Machine learning-assisted investigation on the thermal transport of β-Ga2O3 with vacancy

beta-Ga2O3 is a promising ultra-wide bandgap semiconductor in high-power and high-frequency electronics. The low thermal conductivity of beta-Ga2O3, which can be further suppressed by the intrinsic vacancy, has been a major bottleneck for improving the performance of beta-Ga2O3 power devices. However, deep knowledge on the thermal transport mechanism of beta-Ga2O3 with defect is still lacking now. In this work, the thermal transport of beta-Ga2O3 with vacancy defects is investigated using the machine learning-assisted calculation method. First, the machine learning moment tensor potential (MTP), which can accurately describe the lattice dynamics behaviors of pristine beta-Ga2O3 and solves the problem of low computational efficiency of existing computational models in beta-Ga2O3 large-scale simulations, is developed for studying the thermal transport of the pristine beta-Ga2O3. Then, the MTP is further developed for investigating the thermal transport of beta-Ga2O3 with vacancy and the thermal conductivity of beta-Ga2O3 with oxygen atom vacancies, which are evaluated by machine learning potential combined with molecular dynamics. The result shows that 0.52% oxygen atom vacancies can cause a 52.5% reduction in the thermal conductivity of beta-Ga2O3 [100] direction, illustrating that thermal conductivity can be observably suppressed by vacancy. Finally, by analyzing the phonon group velocity, participation ratio, and spectral energy density, the oxygen atom vacancies in beta-Ga2O3 are demonstrated to lead to a significant change in harmonic and anharmonic phonon activities. The findings of this study offer crucial insights into the thermal transport properties of beta-Ga2O3 and are anticipated to contribute valuable knowledge to the thermal management of power devices based on beta-Ga2O3.