the final product of conventional preparation is usually powder that must be consolidated and densified before physical property measurements can be performed. In this work, SPS was used to prepare dense pellets of (W

II. METHODS
Fourteen compositions were chosen based on reported stable phases, 10  tering on an instrument from FCT Systeme GmbH, Germany. In a chamber base pressure of 10 torr of argon, 80 MPa uniaxial pressure was applied using a 9 mm inner-diameter graphite die (EDM-4, POCO graphite). The sample was then heated rapidly to 1473 K at 150 K/min, annealed for 6 min at 1473 K, then rapidly cooled at 180 K/min ( Figure 2).
X-ray diffraction (XRD) of the resulting dense pellets was performed using a laboratory instrument (Philips X'Pert diffractometer, Cu Kα radiation). After initial room-temperature electrical transport measurements, a subset of these samples was subjected to more rigorous study (Figure 3).
High-resolution synchrotron XRD data on finely-ground powder was acquired at 295 K at beamline 11-BM at the Advanced Photon Source (APS), Argonne National Laboratory, using an average wavelength of ∼0.4137Å with a diffractometer that has been described in detail by Wang et al. 11 The precise wavelength was determined using a mixture of Si (SRM 640c) and Al 2 O 3 (SRM 676) NIST standards run in a separate measurement. Samples were contained in 0.8 mm diameter Kapton capillaries and the packing density was low enough that absorption was not noticeable. Rietveld refinement was performed using the TOPAS 5 Academic program suite. 12 Instrument profile parameters were determined using Si (SRM 640c) and LaB 6 (SRM 660b) NIST lineshape standards, run in separate measurements.
Scanning electron microscopy (SEM) was performed on an FEI XL30 microscope using a backscatter electron (BSE) detector; EDX data were collected with an EDAX Si-drift detector. Typical accelerating voltages were 15 keV or 20 keV. To prepare samples for SEM, pellets were mounted in epoxy and polished with successively finer abrasives; the final polishing step was performed with 0.25 µm colloidal diamond suspension on MicroCloth.
A thin layer of Pd-Au, 2 nm to 5 nm, was sputtered on the surface to alleviate charging during imaging. The brightness and contrast of the micrographs presented have not been post-processed in any way. Particle size analysis was performed using ImageJ 13 on 2500× magnification micrographs, areas of approximately 33 µm × 33 µm, to visualize the smallest features, some of which remain sub-pixel diameter even at this magnification. A minimum size cutoff threshold of 0.05 µm 2 was used to minimize false positives.  16 The estimated error for the heat capacity introduced by this method was < 3% at 300 K, and < 5% at 1000 K. The thermal diffusivity (α) was measured using a Netzsch LFA 457 laser flash system. The samples for the thermal diffusivity measurements were machined to be coplanar with a thickness of between 1 mm and 2 mm and a diameter of 8 mm. The samples were then spray coated with colloidal graphite on both sides to ensure maximum optical absorption and emissivity. Thermal diffusivity was measured in an argon gas atmosphere from 308 K to 1050 K at intervals of 50 K. A final measurement at 308 K was taken after cooling to ensure reproducibility of the measurements and to ensure the coatings remained intact. The thermal diffusivity values were determined using the model of Clark and Taylor, 17 which corrects for radiative losses.

III. RESULTS AND DISCUSSION
Powder XRD reveals the presence of elemental, body-centered cubic (BCC) W in many samples, in addition to peaks characteristic of W-Nb-O phases (Figure 4) finement is shown in Figure 5, using monoclinic WNb 12 O 33 (space group C2, with refined parameters very near those reported by Roth et al. 9 ) and BCC W for the model. The fit is able to account for the elemental W but is unable to describe the primary phase reflections, particularly at low Q. The exact W-Nb-O phases were not able to be determined using powder XRD, as their crystal structures have large unit cells with many atomic positions, and structural similarity leads to many coexisting phases that complicate analysis.
Furthermore, the XRD patterns of the many phases can only be distinguished by weak superstructure peaks at very low angles (Q < 1.5Å −1 ) created by O ordering, which contributes only weakly to the total X-ray scattering compared to the more electron-rich W and Nb atoms in the materials.
SEM was performed to investigate the microstructure of the samples, and to reveal the disposition of the multiple phases in each sample. There were no noticeable differences be- has comparable properties to material of the same composition made using traditional solid-state methods (red line).

A. Thermoelectric properties
To determine the thermoelectric performance of the composites, electrical transport Several samples were also made by conventional solid-state reaction of precursors in evacuated silica ampoules at 937 K for 48 hours, 10 followed by densification using the same SPS processing. Samples made using this two-step process displayed similar X-ray diffraction patterns and high-temperature electrical transport properties (electrical resistivity and Seebeck coefficient) as analogous samples made using a one-step process.
Thermal conductivity was determined for a smaller subset of samples ( Figure 9) to determine the influence of the microstructures observed in BSE micrographs ( Figure 6).
Densities used to calculate thermal conductivity were obtained through He pycnometry on consolidated pellets. Although the relative density (compared to the theoretical maximum single-crystal density) cannot be unambiguously known due to the presence of multiple phases and oxygen deficiency, the samples examined here are highly dense; scanning elec-  Taken together with electrical property measurements, thermal conductivity measurements reveal the largest zT in this compositional series is achieved by the x = 0.08 member, where zT = 0.1 at 950 K, the highest temperature measured here ( Figure 10).

IV. CONCLUSIONS
Early transition metal oxides were targeted using a datamining approach, and dense pellets of (W these composites are in the same performance class as CaMnO 3 -based materials, which have been heavily researched ( Figure 11). Physical property measurements reveal that the highest power factor and zT achieved in these samples is for the x = 0.08 member, which achieves zT = 0.1 at 950 K. Although this is the highest temperature measured here, the materials are stable to higher temperatures, where the zT should be higher.
Further studies should lead to improvement in electrical transport properties, and improved thermoelectric performance in this new class of n-type oxides that are stable at high temperature.