III. RESULTS AND DISCUSSION

The loose powders of Bi₂Te₃ were successfully processed in a layer-by-layer approach to produce a final disc-shaped part which was 8 mm in diameter and over 1 mm thick. The disk was formed with eight layers of powder with each layer about 150 µm thick. To form the disk, the optimal laser power, spot size, scan speed, and hatch distance were 25 W, ∼50 µm, 500 mm/s, and 37.5 µm, respectively. These parameters led to a laser energy density E of 1.3 J/mm², according to $E = {p \over {vh}}$, where p is the laser power, v is the scan speed, and h is the hatch distance.^{Reference Ho, Gilbson and Cheung38} The processing parameters were chosen such that the energy density was high enough to melt each layer thoroughly and low enough to avoid sample bending.

Improving density (or minimizing porosity) of the produced parts is a major challenge for SLM, regardless of whether it is applied to metals, ceramics, or semiconductors. Size distribution and particle morphology affect the ability of the powder to flow,^{Reference Wilkes, Hagedorn, Meiners and Wissenbach39–Reference Liu, Marziano, Bentham, Litster, White and Howes41} which influences packing of the powder bed and thus affects density (or porosity). For the case of Bi₂Te₃, the as-received powders had a large size distribution ranging from below 10 µm to larger than 150 µm, and the powder particles generally have irregular shapes (i.e., aspherical, high aspect ratio). Detailed powder characterization including size and morphology (e.g., aspect ratio and circularity) was reported previously.^{Reference Batista, El-Desouky, Crandall, Wang, Yang and LeBlanc42} The large size distribution and irregular particle shapes cause mechanical interlocking and lead to voids which cause low packing density of the powder bed before melting.^{Reference Batista, El-Desouky, Crandall, Wang, Yang and LeBlanc42} Consequently, the Bi₂Te₃ disks formed were not fully dense, as shown in Fig. 1(b). To increase density, we increased the laser power from 16 W to 25 W (changing the energy density from 0.85 J/mm² to 1.3 J/mm², respectively) with other parameters the same, resulting in an effective improvement of the density from 6.27 g/cm³ to 6.81 g/cm³, or 81%–88% relative density when compared to the standard Bi₂Te₃ density, 7.74 g/cm³.^{Reference Haynes43} A comparison of the samples made with low and high energy density is shown in Figs. 1(b) and 1(c). Figures 1(b)–1(d) show multiple geometries, disc and rectangular, made with the SLM process.

FIG. 1.

Optical images of parts built from commercially available Bi₂Te₃ powder using a layer-by-layer SLM approach. (a) Sketch of the SLM process on TE materials. (b and c) Parts in disk shape, with 8 mm in diameter. (d) Part in rectangular shape ∼3 mm × 11 mm. All parts are over 1 mm thick. The laser processing parameters were adjusted to increase the relative density. (b) 16 W of laser power resulted in a relative density ∼81%. (c) 25 W of laser power resulted in a relative density ∼88%.

The high density sample [shown in Fig. 1(c)] was further analyzed to demonstrate the quality of melting in the SLM processing. Figure 2 shows a scanning electron micrograph of the sample’s cross section, in-plane along the build direction (perpendicular to the laser scan direction). No separation between adjacent layers was visible across the entire surface of the cross section. The surface features some spherical pores, the majority of which are approximately 5 µm or larger in diameter. Spherical pores are generated from gas trapped during rapid solidification^{Reference El-Desouky, Carter, Mahmoudi, Elwany and LeBlanc28,Reference Kempen, Thijs, Yasa, Badrossamay, Verheecke and Kruth44,Reference Monroy, Delgado and Ciurana45} or keyhole instability and collapse.^{Reference Katayama, Kawahito and Mizutani46,Reference Lee, Ko, Farson and Yoo47} This sample’s relative density, 88%, is comparable to the control samples manufactured using hot pressing on Bi₂Te₃ (92% relative density), suggesting that the fusion of powders was quite successful through the SLM process.

FIG. 2.

Scanning electron micrograph of a polished cross section of the SLM processed Bi₂Te₃ bulk sample. Each layer is about 150 μm thick; no separation between powder layers is observable.

To explore any chemical or structural transition during the SLM processing, the processed samples were compared to unprocessed control samples using powder XRD. The results presented in Fig. 3 indicate that the two spectra are almost identical in the measured range, and no clear new phases were observed. Both spectra are well-matched with the database values for Bi₂Te₃ (PDF#00-015-0863), suggesting the SLM process preserved the structure for Bi₂Te₃. This result is consistent with the Bi–Te system’s phase diagram where the only possible phase crystallized from the liquid phase at 60 at.% of Te is Bi₂Te₃ at 585 °C.^{Reference Satterthwaite and Ure48} XRD peak intensity differences for Bi₂Te₃ from spark plasma sintering, rolling, and forging were previously reported and interpreted as crystallographic preferred orientation changes.^{Reference Zhao, Zhang, Li, Zhang and Liu49–Reference Euvananont, Jantaping and Thanachayanont51} However, our results do not indicate a change in preferred orientation. We did notice XRD intensity differences from coarse powders crushed from SLM processed samples, indicating a lack of random orientation. However, continuing to grind samples to fine powders led to intensities which were well-matched to the control samples, suggesting that the SLM process does not influence the intrinsic anisotropy for Bi₂Te₃.

FIG. 3.

XRD patterns of Bi₂Te₃ samples for original powders and those from SLM processed sample. Both spectra are well-matched with positions and intensities, and they match the database values for Bi₂Te₃ (PDF#00-015-0863).

We also characterized the thermoelectric and electrical properties of the SLM-produced samples. The room temperature Seebeck coefficient, S, and four-probe electrical resistivity, ρ, were measured in the in-plane direction using a custom-designed vacuum probe. The range of S and ρ values are 85–189 µV/K from five samples and 1–15 mΩ cm from four samples at room temperature, respectively, with the majority falling between 144 and 189 µV/K and 1–5 mΩ cm. A summary of the Seebeck coefficient and electrical resistivity values is shown in Fig. 4. The measured values vary from sample to sample, and the variation is possibly related to varying powder bed layers and defect conditions (as discussed below) which are introduced from the melting and solidification process. The aspherical powder particle morphology leads to variation in the surface condition of each powder bed layer, so there may be a slight variation in the density and porosity of the produced samples. These geometric variations affect the calculations of intrinsic physical properties since only extrinsic properties are measured (e.g., electrical resistivity is extracted from a measurement of electrical resistance). Production and selection of only spherical powder particles could narrow the variation. Our results are consistent with reported experimental values, which range from 130 μV/K to 260 μV/K for Seebeck coefficient and 0.6 mΩ cm to 2.5 mΩ cm for electrical resistivity.^{Reference Goldsmid52} For example, our sample with room temperature S and ρ values of 166 μV/K and 1 mΩ cm, respectively, is in good agreement with the reported values of 160 μV/K and 0.7 mΩ cm on regular p-type Bi₂Te₃ materials.^{Reference Goldsmid33}

FIG. 4.

A summary of the measured Seebeck coefficient and electrical resistivity at room temperature. Samples #1–#6 are the SLM-processed samples; samples #7 and #8 are fabricated through hot-pressed method on pristine powders. The processing parameters for all the samples can be found in Table S1 in the Supplementary Material.

The thermal and electrical properties of the SLM-produced samples were further characterized at high temperatures. The temperature-dependent absolute Seebeck coefficient, electrical conductivity, specific heat, thermal diffusivity, thermal conductivity, and thermoelectric figure of merit ZT were characterized up to almost 500 °C in either vacuum or a helium environment using the Linseis system; the results are shown in Figs. 5 and 6. The sample measured at high temperature (yielding the data in Fig. 5) was not the same sample as that measured at room temperature, but all samples were processed in the same way. The absolute value of the Seebeck coefficient is representative of the material’s ability to convert thermal energy to electrical energy. As temperature increased, S first decreased from about 165 μV/K at 50 °C until it reached close to zero at about 350 °C. Then, the Seebeck coefficient’s absolute value increased with temperature, up to −50 μV/K at 473 °C. The electrical conductivity shows a similar behavior, initially decreasing and then increasing around 300 °C.

FIG. 5.

The absolute Seebeck coefficient and electrical conductivity from 50 to 473 °C for the SLM-processed Bi₂Te₃ sample. The conductivity data were averaged over a set of ten measured values, and the uncertainty bar shown is from the standard deviation. The uncertainty on the measured Seebeck coefficient is ±7%, according to the Linseis technical specifications.

FIG. 6.

The thermoelectric properties for the SLM-processed Bi₂Te₃ sample from 30 up to 500 °C. (a) Both the specific heat (square) and thermal diffusivity (triangle) increase with increasing temperature in the measured temperature range. The uncertainties for specific heat and thermal diffusivity are ±5% and ±2.4%, respectively, according to the Linseis technical specifications. (b) As the temperature increases, the thermal conductivity (filled circle) increases, and the thermoelectric figure of merit, ZT (open circle), first decreases until 300 °C then increases.

Both the thermoelectric and electrical behaviors are in agreement with previous reports. Unlike thin films which can be p-type or n-type,^{Reference Zou, Rowe and Min53–Reference Kim, Choi, Bae, Kim and Oh55} both single crystals and polycrystalline bulk samples of Bi₂Te₃ are p-type at room temperature^{Reference Satterthwaite and Ure48,Reference Goldsmid52,Reference Vasenin56–Reference Nolas, Sharp and Goldsmid58} if the sample is grown slowly from a stoichiometric melt without intentional doping. There are consistent reports that Bi₂Te₃ bulk samples undergo a p-to-n conversion while heated at high temperatures if the annealing time is short (less than 4 h, e.g., Ref. Reference Schultz, McHugh and Tiller59). The magnitude of the negative S values increases as the temperature increases,^{Reference Vasenin56,Reference George, Sharples and Thompson60} which is consistent with our results (see Fig. 5). Even thin film samples which are n-type at room temperature show the same trend: an increasingly negative S and ρ that first increases and then decreases with increasing temperature.^{Reference Kim, Choi, Bae, Kim and Oh55}

There are differing perspectives about the cause of the p-to-n transition. One theory explains the transition based on crystalline vacancy defects.^{Reference Schultz, McHugh and Tiller59} Point defects act as donors or acceptors depending on their energy level relative to the Fermi level. In this theory, tellurium vacancies are the electron donors and the predominant active point defects in the Bi₂Te₃ system. The original p-type material is driven to n-type because the higher temperature effectively generates additional Te vacancies for short-time heating. The conversion from p- to n-type reflects the relative concentrations and position of the Te and Bi vacancy energy levels. George et al. also suggested that an increase of vacancy concentration, due to both thermal excitations and sintering effects,^{Reference Bernard61} will result in a n-type material because the vacancies act as hole traps.^{Reference George, Sharples and Thompson60} The idea of vacancies in Bi₂Te₃ playing a significant role was supported by several studies.^{Reference Miller and Li62–Reference Drasar, Lostak and Uher67} However, Hashibon and Elsässer^{Reference Hashibon and Elsässer68} suggested that antisite defects, Te_Bi and Bi_Te, instead of vacancies, play a dominant role in the electrical and thermoelectric properties of Bi₂Te₃. The antisite defect Bi_Te leads to p-type conduction as an acceptor, and Te_Bi leads to n-type conduction as a donor. The authors also rationalized the original formation of p-type Bi₂Te₃ by claiming that there are more Te sites than Bi sites in the lattice (3:2 ratio), so the excess of Bi_Te defects leads to p-type semiconducting behavior. Kim et al.^{Reference Kim, Choi, Bae, Kim and Oh55} explained their experiment results using this antisite defect model.

In summary, changes in electrical and thermoelectric properties can occur in binary compounds like Bi₂Te₃, depending on the thermal history of the material.^{Reference Schultz, McHugh and Tiller59} Although there are different explanations, a common theme is the electrical and thermal behavior of Bi₂Te₃ depends heavily on crystalline point defects, be they vacancy defects, antisite defects, or other types of point defects. To identify which effect dominates is beyond the scope of this work; however, we propose a general framework to explain our experimental results. The original sample is p-type with holes as the majority carriers. As temperature increases, increasing point defects act as donors until the material eventually becomes n-type. During this process, the material experiences a carrier density decrease from the annihilation of the promoted electrons and existing holes. The large decrease in the carrier concentration is accompanied by a slight increase in the carrier mobility, as supported by the findings of other studies.^{Reference Kim, Choi, Bae, Kim and Oh55,Reference Kim and Lee69} The end effect is an increase in ρ with the initial increase of temperature, as shown in Fig. 5. As the temperature further increases to above 300 °C, the mobility of carriers sharply increases, similar to the results of Kim et al.^{Reference Kim, Choi, Bae, Kim and Oh55} This improvement in the carrier mobility probably comes from grain growth of the material and a reduction in scattering centers during the high temperature treatment.^{Reference Kim and Lee69} Grain growth in Bi₂Te₃ was directly observed and supported by Schultz et al.^{Reference Schultz, McHugh and Tiller59} No grain growth was observed in the bulk Bi₂Te₃ specimens at 325 °C, but grain growth was observed at 400 and 500 °C. Higher carrier mobility leads to larger electrical conductivity, which is consistent with our observations.

To verify the conjecture that annealing leads to a transition from p- to n-type semiconductor, the pristine Bi₂Te₃ powders were hot-pressed to bulk format at 500 °C in a N₂ environment and then measured at room temperature. Two hot press runs, one for 10 min and another for 3 h, generated bulk samples with densities of 6.1 g/cm³ and 7.1 g/cm³ (about 79% and 92% relative density), respectively. The thermoelectric and electrical properties of both samples agree. The higher density sample had an S of −108 μV/K and ρ of 0.58 mΩ cm while the other sample’s properties were −90 μV/K and 0.6 mΩ cm, respectively. Compared with the positive S values from the SLM-produced samples, these results confirm high temperature annealing changed Bi₂Te₃ from a p-type to n-type semiconductor.

Figure 6(a) shows the measured specific heat and thermal diffusivity in the cross-plane direction; both properties increase with temperature. The thermal conductivity was then calculated according to them. The thermal conductivity increased with increasing temperature over the entire measurement region as shown in Fig. 6(b). With measured properties, the thermoelectric figure of merit ZT was calculated, as shown in Fig. 6(b). The behavior of ZT with temperature is primarily dominated by the electrical conductivity and the Seebeck coefficient. ZT first decreased with increasing temperature until about 300 °C, slightly increased between 300 and 400 °C, then increased more sharply with increasing temperature. The highest ZT value was about 0.11 at 50 °C. The S and ZT values of the SLM-processed bulk Bi₂Te₃ are in the same range as other p-type or doped n-type bulk samples.^{Reference Wright70} These results indicate that the SLM method is capable of producing Bi₂Te₃ samples with thermoelectric properties comparable to samples produced by traditional methods.