Thermoelectric properties of Fe and Al double substituted MnSiγ (γ~1.73)

Two series of Fe and Al double substituted MnSiγ chimney ladders with a nominal valence electron count, VEC1⁄414 per transition metal were prepared (γ1⁄41.75). Simultaneous replacement of Mn with Fe and Si with Al yielded the Mn1 xFexSi1.75 xAlx series while the second Mn1 xFexSi1.75–1.75xAl2x series follows the pseudo-binary between MnSi1.75 and FeAl2. Scanning electron microscopy and elemental mapping revealed that 60% of the nominal Al content ends up in the product with the remainder lost to sublimation, and that up to 7% Al can be substituted in the main group sublattice. Profile analysis of Xray powder diffraction data revealed gradual changes in the cell metrics, consistent with the simultaneous substitution of Fe and Al in a fixed ratio. All samples are p-type with VECE13.95 from the structural data and 1 10 holes cm 3 from variable temperature Seebeck measurements. The substituted samples have lower electrical resistivities (ρ300 K1⁄42–5 mΩ cm) due to an improved microstructure. This leads to increased thermoelectric power factors (largest S/ρ1⁄41.95 mW m 1 K ) compared to MnSiγ. The thermal conductivity for the Mn0.95Fe0.05Si1.66Al0.1 sample is 2.7 W m 1 K 1 between 300 and 800 K, and is comparable to literature data for the parent material. & 2015 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Thermoelectric energy converters can be used to increase the efficiency of any heat generating process. However, the lack of cost-effective thermoelectric materials has so far limited their use to certain niche applications [1]. The efficiency of a thermoelectric material is defined by its dimensionless figure of merit, ZT ¼(S 2 / ρκ)T, where S is the Seebeck coefficient, ρ is the electrical resistivity, κ¼κ el þ κ lat is the sum of the electronic and lattice thermal conductivities, and T is the absolute temperature [1]. Over the past decades a large number of material systems have been investigated for thermoelectric power generation, including lead tellurides, skutterudites, half-Heuslers and silicon-germanium alloys [2][3][4][5][6]. Another group of materials under investigation are intermetallic compounds with the Nowotny chimney-ladder (NCL) structure. These form for a variety of combinations of group 4-9 transition metals and group 13-14 main group elements with VEC E14 [7][8][9][10]. When the VEC is close to 14, the Fermi energy is located in the middle of a narrow band gap and the chimney ladder compounds exhibit semiconducting properties.
In particular, manganese silicides with composition MnSi γ (1.70 rγ r1.75) have attracted significant interest because of their promising ZT ¼0.5 but also because of their low cost, environmental friendliness, chemical and mechanical stability and resistance to oxidation at high temperatures [11,12]. The MnSi γ structure consists of two tetragonal subsystems: a [Mn] chimney and a [Si] ladder which share a common a axis but have different and often incommensurate repeat periods along the c-direction. A convenient structural description uses a 3þ 1 dimensional model (superspace group I4 1 /amd(00γ)00ss) consisting of a basic tetragonal cell a Â a Â c Mn (a E5.5 Å, c Mn E4.4 Å) and a modulation parameter (γ ¼c Mn /c Si ) [12,13] [12], while recently ZT ¼0.5 was reported for MnSi γ grown using vapour transport [18]. A large number of chemical substitutions have been reported for MnSi γ . Most of these focus on the transition metal sublattice and include Mn 1 À x Fe x Si γ (ZT¼0.45 at 800 K) [19,20], Mn 1 À x Cr x Si γ (ZT¼0.45 at 900 K) [20][21][22], co-doping with Cr and Ru (ZT¼0.6 at 850 K) [21], Mn 1 À x Ru x Si γ (ZT¼0.75 at 875 K for x¼ 0.1) [23,24] and Mn 1 À x Re x Si γ (ZT¼ 0.6 at 800 K) [25]. Substitutions into the main group sublattice have also been explored, and in particular hole doping using Al has been found to be a promising route towards obtaining good figures of merit (e.g. ZT ¼0.65 at 850 K) [26,27]. Germanium has been used to improve the power factor (S 2 /ρ) [28], and simultaneous substitution of Al and Ge has been reported recently (ZT¼ 0.6 at 823 K) [29]. In all these studies, relatively small amounts of substitution (o1% Al and o2% Ge) into the main group sublattice were reported before segregation was observed. In an attempt to achieve higher substitution levels we have exploited the co-doping of Fe and Al to maintain a constant VEC ¼ 14. Two series of materials were prepared: the first (Mn 1 À x Fe x Si 1.75 À x Al x ) was created by 1:1 substitution of Mn by Fe and Si by Al to maintain an overall AX 1.75 stoichiometry. The second series (Mn 1 À x Fe x Si 1.75-1.75x Al 2x ) connects two binary phases that obey the VEC ¼14 rule: MnSi 1.75 and FeAl 2 . This series has an increasing main group to transition metal ratio (AX 1.75 þ 0.25x ). The structure and microstructure of these samples were investigated using X-ray powder diffraction and scanning electron microscopy and energy dispersive X-ray elemental (EDX) mapping, while the thermoelectric performance was determined through measurement S, σ and κ.

Experimental procedure
Three gram polycrystalline samples with nominal compositions Mn 1À x Fe x Si 1.75 À x Al x (0rxr0.25) and Mn 1À x Fe x Si 1.75-1.75x Al 2x (0rxr0.1) were synthesised by arc-melting and annealing. Stoichiometric amounts of Mn (99.6%), Fe (99.9þ%), Si (99.99%) and Al (99.97%) were mixed using a mortar and pestle and pressed into 13 mm diameter pellets. The pellets were arc-melted on a water cooled copper plate under an argon atmosphere. Samples were melted 2 or 3 times and after each melting process they were flipped upside down in order to promote homogeneous mixing. The arcmelted ingots were subsequently wrapped in tantalum foil and annealed in evacuated, sealed quartz tubes at 900 1C for 5 days. A black residue coated the inside of the arc-melting chamber, while minor discoloration of the quartz tubes was observed after annealing. The overall weight loss was r2 wt%. X-ray powder diffraction patterns (XRD) were recorded to check the phase purity and determine the lattice parameters for all samples. Data were collected on a Bruker D8 Advance instrument with monochromated CuKα 1 radiation. The X-ray diffraction patterns were analysed using the JANA2006 software package [30]. The microstructure and chemical composition of a subset of the samples was analysed using a Quanta 650 FEG ESEM equipped with an Oxford Instruments X-max 150 N detector for elemental mapping. The working distance, beam spot size and collecting time were 10 mm, 4.5 and 4-10 frames, respectively. Prior to analysis the samples were polished using fine Al 2 O 3 sand paper down to 0.5 μm roughness. Rectangular bars were cut using a diamond saw and used for transport measurements. High temperature (300-1100 K) measurements of S and ρ were carried out using a Linseis LSR-3 instrument. The thermal conductivity (κ) of the Mn 0.95 Fe 0.05 Si 1.662 Al 0.1 sample was calculated from the measured thermal diffusivity α, specific heat capacity C p , and density d using the relation κ¼αC p d. The thermal diffusivity and heat capacity were measured using a Netzch LFA 457 and Perkin Elmer DSC 8500, respectively. The accuracy of the thermal conductivity is estimated to be 77.5%. The sample densities were calculated from the mass and volume of the bars used in the transport measurements.

Phase analysis
The powder XRD patterns of all samples are given in Fig. 1. Starting with the AX 1.75 series, the common impurity phase MnSi was observed for x¼0 and x¼0.05 and again at the highest substitution levels (x¼0.2 and 0.25). For x¼ 0.25 a new phase emerges in addition the NCL reflections. We have not been able to index this phase but it does not correspond to any known binary or ternary phase, including FeAl 2 for which a triclinic structure is reported [31,32]. The AX 1.75 þ 0.25x series with increasing main group content has a smaller substitution window and could only be prepared for x¼ 0.05 (0.1 Al). For x¼ 0.1, a second phase which is very similar in appearance to that observed in the AX 1.75 series is observed. The lattice parameters, volume of the basic cell and the main group content (γ) for the Mn 1À x Fe x Si 1.75 À x Al x and Mn 1À x Fe x Si 1.75-1.75x Al 2x series are shown in Fig. 2, and are listed in Table S1. Literature data for the Mn 1À x Fe x Si γ series in which Mn is replaced by Fe, leading to electron doping, is included in Fig. 2 [19]. In this reference series, a shrinkage of the a-and c axes is observed. For the AX 1.75 series, a small reduction in the a axis and an almost constant c axis are observed, which is consistent with the simultaneous replacement of Mn with smaller Fe and Si with larger Al. The AX 1.75 þ 0.25x series has increasing lattice parameters, in keeping with the larger main group content. Both the x¼0.25 sample from the AX 1.75 series and the x¼ 0.1 sample from the AX 1.75 þ 0.25x series (open symbols in Fig. 2) follow the trends set by the pure samples without the competing phase, suggesting that the solubility limit may not be reached. The main group content (γ) for the AX 1.75 series decreases gradually upon Fe and Al substitution (Fig. 2). The main group content for the AX 1.75 þ 0.25x series is larger and is increasing, in keeping with the expected trend. Extrapolation towards x¼0 suggests an estimated γ¼1.725 for the AX 1.75 series, and γ¼1.730 for the AX 1.75 þ 0.25x series.
This reveals a subtle step change reduction in main group content compared to the MnSi 1.737 parent phase (ΔγE À 0.01).
Scanning electron microscopy and elemental mapping were used to investigate the microstructure and chemical compositions for the samples upon which property measurements were undertaken. These are: MnSi 1.75 , x ¼0.1 and 0.2 from the AX 1.75 series and x ¼0.05 from the AX 1.75 þ 0.25x series. The results are summarised in Fig. 3 and Table 1. The MnSi 1.75 sample has the most surface structure while the other samples appear better sintered, and had a better finish after polishing. Elemental mapping for MnSi 1.75 (not shown) showed a homogenous distribution of Mn and Si, and also revealed small amounts of elemental Si impurities not evident from the X-ray powder diffraction. An average of three large area elemental maps (212 Â 146 μm 2 ) taken on different phase pure regions of the ingot yielded a chemical composition MnSi 1.67 (3) , where the standard deviation is the variation between the three datasets. Similar analysis on homogenous areas of the Fe and Al substituted samples yielded the following chemical compositions: Mn 0.90 Fe 0.098(7) Si 1.60(1) Al 0.063 (7) , Mn 0.80 Fe 0.18(1) Si 1.48(2) Al 0.11 (1) , and Mn 0.95 Fe 0.049(1) Si 1.64(1) Al 0.059 (2) . The transition metal content is therefore in excellent agreement with the nominal values, while the total main group content is systematically underestimated. In terms of Al content, the EDX data consistently suggest this is $60% of the expected content for both series. This large reduction is consistent with the weight loss observed during arc-melting ( r2 wt%) which is sufficient to explain the loss assuming Al is preferentially sublimated. A similar loss of Al was observed during directional solidification of Ru 0.27 Re 0.73 (Si 1 À x Al x ) 1.67 alloys [33], and this may be a general feature of materials prepared from the melt. The elemental mapping also picked up residual Al-rich areas in the ingots but these are small and are not sufficient to explain the $ 40% loss. In addition, there is no evidence for the presence of elemental Al in the X-ray powder diffraction data. The Al-rich areas are often linked to porosity in the microscopy images (see Fig. 3b). The Mn 0.80 Fe 0.18(1) Si 1.48(2) Al 0.11 (1) sample is characterised by a large number of pores with diameters up to 50 μm while the areas in between are well sintered and of homogeneous chemical composition (Fig. 3d). This suggests that the subliming Al (melting point 660 1C) serves as a mineraliser promoting sintering but also leads to the presence of porosity due to trapped Al vapour. The competition between improved sintering and porosity is also reflected in the sample densities (Table 1) with the highest values observed for the samples with a nominal amount of 0.1 Al per formula unit.

Thermoelectric properties
The temperature dependence of S, ρ and S 2 /ρ for MnSi 1.75 , x¼0.1 and 0.2 from the AX 1.75 series and x¼ 0.05 from the AX 1.75 þ 0.25x series are given in Fig. 4. The observed S(T) are positive and are characterised by a linear increase up to a broad maximum at 700-800 K, beyond which S is reduced. This behaviour is typical for degenerate semiconductors with a contribution from minority carriers (n-type) at higher temperatures. The linear part of S(T) suggests a fixed number of charge carriers and was fitted using S¼ (8π 2 k B 2 /3eh 2 )m n (π/3n) 2/3 T [2], where k B is Boltzmann's constant, e is the electronic charge, h is Planck's constant, m n is the effective mass and n is the carrier concentration. The carrier concentrations are in the range of 1-1.4 Â 10 21 cm À 3 (Table 1) using a hole effective mass of three times the bare electron mass [34]. The thermal band gaps (E g ) were estimated using E g ¼2eS max T max where S max and T max are the maximum Seebeck coefficient and the temperature at which it occurs [35]. Band gap values of 0.38 eV were found for all compositions ( Table 1). The ρ(T) data also show a linear increase up to a maximum which occurs at slightly higher temperatures (700-900 K) compared to the maximum in S(T). This again reflects the contribution of minority charge carries at elevated temperatures. The most striking feature of the ρ(T) data is the scatter in magnitude, which varies between 2 and 10 mΩ cm at room temperature. Given the similar S(T) these variations cannot be due to changes in carrier concentration but must reflect microstructure factors. The microscopy data suggest that Al acts as a mineraliser leading to improved sintering which is expected to remove grain boundary contributions to the electrical transport. The presence of porosity due to trapped Al vapour is not expected to affect ρ(T) significantly as the concentration of pores The reference data for the Mn 1 À x Fe x Si γ series were taken from Ref. [19]. Samples that contain the competing phase (see text) are represented by open symbols. remains low. Indeed, all substituted samples have significantly lowered ρ(T) values (2oρ 300 K o5 mΩ cm) compared to the parent material (ρ 300 K ¼ 10 mΩ cm), in keeping with a reduction of grain boundary contributions. The temperature dependence of S 2 /ρ is similar for all samples with a broad maximum centred on 700-800 K. The variation in magnitude is accounted for by the observed scatter in ρ(T). A maximum S 2 /ρ¼ 1.9 mW m À 1 K À 2 is observed for the x¼ 0.1 sample from the AX 1.75 series. The κ(T) of the x¼0.05 sample from the AX 1.75 þ 0.25x series was measured as a representative sample. The κ(T) is almost temperature independent up to 800 K (κ¼2.7 W m À 1 K À 1 ) beyond which it increases rapidly (Fig. 4d). This sudden increase is linked to a bipolar contribution due to the presence of both p-and n-type charge carriers. It is therefore not possible to extract a reliable lattice thermal conductivity using the normal subtraction of a  Table 1 EDX composition, total main group content (γ), valence electron count, density, resistivity (ρ) and Seebeck coefficient (S) at 315 K, thermal band gaps (E g ) and carrier concentrations (n p ) for selected Mn 1 À x Fe x Si 1.75 À x Al x and Mn 1 À x Fe x Si 1.75 À 1.75x Al 2x compositions.

Discussion
X-ray powder diffraction and microscopy imaging confirm that Fe and Al double substitution is an effective route to increase the amount of Al substitution in the MnSi γ chimney ladder structure. EDX elemental analysis suggests that about 60% of the nominal Al content is found in the final product while the remainder is lost due to sublimation. This constant percentage loss is supported by the systematic changes in the cell metrics which suggest successful codoping of Fe and Al in a fixed ratio. The highest substitution level is 7% of the main group lattice, and was observed for the x¼ 0.2 sample of the Mn 1À x Fe x Si 1.75À x Al x series. The measured S(T) are nearly identical and demonstrate that all samples have similar carrier concentrations. The lack of significant doping suggests that the Si/Al ratio and total main group content (γ) adjust to maintain a carrier concentration close to that of the parent material (VEC¼ 13.95). This is confirmed by taking the Si/Al ratio from EDX and the total main group content (γ) from diffraction and calculating the VEC for the substituted samples (Table 1). This leads to VEC¼13.92 (x¼0.1; AX 1.75 series), VEC¼13.94 (x¼ 0.2; AX 1.75 series) and VEC¼13.91 (x¼ 0.05, AX 1.75 þ 0.25x series).
This suggests that the changes in Si/Al ratio and γ are purely driven by electronic origins, and may not be caused by the rapid nature of the arc-melting process itself. For both series a competing phase was observed as x was increased, signalling that there is a limit to the amount of Al ($ 7% from our EDX data) that can be incorporated in the NCL structure. To test whether the presence of Al is the limiting factor in the phase stability, the two-phase x¼0.25 sample from the AX 1.75 series was prepared with a gradually decreasing Al content. This yielded the Mn 0.75 Fe 0.25 Si 1.6875-0.75y Al y series (0ryr0.25) whose X-ray powder diffraction patterns and lattice parameters are given in Figs. S2-S3 and Table S2. Inspection of the diffraction data reveals that the replacement of Al with Si eliminates the competing phase for yr0.1. This suggests that the presence of Al is indeed the limiting factor in the phase stability of these materials. The electron microscopy, sample densities and variations in ρ(T) all suggest that Al serves as a mineraliser leading to improved sintering but also to porosity due to trapped vapour. This leads to ρ(T) and S 2 /ρ values for Mn 0.9 Fe 0.1 Si 1.65 Al 0.1 that are comparable to the best reported MnSi γ based materials [18,26]. Elimination of the porosity due to trapped Al vapour may enable further improvements in ρ(T) and S 2 /ρ. The inset to Fig. 4d shows the temperature dependence of ZT for the x¼0.05 sample from the AX 1.75 þ 0.25x series, and this shows a maximum ZT¼ 0.3 at 825 K. Assuming a similar κ(T) for the sample with the largest S 2 /ρ values, leads to an estimated ZT¼ 0.5.
To conclude, double substitution of Fe and Al is a route to increase the Al content in the MnSi γ chimney ladder structure. The maximum observed Al content is 7% of the total main group sublattice. The prepared samples adjust their Si/Al ratio and total main group content (γ) in order to maintain a constant valence electron count near 13.95 electrons per formula unit. In terms of the thermoelectric performance, the Al substituted samples have improved electrical resistivities and power factors up to 1.95 mW m À 1 K À 2 are observed for a nominal Mn 0.9 Fe 0.1 Si 1.65 Al 0.1 sample at 800 K.