Electrode interface optimization advances conversion efficiency and stability of thermoelectric devices

Thermoelectric generators have seen wide deployment in space missions and are being pursued for industrial waste heat recovery. While thermoelectric material performance (high ZT) and device topology are crucial, the electrode interface strongly influences both conversion efficiency and long-term reliability due to interfacial electrical resistance and degradation via reaction/diffusion. For CoSb3-based filled skutterudites, a leading mid-temperature class, device stability is hampered by interfacial reactions and Sb diffusion that form interfacial reaction layers (IRLs) and decomposition layers (DLs), raising resistance and inducing stress. Conventional barrier layer selection relies on qualitative guidelines (thermal expansion and work function matching) and trial-and-error, lacking predictive models for long-term interfacial evolution. The research question is how to predict and optimize barrier layers to achieve low interfacial resistivity and high stability by linking atomistic kinetics (reaction energy and diffusion barriers) to macroscopic IRL growth and device performance.

The paper situates the work within advances in TE materials (filled skutterudites, half-Heuslers, PbTe-based, Zintl phases) with ZT values exceeding 1.5–2.0, and device efficiency records reaching ~9% for single-stage and >12% for segmented/cascaded modules. Prior efforts addressed interface selection using thermal expansion/work function matching and various metals/alloys as barriers (e.g., Ti, W–Cu), but lacked quantitative, predictive screening of interfacial reaction/diffusion and long-term reliability. Previous observations indicate Sb dominates diffusion in CoSb3 interfaces; Yb filler participation is minimal. Models such as reaction–diffusion theory and Arrhenius kinetics have been used to describe interfacial growth qualitatively, but not tied to a practical screening criterion for barrier materials.

• Theoretical screening criterion: Modeled initial interfacial chemical reactions between CoSb3 and transition metals X forming CoSb2 and XSb2 (e.g., NbSb2 for X=Nb). Defined interfacial reaction energy EIR = E(CoSb2)+E(XSb2)−E(CoSb3)−E(X). Required EIR<0 to ensure bonding. For long-term growth, considered Sb diffusion as dominant; computed activation energy barrier for Sb migration EMig in XSb2 via CI-NEB.
• First-principles calculations: DFT using VASP with PAW pseudopotentials and PBE-GGA, 370 eV cutoff, gamma/2×2×2 k-point grids, energy convergence 1e−8 eV, forces <0.005 eV/Å. Formation enthalpies and Sb migration barriers computed; migration path evaluated with 8 CI-NEB images.
• Experimental joints: Fabricated CoSb3-based joints with Nb or Zr barrier foils (~25 μm) via one-step hot pressing at 923 K, 60 MPa, 30 min. TE materials: n-type Yb0.3Co4Sb12 and p-type Ce0.85Fe3CoSb12 prepared by melting–quenching–annealing. Joints diced (4×4×4.5 mm), vacuum-sealed, and isothermally aged at various temperatures for accelerated tests.
• Characterization: SEM/EDS mapping to resolve interfacial layers and compositions; measured IRL and DL thicknesses (averaged over >10 locations, multiple specimens). Interfacial resistivity measured by four-probe platform; decomposed total interfacial resistivity Rtotal into RIRL (IRL), RDL (DL), and contact resistances RC between layers.
• Kinetic modeling: Applied Deal–Grove-type relation t = x·(x−x0)/k1 + (x−x0)/k0 with Arrhenius temperature dependence for chemical constant k0 and diffusional constant k1. Identified reaction-controlled (linear) and diffusion-controlled (parabolic) regimes; defined critical thickness/time (x1/2, t1/2) for mechanism transition. Fitted ko, k1, E0 (reaction activation), E1 (diffusion activation) from time–thickness data at multiple temperatures; generated predictive surfaces of IRL thickness versus time and temperature.
• Module fabrication and testing: Built 8-pair module (20×20×14.5 mm) with Nb barrier, p-legs (5×4 mm) and n-legs (4×3 mm), hot-side connections to Mo0.6Cu0.4 via Ag–Cu–Zn brazing; cold-side to Cu-metallized ceramic via Pb–Sb solder; glass fiber spacers. Measured I–V, power, internal resistance, and efficiency on a home-made setup across operating temperatures; conducted long-term service test (Th=818 K, Tc=308 K, 846 h), thermal shock, and current kick tests.

• Screening criterion: Combining negative interfacial reaction energy (EIR<0) with high Sb migration barrier EMig in XSb2 defines a “sweet spot” for barrier layers. Elements like Nb, Ti, Zr have negative EIR; Cu and W do not bond (EIR>0). High EMig correlates with stronger X–Sb bonding and higher charge density between Sb atoms.
• Interfacial structure: Both Nb and Zr joints form dense IRLs (NbSb2 or ZrSb2) and a CoSb2 decomposition layer (DL). Yb is largely absent from IRL growth, detected only at DL/IRL boundary at trace levels.
• Kinetics: Early-time growth is reaction-controlled (t ~ x), transitioning to diffusion-controlled (t ~ x^2) after ~50 h (e.g., Nb@923 K, Zr@858 K). Fitted parameters show Zr joints have much higher ko and k1 and lower activation barriers (faster reaction and Sb diffusion) than Nb joints. Predicted IRLs in Nb joints remain much thinner over time/temperature than Zr.
• Critical regime: The critical time t1/2 is typically <500 h at the tested temperatures; thus, long-term service is diffusion-dominated.
• Interfacial resistivity: RC is very small (<0.6 μΩ·cm^2). DL resistivity ρDL ~2900 μΩ·cm is much larger than ρIRL, making RDL the dominant contributor to Rtotal after aging (e.g., Nb joint at 873 K: after 15 days, RDL ≈80% of Rtotal; Zr at 873 K: after 5 days, RDL ≈97%).
• Resistivity evolution: After 100 days at 873 K, fitted interfacial resistivity Rtotal remains <10 μΩ·cm^2 for Nb joints (≈2% rise in internal resistance), but reaches ~80 μΩ·cm^2 for Zr joints (≈16% rise).
• Device performance: The 8-pair Nb-barrier module achieves a record single-stage skutterudite efficiency of 10.2% at Th=872 K (ΔT=574 K), with Pmax=4.1 W. After 846 h service (Th=818 K, Tc=308 K), both output power and internal resistance changed by <1%; post-test ηmax remains 10.4% at Th=873 K (ΔT=575 K). A high 9.5% efficiency is achieved at Th=823 K, favorable for long-term stability.
• CTE matching: Nb (7.6×10−6 K−1) and NbSb2 (8.4×10−6 K−1) have thermal expansion close to SKD (~9.1×10−6 K−1), aiding mechanical reliability.

The study addresses the longstanding challenge of predicting and minimizing interfacial degradation in skutterudite thermoelectric devices by linking atomistic energetics and diffusion kinetics to IRL growth and interfacial resistivity. Requiring EIR<0 ensures robust bonding, while a large EMig suppresses Sb diffusion, slowing IRL thickening and resistivity growth. Nb satisfies both criteria, yielding the slowest IRL growth, the smallest interfacial resistivity, and superior long-term stability relative to Zr and previously used Ti barriers. The reaction–diffusion model and fitted Arrhenius parameters allow quantitative prediction of IRL thickness and interfacial resistance over time and temperature, informing device design and service conditions. Implementing Nb as a barrier in an optimized 8-pair module results in record single-stage efficiency and negligible degradation over extended high-temperature operation, demonstrating the practical impact of the screening criterion on device integration and reliability.

This work introduces a simple, predictive criterion—negative interfacial reaction energy combined with high Sb migration barrier in the reaction layer—to screen barrier materials for CoSb3-based skutterudite devices. Guided by this, Nb is identified and validated as an optimal barrier, producing the slowest IRL growth and lowest interfacial resistivity among reported joints. An 8-pair Nb-barrier module achieves a record 10.2% efficiency at 872 K and exhibits exceptional stability over 846 h. The approach bridges microscopic kinetics and macroscopic performance, offering a generalizable pathway for interface optimization in thermoelectric device integration. Future research could extend this framework to other TE material systems, refine diffusion modeling beyond Sb-only assumptions, and explore additional barrier candidates within the identified sweet spot to balance bonding, blocking, and conducting functions.

• Model simplifications: Yb-filled CoSb3 was simplified to CoSb3; filler participation in IRL growth (e.g., Ba) is not comprehensively treated. Yb was detected only in trace amounts at interfaces, but its full role remains unresolved.
• Diffusion focus: Only Sb diffusion in XSb2 was considered; other species (Co, X) diffusion and effects of CoSb2 diffusion were assumed negligible.
• Threshold ambiguity: A precise EMig cutoff was not defined; Ti’s performance was used as a heuristic reference.
• Experimental scope: Validation focused on Nb and Zr; broader experimental screening of other predicted candidates (e.g., Ta, Y) was not performed.
• DFT approximations: Use of PBE-GGA and supercell limitations may affect absolute values of EIR and EMig; E1 (experimental diffusion activation) encompasses processes beyond the calculated EMig.
• Long-term extrapolation: IRL and resistivity predictions rely on Arrhenius fits and Deal–Grove-type kinetics; real service environments may introduce thermal cycling, mechanical stress, and chemical exposures not fully captured.