Quantitative modeling of phonon-bridge modulated heat transport at GaN/Al interfaces using machine learning potentials

Efficient thermal management is critical for high-power gallium nitride (GaN)-based electronic devices. However, accurately modeling thermal transport across the GaN/Al interface, particularly in the presence of phonon bridges, remains challenging for molecular dynamics (MD) due to the limited fidelity of conventional interatomic potentials. Here, we develop a high-accuracy machine learning interatomic potential (MLIP) to systematically investigate phonon-bridge-mediated interfacial thermal conductance (ITC) of the GaN/Al heterointerface. We unveil a pronounced dichotomous dependence of ITC on the phonon-bridge thickness. While the pristine GaN/Al interface exhibits an ITC of 0.35 GW/m 2 K, introducing a 10 Å AlN intermediate layer maximizes the ITC to nearly 0.6 GW/m 2 K. Notably, as the bridge thickness increases, the ITC gradually degrades, reaching a critical thickness at approximately 50 Å where it converges to the pristine value. Comprehensive analyses of the interfacial phonon density of states and spatial temperature profiles elucidate the underlying mechanism: this critical thickness is governed by a trade-off between the enhanced interfacial phonon spectral matching and the increasing intrinsic bulk thermal resistance of the AlN layer. Furthermore, we propose an analytical model that quantitatively captures this modulation of interfacial heat transport. Our findings provide theoretical guidelines for engineering phonon bridges in next-generation thermal management architectures.