How to Measure Thermoelectric Properties Reliably

Tian-Ran Wei,Mengjia Guan,Junjie Yu,Tiejun Zhu,Lidong Chen,Xun Shi

DOI: https://doi.org/10.1016/j.joule.2018.10.020

IF: 46.048

2018-01-01

Joule

Abstract:Tian-Ran Wei is currently an assistant professor at the School of Materials Science and Engineering, Shanghai Jiao Tong University. He obtained his PhD from Tsinghua University in 2017. He worked at Shanghai Institute of Ceramics, Chinese Academy of Sciences from 2017 to 2018 before joining Shanghai Jiao Tong University. His current research focuses on advanced thermoelectric materials.Mengjia Guan is a PhD student at the Shanghai Institute of Ceramics, Chinese Academy of Sciences. She obtained her BS from Central South University in 2014. Her research focuses on superionic copper chalcogenide thermoelectric materials.Junjie Yu is a PhD student at the School of Materials Science and Engineering, Zhejiang University. He obtained his BS from Zhejiang University in 2015. His research focuses on half-Heusler thermoelectric materials.Tiejun Zhu is the Qiushi Distinguished Professor of Materials Science and Engineering at Zhejiang University. He obtained his PhD from Zhejiang University in 2001. He was a research fellow at Singapore-MIT Alliance from 2002 to 2004, and a visiting scientist at California Institute of Technology in 2011. He has been a faculty member at Zhejiang University since 2004. His current research focuses on the advanced fabrication, microstructure, and transport mechanisms of high-efficiency thermoelectric materials.Lidong Chen is a Professor of Shanghai Institute of Ceramics, Chinese Academy of Sciences (SICCAS), and the Co-Editor-in-Chief of npj Computational Materials. He received his BS (1982) from Hunan University and doctorate (1990) from Tohoku University. After being chief engineer and postdoctoral appointment at Riken Cooperation and Japan National Aerospace Laboratory, respectively, he worked at the Institute for Materials Research, Tohoku University, as research associate and associate professor. He joined SICCAS as a professor in 2001. His research mainly focuses on the design, synthesis, and characterization of thermoelectric materials and devices.Xun Shi is a professor at the School of Materials Science and Engineering, Shanghai Jiao Tong University. He received his BS (2000) in Tsinghua University and PhD (2005) in the University of Chinese Academy of Sciences. He worked at the University of Michigan (USA) as a postdoc from 2007 to 2009. He then joined the R&D center in General Motors. Since 2010 he has been a professor at the Shanghai Institute of Ceramics, Chinese Academy of Sciences (SICCAS). His current research focuses on advanced thermoelectric materials and flexible semiconductors. Tian-Ran Wei is currently an assistant professor at the School of Materials Science and Engineering, Shanghai Jiao Tong University. He obtained his PhD from Tsinghua University in 2017. He worked at Shanghai Institute of Ceramics, Chinese Academy of Sciences from 2017 to 2018 before joining Shanghai Jiao Tong University. His current research focuses on advanced thermoelectric materials. Mengjia Guan is a PhD student at the Shanghai Institute of Ceramics, Chinese Academy of Sciences. She obtained her BS from Central South University in 2014. Her research focuses on superionic copper chalcogenide thermoelectric materials. Junjie Yu is a PhD student at the School of Materials Science and Engineering, Zhejiang University. He obtained his BS from Zhejiang University in 2015. His research focuses on half-Heusler thermoelectric materials. Tiejun Zhu is the Qiushi Distinguished Professor of Materials Science and Engineering at Zhejiang University. He obtained his PhD from Zhejiang University in 2001. He was a research fellow at Singapore-MIT Alliance from 2002 to 2004, and a visiting scientist at California Institute of Technology in 2011. He has been a faculty member at Zhejiang University since 2004. His current research focuses on the advanced fabrication, microstructure, and transport mechanisms of high-efficiency thermoelectric materials. Lidong Chen is a Professor of Shanghai Institute of Ceramics, Chinese Academy of Sciences (SICCAS), and the Co-Editor-in-Chief of npj Computational Materials. He received his BS (1982) from Hunan University and doctorate (1990) from Tohoku University. After being chief engineer and postdoctoral appointment at Riken Cooperation and Japan National Aerospace Laboratory, respectively, he worked at the Institute for Materials Research, Tohoku University, as research associate and associate professor. He joined SICCAS as a professor in 2001. His research mainly focuses on the design, synthesis, and characterization of thermoelectric materials and devices. Xun Shi is a professor at the School of Materials Science and Engineering, Shanghai Jiao Tong University. He received his BS (2000) in Tsinghua University and PhD (2005) in the University of Chinese Academy of Sciences. He worked at the University of Michigan (USA) as a postdoc from 2007 to 2009. He then joined the R&D center in General Motors. Since 2010 he has been a professor at the Shanghai Institute of Ceramics, Chinese Academy of Sciences (SICCAS). His current research focuses on advanced thermoelectric materials and flexible semiconductors. Thermoelectric (TE) materials, which can convert energy directly between heat and electricity, have gained a lot of attention recently.1Snyder G.J. Toberer E.S. Complex thermoelectric materials.Nat. Mater. 2008; 7: 105-114Crossref PubMed Scopus (8002) Google Scholar The energy conversion efficiency is determined by a material's TE performance, which is governed by the dimensionless figure of merit zT (= S2σT/κ, where S is the Seebeck coefficient, σ is the electrical conductivity, T is the absolute temperature, and κ is the thermal conductivity). A high conversion efficiency requires a high zT, i.e., large S and σ but low κ. The zT is difficult to directly measure in experiments, but can be calculated using measured S, σ, and κ. S and σ can be directly measured at both low and high temperatures. The κ below room temperature can also be directly measured by using instruments such as the physical property measurement system (PPMS). However, the κ above room temperature is usually calculated via κ = CPdλ through measuring the material's heat capacity CP, density d, and thermal diffusivity λ. The measurement principles and methods for TE properties are shown by Borup et al.2Borup K.A. de Boor J. Wang H. Drymiotis F. Gascoin F. Shi X. Chen L. Fedorov M.I. Müller E. Iversen B.B. et al.Measuring thermoelectric transport properties of materials.Energy Environ. Sci. 2015; 8: 423-435Crossref Google Scholar Usually the uncertainties of commercial instruments are ±3% for σ, ±4% for S, ±1% for d, ±3% for λ, and ±5% for high-temperature CP. Thus, the error for zT is more than ±15%. However, the repeatability, which is the closeness of agreement between successive measurements carried out under the same condition, can be much better than the accuracy when using those well-established commercial instruments. Thus repeatability is more important and useful when comparing TE properties in experiments. Over the past 20 years many publications have shown that zT values are significantly improved in many types of TE materials, which is mainly ascribed to the deep understanding on TE transport mechanisms and quick technology innovation.1Snyder G.J. Toberer E.S. Complex thermoelectric materials.Nat. Mater. 2008; 7: 105-114Crossref PubMed Scopus (8002) Google Scholar However, because it requires measuring a few parameters to calculate zT, the zT values are greatly divergent in some types of TE materials. This gives rise to the question of what are the true zT values in these materials. Furthermore, all the related insightful analyses and discussions should be based on correct experimental data, which relies on reliable measurements. In this commentary, we summarize a few key concerns about TE property measurements covering all electrical and thermal transport parameters. Various possibilities in the measurements will be given to show how much the properties deviate from their exact values. These concerns should be taken seriously during measurements in order to obtain reliable zT values. Figure 1A is the schematic map for S and σ measurements in commercial instruments such as Ulvac-Riko ZEM and Linseis LSR systems. For a specimen under temperature gradient, the temperatures at A and B are measured by two thermocouples to calculate the temperature difference (ΔT = TB − TA). The voltage (ΔV) between A and B is measured by using the positive (or negative) wires of two thermocouples. Usually at least three pairs of ΔV and ΔT are collected to plot the ΔV∼ΔT curve. The curve slope is measured Seebeck coefficient, SM, at temperature (TA + TB)/2. The metal wires themselves also have Seebeck coefficients (SW), which has to be subtracted from SM to obtain the material's S. Such correction is automatically implemented in commercial instruments but is often neglected in home-made instruments. Frequently, ΔV is not zero at ΔT = 0 K. This will introduce noticeable voltage offset for SM.2Borup K.A. de Boor J. Wang H. Drymiotis F. Gascoin F. Shi X. Chen L. Fedorov M.I. Müller E. Iversen B.B. et al.Measuring thermoelectric transport properties of materials.Energy Environ. Sci. 2015; 8: 423-435Crossref Google Scholar, 3Iwanaga S. Toberer E.S. LaLonde A. Snyder G.J. A high temperature apparatus for measurement of the Seebeck coefficient.Rev. Sci. Instrum. 2011; 82: 063905Crossref PubMed Scopus (224) Google Scholar If S is directly calculated by ΔV/ΔT, SM may significantly shift from its true value.4Liu Y. Fu C. Xie H. Zhao X. Zhu T. Reliable measurements of the Seebeck coefficient on a commercial system.J. Mater. Res. 2015; 30: 2670-2677Crossref Scopus (6) Google Scholar The ΔV∼ΔT curve may deviate from linearity if ΔT is too large (ΔT > 10 K) and has appreciable errors if ΔT is too small (ΔT < 1 K). Thus the ideal ΔT range (ΔTmax–ΔTmin) should be between 3 and 5 K. The S∼T plot of Nb0.85CoSb is shown in Figure 1B. A small ΔT range (squares, ∼1.2 K) gives a zigzag curve, while a large ΔT range (circles, ∼3.2 K) gives a very smooth curve. Correspondingly, the measurement error for the large ΔT range, marked by the shadow, is obviously smaller than the small ΔT range. As indicated in the inset of Figure 1B, when ΔT range is small, the error is large and the derived S deviates much from the real value. On the contrary, if a reasonably larger ΔT range is applied, the measured S value is close to the real one. It is noteworthy that people usually set a fixed temperature difference (∼40 K) between two electrodes (Figure 1A), but the ΔT range may become smaller and smaller when chamber temperature increases because of severe heat radiation, resulting in noticeable fluctuations and errors in the measurement (see the data above 1,000 K in Figure 1B). Therefore, the applied temperature gradient between two electrodes should be carefully programmed to maintain a proper ΔT range during the whole measurement. In addition, several other factors such as the differences in thermocouples, reactions between thermocouple and samples, bad contact, and cold finger effect2Borup K.A. de Boor J. Wang H. Drymiotis F. Gascoin F. Shi X. Chen L. Fedorov M.I. Müller E. Iversen B.B. et al.Measuring thermoelectric transport properties of materials.Energy Environ. Sci. 2015; 8: 423-435Crossref Google Scholar, 5Yazdani S. Kim H.Y. Pettes M.T. Uncertainty analysis of axial temperature and Seebeck coefficient measurements.Rev. Sci. Instrum. 2018; 89: 084903Crossref Scopus (6) Google Scholar can also affect the measured S. In both commercial and home-made instruments, the thermocouples may become polluted and dirty after running for a long time. Thermocouples may react with materials during measurements to change S and σ. Figures 1C and 1D compare TE properties of Ba0.08La0.05Yb0.04Co4Sb12 measured by the dirty and new thermocouples. The deviations for S and σ are ∼10% around 800 K, leading to a deviation near 30% for power factor (PF). The dirty thermocouples give a zT over 2.1, 20% higher than the reported 1.7.6Shi X. Yang J. Salvador J.R. Chi M. Cho J.Y. Wang H. Bai S. Yang J. Zhang W. Chen L. Multiple-filled skutterudites: high thermoelectric figure of merit through separately optimizing electrical and thermal transports.J. Am. Chem. Soc. 2011; 133: 7837-7846Crossref PubMed Scopus (1135) Google Scholar Therefore, timely cleaning and periodical replacement of thermocouples as well as clocked calibrating on standard samples are indispensable for reliable measurements. In addition, reasonably large pressure is required for good contacts between thermocouples/electrodes and thermoelectric materials, but too large a pressure may lead to material's bending at high temperatures. The standard four-probe method is widely used in commercial instruments for σ measurement (Figure 1A), which is calculated by σ = L/(RAS) where R is the electrical resistance between points A and B, AS is the cross-sectional area, and L is the probe distance. This requires an accurate probe distance. However, in some commercial instruments, the probes are fixed with a default probe distance that is not automatically calibrated for every measurement. After running for a long time and thermal cycling, L may be greatly shifted from its default value due to thermal expansion or aging. Therefore, probe distance must be measured for each measurement. Low-temperature κ is usually measured by the thermal transport option (TTO) in PPMS. The measurement principle of two-probe configuration is schematically depicted in Figure 2A. Heat pulse is applied on one side of the sample by running current through the heat shoe. The temperatures of both sides are recorded to calculate the temperature difference ΔT = Thot− Tcold. κ can then be derived via Fourier 's law for the given heat power and sample size. The thermometers (Cernox, low-temperature thermocouple) are easily damaged and their resistances may be shifted from the original calibrations. In this case, incorrect κ values and temperature dependence may be obtained. As shown in Figure 2B, multiple-filled skutterudites exhibit a somewhat amorphous character but are actually typical crystalline compounds. After calibration of thermometers on standard samples, the crystalline behavior in κL-T is observed.7Shi X. Salvador J.R. Yang J. Wang H. Thermoelectric properties of n-type multiple-filled skutterudites.J. Electron. Mater. 2009; 38: 930-933Crossref Scopus (49) Google Scholar Therefore, it is recommended to regularly calibrate thermometers. Sample size also strongly affects the measurements when using PPMS. As recommended by the PPMS manual, for common TE materials with κ ranging from 2 to 5 W/mK at 300 K, bar-shaped samples are recommended, but if κ is smaller than 1.5 W/mK, a disk sample with very small height is required. Thermal diffusivity λ is usually measured by the laser flash method. As shown in Figure 2C, laser heats one side of sample and the heat signal on the back side is simultaneously recorded varying with time. Normally, the back-side signal quickly increases to reach a peak value and then slowly decays. λ is calculated via λ=0.1388z2/t1/2, where z is the thickness of the sample and t1/2 is the time when the signal of the back side reaches the half-maximum value. The raw signal is automatically corrected by the software considering the shift of baseline (left panel of Figure 2D). The correct signal shape should be like that shown in Figure 2C, otherwise it may give extra errors to the measured λ. For materials with large thickness and low κ, the signal correction may fail to give reliable curves. For example, Cu2Se has very low κ and λ. For a disk with a diameter of 10 mm, when its thickness is 1 mm we observe correct back-side signals and the t1/2 is nearly the same for three independent measurements (shots), giving λ = (0.573 ± 0.003) mm2/s, although the signal intensity of the first shot is smaller than that of the second and third. However, when its thickness is 3 mm, the raw signals for all three shots do not have peaks. Furthermore, the baseline correction only works for the first shot and fails for the other two. Thus a large discrepancy in λ is obtained for the three shots with the values of 0.537, 0.308, and 0.337 mm2/s. The instrument finally uses the averaged value, which deviates considerably from the true value. Therefore, the back-side signal should be checked for each measurement. To avoid failure of signal correction, low-κ materials should have very small thickness while high-κ materials should have reasonably large thickness. In addition, sensor contamination or lack of liquid nitrogen can cause poor signals. Of course, the samples must not be tilted and the size of the sample should fit the holder well to avoid the leakage of light. Differential scanning calorimetry (DSC) and PPMS are more frequently used for high- and low-temperature CP measurements, respectively. Similar to low-temperature κ measurement, the calibration in PPMS must be often performed to avoid incorrect CP measurement. For DSC measurement, pellets, rather than powders, should be used to avoid thermal inhomogeneity and gas desorption on the particle surface. Unfortunately, for many TE materials, although a standard measurement process is carried out, the CP measured by DSC may have bad repeatability from time to time and from person to person even when using the same sample and instrument. High-temperature CP can also be measured together with λ by a commercial laser flash instrument, but the results are usually unreliable.8May A.F. Toberer E.S. Saramat A. Snyder G.J. Characterization and analysis of thermoelectric transport in n-type Ba8Ga16-xGe30+x.Phys. Rev. B. 2009; 80: 125205Crossref Scopus (346) Google Scholar Therefore, it is strongly suggested to use theoretically calculated CP values to compare with the experimental data (Figure 2E). The calculated CP using the Dulong-Petit value is too rough, but is reasonably acceptable when including the contribution from the material's thermal expansion. Thus the total CP is a sum of Cph,H, Cel, and CD, which are contributed by phonons, electrons, and dilation of the lattice, respectively.9Delaire O. May A.F. McGuire M.A. Porter W.D. Lucas M.S. Stone M.B. Abernathy D.L. Ravi V.A. Firdosy S.A. Snyder G.J. Phonon density of states and heat capacity of La3-xTe4.Phys. Rev. B. 2009; 80: 184302Crossref Scopus (81) Google Scholar For anisotropic materials, all the properties must be measured along the same direction, which is usually ignored for new anisotropic materials. If the direction of low κ and the direction of high PF are combined, the zT will be overestimated. For example, the anisotropy of polycrystalline Bi2Te3-based materials is largely related to its texture degree, which strongly depends on fabrication processing. Samples with high texture degree can show unreasonably high zT if the electrical and thermal properties are measured along different directions.10Shen J. Hu L. Zhu T. Zhao X. The texture related anisotropy of thermoelectric properties in bismuth telluride based polycrystalline alloys.Appl. Phys. Lett. 2011; 99: 124102Crossref Scopus (112) Google Scholar As shown in Figure 3A, even if the anisotropy is not very strong for a hot-pressed (HP) sample, the zT can still be ∼30% higher than the real value. Many measurement errors also come from poor quality of materials, i.e., macroscopic and microscopic inhomogeneity, and chemical and thermal instability. For large-scale samples, chemical composition may be macroscopically non-uniform. As we know, TE properties are very sensitive to carrier concentrations, which may be significantly changed for a tiny composition variation. Taking Cu1.97Se0.5S0.5 as an example, the ingot (the size is Φ10 × 20 mm) after direct melting and annealing processes (Figure 3B) is not homogeneous. The chemical compositions vary from the top to bottom. We cut the pieces from both top and bottom parts to run TE property measurements with the data shown in Figures 3C and 3D. The top has a low PF and κ while the bottom has a high PF and κ. The calculated zTs using different parts are definitely incorrect and greatly shifted from the true value11Zhao K. Qiu P. Song Q. Blichfeld A.B. Eikeland E. Ren D. Ge B. Iversen B.B. Shi X. Chen L. Ultrahigh thermoelectric performance in Cu2-ySe0.5S0.5 liquid-like materials.Mater. Today Phys. 2017; 1: 14-23Crossref Scopus (117) Google Scholar (zT ∼ 1.2, see Figure 3D). Therefore, to obtain correct/reliable zT for large-scale or inhomogeneous materials, it is recommended that specimens from the same part (as close as possible) are used. When using the spark plasma sintering (SPS) technique to sinter TE materials with low melting point and high vapor pressure elements such as In, Ga, and Zn, the large electric current may destroy the local compositions and microstructures to introduce some inhomogeneity if the insulating coating is not good enough. As shown in Figure 3E, In4Se3 prepared by SPS exhibits an “abnormal” transition in S between 400 and 600 K. In contrast, the HP sample has very smooth curves. This phenomenon is probably related to the inhomogeneous distribution of In in an SPS sintered sample. The material's instability may also obviously affect TE property measurements. For composite Ba0.44Co4Sb12/C60, C60 reacts with Ba to form Ba6C60 and reduce Ba content in skutterudites. If the reaction is not completed during measurements, extra chemical potential is generated. Such potential has weak contributions to σ and κ, but sharply increases material's S, as shown in Figure 3F. The S dramatically increases with increasing T, reaching ∼−600 μV/K at 823 K, giving a zT of around 10. However, after annealing the sample at high temperatures for a week, the S is lowered back the normal values, giving a zT a value slightly larger than 1.0.12Shi X. Chen L.D. Bai S.Q. Huang X.Y. Zhao X.Y. Yao Q. Uher C. Influence of fullerene dispersion on high temperature thermoelectric properties of BayCo4Sb12-based composites.J. Appl. Phys. 2007; 102: 103709Crossref Scopus (65) Google Scholar In addition, element volatilization and material deformation occur in many materials at high temperature, which cause considerable effects on sample geometry, density, and composition, leading to additional measurement errors. Therefore, to obtain reliable data samples homogeneity and stability should be ensured, probably by annealing materials for enough time. Repeated or cycling measurements can be carried out to check the material's stability. Furthermore, many materials have phase transitions. Commercial instruments can easily give incorrect data during the phase transition. Thus, the TE properties during phase transitions should be carefully considered and usually removed if they are measured by commercial instruments. We have summarized several key concerns about the measurement of TE properties. A few examples are given to show the extent of deviation that occurs in these TE properties by proper and improper measurements. If abnormally measured values are obtained, additional checks of original signals, raw data, measurement conditions, instrument parameters, and materials' quality is strongly recommended. Although the error of the calculated zT is very large (more than ±15%) when simply accumulating the uncertainty of each measured TE parameter, its measurement repeatability should be limited within ±5% on the same well-established instrument if these concerns are taken into account. After one obtains reliable TE properties, various analyses are performed to understand the underlying thermoelectric transports, thus to provide useful and correct information for future study.