Molybdenum Oxides MoOx: Spark-Plasma Synthesis and Thermoelectric Properties at Elevated Temperature.

Molybdenum oxides MoOx in the composition range 2 ≤ x ≤ 3 were synthesized and compacted by the solid-state reaction of powdered α-MoO3 with Mo in the spark-plasma synthesis (SPS) process at temper...


■ INTRODUCTION
Being versatile structures with a wide range of physical and chemical properties, molybdenum oxides gain ongoing attention for the implementation into optical, electronic, catalytic, bio, and energy systems. 1,2 Most interesting for thermoelectric (TE) application is the variation of the electric and thermal transport properties of MoO x (2 ≤ x ≤ 3), which may be similarly tunable as those of WO x compounds. 3 The transition metal oxide's (TMO's) adaptive structures are presumed to reduce the lattice thermal conductivity due to enhanced phonon scattering caused by structural and bonding inhomogeneities. 4 For x = 3, the stable modification is the layered wide-band ntype semiconductor α-MoO 3 with an indirect band gap in the range 2.81−3.03 eV. 5−8 Herein, edge-sharing [MoO 6 ] octahedra are condensed in zig-zag chains, which are interconnected via corners and form slabs with AB stacking in the [100] direction, which are bonded by van der Waals interactions (Figure 1a).
The removal of oxygen and accompanied partial reduction of Mo(VI) introduce crystallographic shear (CS) while the layer structure is maintained, resulting in triclinic Mo 18 O 52 (x = 2.889, Figure 1b). Based on the crystal structure, Mo 18 O 52 is presumed to be a pseudo-1D conductor but, except from a note by Kihlborg on its existing electrical conductivity, no evidence is known. 9,10 For the same composition, a monoclinic high-temperature modification Mo 9 O 26 (Figure 1d) is reported. 10 This phase is structurally derived from the metastable ReO 3 -type modification β-MoO 3 (Figure 1c) by introduction of CS as originally described by Magneĺi,11 resulting in a representative of the homologous series Mo n O 3n−1 with n = 9. From calculations of the enthalpy of formation, the monoclinic modification should be thermodynamically stable. 12 Another member of the same series with n = 8 is the monoclinic Mo 8 O 23 (x = 2.875), which differs from Mo 9 O 26 in the distance between the CS planes ( Figure 1e).
Further reduction (x = 2.765) yields the pentagonal column (PC) phase Mo 17 O 47 (Figure 1f) wherein the PC structural motif (orange) exhibits an unusual terminating oxygen atom within the tunnels.
Another reported PC phase Mo 5 O 14 (x = 2.800) is presumed to be a metastable intermediate during the formation of Mo 17 O 47 . It might not be thermodynamically stable in the binary Mo−O system but is stabilized by Mo substitution with V, Ti, Nb, or Ta, whereas W does not seem to influence the stability range. 13,14 Interconnected by [MoO 4 ] tetrahedra, ReO 3 -type slabs are oriented in alternating fashion in the orthorhombic γ ( Figure  1g) and in parallel fashion in the monoclinic η modification (Figure 1h) of Mo 4 O 11 (x = 2.750). Both are known for their low-temperature transitions around 100 K, which are interpreted as charge density wave (CDW) formations in these pseudo-2D conductors. 15,16 The structural γ → η transition was reported to be sluggish in the solid state, and the transition temperature remains indefinite at around 873 K 10 (Figure 2a).
In the monoclinic distorted rutile-type MoO 2 , chains of edge-sharing [MoO 6 ] octahedra are interconnected via corners (Figure 1i). MoO 2 is reported to be a metallic conductor at room temperature with large variation of the electrical conductivity from 3.35 × 10 5 to 11 × 10 5 Sm −1 . 17−19 Hightemperature data is not available to the best knowledge of the authors.
Research on molybdenum oxides since the 1950s was focused on structure determination and physical properties at low temperatures. Thus, TE properties at elevated temperatures and the respective potential for high-temperature application are mostly unknown. Moreover, research was often constrained to single crystals obtained from solid-state reactions 10,20−23 and chemical vapor transport (CVT) 13,24−26 or to diverse nanostructures. 2 During spark-plasma synthesis (SPS), sample powder is encapsulated in an electrical conducting die. The application of a voltage results in high currents (order of 500 A) and a Joule heating of the die. 27 Apart from the fast solid-state synthesis with simultaneous compaction of gram-sized amounts of oxide materials, 28 SPS is an electrochemical method whose electric field causes a different diffusion behavior of anions and cations and is supposed to be responsible for the formation of new compounds, for example, alkali-metal silicon clathrates. 29 Besides the analysis of the phase formation under SPS conditions, this allows for the measurement of electric and thermal transport properties, which are presented here for single-phase polycrystalline specimens.

■ EXPERIMENTAL SECTION
Starting Materials. Mixtures of MoO x for solid-state reduction in the composition range 2 < x < 3 were prepared according to the general redox reaction from α-MoO 3 (ChemPur, 99.9%) and Mo (ChemPur, 99.95%) in powder form by mixing them for 15 min in an inert atmosphere. Samples of MoO 3 and MoO 2 (Alfa Aesar, 99.9%) were directly compacted from the starting material. Prior to mixing, powder X-ray diffraction (PXRD) showed all three starting materials to be free from crystalline contaminations. Optical emission spectrometry with inductively coupled plasma (ICP-OES) expectably yields minor amounts of nitrogen in both oxides, and common impurities in all starting materials are Na (<0.223 wt %), Ta (<0.1 wt %), C (≈0.02 wt %), and W (<0.013 wt %) (for details, see Table S1, Supporting Information). Spark-Plasma Synthesis. The spark plasma synthesis (SPS) was used to perform the redox reaction (eq 1) in the MoO 3 + Mo powder mixtures, as well as for compaction of the polycrystalline products into solid pellets for physical property measurements (Figure 2b). Details on the SPS process were published elsewhere. 3,27 Graphite dies with diameters of 10 mm and a graphite lining were filled with ∼1.5 g of starting mixture in an inert atmosphere and processed with an SPS-515 ET Sinter Lab apparatus (Fuji Electronic Industrial Co. Ltd., Tsurugashima, Japan 30 ). Fast linear heating (50 K min −1 ) was performed in vacuum (<0.4 mbar) under uniaxial pressure (80 MPa), followed by a dwell time t dwell at the maximum temperature T max and free cooling (Figure 2c). The maximum temperature for all syntheses was constrained to T max = 973 K to avoid deviation from the nominal compositions (see section Synthesis of MoO x Compounds). After every experiment, the sample pellet was polished and crushed to   Chemistry of Materials pubs.acs.org/cm Article check the phase formation by powder X-ray diffraction (PXRD). This procedure was repeated for n cycles until PXRD patterns did not show further changes and phase equilibrium might be assumed. Depending on the necessary n, the product yield ranged from 63% (MoO 2.889 ) to 25% (MoO 2.765 ) with respect to the starting mass. Within this work, single experiments are annotated in the form [n × T max /t dwell ]. Parameters of all syntheses (step "S") together with the parameters for subsequent compaction (step "C") are summarized in Table 1.
Powder X-ray Diffraction. Starting materials and powdered SPSprocessed samples were examined with powder X-ray diffraction (PXRD) on a G670 Guinier camera (HUBER Diffraktionstechnik GmbH & Co. KG, Rimsting, Germany) with Cu Kα1 radiation (λ = 1.54056 Å, graphite monochromator, 5°≤ 2θ ≤ 100°, Δ2θ = 0.005°). Qualitative phase analysis was performed by comparison of the measured patterns and theoretical patterns of known phases from the ICSD database. PXRD patterns of single-phase materials were measured with the internal standard LaB 6 (NIST SRM 660a, a = 4.1569162 ± 0.0000097 Å). Subsequent refinement of the unit cell parameters was performed using reflection positions with the leastsquare method WinCSD software package. 31 Thermal Analysis. For the characterization of the physical properties, all samples were SPS compacted (Table 1, step "C"). From differential scanning calorimetry (DSC) on a DSC 8500 (PerkinElmer, Inc., Waltham, MA, USA) in an argon atmosphere, the experimental c p values are found to equal the theoretic values from the NIST-JANAF database 30 for MoO 3 (room temperature: 75 J mol −1 K −1 ) and MoO 2 (room temperature: 56 J mol −1 K −1 ) with respect to the measurement error (±5%, Figure S4, Supporting Information). An exception is found for MoO 2.750 , which will be discussed later on. Thermogravimetry (TG) of MoO x samples (2 < x ≤ 3) was performed on an STA 449 F3 Jupiter (Netzsch GmbH & Co. KG, Selb, Germany) in vacuum with a heating rate of 10 K min −1 , resembling the SPS conditions, and a maximum temperature T max of 1011 K.
Thermoelectric Properties. The thermal diffusivity D(T) was measured by laser-flash analysis (LFA) on a LFA 457 MicroFlash (NETZSCH GmbH & Co. Holding KG, Selb, Germany) in a dynamic helium atmosphere (50 mL min −1 ) between room temperature and 773 K in steps of 25 K to calculate the total thermal conductivity κ tot (error, ± 3%) as The mass density ρ was determined after LFA with the Archimedean method.
Specimens with the approximate dimensions 1.5 × 1.5 × 6 to 8 mm 3 were cut from the SPS pellets using a wire saw, whereby the long edge was perpendicular to the pressure direction in SPS.
Measurements of the electrical conductivity σ(T) and Seebeck coefficient α(T) were performed along the long edge on a ZEM-3 (ULVAC RIKO, Inc., Saito, Japan) under low pressure of helium (150 mbar) from RT to 760 K in steps of 25 K (error ± 10%).
The electronic contribution κ el to the total thermal conductivity and consequently the lattice contribution κ lat were calculated with the Wiedemann−Franz (WF) equation For the calculation, κ tot values were fitted with fourth order polynomials. Since for semiconductors the Lorenz number L can strongly deviate from the table value, the approximation was used, which is based on the single parabolic band model of acoustic phonon scattering. 33 Comparative results obtained with constant L = 2.4453·10 −8 WΩ·K −2 show negligible deviations, and all trends remain the same. For a comparison with other TE materials, the power factor was calculated according to with a maximum error for α of +1%/−13.2% earlier established for the used setup. 34 Measurements of σ(Τ) and α(T) were performed for at least two heating and cooling cycles to verify reproducibility. Since α(T) of MoO 2.750 showed some unsystematic aberration during the first heating cycle from the following cycles ( Figure S7 The phase Mo 5 O 14 was not obtained via SPS because by now, this phase has been prepared only with Mo substitution by V, Ti, Nb, or Ta and might not be stable in the binary system. 13,14 Obtained single-phase specimens confirm the SPS technique to be suitable for the preparation of MoO x phases despite small composition differences between them (≥0.1 at % oxygen, see also section Phase Homogeneity Ranges and Composition Control). Reduction due to the usage of graphite-lined dies at high temperatures (T ≤ 973 K) in an evacuated chamber (<0.4 mbar) is visible at the sample surface in graphite contact but does not proceed into the bulk within the applied dwell times (≤1 h per cycle). For maintaining the target stoichiometry during multiple-step synthesis, polishing of the compacted samples is inevitable. Thus, the product yield with respect to the starting mass of the educt mixture is reduced to a minimum of 25% in the case of Mo 17 O 47 whose  n, number of cycles; S, synthesis; C, compaction. b ρ rel = (ρ arch /ρ diff )·100% (ρ arch , Archimedean density; ρ diff , theoretical density from X-ray diffraction results (Table 2)). c Overestimation of ρ arch due to residues of secondary phase MoO 2 ( Figure 3).
Chemistry of Materials pubs.acs.org/cm Article low formation temperature (<550 K) necessitates seven synthesis cycles.
As-received α-MoO 3 powder was compacted directly in onestep SPS treatment. Unlike in WO 3,3 no reduction of the sample bulk to lower oxides is observed according to PXRD due to the much lower applied temperatures: an increase of the SPS chamber pressure showed significant decomposition to start in a temperature range of 1007−1034 K in several attempts, while from comparative thermogravimetry (TG) under resembling conditions (vacuum, heating rate: 10 K min −1 ), a considerably lower decomposition onset at 932−934 K is found ( Figure 4 and Figure  increased by ∼30 K with respect to that of α-MoO 3 . Also, the lower quality of the SPS vacuum (<0.4 mbar) must cause a higher oxygen vapor pressure above the oxide surface and may be a further explanation for the increased decomposition temperature compared with that in TG vacuum (4 × 10 −3 mbar). It is worth mentioning that the grain size of the as-received powder changed by three orders of magnitude within only 120 min of SPS ( Figure S3, Supporting Information).
Samples with 2 < x < 3 required multiple SPS cycles (n in Table 1) with intermediate crushing to increase concentration gradients between grains, homogenize grain sizes and thus reach phase equilibrium faster. ICP-OES measurements on MoO 2.889 powder after three successive SPS steps reveal no change of the contamination concentrations due to the polishing and crushing (Table S2, Supporting Information).
In all synthesis experiments, according to eq 1, the triclinic Mo 18 O 52 (x = 2.889) is observed after the initial heating cycle, although from calculations of the enthalpy of formation, the monoclinic Mo 9 O 26 is predicted to be the more stable modification. 12 However, due to the constrained T max = 973 K, the suggested formation temperature of  (Table 2).
In contrast, all MoO 2.875 specimens that were prepared in the suggested stability range of Mo 8 O 23 (x = 2.875) 10 (Figure 3, red). The low-temperature η-modification, which should exist at temperatures <873 K, was never observed, as discussed more detailed in the section Phase Formation.
Abundant crystallographic data is available for γ-Mo 4 O 11 ( Table 2). The obtained lattice parameters b and c (space group: Pna2 1 ) agree well with both, single-crystal (-sc) and powder (-p) data. The a parameter is significantly shorter in the polycrystalline samples of our work (24.4760(10) Å) and by Fun et al. (24.4756 Å) in comparison to single-crystals by Kihlborg et al. (≈24.49 Å) and a single crystal, which we synthesized (24.501(2) Å, details will be published elsewhere). Since single-crystal data is generally not standardized, it   (Figures 5b and 6b).  Most recent literature data is quoted for MoO 2 and α-MoO 3 . Diffraction data was obtained from single-crystals (-sc) or powder (-p) in the asreceived state (AR) or after preparation via SPS, solid-state reaction of MoO 3 and Mo or MoO 2 in vacuo (SSR), chemical vapor transport (CVT), electrocrystallization (EC), and sublimation growth (SG). Powder from SPS was annealed (-a) for 1 week at 573 K. b Originally published setting changed for better comparison of the axis lengths. c Secondary phase Mo 8 O 23. d I 2 was used as a transport agent; 5 TeCl 4 used as a transport agent. e ZnI 2 was used as a transport agent. f TeCl4 was used as a transport agent.
within only 5 min (not shown). The synthesis of the η polymorph is usually performed directly at temperatures below the γ stability range, which excludes the possible necessity of supercooling γ to reach a γ → η transition. 10,13,22,25,26,39 Thus, under SPS conditions, the equilibrium may be influenced by the applied electric field, which introduces a driving force additional to the concentration gradient and therefore affects the diffusion paths of mobile charged species, that is, Mo cations and O anions. For comparison, reports on SPS processed hyperstoichiometric UO 2+x show the uranium oxidation state to gradually increase toward the anode (+) due to directional migration of oxygen defects. 40 Based on this, an increase of the Mo oxidation state is expected at the anode (+) leading to the formation of More extensive studies on the still not understood γ → η transition kinetics during both SPS and classical solid-state reaction are ongoing.
Phase Homogeneity Ranges and Composition Control. The complex structures in the MoO x compounds form within a narrow composition range (2 ≤ x ≤ 3) with composition differences as low as Δx ≈ 0.015 (≅ 0.1 at % oxygen) between the Mo 4 O 11 polymorphs (x = 2.750) and Mo 17 O 47 (x = 2.765). Thus, it is reasonable to ask if the small stoichiometric differences of the starting mixtures can be maintained during the SPS process.
To study this issue, two mixtures with x = {2.740, 2.760} were SPS-treated additionally to the aforementioned x = 2.750 (compare Figure 3, red) (Table 3). This supports the statement that none of the MoO x phases has a homogeneity range, 10,20 and challenges on stoichiometric deviations as reason for different resistivity transitions in γ-Mo 4 O 11 crystals under high pressure were reported pre-   12 The electrical conductivity (σ = neμ) of MoO 2.889 shows the temperature dependence of an intrinsic semiconductor (n, carrier concentration; e, elementary charge; μ, carrier mobility). Assuming that μ is temperature independent, the activation energy E a can be extracted from Arrhenius plots due to the proportionality σ(T) ∝ exp[−E a /2k B T] (k B , Boltzmann constant) as exemplary shown in Figure 8 (inset). From two heating cycles of two MoO 2.889 samples each, this yields E a = 0.30(3) eV and indicates Mo 18 O 52 to be a narrow-gap semiconductor. Since no band structure and theoretic band gap E g are known, E a is not necessarily associated with E g but can also result from an energy difference between impurity states and the valence band. A subtle hysteresis of σ(T) between heating and cooling may be a real effect but is not significant with respect to the error of our measurement ( Figure S5, Supporting Information).
Similar, for the bad-metallic MoO 2.765 (Mo 17 O 47 ), there is found an insignificantly decreased σ(T) below 500 K in the first heating cycle ( Figure S6, Supporting Information).
At room temperature, for MoO 2.750 , the electrical conductivity reported for single crystals is strongly anisotropic and varies from 0.23 × 10 5 Sm −1 parallel to the bc plane 16 to 5−13 × 10 5 Sm −1 perpendicular to it. 16,25,41,42 A value of σ = 3.12 × 10 5 Sm −1 measured for our polycrystalline samples ranges in between, which is plausible because of approximately isotropic distribution of crystallite directions and increased scattering at grain boundaries.
The electrical conductivity of MoO 2 at room temperature is close to a value of 5 × 10 5 Sm −1 reported for 298 K. 19 The large variation of published room-temperature conductivities σ = 3.35 × 10 5 to 11 × 10 5 Sm −117−19 may be traced back to a neglection of crystallographic directions in these measurements: MoO 2 is expected to be highly anisotropic due to alternating short (2.51 Å) and long (3.11 Å) Mo−Mo distances along the a axis. 32 In our polycrystalline samples, an average conductivity is plausible, assuming nonpreferred orientation of the grains.
A reproducible bump at 450 K in thermal conductivity of γ-Mo 4 O 11 (x = 2.750, arrow) is caused by an increase of the specific heat c p (Figure 4, inset), which is not considered in the κ tot (T) calculation from tabulated theoretic c p values 32 (eq 2). In our DSC studies, we observed this reversible endothermal effect in both SPS-processed powder and single crystals grown from chemical vapor transport (will be published elsewhere). Thus, we assume a hitherto unknown phase transition, which requires further studies. Previously reported κ tot for single crystals along one unspecified short axis (20 Wm −1 K −1 at 298 K 48 ) is expectably higher than that of our polycrystalline material with macroscopic isotropy and grain boundaries acting as scattering centers.
Almost temperature-independent κ lat is observed in Mo 17 O 47 (x = 2.765). The highest κ lat is found for MoO 2 (9−20 Wm −1 K −1 ), which is the structure with most condensed [MoO 6 ] octahedra (Figure 1i). The general trend of increasing κ lat with decreasing O/Mo ratio x is attributed to the proceeding condensation of [MoO 6 ] octahedra (Figure 9b). Temperature dependence increases for x ≤ 2.750, but with respect to the approximative character of the calculation from the WF law, these κ lat values should not be overinterpreted.
All MoO x samples with 2 ≤ x ≤ 2.765 exhibit a negative Seebeck coefficient α according to the expected n-type conductivity ( Figure 11). Seebeck coefficients of MoO 2 and γ-Mo 4 O 11 are low due to the presumed high charge carrier concentration in these bad-metal-like materials. For MoO 2 , the values of α range from −6 μV K −1 to a local minimum of −9 μV K −1 at 540 K. Cycling of three independent γ-Mo 4 O 11 (x = 2.750) samples yields values of −12 to −21 μV K −1 and a linear temperature dependency α(T) ∝ −0.017(1) μV K −2 ·T (exemplary depicted in the inset of Figure 11), which is typical for degenerate semiconductors. Minor unsystematic deviation from this linearity occurred in the first heating cycle ( Figure  S7, Supporting Information). A calculation of the carrier mobility was not possible yet due to a not yet understood oscillatory behavior of the Hall voltage with temperature observed in Hall-effect measurements. With −44 to −70 μV K −1 , Mo 17 O 47 (x = 2.765) exhibits the highest negative α values with no maximum in the measured temperature range.
In contrast, a positive and exceptionally high Seebeck coefficient in the range α = +140 μV K −1 is found for Mo 18     phase is assumed to be a narrow-gap p-type semiconductor below 440 K, which shows exceptionally low thermal conductivity of κ tot ≈ κ lat = 0.5−0.9 Wm −1 K −1 in the whole temperature range. Also, a large positive Seebeck coefficient of +140 μV K −1 is observed at room temperature, which however is relativized due to a presumably low charge carrier concentration. For heating above 440 K, a p−n transition occurs and the Seebeck coefficient is distinctly decreased (−38 μV K −1 at 763 K).
The resulting figure-of-merit ZT is low for all MoO x phases (2 ≤ x < 3) with the following maxima: ZT = 5 × 10 −3 , 39 × 10 −3 , and 14 × 10 −3 at a maximum temperature of 763 K for Mo 18  The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.chemmater.9b05075. Figure S1: decomposition of α-MoO 3 starting material during SPS; Figure S2: decomposition onsets of α-MoO 3 as found from thermogravimetry and from the chamber pressure during SPS; Figure S3: light microscopic images of α-MoO 3 after SPS compaction; Figure  S4: specific heat of MoO x samples from DSC measurements; Figure S5: cycling of electrical conductivity measurements on MoO 2.889 ; Figure S6: cycling of electrical conductivity measurements on MoO 2.765 ; Figure S7: cycling of Seebeck coefficient measurements on MoO 2.750 ; Figure S8: estimation of the secondary phase content in practically single-phase MoO 2.889 from the X-ray diffraction pattern; Table S1: chemical compositions of the starting materials;   Synthesis and physical property measurements including data analysis were performed by F.K. M.S. performed specific heat measurements including the data analysis. All authors contributed equally to the data interpretation and discussion. The manuscript was written by F.K. and revised by Y.G., I.V., and M.S. All authors have given approval to the final version of the manuscript.