Advancing Thermal Management with Machine-Learning Potentials on Boron Nitride (BN) and Other Group 13 Nitrides

Abstract

To achieve seamless heat dissipation, it is essential to use materials with high thermal conductivity to improve thermal management. In this study, we have utilized ab initio and machine learning techniques to systematically explore the lattice thermal conductivity of BN and other group 13 nitride based bulk and bilayer materials. By employing data-driven training of potentials of different atomic configurations at different time steps obtained from the AIMD data, we have demonstrated the comparability of the results obtained from the machine-learned potentials compared to the density functional theory (DFT) based calculations on thermal conductivity. Furthermore, we examined the significance of four phonon interactions in group 13 nitrides by comparing the calculated values with the available experimental values. Notably, bilayer AlN exhibits a high thermal conductivity (881 W m-1 K-1) due to its stronger covalent bonding, which contradicts the trend observed from B to In. Our study highlights that the machine learning potential-based approach can provide more accurate results than DFT, paving the way for future robust investigations of materials using high-throughput screening techniques. © 2023 American Chemical Society.