Interfacial Thermal Transport Analysis using Machine Learning Potential at AlN/Cu van der Waals Interface

Three-dimensional integration is critical for next-generation integrated circuits, yet thermal management remains challenging due to complex interface environment. Owing to its high thermal conductivity, aluminum nitride (AlN) is a promising insulating layer for through-silicon via structures, yet its interfacial thermal behavior with copper (Cu) electrodes remains insufficiently understood. Here, we investigate phonon transport across non-bonded AlN-Cu interfaces governed by van der Waals (vdW) interactions using a machine learning potential (MLP) trained by density functional theory (DFT) data. The MLP accurately reproduces DFT-level phonon dispersion, energy-volume relationships, and interfacial binding energies. Non-equilibrium molecular dynamics simulations reveal total thermal resistance values of 10-14 m 2 K/GW at room temperature, approximately five times smaller than values obtained with empirical potentials and in excellent agreement with experimental data for similar semiconductor-metal interfaces. Unlike empirical potentials, the MLP predicts increasing thermal resistance with temperature, attributed to phonon softening and vibrational mismatch. Spectral decomposition analyses confirm reduced phonon thermal conductance at elevated temperatures. This work provides fundamental insights into weakly bonded semiconductor-metal interfaces and guidance for thermal management in 3D integrated systems.