Corrugating van der Waals gaps for decoupling heat and charge transport in layered thermoelectrics Authors: Chuan-Dong Zhou, Jianfeng Cai, Minhui Yuan, Rong Huang, Dongxiang Lv, Yue Wu, Rui-Jie Li, Zongwei Zhang, Xiaojian Tan, Guo-Qiang Liu, Qiang Zhang, Bo Liang, Jun Jiang Published: 2026 ABSTRACT Layered thermoelectric materials face intrinsic challenges in disentangling phonon and electron transport due to their anisotropic bonding networks. Here, we introduce a van der Waals gap engineering strategy that deliberately imposes out-of-plane stress on n-type bismuth telluride (Bi2Te3). Selective interlayer doping creates local charge imbalance, which in turn drives ripple-like lattice corrugations. These structural undulations mimic substrate-induced strain fields, renormalize phonon dispersion, and substantially reduce phonon velocity, thereby suppressing lattice thermal conductivity. The corrugation amplitude is far smaller than the electronic mean free path, ensuring negligible additional electron scattering. As a result, the material reaches a peak zT of 1.43 at 350 kelvin, while the fabricated module delivers a conversion efficiency of 7.5% under a 250-kelvin temperature gradient—both representing state-of-the-art performances for n-type Bi2Te3 systems. More broadly, this work establishes interlayer stress as a general strategy to manipulate phonons in van der Waals solids, providing previously unidentified design principles for high-efficiency thermoelectrics. INTRODUCTION Thermoelectric (TE) materials provide a solid-state solution for converting heat into electricity and vice versa, with potential applications in waste-heat recovery, cooling, and distributed power generation (1–3). Their efficiency is quantified by the dimensionless figure of merit zT = S2σT/(κe + κl), where S is the Seebeck coefficient, σ is the electrical conductivity, T the absolute temperature, and κe and κl are the electrical and lattice thermal conductivities, respectively (4–6). However, the intrinsic interdependence of these parameters makes it difficult to improve one without compromising another (7–11). In systems with relatively weak electron-phonon coupling, reducing lattice thermal conductivity κl appears most promising, but conventional strategies based on defects or nanostructuring often degrade carrier mobility, limiting their effectiveness (12–17). Bismuth telluride (Bi2Te3), as well as its alloys, remains the most important TE material around room temperature because of its relatively high zT and wide use in practical modules (18–21). Its crystal structure consists of quintuple layers connected by van der Waals (vdW) interactions along the c axis (22–24). This layered structure results in highly anisotropic transport properties, making it particularly challenging to decouple phonon and electron transport (25, 26). Nanostructuring strategies have successfully improved the performance of p-type (Bi,Sb)2Te3, achieving peak zT up to ~1.4 (27), but have proven ineffective in n-type Bi2(Te,Se)3 because of its more strongly anisotropic transport (28, 29). Consequently, the TE performance of n-type Bi2Te3 alloys remains notably lower than that of their p-type counterparts (30–33). Recent studies on two-dimensional semiconductors have shown that microscale corrugations can suppress in-plane phonon transport while leaving σ nearly unaffected or even enhanced (34–38). Wu and colleagues demonstrated that MoS2 thin films grown on rippled substrates exhibited increased mobility due to reduced electron-phonon scattering (39). Corrugation-like effects have been introduced in bulk Bi2Te3, even though the authors did not announce this concept. Sui and colleagues applied a “layered crystal stacking hot deformation” process to generate planar deformation, yielding reduced thermal conductivity and a peak zT of 1.2 (40). Beyond mechanical means, chemical doping can also introduce internal stress. In Bi2Te3, Cu shows distinct site preferences—substituting Bi sites in p-type but occupying vdW gaps in n-type samples—where it interacts electrostatically with quintuple layers to stabilize the lattice (40–45). In n-type Bi2Te3, Cu doping has been reported to partially decouple the Seebeck coefficient and electrical conductivity, resulting in a considerably enhanced power factor (PF), with values reaching up to 63.5 μW cm−1 K−2 (43). Meanwhile, Cu doping induces an expansion of the c-axis lattice parameter while leaving the a and b axes nearly unchanged (46–48). These effects are closely linked to its intercalation within the vdW gaps in n-type samples, and they underscore the potential of such interlayer doping to modify the anisotropic lattice framework. Consequently, this doping strategy offers a platform to emulate substrate-induced corrugation in bulk-layered materials. This study demonstrates a vdW gap engineering strategy to decouple phonon and electron transport in layered n-type Bi2Te3, as illustrated in Fig. 1A. Interlayer incorporation of Cu+ ions enlarges the vdW spacing, and subsequent Ga3+ doping introduces local charge imbalance that generates out-of-plane stress and drives ripple-like corrugations of the quintuple sheets. The resulting structural distortion breaks the in-plane translational symmetry, renormalizes the phonon spectrum, and reduces phonon velocity, thereby markedly lowering κl. Meanwhile, the corrugation amplitude remains far below the electronic mean free path, so in-plane electrical transport is only weakly affected. Consequently, Cu/Ga co-doped samples exhibit simultaneously enhanced weighted mobility μW and lower κl compared with recent reports (Fig. 1B) (25, 32, 40, 42, 44, 49, 50). In particular, Bi1.997Ga0.003Te2.7Se0.3 + 0.08 wt % CuI achieves a peak zT of 1.43 at 350 K and a high average zTave of 1.21 over 300 to 500 K (Fig. 1C) (25, 31, 32, 40, 42, 43, 51–54). Furthermore, a 17-pair module constructed from this material delivers a conversion efficiency η of 7.5% under a temperature difference ΔT of 250 K (Fig. 1D) (25, 40–42, 51, 54, 55). Mechanism and results of vdW engineering. (A) Schematic illustration of the structural modification by vdW engineering. (B) Comparison of weighted mobility μW and lattice transport 1/κl at 300 K (25, 32, 40, 42, 44, 49, 50). (C) Temperature-dependent zT for the Bi1.997Ga0.003Te2.7Se0.3 + 0.08 wt % CuI as well as recent reports (25, 31, 32, 40, 42, 43, 51–54). (D) Conversion efficiencies of the constructed TE modules for comparison (25, 40–42, 51, 54, 55). HD, LCSHD, LSM, LTDS, SPS, and ZM refers to Hot Deformation, ..., Sparks Plasma Sinter, and Zone Melting, respectively. RESULTS TE properties The n-type Bi2Te3 samples in this study were prepared using the temperature-gradient method. CuI was first introduced to regulate the carrier concentration (nH) and to expand the vdW gap, followed by partial substitution of Ga for Bi to generate out-of-plane stress while preserving optimized electronic transport. The thermal transport properties of Bi2-yGayTe2.7Se0.3 + x wt % CuI are summarized in Fig. 2, measured along the in-plane direction (out-of-plane data are shown in fig. S1). Thermal transport properties and analysis of Bi2-yGayTe2.7Se0.3 + x wt % CuI samples. (A) Temperature dependence of total thermal conductivity. (B) Temperature dependence of sum of bipolar and lattice thermal conductivities. (C) Comparison of sound velocity between Bi2Te2.7Se0.3 + 0.08 wt % CuI and Bi1.997Ga0.003Te2.7Se0.3 + 0.08 wt % CuI. (D) The variation of CTE (coefficient of thermal expansion) with composition. (E and F) Raman spectra comparing Bi2Te2.7Se0.3 + 0.08 wt % CuI and Bi1.997Ga0.003Te2.7Se0.3 + 0.08 wt % CuI show quenched in-plane and out-of-plane vibration modes. (F) Comparison of sound velocity between Bi2Te2.7Se0.3 + 0.08 wt % CuI and Bi1.997Ga0.003Te2.7Se0.3 + 0.08 wt % CuI. a.u., arbitrary unit. Figure 2A shows the total thermal conductivity (κtot), while the corresponding thermal diffusivity, heat capacity, and Lorenz number are shown in fig. S2. For Ga-free samples (y = 0), κtot increases monotonically with CuI content due to the enhanced electronic contribution. Upon Ga incorporation, however, room-temperature κtot exhibits a nonmonotonic dependence on y, reaching a minimum of 1.2 W m−1 K−1 for y = 0.003. Figure 2B further separates the lattice and bipolar contributions (κl + κb), highlighting a minimum of 0.48 W m−1 K−1 at 300 K. To understand the decrease of lattice thermal conductivity, we measured acoustic velocities (Fig. 2C) for several samples. Cu doping alone produces negligible changes, but Cu/Ga co-doping reduces the transverse sound velocity from ~1570 to ~1258.6 m/s, together with a moderate softening of the longitudinal mode. Since the lattice wrinkles in this system are introduced by extremely dilute dopants, it is practically impossible to construct a sufficiently large supercell and obtain a reliable phonon spectrum from first-principles calculations. Instead, we discuss the change in sound velocity using a simple one-dimensional atomic-chain model. Within a perturbative treatment, the vibrational frequency of a rippled chain can be written asω(k)=2βcos2θ/msin(12ak)where β, m, a, and k denote the harmonic force constant, atomic mass, lattice constant, and phonon wave vector, respectively, and θ is the twisting angle associated with the wrinkle. Taking the limit (k → 0) gives the acoustic sound velocity, which is reduced by a factor of cosθ compared with the flat chain. This simple expression shows that any bond twisting necessarily lowers the sound velocity, but the reduction is modest when θ is small. Therefore, the pronounced decrease in sound velocity implied by Fig. 2C requires relatively large local bond-bending angles. At the same time, the effective in-plane harmonic force constant becomes βcos2θ; the weakening of the harmonic term enhances the relative importance of higher-order force constants, i.e., lattice anharmonicity. This, in turn, suggests that the in-plane thermal expansion coefficient (CTE) should increase in the presence of strong local wrinkles. Motivated by this analysis, we measured the in-plane CTEs of several samples, as summarized in Fig. 2D. CuI doping alone leads to a slight enhancement of the in-plane thermal expansion. Upon further introducing Ga, the thermal expansion first increases and then decreases. This behavior is consistent with our picture: Ga creates stronger local distortions than Cu and therefore has a much more pronounced effect on thermal expansion; however, when the Ga content becomes too high, the vdW gap tends to expand more uniformly rather than forming larger local wrinkles, so the effective anharmonicity and thermal expansion start to drop. We observe a similar increase of the in-plane CTE in Bi2Te2.7Se0.3 + CuI + Zn and Bi2Te2.7Se0.3 + CuI + Cd. The lattice thermal conductivity can be expressed as κl = 1/3Cvvsl, where Cv, vs, and l represent the heat capacity, group velocity, and mean free path, respectively. Given that the CTE is proportional to the Grüneisen parameter (γ), and the mean free path is inversely proportional to the square of the γ, the relative change in κl can be estimated from the sound velocity and CTE data presented in Fig. 2 (C and D). Based on the measured results, the estimated relative κl (300 K) of the 0.3% Ga-doped sample compared to the Ga-free sample is ~0.9 × (1.3/1.62)2 = 58%, where the factor 0.9 accounts for the relative change in group velocity. This estimated value agrees well with the experimental data shown in Fig. 2B. Unlike conventional dilute alloying, which scatters high-frequency phonons and is effective mainly at elevated temperatures, light Ga doping produces a pronounced reduction of κl even at room temperature. This indicates a fundamentally different mechanism. Raman spectra provide direct evidence: While Cu-doped samples show modes consistent with earlier reports (56), Cu/Ga co-doped samples exhibit obvious broadening of the A1g2 mode (Fig. 2, E and F). This mode corresponds to vibrations along the c axis, and its broadening reflects fluctuations of local stress and perturbations to out-of-plane force constants. Such broadening is a sign of phonon renormalization, suggesting that Ga-induced charge imbalance in the vdW gap distorts the lattice periodicity and alters phonon dispersions. The Raman spectra of other samples are provided in fig. S3. Figure 3 assembles multiscale evidence that Ga/Cu co-doping drives pronounced local corrugation of the vdW-layered structure in n-type Bi2Te3. The electron backscatter diffraction (EBSD) maps shown in Fig. 3 were collected on a polished section perpendicular to the c axis, so that the in-plane grain morphology and orientation distribution could be directly resolved. At the millimeter scale, EBSD inverse pole figure maps show that Cu doping alone produces giant, plate-like grains with nearly straight boundaries, whereas additional Ga breaks these plates into narrower, stripe-like domains with visibly corrugated grain boundaries (Fig. 3, A and D). At the micrometer scale, cross-sectional scanning electron microscopy (SEM) images reveal a similar trend: the Cu-doped sample maintains parallel, flat lamellae, while the Cu/Ga co-doped sample develops wavy, undulating stacks (Fig. 3, B and E). Atomic-resolution scanning transmission electron microscopy (STEM) analyses further confirm that the vdW gaps in the Cu/Ga sample host interlayer atoms, evidencing the ripple-like lattice corrugations (Fig. 3, C and F). Together, these observations consistently point to Ga-induced corrugation of the vdW gaps from the grain scale down to the atomic scale. Microstructural characterization of Bi2-yGayTe2.7Se0.3 + x wt % CuI samples. (A and D) EBSD inverse pole figure maps for the Cu-doped sample (x = 0.08, y = 0) and the Cu/Ga co-doped sample (x = 0.08, y = 0.003), respectively, showing giant plate-like grains in the Cu-doped sample and refined stripe-like domains with wavy grain boundaries in the co-doped sample. (B and E) Cross-sectional SEM images of the corresponding samples, where x = 0.08, y = 0 exhibits flat and parallel lamellae, while x = 0.08, y = 0.003 shows pronounced bending and undulation of the layered structure. (C and F) Aberration-corrected HAADF-STEM images highlighting uniform vdW gaps and flat quintuple layers in the Cu-doped sample and rippled atomic layers in the Cu/Ga co-doped sample; in (F), an overlaid atomic model is provided as a guide to the eye. Concomitant with this structural evolution, the average grain size changes systematically: Cu intercalation in the vdW gaps stabilizes the stacking and promotes grain coarsening along the c axis, yielding millimeter-scale plates, whereas Ga3+ co-doping introduces local charge imbalance and out-of-plane stress that subdivide these giant grains into finer, stripe-like domains. This grain refinement, together with the mesoscale corrugations and atomic-scale interlayer distortions, provides a direct structural origin for phonon renormalization. The enhanced phonon scattering and reduced sound velocity naturally account for the pronounced suppression of lattice thermal conductivity and thus contribute to the improved TE performance of the Cu/Ga co-doped samples. Consistent microstructural evolution is also observed across a broader composition space: Supplementary x-ray diffraction (XRD) and EBSD data for samples doped with CuI + Cd and AgI + Cd (figs. S4 to S13) likewise document a transition from giant plate-like grains to corrugated stripe-like domains, accompanied by a monotonic decrease in κl. Additional cross-sectional SEM and aberration-corrected STEM/GPA (Geomteric Phase Analysis) analyses for various compositions (figs. S14 to S23) reveal analogous changes from flat, parallel lamellae to rippled stacks with interlayer intensity maxima and long-wavelength shear bands, further confirming that vdW-gap corrugation is a robust structural motif throughout this doping series. Phonon renormalization and ripple-induced velocity reduction Figures 2 and 3 demonstrate that Cu/Ga co-doping effectively suppresses lattice thermal conductivity through microstructural corrugation and phonon renormalization. Figure 4 summarizes the corresponding electronic transport and overall TE performance of Bi2-yGayTe2.7Se0.3 + x wt % CuI. As shown in Fig. 4A, electrical conductivity (σ) in Ga-free samples increases with CuI content, consistent with the rise in carrier concentration (nH). Introducing Ga initially reduces σ due to impurity perturbation but further Ga incorporation restores and even enhances conductivity. The Seebeck coefficient (S) is slightly increased by Ga doping (Fig. 4B), despite the nearly unchanged nH (fig. S24). Above 400 K, |S| decreases in all samples owing to the bipolar effect. The combined impact yields the highest PF of 56.8 μW·cm−1 K−2 at 350 K in Bi1.997Ga0.003Te2.7Se0.3 + 0.08% CuI (Fig. 4C), indicating that Ga doping not only lowers κtot but also enhances PF, thus synergistically improving TE performance. Electrical transport behavior and TE performance of Bi2-yGayTe2.7Se0.3 + x wt % CuI samples. (A) Temperature dependence of electrical conductivity. (B) Temperature dependence of Seebeck coefficient. (C) Temperature dependence of PF. (D) XRD patterns of bulk samples. (E) DOS calculated for representative compositions. (F) Seebeck coefficient plotted against carrier concentration. (G) Weighted mobility (μW) at 300 K and orientation factor calculated from XRD. (H) Correlation between μW and 1/κl at 300 K. (I) Temperature dependence of zT values (31, 43, 50–52, 54, 55). To clarify the origin of the enhanced S, the room-temperature XRD patterns (Fig. 4D) reveal a systematic left shift of diffraction peaks, corresponding to an expansion of the c-axis lattice constant from 30.11 Å in undoped Bi2Te3 to 30.47 Å in the Cu/Ga co-doped sample, while in-plane constants remain unchanged. Such anisotropic expansion suggests strengthened Coulomb attraction along the c axis. Previous studies have shown that the high PF of Bi2Te3 arises from its structural anisotropy: Dispersive in-plane bands yield small transport mass, while flat out-of-plane bands contribute large density-of-states (DOS) mass (57–59). Density functional theory calculations using the experimentally determined c constants (Fig. 4E) confirm that the enlarged c flattens the band dispersion along kz and increases DOS near the conduction band edge, thereby explaining the observed enhancement in S. Pisarenko analysis (Fig. 4F) further indicates an increased DOS effective mass md∗ from 1.74 m0 in Cu-doped samples to 2.25 m0 with Cu/Ga co-doping (31, 43, 50–52, 54, 55) Besides band anisotropy, the orientation factor is also improved by the Coulomb attraction along the c axis. As shown in Fig. 4G, weighted mobility, μW=(md∗/me)1.5μ0, increases in Cu- and Cu/Ga-doped samples, accompanied by orientation factor rising from 0.80 (undoped) to 0.94 (Cu/Ga co-doped). EBSD pole figures also corroborate improved grain alignment (figs. S25 and S26). Owing to the enhanced structural anisotropy and orientation factor, Cu/Ga co-doped samples exhibit the superior weighted mobility μW. Figure 4H illustrates the decoupling of electrical and thermal transport. Cu doping increases μW but also elevates κtot. Similar phenomena have been observed in previous studies on Cu-doped Bi2Te2.7Se0.3 (40). By contrast, Ga co-doping induces phonon renormalization and effectively reduces κl, thereby reconciling high μW with low thermal conductivity (figs. S27 and S28). Consequently, as shown in Fig. 4I, Bi1.997Ga0.003Te2.7Se0.3 + 0.08%CuI achieves a peak zT of 1.43 at 350 K, with average zT surpassing recent reported values (fig. S29). To the best of our knowledge, this represents the highest reported value for n-type Bi2Te3 systems. TE module performance To further evaluate the enhanced performance of the optimized Bi2(Te,Se)3 material, 17-pair TE modules were fabricated (details in fig. S30), consisting of n-type Bi1.997Ga0.003Te2.7Se0.3 + 0.08%CuI and p-type Bi0.4Sb1.6Te2.95Se0.05 + 0.1% Ag8GeTe6 (fig. S31). Theoretical efficiency was first estimated by three-dimensional finite element simulations using the COMSOL Multiphysics software (fig. S32), which suggested that the maximum efficiency ηmax becomes nearly constant when the leg area ratio Ap/An lies between 0.9 and 1.2 (Fig. 5A). For practical fabrication, p- and n-type legs were prepared with identical cross sections, and the final leg dimensions were set to 5.0 mm by 2.3 mm by 2.3 mm. TE module performance of Bi1.997Ga0.003Te2.7Se0.3 + 0.08%CuI alloy. (A) The maximum efficiency ηmax as a function of the cross-sectional area ratio of the p- and n-type legs (Ap/An) and the ratio of height to the total cross-sectional area of the legs (H/Apn) as simulated by finite element calculations. (B) Optical image of the fabricated 17-pair TE modules. (C) The measured data by changing the current in the circuit: Load voltage and output power. (D) The measured data by changing the current in the circuit: Heat flow. (E) The measured data by changing the current in the circuit: conversion efficiency at different temperature differences for module no. 1. (F) A comparison of the maximum conversion efficiency between module no. 1 and those reported in other studies (26, 48, 49, 52, 55–57). The output characteristics of the fabricated module were measured using a homemade test system (fig. S33). Figure 5 (B to D) shows the voltage U, output power P, and conversion efficiency η as a function of current under different temperature differences. The open-circuit voltage reaches 1.71 V at ΔT = 250 K (Fig. 5B). The output power first increases and then decreases with current, giving a maximum P of 1.22 W at ΔT = 250 K (Fig. 5C). The conversion efficiency peaks when the external load resistance matches the internal resistance of the module (Fig. 5D). At ΔT = 250 K, the module achieves a maximum efficiency ηmax of 7.5% (Fig. 5E), which surpasses most previously reported Bi2Te3-based devices (25, 51, 52, 55, 60–62). The details of the fabricated module are provided in the SI, including the internal resistance, interface contact resistance, and characterization by SEM (figs. S34 to S36). To further confirm scalability, a second 17-pair module (module no. 2) was fabricated from a large Ø30 mm, 1500-g crystal (figs. S37 to S42). The output power and conversion efficiency of module no. 1 and module no. 2 as functions of ΔT are compared in fig. S43. Module no. 2 exhibits a maximum efficiency of 7.2% at ΔT = 250 K, only slightly lower than module no. 1. These results demonstrate that the present Bi2(Te,Se)3-based modules deliver excellent output power and conversion efficiency, achieving a state-of-the-art performance among n-type Bi2Te3 TE. DISCUSSION In summary, we demonstrate a vdW gap engineering strategy to decouple phonon and electron transport in n-type Bi2Te3. Interlayer Cu stabilizes lattice stacking through long-range Coulomb attraction, while Ga co-doping introduces local stress that ripples the quintuple layers and reshapes phonon dispersions. The resulting phonon renormalization, manifested in broadened Raman modes and reduced acoustic velocities, effectively suppresses lattice thermal conductivity without compromising carrier mobility. This synergistic modulation leads to a peak zT of 1.43 at 350 K in Bi1.997Ga0.003Te2.7Se0.3 + 0.08% CuI sample. The proposed strategy offers a previously unidentified and practical pathway for tailoring transport properties in layered TE materials. MATERIALS AND METHODS Raw materials and sample synthesis Te (5 N) bulk, Se/Bi/Ga particles (5 N), and CuI powder were mixed according to the stoichiometric ratios specified for Bi2Te2.7Se0.3 + x wt % CuI (x = 0.06, 0.08, and 0.1) and Bi2-yGayTe2.7Se0.3 + 0.08 wt % CuI (y = 0.001, 0.003, and 0.005). The mixture was hermetically sealed in quartz tubes with a vacuum level of 10−3 Torr and a diameter of Ø12.7 mm. Subsequently, it was melted for 1 hour and subjected to oscillation at 1023 K for an additional hour. Upon removing the quartz tubes, they were vertically positioned in ambient air and rapidly quenched. Upon reaching room temperature, they were vertically positioned in a zone melting furnace, with the pointed end of the quartz tube placed at the center for 30 min of insulation. The samples were prepared through zone melting with a growth rate set at 25 mm hour−1. Last, the composition exhibiting the highest zT value, Bi1.997Ga0.003Te2.7Se0.3 + 0.08 wt % CuI, was mixed and sealed in quartz tubes with diameters of Ø30 mm, under a vacuum level of 10−3 Torr. Subsequently, samples were prepared following the aforementioned procedure. The samples were designated as Ø30 mm–1500 g. TE performance measurement The samples obtained through zone melting were precision-cut into rectangular strips with dimensions of 3 mm by 3 mm by 10 mm using an electrical discharge machining cutter. The S and σ were concurrently measured using a four-probe method within a low-pressure argon gas environment, using the ZEM-3 system (Ulvac-Riko, Japan). In addition, the samples obtained through zone melting were cut into thin slices with dimensions of Ø12.7 mm and d = 1.8 mm. The thermal diffusivity (D) was measured with an LFA-467 laser flash instrument, while the density was determined through the Archimedes method (Netzsch LFA-467, Germany). The specific heat capacity (Cp) was derived from the Dulong-Petit limit. The density was measured using the Archimedes method, and the densities of all samples are summarized in table S1. The кtot was determined by applying the equation κtot=ρDCp. The Hall coefficient (RH), carrier concentration (nH), and mobility (μH) at room temperature were measured using a comprehensive physical property measurement system (PPMS-9, Evercool), and the carrier concentration and mobility were determined based on the equations nH=1/(eRH) and μH=σRH, respectively. Structural and morphological characterization The crystal structure of the bulk samples in the in-plane direction was analyzed by means of XRD (Bruker D8, Germany). The distribution of elements was analyzed through energy-dispersive x-ray spectroscopy. The microstructures of grains were characterized by EBSD carried out on an SEM (Quanta FEG 250, FEI). The morphology of the microstructure was examined using a high-resolution transmission electron microscope (Talos F200x, Thermo Fisher, USA). The atomic arrangements were characterized by the aberration-corrected STEM (HF5000, Hitachi Tech, Japan) using a high-angle annular dark-field detector (HAADF) and annular bright-field detector. Raman spectra of the specimens at room temperature were measured by a Raman microscope (Horiba Jobin Yvon), which uses a continuous-wave 532-nm laser to excite the sample and detect the Raman scattering and shifts of vibrational frequencies with a spectral resolution of ≈2 cm−1. CTE measurement Linear CTE was measured by thermomechanical analysis in the in-plane direction (parallel to the ab plane). Rectangular bars with dimensions of 4 mm by 4 mm by 15 mm were cut from bulk samples, with the 15-mm axis aligned parallel to the ab plane. The end faces were polished to ensure good parallelism and stable contact with the probe. Measurements were carried out under flowing Ar to minimize oxidation. A constant heating rate of 5 K min−1 was applied over the temperature range used in Fig. 2D. The instrument was calibrated using a certified reference standard, and baseline correction was performed prior to data collection. CTE was calculated as α = (1/L0) dL/dT, where L0 is the initial length at room temperature. For each composition, repeated runs (and, where applicable, multiple specimens) were measured to confirm reproducibility, and the reported curves represent averaged results. The typical uncertainty of CTE is within a few percent. TE module fabrication and measurement The TE module has dimensions of 20 mm by 20 mm and is composed of 17 pairs of p-n legs. The n-type legs consist of samples with sizes of Ø12.7 mm–50 g and Ø30 mm–1500 g, whereas the p-type legs are made of p-type Bi0.4Sb1.6Te2.95Se0.05–0.1% Ag8GeTe6 material. The fabrication process of the TE module based on Bi2Te3 is detailed in fig. S30. Both the p-type and n-type legs are coated with Ni to serve as diffusion barrier layers. Subsequently, they are individually cut into particles with dimensions of 5 mm by 2.3 mm by 2.3 mm (p-type) and 5 mm by 2.3 mm by 2.3 mm (n-type). The p-type and n-type legs are arranged alternately on an alumina plate with copper electrodes attached. They are joined together through soldering using Sn5Pb9.25Ag2.5 at a temperature of 633 K. The fabricated modules are labeled as 1 and 2. Measurement is conducted using a specially designed measurement system. The η is calculated using the formula η=P/P+Qc, where P represents the measured output power, and Qc denotes the measured heat flux at the cold side.