Modeling thermal transport in AlN/GaN superlattices and heterostructures with machine-learned force fields

Thermal transport in superlattices (SLs) is particularly rich because it can exhibit three distinct transport regimes: coherent wave-like, ballistic incoherent, and diffusive propagation, each governed by different physical mechanisms. Coherent transport is especially promising for technological applications because it enables control of heat flow through band-structure engineering, interference effects, and phononic filtering in ways that are not accessible in purely diffusive materials. In this work, we investigate thermal transport across AlN/GaN heterostructure interfaces and SLs using non-equilibrium molecular dynamics simulations and the atomistic Green's function method in combination with a machine-learned force field potential. Our results are compared with analytical models and previous literature. For AlN/GaN interfaces, we find that inelastic phonon transmission contributes a significant fraction of the thermal boundary conductance, and we develop an inelastic extension of the diffuse mismatch model that accounts for all possible three-phonon processes. In the SL thermal conductivity, we observe a minimum at a period of approximately 2 nm, together with multiple signatures of partially coherent acoustic transport persisting even at room temperature. The results further reveal a fundamental contrast in the role of anharmonicity between isolated interfaces and periodic SLs, where it enhances interfacial conductance in the former while suppressing thermal transport in the latter.