Accelerated Discovery of Thermoelectric Materials: Combinatorial Facility and High-Throughput Measurement of Thermoelectric Power Factor

1Accelerated Discovery of Thermoelectric Materials: Combinatorial 2Facility and High-Throughput Measurement of Thermoelectric Power 3Factor 4Jorge García-Canadas, †,∥ Nicholas J. E. Adkins, ‡ Stephen McCain, ‡ Bastian Hauptstein, ‡ Ashley Brew, † 5David J. Jarvis, § and Gao Min* ,† 6 † School of Engineering, CardiffUniversity, The Parade, CardiffCF24 3AA, United Kingdom 7 ‡ College of Engineering & Physical Sciences, The University of Birmingham, Edgbaston, Birmingham B15 2TT, United Kingdom 8 § European Space Agency, Keplerlaan 1, PO Box 299, 2200 AG Noordwijk, The Netherlands 9ABSTRACT: A series of processes have been developed to 10 facilitate the rapid discovery of new promising thermoelectric 11 alloys. A novel combinatorial facility where elements are wire12 fed and laser-melted was designed and constructed. Different 13 sample compositions can be achieved by feeding different 14 element wires at specific rates. The composition of all the 15 samples prepared was tested by energy dispersive X-ray 16 spectroscopy (EDS). Then, their thermoelectric properties 17 (power factor) at room temperature were screened in a specially 18 designed new high-throughput setup. After the screening, the 19 thermoelectric properties can be mapped with the possibility of identifying compositional trends. As a proof-of-concept, a 20 promising thermoelectric ternary system, Al−Fe−Ti, has been identified, demonstrating the capability of this accelerated 21 approach. 22 KEYWORDS: high-throughput, thermoelectric materials, laser processing, combinatorial chemistry, power factor 23 ■INTRODUCTION 24 Nowadays, there is an increasing demand on power generation. 25 In this respect, thermoelectricity has attracted considerable 26 interest over the past decades due to its ability to directly 27 convert heat into electricity and its application in solid-state 28 refrigeration. 1 Most of the research on developing new 29 materials has been focused on maximizing the dimensionless 30 figure of merit ZT, which relates to the efficiency of the 31 materials. It is defined as ZT = S2T/λρ, where Sis the Seebeck 32 coefficient, Tthe absolute temperature, λthe thermal 33 conductivity, and ρthe electrical resistivity. The power factor, 34 defined as S2/ρ, is also a useful parameter to evaluate 35 thermoelectric performance of materials. 36 There is a strong demand on high-ZT bulk materials, 37 specially containing environmentally friendly, sustainable and 38 abundant elements. 2 The search for new thermoelectric 39 materials requires huge amounts of sample synthesis and 40 characterization. For this reason, combinatorial approaches are 41 of significant importance for accelerating the discovery of novel 42 materials. 3 Examples of both experimental 4−7 and theoretical 43 approaches 8−11 have been reported in the literature. Employing 44 this approach, a wide number of compounds combining 45 different elements in a wide range of compositions can be 46 synthesized. Then, the relevant properties have to be measured 47 by means of high-throughput facilities, leading to a library of 48 compositional-dependent properties useful to identify promis49 ing materials and find out compositional trends. 50Here we report an integral combinatorial approach where a 51vast number of bulk alloy samples have been prepared in a 52novel combinatorial facility fed by wires of different pure 53elements. The wires were laser-melted to form the alloys and 54different compositions were achieved by adapting the feed rate 55of each wire. The thermoelectric power factor was then 56screened at room temperature in all the samples using a 57specially designed high-throughput facility. After an initial 58screening of hundreds of different samples, a promising ternary 59system, Al−Fe−Ti was identified. 60 ■EXPERIMENTAL PROCEDURES 61Combinatorial Facility. This facility is based on the 62suspended droplet alloying (SDA) concept, which utilizes a 63laser beam to melt elemental wire feedstock in order to produce 64 f1a small button of bulk alloy material (see Figure 1). 65Compositionally different samples can be synthesized by 66varying the ratio of wire feed rates. The alloying process begins 67when the aligned wires from each element are fed into the 68beam path and melt. The melting process initiates the 69formation of a droplet on the tip of each wire. The elemental 70droplets grow as new material is fed into them. Because of the 71proximity of the wires to one another, the droplets join and Received: November 25, 2015 Revised: April 28, 2016 Research Article pubs.acs.org/acscombsci © XXXX American Chemical Society ADOI: 10.1021/acscombsci.5b00178 ACS Comb. Sci. XXXX, XXX, XXX−XXX jxb00 |ACSJCA |JCA10.0.1465/W Unicode |research.3f (R3.6.i11:4432 |2.0 alpha 39) 2015/07/15 14:30:00 |PROD-JCA1 |rq_5494038 |5/23/2016 11:22:58 |6|JCA-DEFAULT 72 form a single alloy droplet, which is suspended only by the 73 wires feeding into it. The droplet remains suspended on the tip 74 of the wires due to surface tension. Once the mass of the 75 droplet becomes large enough, the gravity force overcomes the 76 force of surface tension and the droplet detaches from the wires 77 falling onto the substrate. As more of these alloy droplets are 78 deposited, the bottom of the sample begins to freeze. However, 79 it is still possible to maintain a molten pool in the upper region 80 of the sample. This droplet deposition sequence is repeated a 81 number of times until the desired sample height is achieved; at 82 which point the laser and wire feeders are simultaneously 83 stopped. The sample is then allowed to cool under argon 84 shielding gas with a flow rate of 5 L/min. The whole process is 85 performed in an argon filled glovebox with oxygen levels of 30 86ppm or less. 87 The key to the SDA process lies in the ability to mix and 88 alloy relatively small volumes of material while it is in contact 89 with only its constituent parts (the elemental wires). SDA is 90 achieved through precise alignment of the wires with one 91 another in a region where they will intersect the beam path and 92 melt. The wire alignment encourages the elemental droplets to 93 join and mix with one another shortly after their formation. 94 Precise control of the wire feed rates and the 100% capture rate 95 of material allows an alloy with the desired stoichiometry to be 96produced with a high degree of accuracy. 97 Feedstock material was purchased from Advent Research 98 Materials (UK). For the case of Al, Fe, and Ti, 1 mm diameter 99 wires with 99.8% purity or higher were used. The substrate 100 material, where the alloy droplets were deposited, was 101 produced from 430-grade stainless steel disc 20 mm in 102 diameter with a thickness of 2 mm. Before alloy synthesis 103 could be commenced some preparatory steps were performed. 104 The wires required to synthesize the thermoelectric alloy were 105 weighed on a four-point balance to ascertain the mass in terms 106 of a linear density (g/m). This allows a target composition in 107 atomic percentage (at. %) to be calculated to a ratio of wire 108feed rates (mm/min) for individual wires. 109 The wires required to produce the bulk material were loaded 110 into the wire feeder assemblies, inserted into a copper delivery 111 nozzle and aligned within the beam path. The feed rates (mm/ 112 min) required to produce the specific target composition were 113 calculated and entered into the control software for each wire 114 feeder. A substrate was positioned underneath the copper 115 delivery nozzle in the center of the beam path on a X,Ytable. A 116 laser power sufficient to melt the wires at the specified feed 117 rates was selected. The laser beam was then fired in continuous 118 wave mode and the wire feeders were activated. Samples are 119typically produced in 2 min. Figure 1. Schematic of wire fed suspended droplet alloying process (not to scale). Figure 2. (a) Scheme of the components of a multifunctional probe. (b) Picture of the power factor screening facility. ACS Combinatorial Science Research Article DOI: 10.1021/acscombsci.5b00178 ACS Comb. Sci. XXXX, XXX, XXX−XXX B 120 For the screening of the thermoelectric properties the 121 samples were cut to a 1−2 mm thick disc perpendicular to the 122 build direction using an AgieCharmilles Cut 20 EDM. This disc 123 was then ground and polished. Discs diameter varied from 6 to 124 15 mm. Sample composition was confirmed using a Hitachi 125 TM3000 Desktop Scanning Electron Microscope (SEM) with 126 Bruker XFlash 4010 Energy Dispersive X-ray Spectroscopy 127 (EDS) detector in conjunction with Bruker Quantax Esprit 1.9 128 software. An agreement of ±2% respect to the compositions 129 from the feed was found. No compositional gradients were 130 identified from EDS measurements performed at different 131 surface points on the sample discs Phase identification was 132 performed in selected samples by X-ray diffraction (XRD) 133 using a Phillips PW1710 Automated Powder Diffractometer 134 with copper (Cu Kα) radiation at 35 kV and 40 mA. The 135 diffractometer was controlled with PW1877 APD version 3.6 136 computer software and initial phase identification was 137 performed using PW1876 PC-Identift version 1.0b software. 138 XRD samples were prepared by grinding the discs with a pestle 139 and mortar to a powder and packing them into an aluminum 140 holder. 141 High-Throughput Power Factor Measurement Fa142 cility. The measurement of the power factor involves the 143 determination of both Seebeck coefficient and electrical 144 resistivity. Unlike thermal conductivity, the power factor can 145 be measured quicker and has been chosen as the main indicator 146 for the initial screening of the thermoelectric performance. 147 Examples of high-throughput tools for the measurement of 148 power factor can be found in the literature, most of them 149 focused on thin films. 5,7,12,13 Although it can also be used for 150 thin films, the facility we developed for the screening of the 151 power factor was designed for bulk samples. The details of the 152 apparatus have been recently reported. 14 As a summary, the 153 equipment measures the electrical resistivity using the Van der 154 Pauw method 15 and the Seebeck coefficient is measured by 155 means of a hot probe. The key aspect of the facility is the use of 156 4 multifunctional probes. Each multifunctional probe consists 157 of a Cu tube with a constantan wire welded right at the tip, f2 158 forming a T-type thermocouple as shown in Figure 2a. In this 159 way, the temperature can be measured right at each probe tip 160and the Cu part can be used to provide electrical contacts for 161the flow of current and the measurement of voltages. 162For the measurement the sample was first placed on a sample 163holder fixed on a motorized stage (Figure 2b). Then, the stage 164was lifted and the 4 multifunctional probes were contacted at 165the edges of the sample, as required by the Van der Pauw 166method. Inside probe A, a heater coil was installed to set its 167temperature ∼3 K above room temperature to allow Seebeck 168coefficient measurements. By consecutively measuring the 169temperatures at the tips of probes A (TA) and D (TD), and 170the open-circuit voltage difference (ΔV) across them through 171their Cu wires, the Seebeck coefficient of the sample was 172calculated as S=ΔV/(TA−TD)+SCu (SCu is the Seebeck 173coefficient of the Cu used in the multifunctional probes 14 ). The 174electrical resistivity was measured directly afterward (keeping 175probe A hot) by applying current across two adjacent probes 176and measuring the induced voltage difference at the other two. 177This was performed varying the polarity and alternating the 178different pairs of probes to obtain the required number of 179measurements to obtain the sheet resistance Rsand then the 180electrical conductivity ρ=Rsd(dis the thickness of the 181sample 14 ). Finally, the stage was moved down to its original 182position and the sample removed to allow the positioning of 183the next one. All the measurements, automated and controlled 184by a computer, were performed in around 20 s. To our 185knowledge, this is the fastest power factor measurement 186reported so far. By adding the time required to place the 187sample and contact the 4 multifunctional probes by means of 188the micropositioners, a total time of 1 to 2 min could be 189expected per sample. 190 ■RESULTS AND DISCUSSION 191Around 500 samples comprising single elements, binary and 192ternary alloys were initially synthesized and their power factor 193screened. From the analysis of the library of results a promising 194ternary alloy system formed by Al−Fe−Ti was identified. The 195ternary diagrams showing the Seebeck coefficient and electrical 196 f3resistivity data are shown in Figure 3. The most negative 197Seebeck coefficient value measured was −57 μV/K, corre198sponding to the Al12.5Fe37.5Ti50 composition (Figure 3a). Not 199very far values were observed for close compositions. Figure 3. Thermoelectric screening of the Al−Fe−Ti ternary alloys: (a) the Seebeck coefficient and (b) electrical conductivity. ACS Combinatorial Science Research Article DOI: 10.1021/acscombsci.5b00178 ACS Comb. Sci. XXXX, XXX, XXX−XXX C 200 Interestingly, samples with positive Seebeck coefficient could be 201 found when Ti content is decreased below 30%, with several 202 compositions showing around 27 μV/K. The lower Ti content 203 of the alloys seems to be the most significant parameter 204 affecting the transition from negative to positive Seebeck 205 coefficients, which might cause the decrease of the mean free 206path with the electron energy. 16 It should be noted that the 207 existence of both positive and negative values of Seebeck 208coefficient is highly beneficial, since both are required when the 209 materials are assembled to form a device and this characteristic 210is not always present in thermoelectric bulk materials. 211The electrical resistivity is shown in Figure 3b. They lie in the 212 order of 10−4Ωcm, an order of magnitude lower than well213established materials such as Bi2Te3. Slightly lower values were 214observed in the area of positive Seebeck coefficients (lower Ti 215 f4content). The results of the power factor are shown in Figure 4. 216The highest power factor (13.3 ×10−4W/m K2) was observed 217 in Al12.5Fe37.5Ti50, corresponding to the composition of the 218most negative value of the Seebeck coefficient mentioned 219 above. The highest power factor observed for positive Seebeck 220coefficient samples (7.0 ×10−4W/m K2) was obtained from 221Al30Fe55Ti15. It can be seen that the large power factors are 222 located around the above-mentioned compositions, and away 223from these compositions, the power factors were much lower. 224Although the screened values of the power factor are lower 225 than Bi2Te3based alloys (typically around 30 ×10−4W/m K2), 226they are comparable to the room temperature values of 227 established high temperature thermoelectric materials, such as 228skutterudites and half heuslers. It should be noted that Al, Fe, 229and Ti are nontoxic and among the most earth-abundant 230 elements of the periodic table. 2 Among a wide range of 231thermoelectric materials only silicides and oxides exhibit similar 232 abundance. The thermal conductivity of the best samples was 233also measured to be in the range of 3−9 W/m K. Clearly, the 234Al−Fe−Ti system shows promising thermoelectric properties, 235which is being further investigated. Figure 4. Results of thermoelectric screening of the power factor of the Al−Fe−Ti ternary system. Samples n1−n3 and p1−p3 are the best nand p-type samples identified, respectively. Figure 5. (a) From top to bottom, XRD of n1, n2, and n3 samples (see Figure 4). (b) From top to bottom, XRD of p1, p2, and p3 samples (see Figure 4). ACS Combinatorial Science Research Article DOI: 10.1021/acscombsci.5b00178 ACS Comb. Sci. XXXX, XXX, XXX−XXX D 236 XRD of samples with promising properties were carried out 237 to identify the crystalline phases present and to compare them f5 238 to those expected from literature phase diagrams. Figure 5a 239 shows the results of materials that exhibit n-type behavior, 240 whereas Figure 5b shows p-type samples. When comparing the 241 compositions of the 3 n-type materials to the isothermal 242 sections for this alloy system, 17 they are expected to exhibit a 243 base centered cubic (bcc) phase but may also contain the 244 hexagonal Laves (C14) phase. The presence of two peaks at 245 42°and 61°in the XRD of n2 and n3 samples (bottom and 246 middle, Figure 5a) shows that only the bcc phase is present in 247 these materials and they exhibit A2 ordering. As there are no 248 peaks to imply B2 order, no further atomic ordering in the 249 phase of these materials can be concluded. 18 Material n1 (top, 250 Figure 5a) contains an A2 bcc phase along with a minority C14 251 hexagonal phase, signified by diffraction peaks which were not 252 previously assigned to the bcc A2 phase. The presence of two 253 phases in this material is expected by considering its position on 254 the isothermal sections of this ternary system: it is located 255 between the cubic and hexagonal stable phases. 17 The three p256 type materials have compositions which are expected to 257 crystallize to form bcc structures, as shown from the isothermal 258 sections. 17 Either a disordered A2, a partially ordered B2 or full 259 Heusler bcc phase with all atoms in defined positions should be 260 expected. Our XRD results in Figure 5b imply that all three p261 type materials exhibit only a bcc phase. The peaks at 44°and 262 64°indicate that this phase presents at least A2 ordering, peaks 263 at 30.5°, 54.5°, and 72.5°show that B2 ordering is present and 264 no peaks for the L21full Heusler phase are observed. 18 The 265 above results indicate good correlation between the phases 266 identified and those expected from reported isothermal 267 sections. 268 The identification of the Al−Fe−Ti system as a promising 269 thermoelectric candidate was a surprise and the discovery of 270 both n-and p-type existing in this alloy system was completely 271 unexpected. These results demonstrate the validity of using the 272 above-reported high-throughput techniques as an effective 273 approach for accelerated discovery of thermoelectric materials. 274 They can offer the capability of high-speed discovery of 275 advanced materials, particularly among a large number of 276 ternary or quaternary intermetallic compounds. 277 ■CONCLUSIONS 278 An integrated combinatorial approach has been developed for 279 accelerated discovery of thermoelectric materials based on a 280 laser melting technique for materials preparation and a 281 multifunctional-probe facility for thermoelectric characteriza282 tion. Hundreds of samples have been synthesized and their 283 room temperature power factors were characterized. This initial 284 work has led to the identification of a promising ternary system, 285 Al−Fe−Ti, which exhibits a maximum power factor of 13.3 × 286 10−4W/m K2. This value is comparable to those of the current 287 high temperature thermoelectric materials. In addition, they are 288 abundant and nontoxic and both nand p-type exist. The 289 discovery of high power factor in Al−Fe−Ti system was 290 unexpected, which demonstrate the validity and effectiveness of 291 the high-throughput approach/techniques (reported in this 292 paper) for the development of advanced materials. 293 ■AUTHOR INFORMATION 294 Corresponding Author 295 *E-mail: min@cardiff.ac.uk. 296Present Address 297 ∥ J.G.-C.: Department of Industrial Systems Engineering and 298Design, Universitat Jaume I, Campus del Riu Sec, 12071 299Castellon, Spain. 300Author Contributions 301The manuscript was written through contributions of all 302authors. All authors have given approval to the final version of 303the manuscript. 304Funding 305The authors wish to acknowledge financial support from the 306Accelerated Metallurgy Project, which is cofunded by the 307European Commission in the seventh Framework Programme 308(contract NMP4-LA-2011-263206), by the European Space 309Agency and by the individual partner organizations. 310Notes 311The authors declare no competing financial interest. 312 ■ACKNOWLEDGMENTS 313Granta Design Ltd. is acknowledged for providing the database 314and results library. 315 ■REFERENCES (1) 316Martín-Gonzalez, M.; Caballero-Calero, O.; Díaz-Chao, P. 317Nanoengineering Thermoelectrics for 21st Century: Energy Harvest318ing and Other Trends in the Field. Renewable Sustainable Energy Rev. 3192013,24 (0), 288−305. (2) 320Gaultois, M. W.; Sparks, T. D.; Borg, C. K. H.; Seshadri, R.; 321Bonificio, W. D.; Clarke, D. R. Data-Driven Review of Thermoelectric 322Materials: Performance and Resource Considerations. Chem. Mater. 3232013,25 (15), 2911−2920. (3) 324Curtarolo, S.; Hart, G. L. W.; Nardelli, M. B.; Mingo, N.; Sanvito, 325S.; Levy, O. The High-Throughput Highway to Computational 326Materials Design. Nat. Mater. 2013,12 (3), 191−201. (4) 327Funahashi, R.; Urata, S.; Kitawaki, M. Exploration of N-Type 328Oxides by High Throughput Screening. Appl. Surf. Sci. 2004,223 (1− 3293), 44−48. (5) 330Otani, M.; Lowhorn, N. D.; Schenck, P. K.; Wong-Ng, W.; 331Green, M. L.; Itaka, K.; Koinuma, H. A High-Throughput Thermo332electric Power-Factor Screening Tool for Rapid Construction of 333Thermoelectric Property Diagrams. Appl. Phys. Lett. 2007,91 (13), 334132102. (6) 335Watanabe, M.; Kita, T.; Fukumura, T.; Ohtomo, a.; Ueno, K.; 336Kawasaki, M. Combinatorial Synthesis and High Throughput 337Evaluation of Thermoelectric Power Factor in Mg-Si-Ge Ternary 338Compounds. Appl. Surf. Sci. 2007,254 (3), 777−780. (7) 339Potyrailo, R.; Rajan, K.; Stowe, K.; Takeuchi, I.; Chisholm, B.; 340Lam, H. Combinatorial and High-Throughput Screening of Materials 341Libraries: Review of State of the Art. ACS Comb. Sci 2011, 579−633. (8) 342Hautier, G.; Jain, A.; Ong, S. P. From the Computer to the 343Laboratory: Materials Discovery and Design Using First-Principles 344Calculations. J. Mater. Sci. 2012,47 (21), 7317−7340. (9) 345Bera, C.; Jacob, S.; Opahle, I.; Gunda, N. S. H.; Chmielowski, R.; 346Dennler, G.; Madsen, G. K. H. Integrated Computational Materials 347Discovery of Silver Doped Tin Sulfide as a Thermoelectric Material. 348Phys. Chem. Chem. Phys. 2014,16, 19894−19899. (10) 349Carrete, J.; Li, W.; Mingo, N.; Wang, S.; Curtarolo, S. Finding 350Unprecedentedly Low-Thermal-Conductivity Half-Heusler Semicon351ductors via High-Throughput Materials Modeling. Phys. Rev. X 2014, 3524, 011019. (11) 353Yan, J.; Gorai, P.; Ortiz, B.; Miller, S.; Barnett, S. a.; Mason, T.; 354Stevanovic, V.; Toberer, E. S. Material Descriptors for Predicting 355Thermoelectric Performance. Energy Environ. Sci. 2015,8, 983−994. (12) 356Yan, Y. G.; Martin, J.; Wong-Ng, W.; Green, M.; Tang, X. F. A 357Temperature Dependent Screening Tool for High Throughput 358Thermoelectric Characterization of Combinatorial Films. Rev. Sci. 359Instrum. 2013,84 (11), 115110. ACS Combinatorial Science Research Article DOI: 10.1021/acscombsci.5b00178 ACS Comb. Sci. XXXX, XXX, XXX−XXX E (13)360 Watanabe, M.; Kita, T.; Fukumura, T.; Ohtomo, A.; Ueno, K.; 361 Kawasaki, M. High-Throughput Screening for Combinatorial Thin362 Film Library of Thermoelectric Materials. J. Comb. Chem. 2008,10 363 (2), 175−178. (14) 364 García-Canadas, J.; Min, G. Multifunctional Probes for High365 Throughput Measurement of Seebeck Coefficient and Electrical 366 Conductivity at Room Temperature. Rev. Sci. Instrum. 2014,85 (4), 367 043906. (15) 368 Van der Pauw, L. J. A Method of Measuring Specific Resistivity 369 and Hall Effect of Discs of Arbitrary Shape. Philips Res. Reports 1958, 370 13 (1), 1−9. (16) 371 Kasap, S. O. Thermoelectric Effects in Metals: Thermocouples; 372 University of Saskatchewan: Saskatchewan, Canada, 2001 (17) 373 Palm, M.; Lacaze, J. Assessment of the Al−Fe−Ti System. 374 Intermetallics 2006,14 (10−11), 1291−1303. (18) 375 Yin, M.; Hasier, J.; Nash, P. A Review of Phase Equilibria in 376 Heusler Alloy Systems Containing Fe, Co or Ni. J. Mater. Sci. 2016,51 377 (1), 50−70. ACS Combinatorial Science Research Article DOI: 10.1021/acscombsci.5b00178 ACS Comb. Sci. XXXX, XXX, XXX−XXX F