Chemical bonding manipulation unlocks high performance ionic-bonded thermoelectrics

Abstract

Developing a strategy to enhance the thermoelectric performance of ionic semiconductors is challenging because of the pronounced electron localization inherent to the bonding nature. Herein, we demonstrate a bond manipulation approach in ionic-bonded compound MnTe achieved by the introduction of diverse atomic species, which effectively delocalizes electrons and modifies phase composition, leading to a comprehensive optimization in thermoelectric performance. As a result, we realize a peak zT value of ~1.6 at 773 K and an average zT value of ~0.9 from 300 K to 773 K. In addition, this chemical bonding engineering induces bond softening and forms multiscale hierarchical structures, resulting in a significant reduction in lattice thermal conductivity. Consequently, the segmented module fabricated from this p-type MnTe-based material achieves a thermoelectric conversion efficiency of 11% at a temperature difference ∆T = 473 K. Our findings establish bond engineering as an effective paradigm for enhancing the thermoelectric performance of ionic compounds.

Researchers boosted the thermoelectric performance of the ionic-bonded compound MnTe by manipulating chemical bonding, achieving improved TE conversion efficiency and zT values via increased electron delocalization and suppressed phonon transport.

Introduction

Escalating energy losses during conversion and operational processes pose a serious global challenge and have garnered increasing attention in recent years¹. Thermoelectric technology holds substantial potential in power generation from waste heat due to its ability to convert heat into electricity. The thermoelectric conversion efficiency is determined by the dimensionless figure of merit (where $\sigma$, S, T, ${\kappa }_e$, and ${\kappa }_L$ represent the electrical conductivity, the Seebeck coefficient, the absolute temperature, the electronic, and lattice thermal conductivity, respectively)². All of these parameters can be optimized by tailoring band structures and microstructures through a variety of established strategies such as band convergence^3,4, resonant level^5,6, alloying⁷, dislocations^8,9, nanostructure¹⁰, stacking faults (SFs)¹¹, and anharmonicity¹². Despite these approaches having significantly different effects on electrical and thermal properties, the overall aim is to improve electrical transport and suppress the thermal conduction.

Bond engineering has recently emerged as a promising strategy for modulating the chemical bonding environment in solids by adjusting parameters such as bond length, bond angle, bond strength, bond polarity, and electronic occupancy. This is typically achieved through methods including atomic substitution¹³, defect engineering¹⁴, heterostructure construction¹⁵, or external stimuli such as temperature¹⁶, strain¹⁷, or electric fields¹⁸. As a versatile strategy, bond engineering bridges atomic-scale chemistry with macroscopic material performance across diverse fields, including thermoelectrics¹⁹, energy storage²⁰, and superconductivity²¹. For thermoelectric materials in particular, bond manipulation offers an effective route to enhance thermoelectric performance by modifying chemical bonding mechanisms to alter phase composition and band structure, resulting in the improvement of electrical transport^22,23. Manipulating the thermal properties to impede heat transport by softening chemical bonding to enhance lattice anharmonicity and reduce phonon group velocity²⁴. On the basis of this bond engineering, several covalently bonded compounds have demonstrated improved thermoelectric performance by modulating bonding nature from covalent to metavalent character^25,26, which facilitates a more favorable balance between electrical and thermal transport. Additionally, in compounds with inherently strong chemical bonds, deliberate bond weakening effectively reduces bonding stiffness^27–29, leading to decreased ${\kappa }_L$ and enhanced zT values. Building on these findings, an intriguing question arises regarding the applicability of bond manipulation strategy to ionic-bonded compounds, which are characterized by strong ionicity and pronounced electron localization. For instance, hexagonal MnTe (space group P63/mmc) exhibits strongly ionic bonding characteristics, as indicated by Phillip’s ionicity scale^30–33. This intrinsic bonding nature leads to a wide bandgap and strong optical phonon scattering³⁴, which together strongly limit carrier concentration and mobility, ultimately resulting in poor electrical performance.

Herein, we demonstrate that the utilization of chemical bonding engineering advances thermoelectric performance in MnTe through alloying with GeTe and AgSbTe₂, as illustrated in the schematic diagram (Fig. 1a). This strategy aims to enhance electrical performance while suppressing thermal transport by modulating electron distribution and introducing additional phonon scattering centers. Regarding the electrical transport, this approach effectively decouples the interrelated nature of the electrical parameters, namely, carrier concentration n, carrier mobility μ, and effective mass m^*, achieving improvements across different parameters and resulting in a significant enhancement in PF. Thermally, the lattice softening induced by increasing the bond length and the formation of hierarchical defects, including numerous grain boundaries, high-density SFs, and abundant point defects, efficiently reduce the phonon group velocities and scatter a wide range of phonons, leading to an ultra-low ${\kappa }_L$.

Fig. 1. Enhancing the thermoelectric performance of MnTe-based materials and modules through chemical bond engineering.

a A schematic diagram of electron delocalization via the incorporation of diverse atomic species. b zT value as a function of temperature and average zT value (inset) for the p-type MnTe investigated in this work, alongside previously reported zT values of p-type MnTe-based materials for comparison^35–38. c Conversion efficiencies (η) as a function of temperature difference ($\mathrm{\Delta }$T ) for the segmented thermoelectric module in this work, along with reported results of MnTe-, GeTe-, PbTe-, and SnTe-based modules from the literature^36,39–43.

As a consequence, we eventually obtain a peak zT value of ~1.6 at 773 K and an average zT of ~0.9 from 300 to 773 K, both of which are the highest reported values in all p-type MnTe systems^35–38 (Fig. 1b). Assembled into a segmented device, an energy conversion efficiency of 11% is attained under a 473 K temperature gradient (Fig. 1c), which is comparable to the performance of some IV–VI group thermoelectric systems^36,39–43. The high thermoelectric performance demonstrated here offers a promising approach for optimizing ionic-bonded compounds and highlights their great potential in power generation applications.

Results and discussion

Bond manipulation and mechanism

Herein, we engineer the chemical bonds by alloying GeTe and AgSbTe₂ into MnTe to relieve the electron localization associated with the intrinsic ionic bonding nature of MnTe. Electron localization function (ELF) and Bader charge analyses are performed to investigate the chemical bond characteristics of hexagonal MnTe, cubic GeTe-alloyed MnTe, and cubic Ge/Ag/Sb-alloyed MnTe, as illustrated in Fig. 2a–c and Supplementary Table 1, respectively. For hexagonal MnTe, the ELF results reveal regions of high electron density surrounding Te²⁻ anions (green areas) and regions of low electron density between Mn sites and adjacent Te sites (blue areas). This observation is consistent with the Bader charge analysis, which reveals substantial charge transfer of ~0.94 e from Mn to Te in hexagonal MnTe. Taken together, these results demonstrate that electrons are predominantly localized on the Te²⁻ anions, confirming the strongly ionic nature of the Mn-Te bond. In stark contrast, the incorporation of Ge and Sb significantly increases the ELF values between anions and cations, accompanied by a reduced magnitude of Bader charge transfer relative to Mn. The simultaneous increase in ELF and decrease in charge transfer indicate weakened bond ionicity and enhanced electron delocalization. To further substantiate these bonding evolutions from a quantum-chemical perspective, crystal orbital Hamilton population (COHP)⁴⁴ analyses are performed to examine pairwise atomic interactions in hexagonal MnTe, cubic GeTe-alloyed MnTe, and cubic Ge/Ag/Sb-alloyed MnTe, as illustrated in Fig. 2d–f. The COHP method partitions the band structure energy into individual orbital-pair contributions, thereby providing a quantitative description of the bonding characteristics between specific atomic pairs. Negative COHP values correspond to bonding states, while positive values indicate antibonding interactions. In addition, the integrated COHP (ICOHP) values offer a measure of bond strength, where larger absolute ICOHP values correspond to stronger bonds. The insets in Fig. 2d–f present the -ICOHP values of representative bonds. Among all interactions, the Mn-Te bond exhibits a relatively low magnitude of -ICOHP, reflecting its intrinsically weak and highly ionic nature. This observation is consistent with the ELF and Bader charge results, both of which reveal minimal charge density between Mn²⁺ cations and Te²⁻ anions, confirming the main ionic character of the Mn-Te bond. Upon the introduction of Ge and Sb, the -ICOHP values enhance substantially, indicating the formation of stronger Ge-Te and Sb-Te bonds relative to Mn-Te. This enhancement in bond strength arises from the increased covalent character resulting from the redistribution of charge density between Ge²⁺/Sb³⁺ cations and Te²⁻ anions, as reflected in the elevated ELF values and reduced Bader charge transfer. Collectively, these results suggest that the incorporation of Ge and Sb elements effectively modulates the electron distribution and local bonding environment of MnTe, underpinning its enhanced electrical performance. Experimentally, the bond manipulation strategy not only leads to a considerable increase in n but also facilitates a phase transition from hexagonal to cubic structure (Fig. 2g and Supplementary Fig. 1). X-ray diffraction (XRD) data clearly demonstrates that the samples exhibit a single-phase rock-salt structure when the Ge content reaches 0.4. Additionally, X-ray photoemission spectra (XPS) analysis of the Mn-2p core state, including Mn-2p_1/2, Mn-2p_3/2, and a satellite peak of Mn-2p_3/2, indicates that Mn exists exclusively in the Mn²⁺ state⁴⁵, ruling out the presence of Mn^4+/3+ states across all samples (Supplementary Fig. 2). It is known that this phase transformation is intrinsically dominated by the underlying chemical bonding mechanisms⁴⁶. In this work, we successfully modify the chemical bonds in MnTe through introducing GeTe and AgSbTe₂, as evidenced by enhanced electron delocalization. Furthermore, our calculations indicate that the alteration of chemical bonds lowers the formation energy of the cubic structure relative to the hexagonal structure when the Ge content approaches 0.4 (Fig. 2g and Supplementary Tabel 2), ultimately inducing a phase transition from the hexagonal to the cubic phase (Fig. 2h). Additionally, thermogravimetric (TG) and differential scanning calorimetry (DSC) analyses are also performed on (Mn_0.6Ge_0.4Te)_0.85(AgSbTe₂)_0.15 over the temperature range from 300 to 773 K (Supplementary Fig. 3). The obtained results substantiate the thermal stability of this material, with no detectable phase transitions observed within the measured temperature range.

Fig. 2. Bond characterizations and phase transition.

a–c Electron localization function (ELF) of hexagonal MnTe (a), cubic Mn_0.6Ge_0.4Te (b), and cubic (Mn_0.6Ge_0.4Te)_0.85(AgSbTe₂)_0.15 (c). d–f Negative crystal orbital Hamilton (-COHP) curves for different chemical bonds in hexagonal MnTe (d), cubic Mn_0.6Ge_0.4Te (e), and cubic (Mn_0.6Ge_0.4Te)_0.85(AgSbTe₂)_0.15 (f). Insets: corresponding negatively integrated COHP (-ICOHP) values. g Calculated formation energy differences between the cubic and hexagonal phases, alongside the carrier concentration n_h at 300 K. h Schematic diagram of the phase transition.

Thermoelectric properties

To investigate the effects of bond-driven structural transitions on the electrical and thermal transport properties, thermoelectric measurements are conducted on Mn_1-xGe_xTe materials (x = 0 to 0.4) (Supplementary Fig. 4a–f ). Electrical conductivity rises markedly with x across the entire temperature range and reaches ~400 S cm⁻¹ at 300 K for x = 0.4. In contrast, the room-temperature Seebeck coefficient shows a pronounced decrease at x = 0.1, followed by a stabilization when x ≥ 0.2. The trends in the electrical conductivity with x arise from the combined contributions of n and μ. Room-temperature Hall measurements demonstrate that the introduction of GeTe induces a notable increase in both n and μ (Fig. 3a). This considerable enhancement in n is related to the increased electron delocalization caused by Ge alloying. Furthermore, μ exhibits a consistent improvement despite the simultaneous rise in n. This enhancement in μ can be ascribed to the structural transition from hexagonal to cubic symmetry and reduced bond polarity. As illustrated in Fig. 3b, the total carrier scattering rate decreases during the transition from the hexagonal to the cubic MnTe phase. This reduction is attributed to the suppressed acoustic deformation potential (ADP) scattering in the cubic phase. Additionally, the strong polarized optical phonon (POP) scattering, arising from the large electronegativity difference between Mn (1.55) and Te (2.10), remains a dominant factor limiting the μ in MnTe^34,35,47 (Fig. 3b). The addition of Ge, possessing a higher electronegativity (2.01), facilitates electron delocalization, effectively diminishing bond polarity and mitigating optical phonon scattering, thereby leading to a further enhancement in μ. To further elucidate the underlying mechanisms in electrical transport properties, density functional theory calculations are employed to examine the evolution of electronic band structure upon Ge alloying (Fig. 3c–e). The results reveal a notable increase in the DOS near the valence band maximum (VBM) upon GeTe alloying, indicating that the incorporation of GeTe can significantly modulate the electronic band structure and induce an increase in the density-of-states effective mass (m^*), and this tendency aligns with the experimental results (Fig. 3a). Further projected DOS analysis on cubic Mn₁₄Ge₁₀Te₂₄, shown at the bottom of Fig. 3c, provides additional evidence that the substantial DOS enhancement primarily originates from the contribution of Ge. Consequently, the electron delocalization and band modifications achieved through the chemical bond engineering effectively decouple the interdependencies among key electrical parameters (n, μ, and m^*), facilitating concurrent improvements across these factors. As a result, the chemical bond engineering results in a pronounced enhancement in PFs compared to the pristine MnTe and previously reported p-type MnTe thermoelectrics^35,36,47,48 over a large temperature range (Fig. 3f).

Fig. 3. Electrical transport properties.

a Room-temperature n, m^*, and μ of MnTe and Mn_0.6Ge_0.4Te. b Calculated carrier scattering rates for hexagonal and cubic MnTe at 300 K. c Density of states for hexagonal Mn₂₄Te₂₄ and cubic Mn₁₄Ge₁₀Te₂₄ (top) and the projected density of states (PDOS) of cubic Mn₁₄Ge₁₀Te₂₄ (bottom); the valence band maximum (VBM) has been shifted to 0 eV. d, e Band structures of hexagonal Mn₂₄Te₂₄ and cubic Mn₁₄Ge₁₀Te₂₄. f Comparison of the PF from 300 to 773 K among MnTe, Mn_0.6Ge_0.4Te developed in this work, and other high-performance p-type MnTe thermoelectrics^35,36,47,48.

While the bond manipulation effectively enhances the electrical performance, it concurrently increases ${\kappa }_L$ due to the elevated crystal symmetry (Supplementary Fig. 4e). To offset this negative effect, alloying with AgSbTe₂ is strategically implemented to suppress thermal transport and further boost zT. Figure 4 illustrates the temperature-dependent thermoelectric properties for (Mn_0.6Ge_0.4Te)_1-y(AgSbTe₂)_y samples (y = 0, 0.05, 0.10, and 0.15). With increasing AgSbTe₂ content, the $\sigma$ initially rises and subsequently declines, while the positive S values exhibit an inverse trend related to $\sigma$. According to the single parabolic model (SPB), the Pisarenko relation plotted in Fig. 4c demonstrates a continuous enhancement in the density-of-states effective mass (m^*) upon AgSbTe₂ alloying. Notably, m^* reaches a maximum of 11.45 m_e in (Mn_0.6Ge_0.4Te)_0.85(AgSbTe₂)_0.15, which is much higher than other reported MnTe alloys^36,49,50, highlighting the significant impact of AgSbTe₂ alloying on electronic band structure modulation. To further investigate the origin of the pronounced increase in m^*, DFT calculations are performed, as depicted in the inset of Fig. 4c and Supplementary Fig. 5. The results reveal an obvious increase in the DOS near the VBM with increasing y, indicating that the addition of AgSbTe₂ induces modifications in the electronic band structure and enhances the m^* of Mn_0.6Ge_0.4Te. Additionally, AgSbTe₂ alloying also cause a pronounced suppression in both the total thermal conductivity (Fig. 4d) and the lattice thermal conductivity (Fig. 4e). Notably, the ${\kappa }_L$ in AgSbTe₂-alloyed samples is largely reduced and reaches an ultralow value of ~0.66 W m^_1 K⁻¹ when y = 0.15. Owing to the increased m^* and largely suppressed ${\kappa }_L$, the weighted mobility ${\mu }_{\mathrm{w}}$(defined as: μ_w = μ (m^*/m_e)^3/2)⁵¹ and the parameter exhibit significant improvement compared to the samples without AgSbTe₂ (Fig. 4f).

Fig. 4. Thermoelectric properties and Pisarenko relations of the MnTe-based materials.

a Electrical conductivity. b Seebeck coefficient. d Total thermal conductivity, e Lattice thermal conductivity. f Weighted mobility and the value of ${\mu }_w/{\kappa }_{\mathrm{lat}}$ at 300 K. c Carrier concentration-dependent Seebeck coefficient at 300 K compared to reported values of other MnTe-based materials^36,49,50, with density-of-states of cubic Mn₁₄Ge₁₀Te₂₄ and cubic Mn₁₁Ge₇Ag₃Sb₃Te₂₄ given in the inset. The valence band maximum (VBM) has been aligned at 0 eV.

To investigate the origins of the low ${\kappa }_L$ in 15%AgSbTe₂-alloyed Mn_0.6Ge_0.4Te, atom probe tomography (APT) and transmission electron microscopy (TEM) are employed to examine the nanoscale element distribution and microstructural features, respectively. Three-dimensional APT reconstruction maps reveal all constituent elements are spatially uniformly distributed within the matrix, with no indication of phase separation, as illustrated in Fig. 5a. Furthermore, low-magnification TEM image reveals a fine-grained microstructure with a grain size of ~1 μm and a high-density stacking fault within the grains (Fig. 5c), which effectively strengthens the low-frequency phonon scattering. The corresponding energy dispersive spectrum (EDS) maps indicate slight Mn enrichment at the grain boundaries (Supplementary Fig. 6). To gain deeper insights into the considerable SFs at the atomic-scale, a high-angle annular dark-field (HAADF) STEM image is presented in Fig. 5d. Two distinctly oriented SFs are observed in this sample, as indicated by the magnified images presented in Fig. 5e, f. To identify the specific slip planes associated with the SFs, a fast Fourier transformation (FFT) of Fig. 5d is performed, as shown in Fig. 5g. The elongated diffraction spots in the FFT pattern further confirm the presence of SFs, and indicate that the SFs occur on the {1 1 1} slip planes. Except for SFs, Lomer-Cottrell (L-C) locks (marked by red solid circles) are formed through the reaction of leading partial dislocations on two intersecting slip planes. The dense network of grain boundaries and high-density SFs can be ascribed to the ball-milling synthesis process, as well as to the slight reduction in intrinsic stacking fault (ISF) energy and the energy barriers (EUSF) for forming ISF, as presented in Fig. 5b. The generalized stacking fault energies (GSFEs) calculations are performed using a slab-vacuum supercell method, incorporating a 16 Å vacuum layer along the [111] direction. The GSFEs profiles are obtained by incrementally shifting the upper six atomic layers along the [$11\overline{2}$] slip direction, followed by partial optimization of atomic coordinates along the [111] axis. In addition to the 2D SFs and grain boundaries, the dense 0D point defects resulting from the high alloying concentration also act as effective phonon scattering centers, and generate significant mass and strain field fluctuations, substantially impeding phonon transport. Geometric phase analysis based on the STEM-HAADF imaging of Fig. 5d reveals that the extensive SFs and dense point defects induce pronounced strain fluctuations (Fig. 5h), leading to intensifying phonon scattering. Furthermore, the addition of AgSbTe₂ leads to an increase in the bond length between anions and cations (Supplementary Tables 3 and 4), significantly weakening the bond stiffness and resulting in a decreased phonon group velocity (Supplementary Table 5).

Fig. 5. Microstructural characterization of 15%AgSbTe₂-alloyed Mn_0.6Ge_0.4Te.

a 3D APT reconstruction displaying the spatial distribution of elements. b Generalized stacking fault energies (GSFE) landscapes of Mn₂₉Ge₁₉Te₄₈ and Mn₂₂Ge₁₄Ag₆Sb₆Te₄₈ along the $\left[11\overline{2}\right]$ direction. Insets: Atom configurations of the ideal and intrinsic stacking fault structure. USF and ISF denote the unstable stacking fault and intrinsic stacking fault, respectively. c Low-magnification TEM image. d High-magnification HAADF-STEM image. e, f Magnified version of the region as marked in (d). g Fast Fourier transformed (FFT) patterns of (d). h Strain mapping of (d) along xy plane.

Consequently, the bond softening, combined with a multiscale hierarchical structure (Supplementary Fig. 7) including atomic-scale point defects, nanoscale SFs, and micron-scale grain boundaries, leads to a significant reduction in ${\kappa }_L$, decreasing it from 1.24 W m⁻¹ K⁻¹ in Mn_0.6Ge_0.4Te to 0.66 W m⁻¹ K⁻¹ in (Mn_0.6Ge_0.4Te)_0.85(AgSbTe₂)_0.15 at 300 K (Fig. 4e).

zT and device performance

Owing to the enhanced power factor and suppressed thermal conductivity, the peak zT value considerably enhances from ~0.5 to ~1.6 at 773 K, as illustrated in Fig. 6a. This bond engineering strategy not only elevates the peak value but also yields substantial enhancements in zT across the entire temperature range, with the average zT rising from approximately 0.1 to 0.9 between 300 and 773 K, attributed to a transition in electron distribution behavior from localization state in MnTe to partial delocalization state in 15%AgSbTe₂-alloyed Mn_0.6Ge_0.4Te. Moreover, to ensure data reliability, two independent batches with identical compositions are synthesized and measured, exhibiting good reproducibility of thermoelectric properties (Supplementary Fig. 8). In addition to achieving high thermoelectric performance, the enhancement of mechanical properties is also critical for practical applications. In this work, the Vickers hardness of our MnTe-based samples is significantly enhanced from 120 to 250 Hv (Supplementary Table 6), which surpasses many reported thermoelectric materials with similar compositions^52–57 (Fig. 6b). This substantial improvement is primarily attributed to solid solution strengthening induced by multi-element alloying, as well as microstructural modifications, including the introduction of abundant SFs and grain refinement. To further demonstrate the superior thermoelectric performance, a two-pair segmented thermoelectric module with 10 $\times$ 10 mm² ceramic substrates on both the top and bottom surfaces was fabricated. The p-type leg consists of a high-temperature segment of MnTe-based material developed in this work and a low-temperature segment of Bi₂Te₃-based material, while the n-type leg is composed of n-type Mg₃Sb₂-based material, as depicted in Supplementary Fig. 9. Detailed synthesis methods for p-type Bi₂Te₃-based and n-type Mg₃Sb₂-based materials can be found in previous resports^58,59. A conversion efficiency evaluation equipment (EPS-101) is employed to measure the voltage (V), output power (P), heat flow (Q), and thermoelectric conversion efficiency (η) as a function of current under various temperature differences ($\mathrm{\Delta }T$). As presented in Fig. 6c, d, the I-U curves exhibit good linearity, with the maximum P and η reaching 0.55 W and 11%, respectively, under $\mathrm{\Delta }T$ = 473 K and a cold-side temperature (T_c) of 300 K.

Fig. 6. Thermoelectric and mechanical performance of MnTe-based materials and modules.

a zT value for (Mn_1-xGe_xTe)_1-y(AgSbTe₂)_y (x = 0 and 0.4, y = 0, 0.05, 0.10, and 0.15) samples. b Comparison of the microhardness of MnTe-based samples in this study with literature-reported values of similar compositions^52–57. c, d Current dependencies of the output voltage and output power (c), and the corresponding conversion efficiencies (d).

To sum up, we develop a bond manipulation strategy to delocalize electrons and modify the phase structure of p-type MnTe materials. Electron delocalization promotes the increase of the carrier concentration, while high alloying concentration induces notable modifications in the band structure and enhances the density of states near the VBM, resulting in a substantial enhancement in PF across a large temperature range. Thermally, bond softening reduces sound velocity, and the presence of multiscale hierarchical defects strongly scatter heat-carrying phonons, which lead to a substantial reduction in ${\kappa }_L$. As a consequence, a peak zT value (~1.6 at 773 K), a high zT_ave (~0.9 from 300 to 773 K), and a high conversion efficiency (11% at $\mathrm{\Delta }$T = 473 K) are realized in this p-type MnTe-based material and its corresponding module. Our demonstration highlights the potential of bond engineering for enhancing the thermoelectric performance of ionic-bonded compounds and for advancing the development of high-efficiency modules.

Methods

Synthesis procedure

High-purity raw materials, specifically Mn (99.95%), Ge (99.999%), Te (99.999%), Ag (99.999%), and Sb (99.999%), were weighed in stoichiometric ratios of (Mn_1-xGe_xTe)_1-y(AgSbTe₂)_y (x = 0, 0.1, 0.2, 0.3, and 0.4; y = 0, 0.05, 0.10, and 0.15). The materials were loaded into a stainless-steel jar under an argon atmosphere to prevent oxidation. Mechanical alloying was conducted via planetary ball milling (SPEX 8000 M) for 8 h. The resulting powders were then compacted in a graphite die with an inner diameter of 12.7 mm and sintered at 823 K under 50 MPa for 15 min using rapid hot-pressing to form dense pellet samples.

Thermoelectric performance measurements

Electrical conductivity (σ) and Seebeck coefficient (S) were simultaneously measured using commercial equipment (CTA-3 Cryoall). The Hall carrier concentration (n_H) and mobility (${\mu }_{\mathrm{H}}$) were measured via the van der Pauw method under a reversible magnetic field of 1.5 T, with calculations based on the formulas: and ${\mu }_{\mathrm{H}}={\sigma R}_{\mathrm{H}}$, where e denotes the electron charge. The total thermal conductivity (${\kappa }_{\mathrm{tot}}$) was derived from ${\kappa }_{\mathrm{tot}}=D\rho C_{\mathrm{p}}$, where thermal diffusivity (D) was measured using laser flash equipment (Linseis, Germany) under a helium flow environment. The uncertainties for the measured electrical conductivity, Seebeck coefficient, and thermal conductivity are 3%, 5%, and 6%, respectively. The heat capacity (C_p) was estimated based on the Dulong–Petit law, and the material density ($\rho$) was obtained using the Archimedes method. Transverse ($v_{\mathrm{T}}$) and longitudinal ($v_{\mathrm{L}}$) sound velocities were determined using an Olympus 38DL PLUS. The thermogravimetric analyzer (STA 449) and the differential scanning calorimeter (DSC 300) were employed to measure weight loss and heat flow, respectively. Samples were heated to 773 K at a rate of 10 K min⁻¹ under a flowing nitrogen atmosphere.

Structural characterization

XRD (Rikagu SmartLab) with Cu ${\mathrm{K}}_{\alpha }$ radiation (λ = 1.5418 Å) was used to characterize the phase structures. Microstructural analysis and elemental mapping were conducted using a high-resolution transmission electron microscope (Talos F200X, FEI, USA), equipped with an energy-dispersive X-ray spectroscopy (EDS) system with four independent silicon drift detectors. X-ray photoelectron spectroscopy (XPS, VG Multilab 2000, Thermo Electron Corporation) was employed to determine the chemical valence states of elements. For APT, the needle-shaped specimen was prepared using a dual-beam focused ion beam/scanning electron microscope (FIB-EBL, Helios 5 CX) via the standard “lift-out” technique. APT measurement was conducted on LEAP 6000 XR (CAMECA) with a UV laser (wavelength = 355 nm), employing a laser pulse energy of 60 pJ and a repetition rate of 158 kHz under a base temperature of 50 K and an average detection rate of 0.2%. APT data was reconstructed and analyzed using AP Suite 6.3 software (Cameca Instruments). Vicker hardness measurements were performed at 300 K using a standard microhardness tester (DuraScan, ZwickRoell GmbH) with a 1 kg load and a dwell time of 10 s.

DFT calculation

DFT calculations implemented in the Vienna Ab-initio Simulation Package (VASP)⁶⁰ were performed to obtain the total density of states and ELF. The Perdew–Burke–Ernzerhof generalized gradient approximation⁶¹ within the projector augmented wave⁶² method was employed to calculate the exchange-correlation energy. 2 × 3 × 4 supercells of cubic Mn₁₄Ge₁₀Te₂₄ and Mn₁₁Ge₇Ag₃Sb₃Te₂₄, and a 2 × 2 × 3 supercell of hexagonal Mn₂₄Te₂₄ were built from the corresponding primitive cells to achieve desired alloying concentrations via the Alloy-Theoretic Automated Toolkit package⁶³. A plane-wave energy cut-off of 400 eV and a sampling spacing of about 2π × 0.025 Å⁻¹ with Γ-centered Monkhorst-Pack scheme in the Brillouin zone were adopted. All the structures were fully relaxed with a force convergence criterion of 10^-3eV/Å before the electronic structure calculation, in which the energy convergence criterion is 10⁻⁶ eV. Dudarev’s DFT + U method⁶⁴ with U = 5 eV was used to include the effect of the electron–electron interactions in spin-polarized systems. In hexagonal and cubic MnTe calculations, a Γ-centered k-point grid with a k-spacing of 0.05 × 2π Å⁻¹ was used for structural relaxations. The convergence criteria for the total energy and ionic forces were set to 10⁻⁵ eV and 0.01 eV/Å, respectively. In the self-consistent calculations, we reduced the k-spacing to 0.04 × 2π Å⁻¹, and used 10⁻⁷ eV for the convergence criterion. To accurately determine the bandgap, the Heyd-Scuseria-Ernzerhof (HSE06) hybrid functional was adopted⁶⁵. Based on first-principles data, we utilized the ab initio scattering and transport package to compute electrical transport properties from ADP), ionized impurity scattering (IMP), and polar optical phonon (POP) scattering^66,67. The GSFEs are calculated using a slab-vacuum supercell approach within DFT. The COHP was calculated using the LOBSTER^68,69 program based on plane-wave DFT output from VASP. The Bader charge analysis was conducted by the Bader^70,71 program. The supercell vectors were defined as ($c_1,c_2,c_3$) = ([$11\overline{2}$], [$1\overline{1}0$], [111]), where $c_1$ and $c_2$ lie within the slip plane and $c_3$ is normal to the slip plane, as illustrated in the insets of Fig. 5b. A 16 Å vacuum layer was introduced along the $c_3$ direction, and periodic boundary conditions were applied in all directions. The supercell contains 96 atoms, and it was fully relaxed prior to constructing the fault structures. For each configuration, the total energy of the optimized fault-free structure was taken as the reference energy $E_0$. The GSFEs were then obtained by incrementally shifting the upper half of the atomic layers along the [$11\overline{2}$] direction, followed by partial optimization of atomic coordinates along the [111] axis and self-consistent energy calculation, yielding $E_s$ at each displacement. The GSFE was calculated using ${\gamma }_{GSF}=\frac{E_s-E_0}{A}$, where A is the area of the slip plane in the supercell.

Module fabrication and efficiency measurements

The synthesis procedures for the low-temperature segment of p-type BiSbTe materials and n-type Mg₃Sb₂ materials can be found in previous reports^59,72. The n-type Mg₃Sb₂-based leg with Fe contact layer and Ni metallization layer was consolidated by SPS at 973 K for 3 min under a uniaxial pressure of 40 MPa. The p-type BiSbTe-based leg, with Ni metallization layer, was fabricatied via rapid hot-pressing at 673 K for 5 min under a pressure of 80 MPa. Additionally, the thermoelectric properties of Bi₂Te₃-based materials are provided in Supplementary Fig. 10. For p-type MnTe-based leg, Ni-based barrier layer was electroplated to the sintered bulk. Finite-element analysis (COMSOL Multiphysics) was employed to optimize the geometric configuration of the two-pair module. All thermoelectric legs were cut to dimensions based on the simulation results. The high-temperature MnTe-based segments were bonded to the low-temperature BiSbTe-based segments using SAC305 solder alloy. Subsequently, both p- and n-type legs were bridged with the Ni-plated copper ceramics substrates using SAC305. AlN ceramic plates served as electrical insulators between the electrodes and the efficiency testing system. The final dimensions of the two-pair module is 10 $\times$ 10 $\times$ 7.76 mm³. A schematic diagram and physical structure of the device are presented in Supplementary Fig. 9. Efficiency measurements were conducted in thermoelectric conversion efficiency evaluation system (EPS-101) under different temperature gradients, and all data were recorded under near-steady-state conditions after thermal equilibrium was achieved. The output voltage (U) and heat flow (Q_c) at the cold side were recorded as a function of applied current (I), enabling calculation of output power (P) and conversion efficiency (ƞ) using the following formulas:

$$P=U\times I\left(\mathrm{S}1\right)$$

$$\begin{array}{c}\hfill \eta =\frac{P}{P+Q_{\mathrm{c}}}\times \hfill \end{array}100\%\left(\mathrm{S}2\right)$$

Author contributions

Haiqi Li, Chen Chen, and Yue Chen designed the concept and established the experimental scheme. Haiqi Li, Kejia Liu, Wenxuan Wang, Dongyi Shen, Huajian Wu, and Chen Chen synthesized the samples and conducted the thermoelectric property measurements. Shuang Lyu and Minglong Wang contributed to the DFT calculations. Yuanhang Xia contributed to the TEM analysis. Haiqi Li and Kejia Liu performed COMSOL calculations. Haiqi Li, Xiaofang Li, Jinxuan Cheng, Chen Chen, and Qian Zhang measured the device performance. Haiqi Li, Qian Zhang, Chen Chen, and Yue Chen analyzed the results. Haiqi Li, Chen Chen, and Yue Chen wrote this manuscript and all authors edited it.

Peer review

Peer review information

Nature Communications thanks Subhajit Roychowdhury and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data generated or analyzed during this study are included in the published article and its Supplementary Information. The data supporting the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70922-1.