Yu Maoa,
Yonggao Yan*a,
Keping Wua,
Hongyao Xiea,
Zekun Xiua,
Jihui Yangb,
Qingjie Zhanga,
Ctirad Uherc and
Xinfeng Tang
*a
aState Key Laboratory of Advanced Technology for Materials Synthesis and Processing, Wuhan University of Technology, Wuhan, Hubei 430070, China. E-mail: yanyonggao@whut.edu.cn; tangxf@whut.edu.cn
bDepartment of Materials Science and Engineering, University of Washington, Seattle, Washington 98195, USA
cDepartment of Physics, University of Michigan, Ann Arbor, Michigan 48109, USA
First published on 18th April 2017
Commercial production of thermoelectric (TE) modules features energy-intensive and time-consuming processes. Here, we propose a rapid, facile and low cost fabrication process for n-type single phase Bi2Te2.7Se0.3 that combines self-propagating high-temperature synthesis (SHS) with the laser non-equilibrium 3D printing method based on selective laser melting (SLM). The optimal SLM processing window for high quality single layers has been determined. Results show that the chemical composition of the sample is very sensitive to the laser energy density (EV) due to the selective vaporization of Se and Te. For energy densities EV of less than 33.3 J mm−3, the composition of the SLM-processed samples is relatively stable. However, as EV exceeds 33.3 J mm−3 and increases further, the vaporization rate of Te and Se significantly increases and is much higher than that of Bi. Empirical formulae relating the chemical composition of the resulting materials with the values of EV are obtained and are used to predict the composition of the SLM-processed material. Most importantly, the temperature dependent TE properties of the SLM-fabricated bulk sample result in a maximum ZT value of 0.84 at 400 K, which is comparable to that of the commercially available material. The work has laid a foundation for the future utilization of this technique for the fabrication of Bi2Te3-based thermoelectric modules.
Unlike the traditional manufacturing methods, selective laser melting (SLM) is an example of the additive manufacturing technology, in which a thin layer of powder is melted using a laser beam and subsequently rapidly solidifies. In this way, 3D objects of full density and virtually any shape can be built by successive steps of powder deposition followed by laser-induced melting.11–13 The SLM process features rapid heating and cooling rates during the non-equilibrium laser processing and as such, the technique is able to produce materials with fine nanostructures. Moreover, the raw materials are utilized more efficiently and there is no need for post-processing. With the above advantages, SLM has been widely employed in the manufacture of metallic parts for the aerospace and automobile industries.12,14–16 If one could use SLM in the parallel production of p- and n-type TE legs, as well as in rapid joining of TE legs to electrodes and insulating substrates, the fabrication of TE modules would be carried out in just one integrated step with much improved efficiency and reduced cost. The key element in the SLM-based technique of TE module manufacturing is the fabrication of TE legs with good control over their chemical composition, phase, microstructure and TE performance.
The SLM technique has been successfully used in the preparation of structural materials, mostly metals and alloys, which have rather high melting points.16–19 Moreover, physical properties of structural materials are mainly affected by the microstructure and the forming quality. In contrast, thermoelectric materials are semiconductors that usually possess a lower melting point, lower thermal conductivity, poor ductility and weak thermal shock resistance. All of the above are likely to contribute to the formation of macro and micro defects during SLM processing of semiconductors. Even more important, since the properties of TE materials are more sensitive to their chemical composition, phase structure and microstructure, it is a considerable challenge to obtain p/n TE legs of desirable composition, free of defects and with high TE performance via the laser non-equilibrium technique. Indeed, there are very few reports in the literature employing the SLM technique in the preparation of thermoelectric materials. In 2016, El-Desouky et al.20,21 tried to process Bi2Te3 by the SLM technique, and explored the effect of various processing parameters on the depth of the molten pool and on the resulting microstructure. However, the chemical composition and thermoelectric properties of their SLM-processed samples were not studied. As the TE properties of Bi2Te3-based materials are very sensitive to their chemical composition, it is of vital importance to assess changes in the chemical composition arising during the SLM process. For example, a slight change in the ratio of Te to Se in n-type Bi2Te3−xSex can greatly affect the final electronic properties of the material.22–24
In this study, n-type Bi2Te2.7Se0.3 powder, prepared by the ultra-fast and low-cost self-propagating high-temperature synthesis (SHS),25 was used in SLM experiments. The influence of the laser volumetric energy density on the macroscopic defects, chemical composition, phase structure and thermoelectric properties was systematically studied. As a result, the optimal processing parameters have been determined. In addition, we have proposed empirical formulas to predict and achieve the precise control over the actual composition of samples following the SLM process. Based on these formulas, the content of Te in the raw material has been modified to fabricate SLM samples having a similar composition and comparable TE performance to n-type Bi2Te3-based materials prepared by the traditional synthesis routes. The work demonstrates the viability of SLM as a rapid fabrication technique of TE materials and perhaps even future TE modules.
![]() | ||
Fig. 1 (a) SEM image and (b) particle size distribution for the SHS-prepared Bi2Te2.7Se0.3 powder used for SLM. |
The particle size was measured by a laser diffraction technique (Marlven, Mastersizer 2000). The surface morphology and microstructure of the samples were characterized by field emission scanning electron microscopy (Hitachi FESEM, SU8020). The phase identification was performed by an X-ray diffraction (PANalytical Empyrean) apparatus operating with a Cu Kα radiation at 40 kV and 40 mA. The actual chemical composition of samples was analyzed by electron probe micro-analysis (JEOL EPMA, JXA-8230). The spatially resolved Seebeck coefficient was measured by a potential-Seebeck-microprobe instrument (Panco PSM) with a spatial resolution of 20 μm. The temperature dependent electrical conductivity (σ) and Seebeck coefficient (α) for SLM bulk samples were measured simultaneously using commercial equipment (ZEM-3, Ulvac Riko, Inc.) under a low pressure He atmosphere in the temperature range of 300–550 K. The thermal conductivity (κ) of the SLM bulk samples was calculated from the relationship κ = DCpd, where D is the thermal diffusivity obtained by the laser flash method (LFA-457, Netzch, German), Cp is the specific heat measured by a differential scanning calorimeter (DSC Q20, TA Instrument, USA), and d is the density measured by the Archimedes method. The Hall coefficient (RH) at room temperature was determined by a Physical Properties Measurements System (PPMS-9, Quantum Design, USA) with the magnetic field of 1 T. The corresponding carrier concentration (n) and carrier mobility (μH) were calculated by the following equations: n = 1/eRH and μH = σRH.
![]() | (1) |
![]() | ||
Fig. 2 SEM images of four typical surface morphologies of SLM-prepared layers: (a) overheated; (b) distorted; (c) flat; and (d) balling. (e) Processing map for the first layer of Bi2Te2.7Se0.3. |
Fig. 2(a)–(d) show four typical morphologies (overheated, distorted, flat and balling) of the single layers (2 × 2 mm2) obtained under different laser power (P = 3–10 W) and scanning speed (ν = 50–500 mm s−1). Meanwhile, the hatch spacing and powder layer thickness were set at 0.05 mm and 0.03 mm, respectively. The scanning strategy is sketched with a red line in Fig. 2(c).
Based on the above four typical morphologies, the processing parameters can be divided into four groups delineated in Fig. 2(e). At high laser powers (P > 8 W) and low scanning speeds (ν < 200 mm s−1), resulting in EV > 33.3 J mm−3, the boundary region between the processed and unprocessed layer of powder develops a distinct concave profile, as shown in Fig. 2(a). When such a high EV is applied, the temperature of the molten pool increases and the heat-affected zone is large. Eventually, the powder outside the laser irradiated region also melts and the concave boundary forms. At the same time, the viscosity of the melt decreases with the increasing molten pool temperature, leading to an increasing liquid flow rate driven by Marangoni effect.27 Therefore, a ridge-like feature forms inside the layer.
When the same laser power (P > 8 W) is coupled with a high scanning speed (ν > 200 mm s−1), EV decreases to around 10–33.3 J mm−3, and the concave boundary layer disappears. However, a raised edge is found at both ends of the laser scanning path, which is mainly due to the fact that the actual scanning speed at the start and end points of the laser path is much lower than the speed over the central region. Consequently, the laser beam has interacted with the powder for a longer time at both ends of the path, the temperature of the molten pool has increased, giving rise to raised edges at both ends, as shown in Fig. 2(b). This phenomenon becomes more pronounced when the scanning speed is higher.
Fig. 2(c) shows a layer with a flat surface finish obtained when the laser power decreased to P ≤ 8 W yet EV was maintained in the range 10–33.3 J mm−3. In this case, the boundary becomes straight with no ridge-like feature observed on the surface. This demonstrates that the laser power is the dominant factor in determining the quality of each layer.
When the laser power is so low or the scanning speed is so high that the energy density falls below 10 J mm−3, the laser beam is unable to melt the powder bed completely. Lower temperature of the melt and its higher viscosity results in a large wetting angle between the molten pool and the substrate and, eventually, it gives rise to balling,28 as seen in Fig. 2(d).
The above results imply that, in order to obtain a high quality n-type Bi2Te3 material, the laser power should be below 8 W, while EV should be controlled between 10 and 33.3 J mm−3. Of course, as the powder layer thickness is altered, the optimal processing window also changes. Nevertheless, the optimal processing window obtained here serves as a good starting point for further optimization to ensure successful fabrication of Bi2Te3-based thermoelectric materials.
Sample | EV (J mm−3) | P (W) | ν (mm s−1) | D (mm) | H (mm) | Actual composition |
---|---|---|---|---|---|---|
1 | 12.5 | 6 | 200 | 0.08 | 0.03 | Bi2Te2.60Se0.23 |
2 | 20.0 | 6 | 200 | 0.05 | 0.03 | Bi2Te2.49Se0.20 |
3 | 26.7 | 4 | 100 | 0.05 | 0.03 | Bi2Te2.44Se0.22 |
4 | 33.3 | 6 | 200 | 0.03 | 0.03 | Bi2Te2.42Se0.21 |
5 | 40.0 | 6 | 100 | 0.05 | 0.03 | Bi2Te2.24Se0.18 |
6 | 44.4 | 8 | 200 | 0.03 | 0.03 | Bi2Te2.27Se0.19 |
7 | 53.3 | 4 | 50 | 0.05 | 0.03 | Bi2Te2.04Se0.16 |
8 | 66.7 | 6 | 100 | 0.03 | 0.03 | Bi2Te1.98Se0.11 |
9 | 80.0 | 6 | 50 | 0.05 | 0.03 | Bi2Te2.05Se0.11 |
10 | 88.9 | 8 | 100 | 0.03 | 0.03 | Bi2Te1.99Se0.10 |
Based on the Langmuir formula,29 the vaporization rate Ji (g cm−2 s−1) of element i in a molten pool of an alloy can be calculated as
![]() | (2) |
![]() | (3) |
![]() | ||
Fig. 3 (a) Calculated vaporization rates of Se, Te and Bi; (b) dependence of parameters α and β dependence on the laser energy density EV. |
Since the vaporization rate of Bi during the SLM process is relatively low, we have chosen the combined total molar ratio of Te and Se in Bi2Te2.7Se0.3 (designated as α) and the relative molar ratio of Se with respect to Te (designated as β) to characterize the chemical composition of each sample. Fig. 3(b) shows the dependence of α and β on EV, where the black and red dashed lines show α (0.6) and β (0.111) of the raw powder, while the blue dashed line is located at EV = 33.3 J mm−3. The results suggest that both α and β of the SLM samples are significantly lower than those of the raw powder due to preferential vaporization of Se and Te during the SLM process. In fact, the data suggest that the loss of Se is more serious than the loss of Te, consistent with the higher vapor pressure of Se compared to Te. When EV is less than 33.3 J mm−3, α and β show little fluctuation because the molten pool's temperature is low under such a low energy density, and the vaporization rate is, consequently, also low. As EV increases above 33.3 J mm−3, α and β decrease rapidly, corresponding to an enhanced vaporization of Se and Te. A linear fit to the data in Fig. 3(b) yields empirical formulas for α and β with EV in the range of EV > 33.3 J mm−3:
α = −0.00113EV + 0.600 | (4) |
β = −0.000712EV + 0.111 | (5) |
![]() | ||
Fig. 4 SEM images showing size and distribution of pores on polished SLM samples under different energy densities EV. |
Fig. 5 shows XRD patterns obtained on the SLM-processed layers at different EV. At low applied EV of 20 J mm−3, a single phase Bi2Te3 is obtained. However, as the laser energy density increases to 40 J mm−3, a second phase with composition Bi4Te5 appears. As EV increases to 66.7 J mm−3 and 80 J mm−3, the vaporization rate increases sharply and α is reduced to 0.53 and 0.51, respectively, resulting in the formation of BiTe. One can conclude that the enhanced vaporization of Se and Te leads to significant deviations in the stoichiometric ratio, which, in turn, brings about the formation of secondary phases and phase segregation. Therefore, as long as the form quality is acceptable, one should try to use the lowest possible laser energy density EV to avoid excessive losses of anion elements.
The Te-compensated n-type Bi2Te2.7Se0.3 bulk sample of 1.5 mm thickness was fabricated by stacking 60 layers on top of each other. The bulk sample shows a 97% relative density and is shown in Fig. S1 of the ESI.† The carrier concentration calculated from the measured RH of this sample is 3.18 × 1019 cm−3 with the Hall mobility of 154 cm2 V−1 s−1, both values very close to those of the SHS-prepared samples.24 Fig. 7(a)–(d) show the temperature dependent TE transport properties of the SLM bulk sample after annealing at 673 K for 36 h. Properties of the SHS-prepared Bi2Te2.85Se0.15 sample and of the ZM Bi2Te2.79Se0.21 sample24 are shown for comparison. Compared to the ZM sample, the SLM bulk sample shows modest electronic properties but a lower thermal conductivity at high temperatures due to limited intrinsic excitation.
![]() | ||
Fig. 7 Temperature dependence of thermoelectric properties of SLM-prepared bulk sample: (a) electrical conductivity, (b) Seebeck coefficient, (c) thermal conductivity, and (d) ZT. The SHS and ZM data are from (ref. 24). |
It is important to note that, the thermal conductivity of the SLM bulk sample was measured parallel to the stacking direction z, while the electrical conductivity was measured in the xy plane (σxy), perpendicular to the stacking direction, as depicted in Fig. 7. The thermal and electrical conductivity along the same direction could not be obtained for the time being due to the limited thickness of samples our customized SLM apparatus can handle. However, by using a relationship curve proposed by Shen et al.,32 the thermal conductivity in the xy plane (κx) could be deduced from the measured thermal conductivity (κz) and the orientation factor F of (00l) diffractions observed on the SLM bulk sample. The XRD pattern (Fig. S2†) collected from the xy plane of the SLM bulk sample shows strong texture along the (110) direction, while the XRD pattern collected from the xz plane shows strong texture along the (00l) direction. Therefore, the measured thermal conductivity κz is almost along the ab plane of Bi2Te3, which is about two times higher than κx along the c axis.32 The calculated ZT value (solid black square in Fig. 7(d)) using κz and σxy thus greatly underestimates the TE performance of our sample. The F value is calculated to be 0.79, approaching the value of ZM Bi2Te3 (F = 1). This highly textured structure is confirmed by the SEM photos (Fig. S3†) of the cross section of the SLM bulk. According to Shen's relationship curve,32 κx/κz = 0.54. κx and ZT for the SLM bulk sample is then calculated and shown in Fig. 7(c) and (d), respectively. The calculated ZT using κx shows a maximum value of 0.84 at 400 K and is higher than the ZT values of the ZM and SHS-prepared samples in the whole temperature range measured. To sum up, the SLM bulk sample exhibits comparable, if not superior, TE performance compared to bulk samples prepared by the traditionally synthesis routes, such as ZM or SHS.24 Moreover, the content of Se in the SLM bulk sample could be adjusted to further optimize its ZT value.
These results demonstrate that the Bi2Te3-based material with a homogeneous composition and excellent TE performance can be prepared via the SLM technique under an appropriate control of the processing parameters and the starting composition.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c7ra02677c |
This journal is © The Royal Society of Chemistry 2017 |