Abstract
Efficient thermal management remains a major challenge in β-Ga2O3 power devices owing to the inherently low thermal conductivity of the material and the large phonon mismatch at its heterointerfaces. In this study, the thermal boundary conductance (TBC) across β-Ga2O3/4H–SiC interfaces is systematically studied using non-equilibrium molecular dynamics simulations implemented in quantumatk. The machine-learned interatomic potential, based on the M3GNet graph neural network framework, is employed to accurately capture interatomic interactions with high fidelity. The simulated TBC values range from 22 MW/m2 K for dielectric Si3N4 interlayers to 73 MW/m2 K for metallic TiN interlayers. Compared to the TBC of ≈40 MW/m2 K obtained for the corresponding direct interface, the results clearly suggest a measurable increase in interfacial heat transport through transition-nitride layers. The analysis of the phonon density of states and cosine similarity S confirms that improved vibrational overlap (SGa2O3/TiN=91%, STiN/SiC = 82% vs SGa2O3/SiC=71%) directly correlates with the increased TBC. Furthermore, atomistic calculation of phonon transmission spectra clearly illustrates that the favorable impact of the TiN interlayer is particularly on the heat-carrying low frequency phonons. These findings highlight the significance of phonon-spectrum matching and the potential role of metallic nitride interlayers (such as TiN) as an effective pathway to mitigate thermal resistance in β-Ga2O3/SiC heterostructures for next-generation power electronics.