Mode-Resolved Phonon Transmittance Across Ga2O3/SiC Interface Using Lattice Dynamics With Machine Learning Potentials

Abstract Gallium oxide (Ga2O3) is emerging as a key material in power electronics due to its wide bandgap, with silicon carbide (SiC) serving as a potential efficient heat-removing substrate in Ga2O3 devices. Traditional methods fall short in accurately predicting thermal boundary conductance (TBC) at the Ga2O3/SiC interface. In this work, we combine machine learning potentials (MLP) and lattice dynamics (LD) to access mode-resolved phonon transmittance across Ga2O3/SiC interface. The MLP, specifically the MACE model, was trained using density functional theory (DFT) data to achieve near first-principles accuracy. The LD approach, grounded in the Landauer formula, significantly reduces computational demands by three orders of magnitude compared to traditional molecular dynamics (MD) methods. Our analysis of atomic displacements near the interface confirmed the reliability of our mode-resolved phonon transmittance. We found the TBC for Ga2O3/SiC to be around 137 MW/m2K, notably lower than predictions by the diffuse mismatch model (DMM), suggesting a pronounced material mismatch. Our approach lays the foundation for predicting the TBC of real material interfaces with first-principles accuracy and offers opportunities for the control of TBC in realistic material interfaces.