Titanium-based thin film metallic glass as diffusion barrier layer for PbTe-based thermoelectric modules

The motivation towards sustainable development has accelerated the progress in green energy technology. Since the 1960s, countless efforts have been invested to improve the performance of thermoelectric (TE) materials and modules, due to the fact that the thermoelectric generator (TEG) is capable of converting waste heat into electricity via the Seebeck effect,^1–3 which provides a mutual solution to both human development and environmental protection. For a TE material, the performance can be evaluated by the figure-of-merit (zT = (S²/ρκ)T), where S, ρ, and κ are the Seebeck coefficient, electrical resistivity, and thermal conductivity, respectively.

As known, the major obstacle for the utilization of mid-temperature TEG is the low conversion efficiency, which can be attributed to the moderate zTs and the low thermal stability of joints/interconnects in a TE module.^4,5 In addition to the optimization of zT, the thermal stability of joints in TE modules is in need of improvement. Given that the hot side operating temperature is ≈700 K for a mid-temperature TEG, the inter-diffusion between the TE substrate and the metal electrode (such as Cu or Ag) could be extremely severe. Specifically, the formation of brittle intermetallic compounds (IMCs) that degrade the electrical conductivity and bonding strength are a primary mechanism of failure in TE devices.^6–11 

Lead-telluride (PbTe), having a rock-salt structure, has been the most well-established mid-temperature TE materials since the 1960s.^12–15 Breakthroughs in zT have been reported in PbTe_1−xI_x, which shows degenerate n-type semiconducting behavior with a peak zT of 1.4 at T > 700 K.¹² Alternatively, the Sr-containing PbTe, which is of p-type conduction, achieves an exceptionally high zT > 2 at T > 900 K attributed to the existence of the all-scale architected microstructure.¹⁵ Regarding the construction of TE modules, various bonding methods have been reported for PbTe-based TEGs, including high-temperature hot-pressing (HP),^16,17 spark plasma sintering (SPS),^18,19 and low-temperature solid-liquid inter-diffusion (SLID)^20,21 via the incorporation of a low-melting-point interlayer (such as Sn and In), etc. No matter which bonding method is considered, the diffusion barrier layer is necessary to suppress the inter-diffusion between the PbTe-based TE material and the metal electrode. Electrodeless Ti²² and Co-P²³ are commonly used as the diffusion layer for mid-temperature TEG; nevertheless, the formation of IMCs and the difference in the CTE (coefficient of thermal expansion) drive the TE community to seek better diffusion barriers.

Attention was drawn to a prior study²⁴ that an extremely low electrical contact resistance (ρ_C ∼ 4.0 × 10⁻¹⁰ (Ω m²)) was measured between the PbTeSe substrate and Ti/Ni/Au contact, due to the utilization of contact metallization. Nevertheless, the results²⁴ are valid only for thin-film devices with operation temperature lower than 200 °C. As for our previous study,²⁵ we first introduced the thin-film metallic glass (TFMG) as an effective diffusion barrier for the AgSbTe₂-based TE module. The grain-boundary-free TFMG only allows bulk diffusion across the TFMG/TE interface, which takes place with an extremely small diffusivity (i.e., slow diffusion). This not only inhibits the inter-diffusion but also guarantees the satisfactory bonding that leads to low electrical and thermal contact resistances at the joint. Herein, we further exploit the Ti-based TFMG as a diffusion barrier for the state-of-art PbTe-based TEG. Single crystal PbTe is selected as the TE substrate and is sputtered with the Cu electrode and Ti-based TFMG (TiZrCuNbCo) to form the reaction couple of Cu/TFMG/PbTe. Upon thermal aging at 673 K for 48 h, the interface of TFMG/PbTe remains free of IMCs, while the electrical contact resistance, which is measured via the transmission line method (TLM), falls in the range of 10⁻⁹ (Ω m²). The electrical contact resistance is 10⁻² times smaller than that value associated with the Ti/PbTe interface under the same condition. As a consequence, the Ti-based TFMG holds great potential for being the diffusion barrier layer in PbTe-based TE modules.

Nominally stoichiometric PbTe substrates were producing by slow cooling of molten elements. Pure Pb (99.999% Alfa Aesar) and pure Te (99.999% Alfa Aesar) totaling 20 g were sealed in fused silica tubes under a vacuum below 5 × 10⁻⁵ Torr. Prior to sealing, the tubes were solution cleaned, flame polished, and carbon coated. The elements were melted at 1248 K in a vertical furnace and held for 4 h. The temperature was lowered to 1123 K at a rate of 1 K/h to crystallize and anneal, after which the furnace was shut off and allowed to cool.

Sputtering targets of the Ti₄₀Zr₁₀Cu₃₆Nb₇Co₇ alloy, pure Ti (99.999%), and pure Cu (99.95%) were used under a vacuum system (less than 1 × 10⁻⁶ Torr). The Ti_45.8Zr_9.5Cu_29.5Nb_8.6Co_6.6 (at. %) TFMG or pure titanium was deposited on the PbTe substrate by radio frequency (RF) magnetron sputtering without the external heating source. The Cu electrode layer was grown, following the deposition of a TFMG or Ti layer to form the sandwich-like reaction couple of Cu/TFMG/PbTe or Cu/Ti/PbTe, respectively. Upon deposition, the working distance between the target and the substrate was 10 cm, while the applied RF power was 100 W under an Ar (5N) gas atmosphere at a pressure of 3 × 10⁻³ Torr. The deposition rate for Ti-based TFMG, Ti, and Cu were 0.117 nm/s, 0.067 nm/s, and 0.25 nm/s, respectively. After the deposition, the reaction couples of Cu/TFMG/PbTe and Cu/Ti/PbTe were again sealed under vacuum and placed in a 673 K furnace for various lengths of time (i.e., 8, 24, 48, and 96 h).

The sheet resistance and the resistivity of Ti-based TFMG and Ti were measured on Si wafers using a resistivity meter (Loresta-AX MCP-T370). The electrical contact resistance and specific contact resistivity of as-deposited and annealed TFMG/PbTe and Ti/PbTe samples were assessed based on the transmission line method (TLM), using the semiconductor parameter analyzer (Agilent B1500A) at room temperature. Stainless-steel shadow mask fabricated the patterned TFMG or Ti deposition with a rectangular area (100 µm × 300 µm) on the PbTe, while those rectangles, which worked as the contact electrodes, were spaced from 150 µm to 580 µm. The current (I) and the voltage (V) were measured on the contact electrodes with different spacing (L). The resistance (R), which was determined from the inverse slope of a linear fit of the I-V plot, can be plotted as a function of contact spacing (L). Furthermore, the contact resistance (R_c) and the transfer length (L_T) can be obtained from the intercepts of the y-axis and the x-axis by curve fitting, respectively. Hereafter, the specific contact resistivity (ρ_c) can be calculated using the following:

where the W refers to width of the deposited contact (i.e., W = 300 µm). The thermal properties of the TFMG are characterized using differential scanning calorimetry (DSC, Netzsch 404 F3 Pegasus) from 300 K to 973 K at a heating rate of 40 K/min under the Ar atmosphere. The crystallographic structures were characterized using a low-glancing-angle X-ray diffractometer (XRD, Bruker D8 Discover) with Cu Kα radiation at 40 kV and 200 mA. The compositions of the TFMG layer and the PbTe substrate were determined by an electron probe micro-analyzer (EPMA, JEOL, JXA-8200). The elemental composition of Ti-based TFMG was identified as Ti_45.8Zr_9.5Cu_29.5Nb_8.6Co_6.6 (in at. %) using EPMA. Transmission electron microscopy (TEM) sample foils were produced by ion beam thinning using 30 kV Ga⁺ beams initially and 5 kV for the final polishing in a dual-beam focused-ion-beam system [FIB, FEI Quanta 3D FEG (field-emission gun)]. Prior to FIB thinning, a thin layer of platinum was deposited on top of the sample to protect the targeted region from ion-damage. TFMG/PbTe and Ti/PbTe cross-sectional microstructures were analyzed by TEM in a FEI Tecnai™ G2 F-20 instrument operated at 200 kV. Selected-area electron diffraction (SAED) with a captured diameter of 200 nm was employed to characterize the structure of local regions in the annealed samples. High-angle annular dark-field (HAADF) imaging and energy dispersive spectroscopy (EDS) in scanning TEM (STEM) mode were used to acquire the compositional profile across each interface.

Figure 1(a) shows the low magnification bright-field (BF) images for the as-deposited Cu/TFMG/PbTe (upper) and Cu/Ti/PbTe (lower) reaction couples, which guarantee the satisfactory bonding prior to thermal treatment. The XRD diffraction pattern of the PbTe substrate [Fig. 1(b)] reveals the major Bragg peaks associated with orientations of (200), (220), and (420), which is identified as the face-centered cubic structure according to JCPDS#38-1435. As for the as-deposited Ti film, two major peaks are detected at 2θ = 38.4° and 70.7°, corresponding to the (002) and (103) planes (JCPDS#44-1294). For the TFMG [Fig. 1(b)], a broad hump is observed in the 2θ of 30°–45°, indicating the amorphous structure at room temperature. The thermal analysis of Ti-based TFMG (Fig. S1 of the supplementary material) suggests that the glass transition temperature, T_g, and crystallization temperature, T_x, are of 732.7 K and 781.9 K, respectively. Hence, the following thermal aging for the reaction couple of Cu/TFMG/PbTe is performed at 673 K, which is below T_g and T_x, to ensure that the TFMG remains in the amorphous structure.

The low-magnification BF images reveal distinct features between Cu/TFMG/PbTe [Figs. 2(a)–2(c)] and Cu/Ti/PbTe [Figs. 2(d)–2(f)]. Upon thermal aging at 673 K for 8 h, the Ti layer is completely consumed, while the multiple layers of Ti-Te-based intermetallic compounds (IMCs) are formed in between the Cu electrode and the PbTe substrate, leaving a large Pb-rich region and eventually formimg Pb precipitates. Similar interfacial reactions are observed with increasing aging time, which suggests that Ti might not be an ideal diffusion barrier for mid-temperature PbTe-based modules. On the contrary, the interfaces of Cu/TFMG and TFMG/PbTe retain free of IMCs up to 24-h annealing, revealing that the TFMG could be effective in inhibiting the inter-diffusion between PbTe and Cu. To verify this observation, cross-sectional STEM images as well as compositional line-scans are conducted for Cu/TFMG/PbTe and Cu/Ti/PbTe couples annealed at 673 K between 8 and 96 h.

The cross-sectional STEM images, SAED patterns, and EDS line-scans of as-deposited Cu/TFMG/PbTe samples and those annealed at 673 K for 8–96 h are summarized in Figs. 3(a)–3(e), where the red arrows indicate the region and direction of the corresponding compositional profiles, which are present on the right-side of Figs. 3(a)–3(e). The line-scan results further suggest that the Te diffuses into the TFMG after 48-h annealing [i.e., Fig. 3(d)], and meanwhile Ti is detected in the PbTe substrate. Nevertheless, no IMCs are observed, and the primary TFMG remains amorphous. It is worth noting that a transition zone between the Cu layer and the TFMG shows partially crystallization upon long-term annealing [e.g., Fig. 3(d)], owing to the fact that the Cu atom exhibits high mobility and could easily diffuse into the TFMG. The crystallization might gradually degrade the amorphous feature of the TFMG and could be suppressed by introducing another buffer layer, to further sustain the endurance of TFMG-incorporated reaction couple.

The inter-diffusion of Pb and Te into the TFMG is more profound when the annealing time prolongs to 96 h [Fig. 3(e)]. After this time, the diffusion species appears to serve as heterogeneous nucleation sites, giving rise to the crystallization of the TFMG and the formation of Ti-Te IMCs. The embedded SAED patterns [the upper-left corner of Figs. 3(a)–3(e)] are obtained from the TFMG layer, showing the evolution from a halo pattern, for the as-deposited amorphous structure, to the spot-decorated ring, for the long-term annealed TFMG with partial crystallization.

Conversely, the polycrystalline Ti thin-film reacts vigorously with PbTe after short-period annealing (Fig. 4). Figures 4(a)–4(e) show the cross-sectional STEM images and EDS line-scan results of as-deposited Cu/Ti/PbTe as well as those annealed for 8–96 h. For the as-deposited Ti film, the corresponding SAED pattern [inset of Fig. 3(a)] reveals a face-centered cubic structure, as identified from the (100), (101), (102), and (110) planes. After 8-h annealing, the Ti layer reacted with PbTe to form Ti_xTe_1−x phases, which could be Ti₃Te₄ or Ti₅Te₄ as suggested by the Ti-Te binary phase diagram.²⁶ Depending on where the local composition falls in the Pb-Te-Ti phase space,²⁷ the following chemical reactions are possible at 673 K:

The three phase equilibrium PbTe-Ti₃Te₄-Pb is expected after long annealing times, ignoring any solubility/reaction with Cu.

Thus, large amounts of Pb are expected in addition to Ti-Te IMCs. The SAED patterns and compositional analysis acquired on the IMCs indicate monoclinic Ti₃Te₄ or tetragonal Ti₅Te₄, in agreement with the proposed reaction scheme. Since the as-deposited Ti film is finite (200 nm in thickness), the IMCs reach the saturated thickness once the Ti film is fully consumed (∼8 h annealing). Additionally, the Pb-forming reactions induce the formation of voids at the IMC/Pb interface, which has been reported in Cu/PbTe and Ag/PbTe reaction couples.^28,29 It is speculated that the formation of voids might relate to the fast diffusion of Pb, which is driven by the high-temperature annealing. With prolong annealing time, the Pb phase shows up at the interface of Cu/IMCs as well as Cu/TFMG, which might have resulted from the low solubility and the fast diffusion of Pb across the IMCs and the partially-crystallized TFMG. It should be noted that the fast diffusion of Pb is essential for the existence of Pb phase adjacent to the IMCs and TFMG; otherwise the oversaturated Pb should form as precipitates. Moreover, the Cu layer peels off from the Ti after 48-h annealing, indicating the poor adhesion that makes the Ti film an inappropriate diffusion barrier layer.

The distinct inter-diffusion behavior observed in the Cu/TFMG/PbTe and Cu/Ti/PbTe couples correlate strongly with the electrical and thermal contact resistivity. Prior to the electrical contact resistivity (ρ_C) assessment of the diffusion couples, the sheet resistance (R_s) and resistivity (ρ) of the as-deposited TFMG and Ti [on Si(100) wafer] were measured using a four-point probe at room temperature. As listed in Table I, the ρ of the TFMG (1.56 × 10⁻⁶ Ω m) falls in a similar range compared with that of the Ti film (1.07 × 10⁻⁶ Ω m) having the same thickness (200 nm). It should be noted that the ρ between the as-deposited films (∼10⁻⁶ Ω m) and crystalline metal bulks (∼10⁻⁸ Ω m) differs by two orders of magnitude, owing to the fact that the smaller grain size produced by sputtering might lead to stronger electron grain-boundary scattering.³⁰ 

Figure 5(a) shows the top-view SEM images for the as-deposited TFMG contacts on the PbTe substrate under different magnifications. These contacts are used for the TLM measurement. The schematic diagram of the TLM measurement [Fig. 5(b)] indicates the meaning of the slope (R_s/W), the intercept of the x-axis (2L_T), and the intercept of the y-axis (2R_c), respectively. The spacing between the deposited contacts (TFMG or Ti) refers to L. With increasing L, the linear slope of the I-V curve decreases, implying that the total resistance (R) increases as shown in Fig. 5(c). Once the contact resistance (R_c) and the transfer length (L_T) are extracted from the R-L plot, ρ_C can be calculated from Eq. (1). For the as-deposited condition, the ρ_C of Ti/PbTe and TFMG/PbTe are of 3.29 × 10⁻⁹ (Ω m²) and 5.70 × 10⁻⁸ (Ω m²), respectively, which fall in a similar range compared with that of macroscopic thermoelectric devices (ρ_C ∼ 10⁻⁹–10⁻⁸ (Ω m²)).³¹ Interestingly, the annealing process (673 K for 30 min) leads to a lower ρ_C (2.48 × 10⁻⁹ Ω m²) of TFMG/PbTe. By contrast, the ρ_C of Ti/PbTe increases to 1.79 × 10⁻⁷ (Ω m²) after a short-period of annealing, which is attributed to the formation of IMCs at the Ti/PbTe interface (Fig. S2 of the supplementary material). Upon long-term aging, the IMC layer grows thicker, and more Pb is produced as the reactions [Eqs. (2) and (3)] proceed towards completion. This increase in reaction products results in higher ρ_C, as summarized in Table I.

The interfacial diffusion of TFMG/PbTe and Ti/PbTe proceeds along distinct paths, owing to the amorphous character of the TFMG and crystallinity of the Ti film. Figure 6 further illustrates a comparison of the inter-diffusion between the TFMG/PbTe (left) and Ti/PbTe (right), respectively. Given that the grain-boundary diffusion is much faster than the bulk diffusion, the increasing density of grain boundaries enhances the inter-diffusion.^25,32 For example, the crystalline Ti film exhibits a higher density of grain boundaries than that of the amorphous TFMG, which provides numerous diffusive paths and gives rise to the severe inter-diffusion between the Ti film and the PbTe. As for the TFMG, despite the large amount of Ti (45 at. %) in the TFMG, the constituent elements do not react with the PbTe until the TFMG becomes crystallized after annealing for 48 h. Although the annealing temperature (673 K) is ∼0.9 of the crystallization temperature (T_x) of the TFMG, the Pb or Te atoms (originating from the PbTe) diffuse gradually into the TFMG, which might provide nucleation sites for crystallization. Yet, the incubation time for crystallization at 0.9T_g in general is less than 1 h,³³ far shorter than 48 h. While the exact mechanism is still unclear why the TFMG remains amorphous after such long period of annealing time at 0.9T_g, this anomaly is also observed in other TFMG studies in various schemes such as solar cell³³ and TE.²⁵ Once the TFMG shows crystallization after long-term annealing, it is no longer an effective diffusion barrier, eventually resulting in the formation of IMCs at the interface. Generally speaking, the optimal operation temperature should be below 0.8 T_g,³³ in order to ensure that the TFMG retains fully the amorphous structure. For the practical use of the TE module consisting of Ti-based TFMG, it is suggested that the operation temperature should be as low as 586 K (i.e., 0.8 T_g) to extend the life of device.

Provided that the TFMG exhibits high thermal stability and superior electrical conductivity, it could be a promising diffusion barrier for mid-temperature PbTe-based devices. In addition to state-of-art bulk TE devices, the TFMG might be an ideal diffusion barrier for flexible thin-film thermoelectric generators, where the constituent layers require high mechanical and electrical stability under repeated tensile and compressive strains.³⁴ 

In summary, two types of reaction couples structured with Cu/TFMG/PbTe and Cu/Ti/PbTe are prepared and annealed at 673 K for 8-96 h, in order to simulate the interfacial reactions of a PbTe-based TE module. Accordingly, the grain-boundary-free Ti-based TFMG inhibits the inter-diffusion between the PbTe substrate and the Cu electrode for at least 48-h annealing before the TFMG crystallizes. Moreover, the electrical contact resistivity (ρ_C ∼ 2.48 × 10⁻⁹ Ω m²) of the TFMG/PbTe couple is comparable with the current macroscopic TE module design requirements, which improves the feasibility of TFMG as a diffusion barrier in TE modules. On the contrary, the formation of multiple Ti-Te IMCs and elemental Pb at the interface of the Ti/PbTe couple suggests that the crystalline Ti film is a less effective diffusion barrier due to the severe inter-diffusion along grain boundaries. As such, the high ρ_C ∼ 1.79 × 10⁻⁷ (Ω m²) of Ti/PbTe is expected. In general, the Ti-based TFMG, having high thermal stability and low electrical contact resistivity when bonded with PbTe, shows great potential as an alternative diffusion barrier layer for mid-temperature PbTe-based TE modules.

See supplementary material for differential scanning calorimetry (DSC) thermogram of TFMG and the cross-sectional micrographs for TFMG/PbTe and Ti/PbTe contact regions.

This work was financially supported by the Applied Research Center for Thin-Film Metallic Glass from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE) in Taiwan. H.J.W. acknowledges the financial support from the Young Scholar Fellowship Program by Ministry of Science and Technology (MOST) in Taiwan, under Grant No. MOST107-2636-E-110-001.