Neuroevolution potential-driven accurate and efficient discovery of Graphene/GaN heterojunctions: From ballistic-diffusive transition to thermal conductivity enhancement

Abstract

Two-dimensional gallium nitride (2D GaN) exhibits outstanding potential for next-generation nanoelectronic and optoelectronic devices due to its high electron mobility and tunable electronic properties. Nevertheless, its relatively low thermal conductivity can lead to localized heat accumulation, which adversely affects device performance. A feasible strategy is to construct 2D graphene/GaN heterojunction presents an effective approach to enhance thermal transport. In this paper, we trained neuroevolution potential (NEP) for accurate and efficient calculate of the thermal properties of GaN/Graphene heterojunction, this approach maintains density functional theory (DFT)-level accuracy while significantly improving computational efficiency. The NEP model achieves root-mean-square errors of 10.22 meV/atom, 203.25 meV/Å, and 60.55 meV/atom for energy, force, and virial predictions, respectively. We comprehensively validate the model through phonon dispersion, radial distribution functions, and thermal conductivity analysis. Furthermore, by integrating nonequilibrium molecular dynamics, homogeneous nonequilibrium molecular dynamics, and spectral heat current methods, we resolve the frequency-dependent phonon transport processes and quantitatively capture the transition from ballistic to diffusive regimes. The key finding is that by studying the spectral energy density and phonon lifetime, we have identified the fundamental reason for the significant alteration in the thermal transport mechanism, which graphene introduces a high-frequency channel, fundamentally enhancing the lattice thermal conductivity of the heterojunction.