Drastic power factor improvement by Te doping of rare earth-free CoSb3-skutterudite thin films

In the present study, we have focused on the elaboration of control of Te-doped CoSb3 thin films by RF magnetron sputtering which is an attractive technique for industrial development of thermoelectric (TE) thin films. We have successfully synthesized sputtering targets with a reliable approach in order to obtain high-quality films with controlled stoichiometry. TE properties were then probed and revealed a reliable n-type behavior characterized by poor electrical transport properties. Tellurium substitution was realized by co-sputtering deposition and allowed obtaining a significant enhancement of the power factor with promising values of PF ≈ 0.21 mW m−1 K−2 near room temperature. It is related to the Te doping effect which leads to an increase of the Seebeck coefficient and the electrical conductivity simultaneously. However, despite this large improvement, the properties remained far from the bulk material and further developments are necessary to improve the carrier mobility reduced by the thin film formatting.


Introduction
Thermoelectricity represents a technological gateway in the current research dynamic focused on sustainable energy development. Through appropriate devices, thermoelectric (TE) materials are attractive for dynamically harvesting energy from surroundings, like body heat, to power Internet of Things (IoT) applications. 1,2 For such applications, the development of thermoelectric materials characterized by large TE performance near room temperature is required in addition to preferable exible formatting. Signicant power harvesting from waste heat requires TE material with a high power factor (PF) and low thermal conductivity. 3 The TE performance of those materials is quantied by the dimensionless gure of merit zT ¼ PF Â Tk À1 ¼ S 2 sTk À1 (T, absolute temperature; S, Seebeck coefficient; s, electrical conductivity; k, thermal conductivity). These properties typically have a tradeoff relationships, and several strategies have been investigated, such as various nanostructuring, 4,5 utilizing interfaces and composites, 6-8 magnetism, [9][10][11] and topological states, 12 to bypass these tradeoffs and enhance the zT.
For the past more than two decades, skutterudite CoSb 3 has been considered as one of the most promising TE material mostly for mid-temperature (300-800 K). [13][14][15][16] Its high carrier mobility and suitable Seebeck coefficient lead to attractive PF, even near room temperature, and both conduction types (p and n) are achieved. Moreover, its relatively complex cubic structure contained two icosahedra voids formed by the CoSb 3 octahedra which allowed the introduction of a foreign element M for acting as a dopant or/and promoting intrinsic low thermal conductivity. Many efforts have been done to obtain further improvement in the electrical transport properties and reduce the lattice thermal conductivity, such as doping and nanostructuring. 14,15,[17][18][19][20][21] More recently, several studies have reported the thin lms synthesis of CoSb 3 skutterudite in order to exploit the effects of interface and grain boundary in the lowdimensional system, that can effectively scatter phonons, for promoting low thermal conductivity. [22][23][24][25][26][27][28][29][30][31][32][33][34] In fact, TE generator based on thin lms is attracting much attention for potential IoT applications as sensors, micro-power sources, exible devices. 1,2 Various thermoelectric materials composed of abundant elements have been fabricated into thin lms. For example, oxides and some very high performance thin lms has been recently reported with Heusler compounds, for example. [35][36][37] Although some promising results have been published regarding CoSb 3 skutterudite thin lms, the electrical properties remain lower than those of bulk materials due to the difficulty in controlling element composition during preparation. 22,29,31,33 Guest ion (rare-earth or alkali-earth metal)-free skutterudites are suitable for thin lms growth because they are not easily oxidized. 27 Since guest ion plays an important role in lowering thermal conductivity, we need to reduce the high thermal conductivity of guest ion-free skutterudite by other methods. 29,30 Substitution of Sb by semimetals elements is a utilized strategy to affect the electronic structure of the CoSb 3 skutterudite by increasing the carrier concentration/electrical conductivity, in addition, to reduce the lattice thermal conductivity by phonon-electron scattering enhancement. Up to this date, several studies on Te-doped or Si-doped CoSb 3 materials have been made in bulk material and seem effective to enhance the electrical and thermal properties. 17,21,[38][39][40][41] Fabrication of thin lms samples based on these materials can be expected to benet us to obtain further improved rare-earth free skutterudite materials.
In the present study, we were focused on the elaboration of rare earth-free CoSb 3Àx Te x skutterudite thin lms by magnetron sputtering which is one of the numerous techniques for preparing thin lms at an industrial scale. We have realized CoSb 3 target for sputtering with a reliable approach for controlling the raw material parameters. Then CoSb 3Àx Te x thin lms were deposited by co-sputtering using CoSb 3 and pure Te targets. Different power was used for the Te target in order to obtain different doping levels and optimize the electronic and thermal transport properties simultaneously.

Target preparation
For the CoSb 3 target, pure elemental Co (powder, $99.9%, #150 mm, Aldrich) and Sb (powder, $99.9%, À100 mesh, Aldrich) were weighed according to the nominal composition of CoSb 3 . The elemental powders were mixed and ground by hand milling in an agate mortar and sealed in a quartz tube under a dynamic vacuum. The mixture was melted/quenched at 1423 K in an electric furnace and annealed 4 days at 973 K. The obtained powder mixture was sieved through >212 mm and loaded into a graphite die with a diameter of 50 mm coated with carbon paper. Densication by spark plasma sintering (SPS -511S, Fuji Electronic Industrial) was processed at 773 K for 5 min (heating and cooling rate of 50 K min À1 ) under a pressure of 40 MPa and vacuum enclosure.
For the Te target, a high purity commercial Te target (99.99%, Furuuchi Chemical Company) was used.

Film preparation
CoSb 3 skutterudite thin lms were prepared by RF co-sputtering deposition technique via AVC-corporation double chamber magnetron sputtering ultrahigh vacuum system. The Pyrex glass was chosen as substrates, which were cleaned by using ultrasonic cleaning for 10 min in acetone, 10 min in absolute ethyl alcohol, and 10 min in deionized water, respectively. The chamber was pumped down to 5.0 Â 10 À6 Pa and the working pressure was kept in 1.0 Pa with Ar ow of 6 cm 3 min À1 . All the samples were deposited with the same sputtering power of 50 W for the CoSb 3 target and 0 W, 5 W, 7 W, and 9 W were used for the Te target, respectively, for T1, T2, T3, and T4 lm. The sputtering time was set to 3 h at 573 K, then annealing steps of 1 h were taken at the same temperature for preventing the lm to peel off. The resulting deposition rates increased with the sputtering power of the Te target and are estimated to be 0.087, 0.088, 0.093, and 0.104 nm s À1 respectively for T1 (0 W), T2 (5 W), T3 (7 W), and T4 (9 W) (S1). Second annealing was realized at 523 K for 1 day under vacuum for crystallizing the skutterudite and improving the lms homogeneity. Two substrates were done for each deposition and a second series with a new CoSb 3 -target was deposited with the same parameters to conrm the results reliability on 4 samples for each condition.

Film characterizations
The crystal structure was investigated by X-ray diffraction (XRD) technique with the prescriptive q-2q mode with the angle of 2q ¼ 10 -100 (Smart Lab3 Rigaku Corporation). X-ray powder diffraction patterns were rened by Rietveld analysis using the FullProf and WinPLOTR soware packages. 42,43 The shape of the diffraction peaks was modeled using a Thompson-Cox-Hastings pseudo-Voigt prole function. 44 Zero-point shi, asymmetry parameters and lattice parameters were systematically rened, and the background contribution was manually estimated. The surface morphology was analyzed by scanning electron microscopy (SEM) (Hitachi SU-8000) and the component analysis was proceeded by energy dispersive spectroscopy (EDS). The thickness of the thin lms was obtained by using a Dektak 6M Stylus Proler measurement system. The carrier concentration and mobility were obtained from Hall effect measurements (Resitest 8300). The electrical conductivity (s) and Seebeck coefficient (S) were simultaneously measured by the four-probe method from 300 K up to 525 K using a ZEM-3 (ULVAK Advance-Riko) device under partial helium pressure. The estimated measurement uncertainties are xing to 6% for the Seebeck coefficient and 8% for the electrical conductivity. 45 The cross-plane thermal conductivity was evaluated by using a picosecond time-domain thermoreectance (TD-TR) instrument (PicoTR, Picotherm Corp.) in a front-heating/frontdetection conguration. A 100 nm-thick Pt thin lm was deposited on the CoSb 3 lm surface by using a DC sputtering system to detect transient temperature changes. A 1550 nm infrared pulsed laser with a repetition frequency of 20 MHz and a pulse duration of 0.5 ps was used as a heat source. A 780 nm probe laser was used to detect the thermoreectance signal. The picosecond TD-TR system was customized to reduce the spot size of the probe laser to ca. 5 mm as described elsewhere. 46 The improved mirror image method for tting all the range of pulse interval 47,48 was used to determine the thermal conductivity value. Here, we assumed the specic heat value of all lms as 3R from Dulong-Petit law (0.235 J g À1 K À1 ).

Results and discussion
The X-ray diffraction (XRD) patterns of the whole thin lms series aer thermal cycles are displayed in Fig. 1a. As can be seen, all patterns exhibited the main diffraction peaks corresponding to the CoSb 3 skutterudite structure (Im 3, a z 9.04 A) with high crystallinity as conrmed by cross-section Scanning Electronic Microscope (SEM) (Fig. 2a). 49 Additional lowintensity peaks are observable in the sample T1 and can be attributed to residual free-Sb. The main thin lms structures were conrmed by XRD renement considering the CoSb 3 skutterudite native structure. Low-reliability factor is obtained independently to the Te target power used during deposition, as presented in Table 1, and conrmed the systematic skutterudite phase formation with high purity. XRD pattern renement allowed to determine the lattice parameter dependency with the Te target power and permitted to estimate the Te doping level in each lm (Table 1 and Fig. 1b). Nonetheless, no conclusion about the atomic distribution of the Te in the Sb site and/or the void site can be drawn from the XRD renement due to the limited resolution of a conventional XRD diffractometer. The increase in the cell parameter led us to the assumption that Sb substitution by Te occurred in accordance with the previous report on bulk but it is not excluded that the Te partially lled the void position in skutterudite. 38,40 Further analysis, using transmission electronic microscope (TEM), is required to elucidate the real position of Te. The pseudo-linear and large increase of the lattice parameter versus Te target power is observable for T3 and T4 in contrast to T2. According to the paper of Li et al., the cell parameter dependence with Te doping level reaches a step at a ¼ 9.053 A corresponding to a Te substitution of 3.13 at%. 38 Nevertheless, in the report of Nagamoto et al., a solubility limit of 6.2 at% is proposed with corresponding cell parameters of a z 9.049 A. 17 In our present study, the cell parameters drastically increased up to a ¼ 9.110 A for T4 suggesting that a higher doping level can be reached by the sputtering method. The composition analysis by Scanning Electron Microscopy using Energy Dispersive X-ray Spectroscopy (SEM-EDS) also sustained this hypothesis with a richest Te composition (>6.2 at%) for the sample T3 and T4 (Tables 2 and  S2 †). It revealed a clear tendency of increasing Te content with the increasing of the Te target power simultaneously with a decreasing of the Sb content following the main substitution between the Sb and the Te. Also, as can be seen from Table 2,   Table 1 Cell parameters at room temperature and reliability factors obtain from Rietveld refinement of CoSb 3 doped Te thin films (T1, T2, T3, and T4) X-ray patterns (l Cu ¼ 1.5406 A) the thickness of the samples increased slightly in good agreement with the theoretical larger deposition rate, provided by the increasing power of Te target, during co-deposition. Although the EDS analysis showed a Te concentration of 12.22% and 22.14% for T3 and T4 lms, respectively, these are likely to include contributions from Sb 2 Te 3 nano particles, as will be shown below. Nevertheless, the large lattice parameters of these samples indicate that the Te concentrations in the CoSb 3 phase of sample T3 and T4 are much higher than the solubility limit so far reported for bulk samples. 17 The substantial Te content affected the surface morphology as shown in Fig. 2. The undoped (T1) and low Te-doped (T2) lms revealed a slightly rough surface composed of irregular particles in the range of 100 nm to 1 mm. These particles are assimilated to the crystallized CoSb 3 -skutterudite as highlighted by the broad diffraction peak (Fig. 1b inset). Nonetheless, the T3 and T4 lms revealed a atter surface covered by nanoscale particles. We ascribed these nanoparticles to a Sb 2 Te 3 phase (R 3mH, a z 4.26 A and c z 30.46 A) which was seeded between the larger CoSb 3 particles during the deposition process. This assumption is supported by the XRD pattern of the T4 which exhibited low intensity and broad peaks located at 2q z 28.20 and 42.40 ( Fig. 1a (#)). It can be assimilated to the highest intensity indexation (015) and (110) of the Sb 2 Te 3 crystal structure. Moreover, the Sb-Te rich composition revealed by SEM-EDS analysis (Tables 2 and S2 †) also supports this assumption. A recent study has shown that the RF-magnetron sputtering method is suitable for obtaining nanosize Sb 2 Te 3 in the range of 5.8-19.6 nm depending on the annealing. 50 The seeding of the Sb 2 Te 3 phase is made propitious by the systematic free-Sb, as observed in the un-doped lm T1, and associated with the oversupply of Te provided by the large target power used for T3 and T4 deposition.
All thin lms are characterized by n-type conduction with large carrier concentrations in the range of 10 21 cm À3 as presented in Fig. 3. It indicates that the deposition process affected the lm composition and promoted an intrinsic electron doping effect. According to the study of Zheng et al., the presence of free-Sb in the undoped lm T1 (Fig. 1a (*)) attested that an Sb deciency involved in the CoSb 3 lm, which thereby induces additional electron carrier leading to the n-type conduction. As an example, the study of P. Fan et al. reported for CoSb 3 thin lm sample annealed at 518 K a carrier concentration of n z 0.8 Â 10 21 cm À3 and an electronic mobility of m z 0.75 cm 2 V À1 s À1 at RT (Fig. 3 (*)). 31 But in the present study, it can be seen that the undoped lm T1 is characterized by slightly larger carrier concentration (n ¼ 2.45 Â 10 21 cm À3 at RT) and a signicantly reduced electronic mobility (m ¼ 0.14 cm 2 V À1 s À1 ). It suggests that our lms are characterized by larger Sb deciency. By using the data reported by P. Fan et al. as reference values, we can see that the carrier concentration signicantly increases with the theoretical Te content up to n ¼ 7.35 Â 10 21 cm À3 at RT for T4 according to the electron donor effect of the Te substitution. 38,39 It is followed by a slight drop of the carrier mobility from m ¼ 0.463 cm 2 V À1 s À1 to 0.362 cm 2 V À1 s À1 , respectively, for T2 and T4 (Fig. 3). The decrease of the electron mobility can likely to be related to the ionized impurity scattering effect enhancement related to the rise of the carrier concentration between the two samples. Other possibilities can be the alloy scattering and the potential uctuation effect due to a non-uniform distribution of Te atoms over the lattice.
The electronic mobilities remained low, by comparison with the bulk CoSb 3 skutterudite, and drastically affected the electrical transport properties as presented in Fig. 4a. The undoped lm T1 exhibited semiconducting behavior with low electrical conductivity s far from the native properties of the CoSb 3 bulk sample. 13,14,18,21 This electrical transport behavior is typical for a heavily doped semiconductor. Therefore, it is consistent with the large carrier concentration and the reduced electronic mobility due to the thin lm formatting effect and small grain size in all lms. 33,34 The Te doping effect allowed to improve the electrical conductivity in the whole temperature range by the increase of the carrier concentration. As can be seen in Fig. 4a, electrical conductivity gradually increased with Te content from s ¼ 4.9 Â 10 3 S m À1 to 3.5 Â 10 4 S m À1 at 300 K, respectively, for T1 and T4. The presence of nanosize Sb 2 Te 3 , in T3 and T4 lms, is assumed to have a negligible effect on the electrical properties considering the small amount detected and the poor electrical mobility of this phase. 50 The Seebeck coefficient measurement conrmed that the whole series is characterized by n-type conduction with a negative Seebeck coefficient. The undoped lm T1 is represented by low values in the temperature range (S ¼ À9.2 to À23.5 mV K À1 from 300 to 525 K; Fig. 4b) comparable with several reports on undoped CoSb 3 thin lms. 31,33,34 However, the Te doping induced a signicant increase of the absolute value of the Seebeck coefficient up to S ¼ À91.5 mV K À1 at 300 K for example in T3. The larger carrier concentration of T1 (n ¼ 2.45 Â 10 21 cm À3 at RT) compared to T2 (n ¼ 1.50 Â 10 21 cm À3 at RT) follow the predictive Pisarenko plot of the CoSb 3 -skutterudite presented in the report of Tang et al. 51 Indeed, it is consistent with the multiple band transport model proposed for the heavily doped n-type CoSb 3 skutterudite and explains the larger absolute Seebeck coefficient of the low Te-doped lm T2 compared to the undoped lm T1. However, it suggests an additional feature leading to the large thermopower in T3 and T4. As presented Fig. 3 Carrier concentration and electron mobility at room temperature as a function of Te content of the CoSb 3 doped Te thin films (T1, T2, T3, and T4) and reference data of CoSb 3 thin film annealed at 518 K (asterisk). 31 before, the large Te amount leads to a change of the crystal structure. The lattice parameter values are signicantly increased for the lms T3 and T4 (a $ 9.0844 A at 300 K, Table 1) by comparison with T1 and T2 (a # 9.0453 A). The large Te amount can be considered to have induced a crystal eld modication by the variation of the Sb-Co-Sb angle and bond length, as described in the report of Hanus et al. 52 The 'lattice effect' can contribute to a band convergence in the CoSb 3 electronic structure and be cumulated with carrier density effect for promoting a large Seebeck coefficient.
Correlated with the conductivity improvement, the Seebeck enhancement leads to a signicant improvement of the PF of the thin lm by 2 orders of magnitude in T4 up to a value of PF ¼ 0.21 mW m À1 K À2 at 300 K (PF ¼ 0.50 mW m À1 K À2 at 525 K) as presented in Fig. 4c. To the best of our knowledge, this is the highest power factor obtained in rare earth-free CoSb 3 -skutterudite thin lms near room temperature and represent an improvement of at least 260% compared to the report of Fan et al. on exible substrates and 840% considering the recent report on Ag-doped CoSb 3 skutterudite thin lm at room temperature. 24,31 The drastic PF improvement, induced by the Te doping effect, constitutes a new record considering the last few years trend of the highest PF obtained on the rare earth-free CoSb 3 -skutterudite thin lms at room temperature (Fig. 5).
To probe the thin lms formatting effect in the thermal transport performance near room temperature, the thermal conductivity of the thin lms was measured by the thermore-ectance method with front detection (S3), 47,48 using a customized focused system. 46,53,54 The results of the calculated total k and lattice k latt thermal conductivity are summarized in Table 3. The electronic thermal conductivity k elec contribution seems negligible compared to the lattice thermal conductivity contribution. The pristine sample (T1) presented a signicantly reduced k latt of 3.8 W m À1 K À1 @ 300 K compared to the corresponding values reported on undoped bulk samples (k bulk z 8-9 W m À1 K À1 @   5 The trend of the highest PF obtained on the rare earth-free CoSb 3 -skutterudite thin films at room temperature. 24,26,30,32,34 This journal is © The Royal Society of Chemistry 2020 RSC Adv., 2020, 10, 21129-21135 | 21133 300 K). 38,[55][56][57] According to the theoretical prediction of Shiga et al., a signicant k reduction of CoSb 3 thin lm should occur for thin lms thickness bellow 500 nm. 58 However, the present thin lms are in the region over 800 nm ( Table 2) which conrmed that even 'thick' lm formatting can efficiently reduce the thermal transport in the CoSb 3 phase by the enhancement of boundary scattering in the surface. This effect can be cumulated with the small grain size (<500 nm) and a higher fraction of grain boundaries (Fig. 2b) which can contribute to lower the thermal conductivity as reported in nanostructured CoSb 3 . 59,60 Te doping provided an additional reduction of the k latt , as already reported in bulk materials. Considering that the samples have a high carrier concentration and low mobility, which indicates a high effective mass, we speculate that phonon-electron scattering can be induced by Te doping and also might have a non-negligible role. 21,38 Ultimately we have obtained a minimum value of k latt ¼ 2.2 W m À1 K À1 @ 300 K for T3. The slightly increased value of k latt ¼ 2.9 W m À1 K À1 @ 300 K for the T4 lm with heavy Te doping might be due to the Sb 2 Te 3 phase which affected the front detection of the thermoreectance analysis by increasing the interfacial thermal resistance. The nal gure of merit varied between zT ¼ 0.02-0.04 near room temperature (300-375 K) and increased up to zT ¼ 0.08-0.1 at 525 K for the T3 and T4 thin lms.

Conclusion
In this study, Te-doped CoSb 3 -skutterudite thin lms were prepared for the rst time by using the co-sputtering of a CoSb 3 target and pure Te target. Tellurium doping was conrmed through XRD renement and SEM-EDS analysis. We revealed that the sputtering method allowed to overcome the Te solubility limit reported of 6.2 at% on bulk samples. The electron doping provided with the substitution of Sb by Te leads to an increase in the carrier concentration and enhanced the poor electrical conductivity of the CoSb 3 thin lms up to s z 3.50 Â 10 4 S m À1 at 300 K for the richest Te content. Simultaneously, the Seebeck coefficient is increased up to S ¼ À93.4 mV K À1 at 525 K for this heavily Tedoped thin lms which provided the best power factor currently reported in rare earth-free CoSb 3 skutterudite lms of PF ¼ 0.21 to 0.50 mW m À1 K À2 , respectively, at 300 K and 525 K. Furthermore, we determined the thermal conductivity of the CoSb 3 -skutterudite thin lms (k ¼ 3.8 W m À1 K À1 @ 300 K for the undoped lm) which highlighted that even thick lm formatting can efficiently reduce the large thermal conductivity of the phase. Finally, we showed that the co-deposition method is suitable for preparing highly doped-CoSb 3 thin lms.