A neuroevolution potential for predicting the thermal conductivity of <b> <i>α</i> </b>, <b> <i>β</i> </b>, and <b> <i>ε</i> </b>-Ga2O3
Zhanpeng Sun, Zijun Qi, Kang Liang, Xiang Sun, Zhaofu Zhang, Lijie Li, Qijun Wang, Guoqing Zhang, Gai Wu, Wei Shen
Abstract
Ga2O3 is an ultrawide-bandgap semiconductor with a variety of crystal configurations, which has the potential for a variety of applications, especially in power electronics and ultraviolet optoelectronics. However, there has been no single interatomic potential reported for Ga2O3 polymorphs in terms of molecular dynamics prediction of thermal conductivity. Here, one interatomic potential has been developed based on neural networks, which has the clear advantages of consuming less computational power than density functional theory and has high accuracy in predicting the thermal conductivity of the three polymorphs of Ga2O3. Using the neuroevolution potential, the thermal conductivity values at 300 K have been predicted. Hence, the κ[average-α] was 67.2% that of β-Ga2O3, and the κ[average-ε] was only 26.4% that of β-Ga2O3. The possible reasons for the discrepancies in thermal conductivity values in various crystal types and orientations have been explored. As a result, it could be shown that the contribution of low-frequency phonons to thermal conductivity was very significant in Ga2O3, and a unit cell with low symmetry and high atomic number would negatively impact the thermal conductivity of the material. In this work, a scheme has been proposed for accurately predicting the thermal conductivity of Ga2O3 and a relatively accurate value of the thermal conductivity of ε-Ga2O3 has been achieved, which could also provide an atomic-scale perspective for the insight into the thermal conductivity differences among α, β, and ε-Ga2O3.