Phonon and Thermal Properties of Silicon Carbide: A Comparison of Empirical and Machine Learning Potentials

Abstract

Silicon carbide (SiC), as a third‐generation semiconductor material, has attracted significant research attention. Various empirical potentials and machine learning potentials have been developed, but there are few comparative studies on phonon and thermal properties. Herein, the Tersoff and Vashishta empirical potentials, as well as the Bayesian force field constructed by the FLARE framework using principled Gaussian process uncertainties (FLARE BFF), for a comparative study, are selected. The phonon dispersion relation, phonon density of states, Grüneisen constants, and the average phonon‐weighted Grüneisen constants are calculated using different potentials, and it is found that the FLARE BFF potential has the highest accuracy with respect to the first‐principles calculations. Furthermore, the thermal conductivity using molecular dynamics simulation with different potentials is calculated. The calculation results using the FLARE BFF potential closely match the experimental reports at high temperature, but the longest computing time is required. This study can facilitate the understanding of thermal properties of SiC.