Investigating thermal transport across the AlN/diamond interface via the machine learning potential

Aluminum nitride (AlN) is an ultrawide-bandgap semiconductor with excellent potential for high-power applications. However, the heat dissipation issue remains a huge challenge for the use of AlN in high-power devices. A promising solution for this problem is to integrate AlN with the diamond heat sink. Therefore, interfacial thermal transport has become a significant bottleneck in thermal management. In this work, one neuroevolution potential is developed based on neural networks, which significantly improves the accuracy of predicting thermal properties compared to traditional potentials and address the issue of inaccurate thermal performance prediction of AlN/diamond heterostructures using traditional potentials. Molecular dynamics simulations based on NEP is adopted to investigate the crystal-orientation-dependent and interfacial-atom-dependent thermal boundary resistance of the AlN/diamond heterostructures after the interfacial bonding. Compared to bonding with Al and C atoms, the TBR decreases by approximately 50 % after bonding with N and C atoms at the interface. Especially for the AlN (0001)-N-diamond (100) heterostructure, the TBR is 0.95 m2·K·GW−1, very close to the theoretical limit of 0.8 m2·K·GW−1 through the diffuse mismatch model (DMM) theory. Finally, the insightful optimization strategies are proposed in this work which could pave the way for better thermal design and management of AlN/diamond heterostructures.