Achieving ultralow and highly isotropic thermal conductivity in coherent Si₃N₄ ceramics heterostructure: A machine learning potential-based molecular dynamics simulation study

Achieving ultralow and highly isotropic thermal conductivity in thin-walled ceramic materials is critical for advanced thermal management and energy applications. In this study, we employ a machine learning potential-based molecular dynamics (MD) simulation to investigate the thermal transport properties of coherent silicon nitride (Si₃N₄) heterostructures. Our non-equilibrium molecular dynamics (NEMD) simulations, performed on samples with effective lengths ranging from 20 to 160 nm and considered temperatures between 300 K and 1200 K, reveal that the coherent α/β‑Si₃N₄ heterostructure exhibits dramatically reduced thermal conductivity compared to Si₃N₄ uniphases. Notably, the coherent α/β‑Si₃N₄ heterostructure demonstrates an ultralow and nearly isotropic thermal conductivity, with an extrapolated infinite-length value of approximately 5.69 W m⁻¹K⁻¹. This performance is attributed to a significantly reduced phonon mean free path resulting from the coherent interfaces, which mitigates the excitation of additional phonon modes. Furthermore, temperature-dependent analyses reveal an anomalous “thermal skip” behavior linked to partial phase transitions that modulate phonon scattering. These findings underscore coherent heterostructuring as a promising strategy for tailoring the thermal properties of Si₃N₄ ceramics for thermoelectric and thermal protection applications.