Over the past two decades several nano-structuring methods have helped improve the figure of merit (ZT) in the state-of-the art bulk thermoelectric materials. While these methods could enhance the thermoelectric performance of p-type Bi2Te3, it was frustrating to researchers that they proved ineffective for n-type Bi2Te3 due to the inevitable deterioration of its thermoelectric properties in the basal plane. Here, we describe a novel chemical-exfoliation spark-plasma-sintering (CE-SPS) nano-structuring process, which transforms the microstructure of n-type Bi2Te3 in an extraordinary manner without compromising its basal plane properties. The CE-SPS processing leads to preferential scattering of electrons at charged grain boundaries, and thereby increases the electrical conductivity despite the presence of numerous grain boundaries, and mitigates the bipolar effect via band occupancy optimization leading to an upshift (by ~ 100 K) and stabilization of the ZT peak over a broad temperature range of ~ 150 K.
Despite its simplistic nature and unique technical merits, thermoelectricity has previously found only a few niche applications mainly due to its low conversion efficiency1,2. For broader applications, it is crucial to develop high performance thermoelectric (TE) materials whose figure of merit (ZT) is in the neighbourhood of 2, where , and α is the thermopower (i.e., Seebeck coefficient), ρ the electrical resistivity, κT the total thermal conductivity, T the absolute temperature, and PF the power factor. The inherent inter-dependence among α, ρ, and κ poses a major roadblock in realizing TE materials with higher ZT values2,3. In this context, tailoring the micro- and nano-structures at multiple length scales using advanced materials preparation methods has to a certain extent been able to decouple the inter-dependence among the TE properties; α, ρ, and κ4,5.
However, with many TE elements (or materials) in a TE device, there are several secondary “limiting factors” besides ZT that must be overcome to achieve high TE device efficiency, viz.: (i) suitability of n- and p-type materials: a similar composition, performance, thermal and mechanical stability for the n- and p-legs of a TE device is required; (ii) a wide operating temperature range for ZT: in practice, a less temperature-dependent and a reasonably high ZT over a broad temperature range is desirable in contrast to a high ZT over a narrow temperature range; (iii) a low electrode contact resistance: for a given PF, a lower ρ is preferred over a higher α since the former results in a lower electrode contact resistance and thus lesser parasitic losses; and (iv) a weak temperature dependence of the compatibility factor (Σ): a weakly temperature-dependent compatibility factor (ratio of the electrical current density to heat flux) is highly desirable6,7,8. Failure to properly address these limiting factors makes the effort of enhancing ZT futile. It is then quite pertinent to explore nano-structuring processes that improve ZT without adversely effecting factors (i)–(iv).
Since its discovery in 1950s9, Bi2Te3 alloys have been probably the most widely studied and device utilized TE materials to date. More recently, Bi2Te3 has gained renewed interest as an intriguing example of a three-dimensional topological insulator10,11. In regards to its TE properties, state-of-the-art Bi2Te3 materials (typically, p-type Bi0.5Sb1.5Te3 and n-type Bi2Te2.7Se0.3) are the only TE materials that can be used for both power generation and refrigeration with a room temperature value of ZT ~ 112. A greater opportunity for an improved performance in these TE materials can be expected if the above limiting factors (i) to (iv) are mitigated. Thus far, researchers have successfully enhanced ZT in p-type Bi2Te3 by subjecting the bulk materials to nano-structuring processes, such as ball milling13, melt spinning14,15, thermal forging16,17 and hydrothermal synthesis18,19, etc., and melt-spun p-type Bi2Te3 ranks high in the list of current state-of-the-art TE materials with a room temperature ZT ~ 1.520. In contrast, similar approaches have been ineffective so far in enhancing ZT in n-type Bi2Te3 due to its strong texture and highly anisotropic transport properties21,22,23,24. Compared to bulk p-type Bi2Te3, bulk n-type Bi2Te3 is easily cleavable25 and traditional nano-structuring processes (e.g., ball milling26) severely deteriorates its in-plane (basal plane) transport that is crucial for realizing potential high ZT values. Furthermore, the mismatch of grain boundaries adversely affects the electrical conduction, and hence an effective nano-structuring process is needed for n-type Bi2Te3 that preserves the in-plane micro-morphology to a maximum extent.
The chemical exfoliation method has been successful in generating quasi-two-dimensional materials, which exhibit the same basal plane properties as that of the parent bulk materials27,28. Upon densification of chemically exfoliated n-type Bi2Te3, one can then control structuring at micro- and nano- length scales, while preserving the basal-plane properties. Ideally, grain boundaries should exhibit multi-tier scattering, i.e., they should scatter: i) phonons more effectively than charge carriers, ii) low energy charge carriers more effectively than high energy charge carriers (i.e., energy filtering), and iii) holes (minority carriers) more effectively than electrons (majority carriers) in the case of n-type materials to suppress the bipolar effect29,30. In this article, we prepared bulk materials comprised of nanostructured n-type Bi2Te3 (Bi2Te2.7Se0.3), and optimized its ZT over a broad range of temperature. We adopted a chemical exfoliation method, which resulted in few-layered n-type Bi2Te3 (Bi2Te2.7Se0.3), and through transport measurements observed the multi-tier preferential scattering described above. Most importantly, we found that a combination of chemical exfoliation and spark plasma sintering (CE-SPS) enables one to attain the multi-tier preferential scattering in n-type Bi2Te3.
A commercial state-of-the-art n-type Bi2Te3 (Bi2Te2.7Se0.3) ingot exhibits a lamellar morphology with layers extending to hundreds of microns (Fig. 1a). Upon chemical exfoliation for 8 hours in N-methyl-2-pyrrolidinone (sample labelled as exf8h), the layer dimensions are reduced to 500–800 nm in length and a few tens of nm in thickness (Figs. 1b–c). Next, the exf8h samples were spark plasma sintered (see Methods section) to yield pellets (exf8h-SPS) whose TE properties and Raman spectra are discussed below. Interestingly, exf8h-SPS exhibited a lamellar micro-structure with an average thickness of 20–50 μm (Fig. 1d) which resulted from the stacking of exfoliated nano-sheets along the c-axis during the SPS process. Thus, the combined CE-SPS process introduced more grain boundaries/interfaces as evidenced in Fig. 1d, and the high-resolution powder x-ray diffraction (XRD) data (see supplementary information, Fig. S1) of exf8h-SPS corresponded to that of single phase Bi2Te2.7 Se0.3 (JCPDS card no. 00-050-0954). The broadening of the (0,0,6) peak (Table S1), as inferred from its full-width at half maximum intensity or FWHM (δθ) for the exf8h sample is consistent with the formation of few-layered nano-sheets (cf. Figs. 1b–c). Furthermore, the average values for the coherence length (δL) showed a significant increase for the exf8h-SPS sample along the (0,0,6) direction while the coherence length did not change significantly along the (1,1,0) direction (see supplementary information, Fig. S1 and Table S1). Such observations imply that the crystallinity or the long-range order is improved due to spark plasma sintering of exf-8h samples despite the use of high-energy processes such as ultrasonication and SPS.
The temperature dependences of α, ρ, and κ of exf8h-SPS samples are compared in Fig. 2 with the corresponding properties of state-of-the-art n-type Bi2Te3 ingot. It is noteworthy that all transport properties were measured in the same direction (perpendicular to the SPS direction; see supplementary information, Fig. S3) and reproduced on at least 3 samples. Furthermore, the low and high temperature measurements performed using two different techniques showed an excellent match for all transport properties. As shown in Fig. 2a, the peak thermopower value (αmax) for exf8h-SPS samples decreased and shifted towards higher temperatures, in agreement with the Goldsmid-Sharp relation31. The upshift in αmax to higher temperature suggests that the bipolar contribution (of holes) to the electrical conduction occurs at a relatively higher temperature in exf8h-SPS, compared to that for the ingot. Although the α values for the exf8h-SPS are relatively lower in magnitude, its metal-like electrical resistivity (Fig. 2d) was found to be lower for T > 100 K making the power factor values (Fig. 2b) comparable to that of the ingot. Indeed, the net reduction of α for the exf8h-SPS samples is related to a decrease in the ρ as expected from the inverse coupled behaviour. It is interesting to note that the ρ of exf8h-SPS samples decreases relative to the commercial ingot despite the increase in the density of grain boundaries. Such a result is unexpected since an increase in grain boundaries is expected to result in increased scattering of charge carriers, and consequently a higher ρ. In order to gain a deeper understanding of the origin of this unexpected decrease in ρ, we performed Hall measurements to estimate the carrier concentration (n) and mobility (μ). Inferring from Figs. 3a–b, the observed decrease in ρ is clearly associated with an increase in the n at T > 100 K. Previously, Teweldebrhan et al.25 reported that the chemical/mechanical exfoliation of bulk Bi2Te3 leads to two kinds of products, those cleaved at the van der Waals gap (see Fig. S2) in which the quintuple units remain intact, and those in which the cleavage occurs within the quintuple units leading to donor-like defects with positively charged TeBi anti-sites/Te vacancies on the surface/boundaries of the exfoliated grains32 (see Fig. 3c). As shown in the schematic in Fig. 3c–e, such localized positive charges on the grain boundaries are known to inject excess electrons into the bulk material, thereby increasing n13. More importantly, any positively charged defects on the grain boundaries (hence forth referred as interfacial charged defects) act as a potential barrier (Fig. 3e), which selectively filters low-energy minority carriers (holes in the case of n-type Bi2Te3) and thereby shifts the onset of bipolar effects (two carrier conduction) to a higher temperature (cf. Figs. 2a, b and e). This bipolar shift also applies to the (κT − κe) of the sample and is discussed below.
Interestingly, the μ of the exf8h-SPS sample shows lower values for T < 100 K despite its improved crystallinity relative to the commercial ingot (Fig. 3b). This decrease in μ is due to additional scattering arising from the charged defects present at the interfaces (Fig. 3d). Depicted in Fig. 3b are various temperature dependent scattering processes that are known to limit μ of electrons according to Matthiessen's rule33. Evidently, the temperature dependence of μ for the exf8h-SPS sample deviates from that of the commercial ingot for T < 100 K, possibly due to an increased contribution to carrier scattering arising from charged defects (~T3/2) and grain boundary scattering (; where φa is the barrier potential due to the interfacial charged defects and κB is the Boltzmann constant). For T < 100 K, the electrical resistivity (ρ = 1/σ, where σ ∝ neμ) of the exf-8h and ingot samples was approximately the same since the increase in n is compensated by a decrease in μ. For T > 100 K, the mobility of both the samples exhibited the same values, implying that the decrease in ρ is entirely due to the increase in n. As a result of the upshift in αmax, the bipolar term in the thermal conductivity is also shifted to higher temperatures (cf. Fig. 4a) and hence the total κ above 300 K shows smaller values for the exf8h-SPS samples relative to the commercial ingot (cf. Fig. 2e). The evident reduction in the magnitude of the κlattice peak (shown by the red down-pointing arrow in Fig. 2e) confirms that new long-range phonon scattering centres (e.g., grain-boundaries/interface) are introduced by exfoliation and the subsequent SPS process, i.e., the CE-SPS process. It is important to measure all thermoelectric properties in the same direction to avoid an overestimation of ZT. For this reason, we consistently measured κ in the same direction as α and ρ (cf. Fig. S3) according to a protocol we developed previously34. Consequently, the upshift due to the bipolar contribution leads to the observed upshift of ZT peak to higher temperatures (Fig. 2c). A detailed discussion and the underlying mechanism for this upshift are presented in the following sections. Furthermore, the concomitant changes in α, ρ, and κ are nearly balanced out in the exf8h-SPS sample for T > 300 K, making ZT values constant (~1.0) in the range of 380–500 K.
The Debye temperature (θD) of pristine Bi2Te3 is known to be ~ 145 K35. At sufficiently high enough temperatures (close to θD), the phonon-phonon scattering dominates due to the Umklapp processes and therefore the lattice thermal conductivity exhibits a (1/T)-temperature dependence as shown in Eq.136 where M is the average mass per atom, V is the average atomic volume, γ is the Grüneisen parameter and h is Planck's constant. At higher temperatures (T > 300 K), the linear increase in the thermal conductivity with temperature can be explained by the dominating bipolar conduction in the system. The bipolar diffusion contribution to κT arises when both holes and electrons contribute to conduction. The bipolar thermal conductivity (κB) can be expressed as , where αe (σe) and αh (σh) are specific contributions of electron and hole carriers to the thermopower (electrical conductivity) respectively37. As shown in Fig. 4a, (κT − κe) is plotted against 1000/T for the temperature range where the U-process dominates (T > 100 K). Since, κT = κL + κe + κB, the temperature dependence of κT − κe can serve as a gauge for the interplay between the lattice and bipolar effects. Clearly, the commercial ingot samples exhibited the expected linear behaviour with respect to T−1 (from phonon-phonon scattering related to κL in Eq. 1) above the red dashed line at ~2.5 K−1 in Fig. 4a, while bipolar effects dominated at higher temperatures (1000/T ~ 2.75 or T > 400 K). Intriguingly, the exf8h-SPS samples showed an evident deviation from the linear behaviour in κL in the intermediate range (100 K < T < 200 K) indicating the presence of an additional scattering process, which is consistent with interfacial charged defect scattering. Furthermore, the net reduction in κL, highlighted by the grey area in Fig. 4a, can be attributed to the increased number of grain boundaries in the exf8h-SPS sample. Most importantly, the ~ 100 K upshift (indicated by the left pointing arrow in Fig. 4a) in the exf8h-SPS sample due to bipolar contributions does not occur until 400 K (1000/T ~ 2.5 in Fig. 4a), which is again consistent with the shift observed in the electrical transport measurements. Finally, it is possible to delineate the individual contributions of the electronic (κe), lattice (κL) and bipolar conduction processes to κT based on their temperature dependences in different temperature regimes. This is based on the assumption that the dominant mechanisms contributing to the phonon conduction in this system are the electronic, lattice and bipolar conduction processes for κT. A summary of estimated values and their percentage contributions to κT for the commercial ingot and exfoliated sample are shown at two different temperatures (Table S2) along with the individual contributions in Figure S3 (graph of the (κT − κe) vs. 1000/T). For details, please see supplementary information section, Figure S3–S4.
Thus far, we have explained the effects of the exfoliation on the TE properties of n-type Bi2Te3. An obvious question that arises is the following: Is it necessary to exfoliate n-type Bi2Te3 to a single-layer level in order to achieve improved TE performance? To address this question, we performed a detailed study in which we monitored the effects of exfoliation time on α, ρ, and κ. Interestingly, we observed that the α and ρ are nearly independent of the exfoliation times greater than 3 hours (from 3 to 8 hours), thereby leading to similar power factor values in the higher temperature range for all the samples (see Fig. S6). This observation of similar power factor values indicates that TE performance saturates beyond 3 hours of exfoliation time suggesting that a single-layer level is not necessary to achieve optimal properties. We have also performed detailed atomic force microscopy (AFM) and micro-Raman spectroscopic measurements in order to further elucidate the effects of exfoliation on layer thickness and the introduction of defects. Our AFM studies (Fig. S7) showed that the layer thicknesses vary from ~ 1–5 μm in the bulk to ~ 120 nm in exf3h samples, indicating that the quintuple cell in n-type Bi2Te3 is cleaved beyond the van der Waals gap during exfoliation. Upon exfoliation, the layer thickness of the exf5h samples reduced to ~ 50 nm, and remained unchanged with extended exfoliation time. These results explain why the TE properties of the exf3h-SPS, exf5h-SPS, and exf8h-SPS samples are nearly the same (Fig. S6). As shown in Fig. 5, the micro-Raman spectra of the exfoliated samples exhibited some interesting features that confirmed the presence of exfoliation-induced defect centers. The low frequency region (60–150 cm−1) showed the presence of Eg2, and A1g2 modes for all samples, and no discernible shift in the Raman peak positions was observed. Notably, we observed the emergence of new modes in the mid-frequency region (200–400 cm−1) exfoliated n-type Bi2Te3. Interestingly, the sharp feature at ~ 760 cm−1 was found to increase in intensity with increasing exfoliation time (inset figure in Fig. S6). These new features in the micro-Raman spectra are tentatively attributed to combination and overtone modes that arise due to defect-induced symmetry breaking38,39. A detailed Raman study will be published elsewhere.
In ideal TE devices, the TE properties must remain optimal over a broad operating temperature range. It is commonplace in practice that different materials are often segmented together, with each material operating in its most efficient temperature range, for an overall optimal efficiency of the TE device. Under such constraints, the optimal efficiency is obviously limited by the properties of the segmented materials6,7. The compatibility factor (Σ) of a TE material is defined in terms of α and ZT as6: Σ for various materials used in the segmentation should be similar in order to achieve high device efficiency. If Σ of the segmented materials differs by a factor of two, the segmentation suffers parasitic losses, thereby lowering device efficiency7. As shown in Fig. 4b, the compatibility factor of exf8h-SPS is appreciably higher than that of the n-type commercial ingot and matches that of n-type PbTe (~3.5 V−1 at 500 K) in the maximum operating temperature range of both the materials7, making it suitable for segmentation without inflicting substantial parasitic losses. Ideally, Σ should be a temperature independent parameter. However, the strong temperature dependence of α and ZT values often leads to a Σ that varies with temperature, making it difficult to optimize the heat and electric current densities along the TE leg that are necessary to take advantage of a large ZT. Therefore, one of the major concerns for developing Bi2Te3-based TE devices is the change in the Σ over the operating temperature range. To account for the variation of Σ as a function of temperature, some materials are often crafted into different shapes to maintain a constant current density throughout the n- or p-type legs. Such shaping effects can result in mechanical instability of the device if Σ varies greatly with temperature. In this regard, it is desirable to have a low fractional change in Σ for robust TE device integration. As shown in Fig. 4c, exf8h-SPS samples exhibit a significantly lower fractional change in Σ compared to a state-of-the-art commercial n-type Bi2Te3 ingot in 300–500 K range indicating that the CE-SPS processed samples are better suited for device integration. Furthermore, the CE-SPS processed samples showed constantly high ZT values from 380–500 K along with Σ values with smaller fractional change from 300–500 K, as well as better compatibility with probable higher temperature TE materials. These advances are critical for the integration of existing non-traditional materials into TE devices.
In summary, we show that a combined chemical exfoliation/SPS technique can mitigate the limiting factors discussed in the introductory paragraph, and can lead to the synthesis of n-type Bi2Te3 with desirable TE properties over a broad temperature range. Particularly, chemical exfoliation allows for the introduction of micro-and nanostructured scattering centres at multiple length scales while preserving the basal plane properties needed for preserving high ZT values. In our method, the possible introduction of interfacial charged defects (TeBi antisites/Te vacancies) resulted in: i) the injection of electrons into the bulk, thereby increasing the carrier concentration, and ii) the introduction of a potential barrier that selectively filtered low-energy minority carriers (holes in the case of n-type Bi2Te3 samples), thereby shifting the onset of bipolar effects. Our micro-Raman spectra exhibited distinct features in the mid- and high-frequency ranges for exfoliated samples confirming the interfacial charge defect-induced symmetry breaking. Another evidence of an additional scattering mechanism in the exf8h-SPS sample is provided by our thermal conductivity data in the intermediate temperature range. An estimation of the bipolar contribution to the thermal conductivity clearly shows a shift in the temperature region where the bipolar effects dominate. The change in the onset of bipolar effects results in the shift of the ZT maxima and at the same time broadens the operating temperature regime. A slight reduction and a shift in the peak of α is balanced by decrease in ρ and κ, leading to peak ZT values, which are same as the commercial ingot. Lastly, the reduction of α while keeping the ZT values constant increases the compatibility factor.
Preparation of nanostructured bulk n-type Bi2Te3 samples
In this work, we achieve nano-structuring via a solvent exfoliation technique of optimized state-of-the-art n-type Bi2Te3 ingot samples (purchased from Marlow industries) with a chemical composition Bi2Te2.7Se0.3. Exfoliated sheets were sintered together to form nanostructured bulk samples using a Spark plasma sintering (SPS) system.
The chemically exfoliated few layer nanosheets of a commercial ingot with a chemical composition Bi2Te2.7Se0.3 were obtained by an ultra-sonication technique. For solvent exfoliation, bulk Bi2Te2.7Se0.3 (~1 g) was dispersed in 100 ml of N-methyl-2-pyrrolidinone (NMP) and sonicated using 1/8” tip sonicator (Branson 250) at 100 W for 1 hr. The resulting dispersion was filtered through a 0.45 μm nylon filter and re-suspended in 100 ml of fresh NMP. Subsequently, the solution was bath sonicated for 3, 5, or 8 hrs and centrifuged at 2000 rpm for 45 min. The supernatant was vacuum filtered using a 0.45 μm nylon filter. Finally, the filtered powder was washed several times using deionized water to remove residual NMP. As a result of ultra-sonication, exfoliated few layer nano-sheets ~ 10–15 nm in thickness and < 1 μm in length were obtained (c.f. Fig. 1d). Further, the exfoliated sheets were compacted using spark plasma sintering (SPS, Dr. Sinter Lab®-515S system) technique. In our previous studies on p-type Bi2Te3, we observed that a higher SPS temperature and a low SPS pressure results in better thermoelectric properties owing to better electrical connectivity and small textural effects34. Hence, we used similar conditions where the samples were sintered at 500°C and 30 MPa pressure under a dynamic vacuum in this study. Before the SPS process, the samples were loaded into graphite dies, and graphite rods were used to apply the pressure. The resulting SPS pellets were 12.5 mm in diameter and 2–3 mm in thickness. The SPS process yielded samples with 98–99% of the theoretical density.
The microstructural and chemical analysis was performed using conventional SEM (Hitachi® S3400N and S4800), TEM (Hitachi® H7500) and high-resolution XRD characterization methods (Rigaku Ultima® IV diffractometer, Cu Kα radiation, λ = 1.5406 Å). Furthermore, micro-Raman spectra was taken on all the samples to observe any symmetry breaking or crystal structure changes using a Dilor XY triple grating monochromator with an excitation wavelength of 514.5 nm.
Transport property measurements
Low temperature (in the range of 20–325 K) resistivity (ρ = 1/σ) and α were measured using a four-probe differential method on a custom-designed system40. In addition, the thermal conductivity (κT) was measured on the same sample using a four probe steady-state method on a custom designed system from 20–320 K41. Both systems used custom designed measurement programs developed in our laboratory using Labview® software. The high temperature electrical resistivity and thermopower were simultaneously measured using the commercial ZEM-2 (Ulvac Riko, Inc.) system under low pressure of a high purity He atmosphere. The high temperature thermal conductivity was calculated using the relation κT = ρDCpD; where ρD is the packing density of the material, D the thermal diffusivity and Cp (≈CV, for solids) the specific heat capacity. The heat capacity measurements were carried out using a NETZSCH® DSC 404 C, thermal diffusivity was measured using a NETZSCH® LFA 457 system and the packing density was measured by the Archimedes principle. The lattice contribution of the thermal conductivity was calculated by subtracting the electronic portion from κT. The electronic thermal conductivity was calculated using the Wiedemann-Franz relation (κe = L0T/ρ, where L0 = 1.66 × 10-8 V2/K2 for the nanostructured sample42).
It is important to note that the laser flash measures the diffusivity along the SPS pressure direction, whereas the low temperature transport properties are measured along the direction perpendicular to SPS direction (in the plane of pellet). The anisotropy in the transport properties along different directions may lead to erroneous interpretation of the ZT values. In order to measure all the properties along the same direction, we cut several bars of our samples and re-stuck them together after rotating the bars by 90 degrees using JB Weld® (a thermally conducting and electrically insulating glue). A detailed study of the measurements in different configurations using this glue can be found in our previous work34. The Hall measurements were performed on a Quantum Design® physical properties measurement system (PPMS) under a magnetic field sweep of ± 5 kOe. All measurements were performed on at least two independent sets of samples to confirm our transport data.
PP, RP, and AMR are grateful to Prof. Malcolm. J. Skove, Department of Physics and Astronomy, Clemson University for valuable discussions on compatibility factor. AMR acknowledges the support from US National Science Foundation grant CMMI-1246800 award. JH acknowledges the support from NSF DMR-1307740 award.