Characterization and modeling of nanocomposite thermoelectric materials system bismuth antimony telluride ((Biy̳Sb1̳-̳y̳)2̳Te3̳) as a function of temperature and magnetic field

Abstract

This thesis looks at (BiySb1-y)2Te3 nanocomposites as an example of the currently available nano systems. In this thesis, (BiySb-y)2Te3 nanocomposites are characterized from ~325K down to ~3K. Advantages of this low temperature regime include the minimization of lattice vibrations and the decreasing of ke with decreasing temperature. As a result, nano effects on IL could be better observed and characterized in this low temperature regime. We are also interested in studying the effect of an applied magnetic field on the conduction carriers in this low temperature regime. We like to find out whether an applied magnetic field could impede the carriers' heat conducting ability more than their current conducting ability. Therefore, a magnetic field effect study is also carried out to see whether any improvement in ZT could be achieved by the applying of a magnetic field. The measurement system used in this thesis is QD PPMS. Only the ACT and TTO options of the QD PPMS apparatus are used for measurements in this thesis. Under the ACT option, Hall and 4-pt p measurements on the same sample are performed. On the other hand, Kti, S, and 2-pt p measurements are performed simultaneously on the same sample under the TTO option. Both the ACT and TTO options use an AC current instead of a DC current during p measurement to eliminate any unwanted Seebeck voltage. Since the ability to perform correct measurements on thermoelectric samples is not a trivial task, benchmarking with known results is a must. In this thesis, I calibrate our QD PPMS against both the manufacturer's results and the published data, and demonstrate that our measurement system gives accurate results. I also benchmark our nt, results under a magnetic field using a pyrex sample. Our results confirm that the QD PPMS apparatus does not introduce artifacts under an applied magnetic field. Thus, any changes observed under the QD PPMS apparatus measurements in an applied magnetic field would be expected to be solely due to the sample. Lastly, no measurable difference is found between our 2-pt p (TTO) and 4-pt p (ACT) measurements. A total of eight (BiySb1-y) 2Te3 samples are measured in this thesis. The sample set includes: (a) one bulk ingot sample manufactured by Marlow (Ingot), (b) four nanocomposite samples (XY21, XY146, XY144, and GJ99) made by collaborators from Boston College (BC) where the letters simply indicate the sample maker's initials, and (c) three nanocomposite samples (0%, 40%, and 100%) made by collaborators from Nanyang Technological University (NTU) in Singapore, where the % denotes the weight % of the nanoinclusions prepared via melt spinning [1] in the sample. All the nanocomposite samples in this thesis are made solely for research uses and are purposely fabricated under conditions different than those used for the best samples previously reported [2, 1]. Although BC and NTU use different starting materials, different fabrication machines, and different fabrication parameters, the resulting densities of the nanocomposites from the ball-milled nanopowders alone (XY21, XY146, XY144, GJ99, and 0%) are almost identical. Moreover, the addition of nanoinclusions prepared via melt spinning decreases the sample mass density somewhat.

(cont.) From the XRD measurement results, we notice that (a) both the NTU and BC samples have the same XRD peak locations, (b) the NTU samples have a lower intensity for peaks (1 0 10) and (1 1 0), and (c) the NTU samples have a higher intensity for peaks (0 0 g) where g is an integer. Comparing the XRD patterns with the reference database, the difference in peak intensities is a good indication that the NTU samples are not completely randomized and have internal preferred orientations. From the SEM images, we notice that the NTU samples and the BC samples are markedly different. For example, the BC samples are shown to have grains in the pm range with a small grain size distribution. On the other hand, the grain sizes of the NTU samples decreases with the addition of nanoinclusions prepared via melt spinning. Moreover, the NTU samples have a wider grain size distribution that ranges from nm to pum. This observed difference is believed to arise from the difference in the fabrication techniques used by the BC and NTJ teams. Temperature-dependent hta, S, and p measurements, along with the carrier concentration measurements, are performed on all samples. All samples are found to be p-type materials. Transport measurements are performed both // and I to the press direction for the nanocomposite samples, and only // to the growth direction for the Ingot sample. Anisotropic behavior is observed in ~tha nd p for all the nanocomposite samples investigated, with the anisotropy being always higher in p than in th . On the other hand, S is found to be isotropic. Thus, care needs be taken during the fabrication process to ensure that no unwanted anisotropic behavior is introduced. Common p and S features among all samples include: (a) a dramatic decrease in the peak value of sth for the nanocomposite samples when compared with the Ingot //'s value, (b) constant slope &S/&T for T < 20K, (c) constant slope &S/ln(T) of ~ 130-140pV/ln(K) for 200K < T < 300K, (d) close-to-zero slope in p for T < 20K, and (e) p cx T'.1 0 for 200K < T < 300K. From the measured stl, S, and p data, the mobility pp, hole mean free path e, and phonon mean free path Eph are computed. It is found that nanocomposite approach decreases lp, fe, and Eph. Moreover, the pp, fe, and E values are always lower in the // direction for the nanocomposite samples than in the I direction. Furthermore, 4e in general is in the nm range while eph ranges from pm to nmn as the temperature increases. Therefore, if one wants to decrease the sl,, a possible solution is to decrease f further. However, in order not to affect the p too much, the lower limit for f should be in the nm range. As a result, decreasing f would have the biggest effect on Kth in the low temperature regime. Using the Kth and - data, KL is extracted through the intercept method (see Section 3.5.3). This method only makes sense if all the samples have similar f. Since pressure is coining from top and bottom during the fabrication process for the nanocomposite samples, my samples are expected not to behave as the same materials system along the // direction. However, for the I direction, they can be considered as a same materials system since no pressure is applied. The 40% and 100% samples are believed to deviate from the results for the 0% sample because of the presence of nanoinclusions in them. rlLL is found to be 0.76W/mn-K for the 40% sample at 297K. When I compare this value with previously determined values for Bi 2Te3 (1.4W/in-K [3]) and (Bio.3Sbo. 7)2Te3 (0.9W/m-K [4]) alloy at 300K, these results confirm that the nanocomposite approach does indeed lower the lattice thermal conductivity. The semi-classical model is then used to interpret the various transport coefficients (o- = 1/p, S, and t;e) and is based on the Boltzmann Transport Equation (BTE) under the relaxation time approximation (RTA). Acoustic phonon scattering, ionized impurity atom scattering, neutral impurity atom scattering, alloy scattering for a 3-atom II"_1II system, point defect scattering, grain boundary scattering, and polar optical phonon scattering are considered for the electrons. On the other hand, boundary scattering, point defect / alloy scattering, and Umklapp scattering are considered for the phonons. We find that for holes, point defect scattering dominates at low T, while acoustic phonon scattering dominates around 300K. As for phonons, boundary and point defect scattering mechanisms dominate at low T, while point defect and Umklapp scattering mechanisms dominate at high T (~300K). We also find that the nano approach increases the crossover temperature Tcross. For the electron model, we observe that the deformation potential (DA) seems to be both process dependent and materials dependent. We see that DA changes from the BC samples to the NTU samples (process dependent). Moreover, DA changes in the NTU samples when going from 0% to 100% (materials dependent). From the electron model, the ionized impurity atom concentration Ni and the neutral impurity atom concentration No reflect the somewhat anisotropic behavior of all the samples investigated. Lastly, f seems to play little role in the determination of p. For the phonon model, we observe that C plays a rather important role in the determination of r1L, especially at low temperatures. The value of C seems to be consistently lower for the BC samples than for the NTU sample (0%) for the nanocomposite samples made solely from ball-milled nanoparticles. We also see that the Umklapp scattering contribution (B') has a materials dependent factor, where B' decreases from 10x101'8s/K for the nanoparticle nanocomposite samples to e 4x1018 s/K for (cont.) the nanocomposite sample made using 100% nanoinclusions prepared via melt spinning.

Furthermore, we see that the point defect contribution (A') reaches the highest value when both the nanoparticles and nanoinclusions prepared via melt spinning are present in the nanocomposite samples (e.g. the 40% sample), similar to the alloying effect on the thermal conductivity. In general, it is desirable to increase the values of A' and B', resulting in a decrease in the KL values. However, care needs to be taken to ensure that the phonon parameters are independent of the electron parameters so that no adverse effect on ZT would result. The determination of L is also carried out based on my electron model findings. We observe that L is isotropic. Moreover, L for each sample investigated reaches the same value of 2.44x10-8 W-Ohm/K2 as T -> OK (completely degenerate limit of (+_ )2). Furthermore, the higher the hole concentration the sample has, the higher its C value is at a given temperature. Lastly, I find that a lower f leads to higher ZT values at 297K for the BC nanocomposite samples measured in the _L direction. On the other hand, a lower f leads to lower ZT values at 297K for the NTU nanocomposite samples measured in both the // and _L directions. From the magnetic field studies on the Ingot and on the 40% samples, few important facts are demonstrated. First, an applied magnetic field can be used to effectively increase the ZT of (BiySbpy)2Te3 , especially at temperatures below 200K. Use of a magnetic field might theoretically extend the effective temperature ranges over which (BiySb-y)2Te3 materials can be used for refrigeration. Second, the data under various applied B fields allow me to confidently calculate the C value below the temperature ranges where a plateau has occurred. Third, the data under various applied B fields serve as an important guideline for both validating any electron model and extrapolating values for L above the plateau occurrence temperatures. As a result, this allows me to get some insights into the temperature dependence of L (see Figure 4-14). Fourth, from the magnetic field dependent transport studies on our samples, we observe that the applied B field pushes away the holes more effectively in the Ingot // than the holes in the nanocomposite samples. We also find that the VvtIplateau values obtained under the magnetic field study serve as a more realistic and practical limit for KL. Lastly, from the magnetic field-dependent studies, we find that having point defects as the dominant scattering mechanism for the carriers results in an increase in ZT under an applied magnetic field. It would be extremely useful if one can make a sample such that the point defect dominant regime is extended to higher temperatures, resulting in a shift of the increase in the ZT ratio regime to a temperature range closer to room temperature (300K). However, care needs to be taken to ensure that such modifications would result in an increase in the ZT values under an applied magnetic field.