Topotactic anion-exchange in thermoelectric nanostructured layered tin chalcogenides with reduced selenium content

Topotactic solution synthesis yields nanostructured tin chalcogenides, SnS1–xSex with controllable composition; spark plasma sintered SnS0.1Se0.9 achieves ZT ≈ 1.16 at 923 K via microstructural texture tuning.


Introduction
Main group metal chalcogenides (MCs) are excellent candidates for thermoelectrics, 1-4 (opto)electronics 5 and photovoltaics 6 due to their outstanding electronic, optical and thermal properties. Bottom-up solution syntheses afford energy-saving means of preparing MC nano/micro-structures with controllable morphology and size. Surface modication using organic surfactants can limit particle growth but such coatings can typically introduce impurities. 7-10 By contrast, synthesis without organic surfactants, solvents or precursors can produce nanostructured MCs with impurity-free surfaces and enhanced electrical performance [11][12][13][14] but require careful experiment design with appropriate synthesis parameters and reagents.
Chemical transformations, including ion exchange, topotactic and pseudomorphic reactions, represent a versatile and effective means to produce new materials with control over crystal structure, composition and morphological complexity. 15 Such transformations can realise prescribed materials that cannot be otherwise prepared. [15][16][17] Indeed, doped compounds, multi-component composites or hetero-structures can be crafted by regulating the progress of transformations, leading to materials with engineered functional properties. [18][19][20][21] When performed in solution, ion exchange can exploit the solubility difference between precursors and products to enable the rapid synthesis of nano/micro-structures (for example, MCs) with predetermined cation and/or anion compositions. 15,18 Combining organic-free synthesis with ion exchange raises the prospect of producing MCs with compositions, crystal structures, morphologies and particle sizes that can be tailored towards delivering high electronic performance.
Thermoelectric materials can be utilised to convert thermal energy directly into electricity and vice versa, thus offering opportunities to refrigerate and to harvest electricity from waste heat via the Peltier and Seebeck effects, respectively. 22,23 Layered tin chalcogenides (LTCs), including SnSe and SnS, have drawn much attention given a formidable combination of excellent thermoelectric conversion efficiency, relatively low toxicity and the Earth-abundance of their component elements. 1,2,[24][25][26] Notably, when p-type SnSe can be grown as a single crystal, it has demonstrated record high ZT values of 2.6 and 2.3 along the b and c crystallographic directions, respectively at 923 K. 1 Polycrystalline SnSe and related doped materials have been prepared in an effort to improve mechanical properties, but ZT values cannot yet emulate those in the single crystalline material. 27 The capacity to synthesise polycrystalline LTCs of premeditated composition to optimise thermoelectric performance is becoming gradually less elusive.  For example, Agdoped SnSe, 28 alkali metal-doped SnSe, [29][30][31][32][33][34] 35 and Sn 1Àx Pb x Se 36 have demonstrated improvements in thermoelectric performance compared to undoped SnSe, pushing ZT to 1.2 at 773 K (Na, K co-doped SnSe) 33 and $1.7 at 873 K (phase-separated Sn 1Àx Pb x Se). 36 Yet, polycrystalline LTCs are primarily fabricated by high-temperature, energy-intensive processes. [27][28][29]31,32,35 Solution syntheses are an attractive alternative but generally involve using organics (solvents and/or surfactants, for example), can produce small sample yields and have offered little opportunity as yet to exert control over composition. [51][52][53][54][55][56][57][58][59][60] For LTCs to be a practicable component of thermoelectric devices, a scalable and cost-effective organic-free synthesis approach to materials with tuneable composition and consistently excellent performance is essential.
In this study, we demonstrate how the combination of two organic-free aqueous solution strategies (anion exchange following direct precipitation) can be utilised to synthesise LTC nano/micro-plates with tuneable chalcogenide composition (e.g. >10 g SnS and SnS 0.1 Se 0.9 , respectively; Fig. S1 and S2 †). The plates can be sintered into textured, dense pellets with competitive thermoelectric performance while partly replacing selenium with less toxic and more Earth-abundant sulfur.

Results and discussion
Characterisation and formation mechanism of anionexchanged SnS 1Àx Se x The synthesis of phase-pure SnS (Fig. 1) involves injection of a Na 2 S aqueous solution into a Na 2 SnO 2 solution that is subsequently boiled for 2 h (Fig. S1 †). Powder X-ray diffraction (PXD) patterns (Fig. 1a) can be indexed exclusively to orthorhombic SnS (ICDD card no. 75-2115). 61 Rietveld renement against PXD data ( Fig. S3; Tables S1 and S2 †) conrms that the SnS product crystallises with orthorhombic space group Pnma; a ¼ 11.2052(4)Å, b ¼ 3.9877(2)Å and c ¼ 4.3242(2)Å. Scanning electron microscopy (SEM) images ( Fig. 1b and S4a-c †) reveal that the product predominantly forms ower-like nano/microstructures that are composed of a series of plates, typically with a lateral size of 3-16 mm and a thickness of 80-650 nm. Imaging and electron diffraction of individual nanoplates were performed using transmission electron microscopy (TEM). Selected area electron diffraction (SAED) patterns collected with the incident beam perpendicular to an isolated plate (Fig. 1c) can be indexed to the [100] zone axis of SnS and conrm the single crystalline nature of the nanostructures. High resolution TEM (HRTEM) images (Fig. 1d) demonstrate a set of lattice spacings of 2.9Å intersecting with an angle of 95(1) , corresponding to the {011} planes of SnS. Energy dispersive X-ray spectroscopy (EDS) (Fig. 1e and S4d †) conrms Sn : S atomic ratios of 51(1) : 49(1). The organic-free SnS plates can be formed into dense pellets at 500 C by either hot pressing or spark plasma sintering (SPS), each leading to strong texturing of the (h00) planes ( Fig. S5 and S6 †). The plate morphology, orthorhombic crystal structure and the overall Sn : S ratio are preserved on pressing (Fig. S7 †). A pellet of the SnS material has an indirect optical bandgap of ca. 1.05 eV (Fig. S8 †) and exhibits a negligible weight loss up to ca. 600 C when heated under Ar gas, indicating excellent thermal stability ( Fig. S9 and S10 †).
Thermoelectric measurements were performed on SnS pellets perpendicular to the pressing direction (Fig. 2). The pellet exhibits high electrical conductivity (s), which rises from ca. 760 S m À1 at 323 K to ca. 2250 S m À1 at 523 K, gradually decreases to ca. 1475 S m À1 at 723 K and once again increases to ca. 1960 S m À1 at 773 K (Fig. 2a). These values are notably much higher than undoped SnS bulk counterparts (e.g. $5-32 S m À1 at 523 K) synthesised by mechanical alloying 25 and hightemperature synthesis. 26 Moreover, they also exceed those of pellets consolidated from solvothermally-synthesised SnS nanorods (e.g. $590 S m À1 at 523 K) 62 and even Ag-doped SnS pellets (e.g. $530 S m À1 at 523 K). 25 The high s values of SnS pellets can be attributed to the organic-free surfaces of the plates, the orientation of the plates, high crystallinity and the high degree of densication ( Fig. S5-S7 †). The positive Seebeck coefficient (S) values for SnS pellets indicate p-type conducting behaviour (Fig. 2b). The Seebeck coefficient for SnS increases nearly linearly from ca. 285 mV K À1 at 323 K to ca. 455 mV K À1 at 723 K before decreasing to ca. 430 mV K À1 at 773 K. The combination of such high electrical conductivity and Seebeck coefficients leads to power factors (e.g. $0.33 mW m À1 K À2 at 573 K and $0.36 mW m À1 K À2 at 773 K) (Fig. 2c) that exceed those of previously reported undoped SnS pellets (e.g. $0.01-0.10 mW m À1 K À2 at 573 K). 25,26,62 The value of thermal conductivity (k) for SnS reduces from z1.918 W m À1 K À1 at 323 K to z0.786 W m À1 K À1 at 773 K (Fig. 2d), to which the lattice thermal conductivity (k L ) is the main contributor (Fig. 2e). The ZT of a SnS pellet was thus calculated, increasing gradually from ca. 0.01 at 323 K to ca. 0.36 at 773 K (Fig. 2f). The ZT at 773 K is higher than the values reported for undoped SnS (e.g. 0.08-0.25 at 773 K) fabricated by various methods, 25,26,62 indicating the signicant potential of the organic-free method in synthesising high-performing metal chalcogenide thermoelectrics.
To demonstrate the utility of the organic-free strategy in obtaining a range of thermoelectrics with prescribed microstructure and tailored composition, we focused on the synthesis of substituted Sn(S,Se) materials, conscious that thermoelectric properties should improve at higher Se concentration. 26,35 We achieved this via an organic-free anion exchange that involved injecting a NaHSe solution into the SnS suspension, followed by boiling for 2 h (Fig. S2 †). Samples were prepared with NaHSe : Na 2 SnO 2 in the appropriate molar ratios with the intention of preparing SnS 1Àx Se x (0.5 # x # 1) (eqn (1) and (2)). (1) The anion exchange is likely driven by the lower solubility product (K sp ) of SnSe in water as compared to that of SnS. 12,15 The PXD patterns of the anion-exchanged products ostensibly resemble that of orthorhombic SnS but with reections shied to lower 2q (e.g. for x ¼ 0.9 in Fig. 1a). 61 SEM ( Fig. 1f and S11a †) reveals that the product nano/micro-"owers" consist of plates of approximately similar size (2-10 mm) and thickness (50- (1) , corresponding to the SnS 0.1 Se 0.9 {011} plane spacings) identify preferred orientation along the h100i direction. Hence both the morphology and crystal structure before and aer anion exchange are essentially unaltered and the process is topotactic (Fig. 3a). 18 In an effort to understand the formation mechanism of SnS 0.1 Se 0.9 , we investigated the outcome of an anion exchange reaction aer only 1 min of boiling (using a Se : Sn molar ratio of 1). PXD shows that the product consists of both SnS 1Àx Se x (x z 0.9) and SnS (  almost uniform distribution of Se and S (Fig. 3c), while the remainder demonstrate an uneven anion distribution. With Se : S ratios varying from 48(1) : 2(1) to 42(1) : 8(1) (Fig. S15 †) the plates in the 2 h anion exchange sample are clearly much richer in Se than the 1 min product. Given the above observations, one can propose a formation process for the SnS 1Àx Se x plates (Fig. 3d). The anion exchange should initiate at the edges and faces of plates. The periphery of a plate can access Se 2À from both the edges and faces and therefore should be richer in Se than the centre. For thicker plates the contribution from anion exchange at the edges should become more pronounced. It is useful to note at this point that both the plate morphology and orthorhombic crystal structure are maintained aer the subsequent high temperature sintering (e.g. at 500 C) by either hot pressing or spark plasma sintering (SPS). Moreover, the overall Se : S ratio remains constant and the distribution of S and Se becomes more even across each plate (Fig. 4a-d).
The NaHSe concentration is a crucial synthesis parameter for the synthesis of SnS 1Àx Se x nano/micro-plates with tuneable S concentration. When the NaHSe : Na 2 SnO 2 molar ratio is increased from 1 through 1.15 to 1.5, it is possible to obtain almost single-phase orthorhombic SnS 1Àx Se x with very high Se concentrations (Fig. S16 †). The Se site occupancy obtained from Rietveld renement increases from 0.91(1) through 0.92(2) to 0.94(2) (Table S5 †) (corresponding to x ¼ 0.91, x ¼ 0.92 and x ¼ 0.94 respectively). On decreasing the Se : Sn molar ratio (and x) from 0.9 through 0.8 to 0.5, the SnS 1Àx Se x products accordingly contain more sulphide together with rising phase fractions of SnS ( Fig. S16; Table S6 †). The lattice parameters and unit cell volumes of these products increase with increasing Se concentration (x) which follows Vegard's law (Fig. S17 †). The products synthesised at various NaHSe concentrations universally take the form of nano/micro-owers consisting of clusters of nano/ micro-plates, reinforcing the premise that the anion exchange is (pseudomorphic and) topotactic ( Fig. S18 and S19 †). As the NaHSe : Na 2 SnO 2 molar ratio is increased, EDS spectra show  that the Se/(Se + S) ratio increases from $0.47, through $0.77, $0.85, $0.89, $0.92 to $0.93 (Fig. S18-S20 †), which is consistent with the Rietveld renement results.

Thermoelectric performance of anion-exchanged SnS 1Àx Se x
The ability to prepare >10 g SnS 1Àx Se x plates, with control over both structure and composition, facilitates the fabrication of textured, dense pellets. Of the various SnS 1Àx Se x materials synthesised, SnS 0.1 Se 0.9 was selected for thermoelectric measurements given previously documented electrical and thermal properties of the equivalent bulk material of the same composition. 35 Pellets with ca. 98% of the SnS 0.1 Se 0.9 theoretical density were consolidated from 2 h anion-exchanged SnS 0.1 Se 0.9 plates via one-step (denoted SnS 0.1 Se 0.9 -1) and two-step (denoted SnS 0.1 Se 0.9 -2) SPS processes. The two-step SPS process can give rise to the superplastic ow of unconstrained samples in the transverse direction (i.e. perpendicular to the pressing direction), in turn producing textured samples. 63 The orthorhombic structure of the materials persists aer sintering and both pellets exhibit strong orientation in the (h00) plane (Fig. S21 †). SnS 0.1 Se 0.9 -2 exhibits much stronger texturing than SnS 0.1 Se 0.9 -1 as evidenced by the higher relative diffraction intensities of the (h00) reections. In fact, the orientation degree (F) for (h00), estimated by the Lotgering method 64 and based on the PXD data (Fig. S21a †), is ca. 0.25 and ca. 0.46, for SnS 0.1 Se 0.9 -1 and SnS 0.1 Se 0.9 -2, respectively, demonstrating the greater texturing in the latter sample. SEM images of the cleavage surfaces (Fig. S22a-d †) and fractured cross-sections (Fig. 4e,  Thermoelectric measurements were performed on both SnS 0.1 Se 0.9 pellets perpendicular to the pressing direction (Fig. 5). The electrical conductivity of SnS 0.1 Se 0.9 -1 (Fig. 5a) increases from ca. 1940 S m À1 at 300 K to ca. 3400 S m À1 at 423 K, gradually decreases to ca. 1740 S m À1 at 673 K and reaches a maximum of ca. 6200 S m À1 at 873 K. The value of s subsequently subsides to ca. 5000 S m À1 at 923 K. This behaviour has been previously observed in SnSe polycrystalline materials. It has been suggested that the reduction in s over the midtemperature range followed by a subsequent increase could be related to a reduction in carrier mobility and a thermal excitation of carriers, respectively. 55,56 SnS 0.1 Se 0.9 -2 exhibits very similar behaviour to SnS 0.1 Se 0.9 -1, but demonstrates higher electrical conductivity below ca. 823 K (and more notably at lower T). As with SnS samples, SnS 0.1 Se 0.9 -1 and SnS 0.1 Se 0.9 -2 demonstrate a combination of clean particle surfaces, pronounced plate orientation, high crystallinity and high density. Coupled with the reduced bandgap engendered by Se doping, the high s values observed can be plausibly rationalised. That SnS 0.1 Se 0.9 -2 exhibits higher conductivity is likely due to the enhanced (h00) texturing induced by the two-step SPS process. Electrical conductivity within the Sn-Se layers (in the bc plane) of SnSe is expected to be much higher than that perpendicular to the layers (along a). 1 Indeed, the room temperature Hall carrier mobility of SnS 0.1 Se 0.9 -2 (evidenced as more strongly textured) is approximately 20.6% higher than that of SnS 0.1 Se 0.9 -1 (Table 1).
Both pellets exhibit p-type conducting behaviour, demonstrated by the positive Seebeck coefficient (Fig. 5b). S for SnS 0.1 Se 0.9 -1 increases almost linearly from ca. 310 mV K À1 at 300 K reaching a maximum at ca. 445 mV K À1 at 673 K. By 873 K, S drops to ca. 320 mV K À1 before a slight upturn to ca. 335 mV K À1 at 923 K. Below ca. 823 K, the Seebeck coefficient for SnS 0.1 Se 0.9 -2 is consistently slightly lower than that for SnS 0.1 Se 0.9 -1 but nevertheless follows the same trend with temperature. The decrease in S above 673 K could be related to the thermal excitation of minority carriers. 55,56 The combination of large s values coupled with high values of S results in exceptional power factors (S 2 s) for SnS 0.1 Se 0.9 -1 and SnS 0.1 Se 0.9 -2 (Fig. 5c) which both show local maxima (of ca. 0.50 mW m À1 K À2 ) at 473 K and (of ca. 0.63 mW m À1 K À2 ) at 873 K. The S 2 s vs. T behaviour hence broadly follows the temperature variation of s, although the power factor for SnS 0.1 Se 0.9 -2 is marginally inferior to SnS 0.1 Se 0.9 -1 above 573 K. Interestingly SnS 0.1 Se 0.9 and SnS show very similar trends in the variation of s, S and S 2 s with temperature ( Fig. S26 †), indicative of the broadly similar crystalline and electronic structures. Ultimately, SnS 0.1 Se 0.9 reaches a higher value of s (Fig. S26a †), probably primarily due to the reduction in the bandgap as Se replaces S.
Both SnS 0.1 Se 0.9 -1 and SnS 0.1 Se 0.9 -2 have higher power factors than SnSe pellets consolidated from surfactant-free nanoplates (increasing from ca. 0.05 mW m À1 K À1 at 300 K to ca. 0.40 mW m À1 K À1 at 550 K; Fig. S26c †), 11 but perhaps particularly remarkable is that S 2 s for SnS 0.1 Se 0.9 -1 at 773 K exceeds those of other SnS 1Àx Se x (x < 1) materials (e.g. S 2 s values at 773 K of z0.24 mW m À1 K À2 for p-type SnS 0.2 Se 0.8 (ref. 26) and z0.40 mW m À1 K À2 for n-type SnS 0.1 Se 0.87 I 0.03 (ref. 35)), and matches or surpasses those for the best examples of doped polycrystalline SnSe (e.g. 0.39-0.48 mW m À1 K À2 for p-type Nadoped SnSe at 773 K). 29,31,32 Such materials require hightemperature methods to produce, so it is clearly possible to replace these synthesis methods by more efficient, energysaving alternatives without sacricing performance.
Microstructural texturing also inuences the thermal conductivity (k) of the pellets (Fig. 5d). The value of k for SnS 0.1 Se 0.9 -1 is low, decreasing from z1.148 W m À1 K À1 at 300 K to z0.464 W m À1 K À1 at 773 K. The values for SnS 0.1 Se 0.9 -2 at the equivalent temperatures are approximately 29.0% and 42.4% larger respectively. The increase in k at 823 K is most likely related to the second order displacive phase transformation of SnSe (from orthorhombic Pnma to orthorhombic Cmcm). 65 As with the electrical conductivity, the thermal conductivity is anisotropic in SnS 1Àx Se x and so the difference in k can be attributed to the different degrees of texturing in SnS 0.1 Se 0.9 -1 and SnS 0.1 Se 0.9 -2. The thermal conductivity along the a-axis (perpendicular to the Sn-Se planes) is much lower than that along b or c. 1 The main contribution to k in the pellets is the lattice thermal conductivity (e.g. k L z 0.427 W m À1 K À1 for SnS 0.1 Se 0.9 -1 and k L z 0.620 W m À1 K À1 for SnS 0.1 Se 0.9 -2 at 773 K) (Fig. 5e) and the magnitude of k and k L demonstrates the degree to which texturing can inuence the thermal properties of polycrystalline materials such as SnSe. It should be noted that k L is still higher than the theoretical minimum for SnSe (k L z 0.26 W m À1 K À1 at 770 K), 35 so one might expect that by further texture control and/or by reducing the plate dimensions, k L could be driven further towards this theoretical minimum. It is also interesting to note that SnS (containing a lighter chalcogen), has a higher k L than SnS 0.1 Se 0.9 across the whole measurement temperature range (Fig. S26e †).
ZT could be calculated for the pellets using the above electrical and thermal transport data (Fig. 5f). The ZT of SnS 0.1 Se 0.9 -1 is the higher across the whole T range, increasing from ca. 0.05 at 300 K to ca. 1.16 at 923 K; the ZT of SnS 0.1 Se 0.9 -2 reaches a maximum of ca. 0.80 at 873 K. The ZT of SnS 0.1 Se 0.9 -1 at 923 K is comparable to SnSe bulk materials with carbon inclusions measured at 903 K. 37 Considering previous reports of the thermoelectric performance of SnSe up to 800 K, it is useful to compare the value of ZT close to this elevated temperature. The value for SnS 0.1 Se 0.9 -1 at 773 K is higher than that of sulfur-free p-type polycrystalline SnSe (ZT z 0.39-0.66) 27,66 and comparable to various metal-doped, p-type polycrystalline SnSe (ZT z 0.5-1.2) 26, [28][29][30][31][32]36 (where all measurements were made perpendicular to the pressing direction at approximately 773 K; e.g. z0.6 for Ag-doped SnSe at 750 K, 28 z0.8 for Na-doped SnSe at 800 K, 31 z1.1 for K-doped SnSe at 773 K). 30 Importantly, this indicates that tin chalcogenides can retain high thermoelectric performance when some of the more toxic Se is removed and replaced by less toxic S via anion exchange. Further systematic study of the thermoelectric performance of SnS 1Àx Se x as a function of x should reveal the optimum materials and processing parameters in the system. Moreover, it may be possible to offset loss of performance with decreasing x by co-doping with appropriate low toxicity, Earth-abundant metals. Although homogeneity of the materials is realised in SPS treated pellets, we note that the extent of anion exchange during synthesis will be governed primarily by reaction kinetics. Both the kinetics and thermodynamics (K sp ) could be modied by replacing water with another solvent, but given the economic and environmental advantages in using aqueous chemistry, modifying reaction temperature (and/or autogenous pressure, hydrothermally) would be the obvious parameter(s) to investigate towards achieving complete topotactic conversion (Fig. 3d). There is thus considerable scope for polycrystalline SnSe materials to achieve (average) ZT values comparable to analogous single crystals. Given the materials design options to reduce k (and notably k L ) further as described above (and for example, via precipitates 30 ) and to increase carrier concentration (to the magnitude of 10 19 cm À3 ) and electrical conductivity, s (e.g. through alkali ion doping 2,24,29,31,33 ), it should be possible to propel ZT to still higher reaches in tin chalcogenides without sacricing green chemistry principles.

Conclusions
In summary, organic-free, scalable, low-cost solution approaches have been developed to produce nanostructured layered tin chalcogenides with tuneable chalcogenide composition. This precipitation-anion exchange protocol yielded SnS 0.1 Se 0.9 plates, which were consolidated into p-type textured pellets, achieving ZT z 1.16 at 923 K through tuning the microstructural texturing. The precipitation-anion exchange

Materials characterisation and testing
PXD data were collected with a PANalytical X'pert Pro MPD diffractometer in Bragg-Brentano geometry (Cu Ka 1 radiation, l ¼ 1.5406Å). Rietveld renement was performed against PXD data using the GSAS and EXPGUI soware packages. 67,68 Imaging and elemental analysis were conducted by SEM (Carl Zeiss Sigma, at 5 and 20 kV respectively) equipped with EDS (Oxford Instruments X-Max 80). Further imaging, elemental analysis and SAED were performed by TEM (FEI Titan Themis 200 equipped with Super-X windowless EDS detector, operated at 200 kV). Thermogravimetric-differential thermal analysis (TG-DTA) of the samples was performed using a Netzsch STA 409 thermal analyser under owing Ar. Optical bandgaps were measured by DR-UV-Vis spectroscopy (Shimadzu, UV-2600). All thermoelectric measurements were made perpendicular to the pressing direction. Following prior determination of thermal stability and pre-heating treatments if necessary, the Seebeck coefficient (S) and electrical conductivity (s) of pellets were measured simultaneously using a Linseis LSR-3 instrument from 300-923 K. The thermal conductivity, k of the pellets was calculated by k ¼ DC p r, where D, C p and r are the thermal diffusivity coefficient, specic heat capacity and density, respectively. D of the SnS 0.1 Se 0.9 pellets was measured using a Netzsch LFA 457 from 300-923 K (Fig. S27a †); C p of SnS 0.1 Se 0.9 (Fig. S27b †) was calculated from the weighted average 26 of reported C p of SnSe 1 and SnS. 25 r of the pellets was measured by the Archimedes method. Electronic thermal conductivity (k e ) was estimated by the Wiedemann-Franz law (k e ¼ LsT, where L is the Lorentz number; L of 1.5 Â 10 À8 V 2 K À2 was applied 1 ), and lattice thermal conductivity (k L ) was calculated by subtracting k e from k. ZT was calculated via ZT ¼ S 2 sT/k. Hall measurements were performed on a nanometrics HL5500 Hall system using a van der Pauw conguration.

Conflicts of interest
There are no conicts to declare.