【Member News】Southern University of Science and Technology Hua Mengyuan Team Developed a Machine Learning Potential Function Model for Complex Multiphase Gallium Oxide

      The fourth generation of ultra-wide band gap semiconductor materials represented by Gallium Oxide (Ga2O3) has excellent physical properties, which is expected to further break through the theoretical limit of existing semiconductor materials, and has important application prospects in the field of electronic power devices, solar blind detection, and high temperature gas sensing. However, due to the high complexity of Gallium Oxide system, the existing computational simulation studies are limited to first-principles calculations at the hundred-atom level. Some important scientific and technological studies related to Gallium Oxide must be based on large-scale computing systems above the 10,000 atomic level, because the required computational amount is too large to carry out systematic studies solely on first-principles calculations.

      Recently, the team of Hua Mengyuan, assistant professor of the Department of Electronic and Electrical Engineering, Southern University of Science and Technology, combined the cutting-edge development trend of Gallium Oxide semiconductor technology and the most cutting-edge research of machine learning, successfully developed a large-scale simulation of the molecular dynamics potential function of Gallium Oxide multiphase symbiosis system, and in-depth study of its growth regulation mechanism and important structural characteristics. It is very important to accelerate the maturity of Gallium Oxide semiconductor technology and solve the key technical problems of the preparation of multiphase symbiosis system. This work is described as “Complex Ga2O3 polymorphs explored by accurate and general-purpose machine-learning interatomic potentials” was published in npj Computational Materials which is a top journal in the field of computational materials.

      In this study, a high-precision first-principles database of Gallium Oxide multiphase systems was established through large-scale first-principles calculations (FIG. 1). Gaussian process regression machine learning algorithm is used to train and fit the database, generate multiple potential function versions of the same generation, and output potential function test results. The accuracy, universality and calculation efficiency of the potential function under each parameter version were tested systematically by the self-developed automatic test software package, and the optimized potential function version after many iterations was obtained.

FIG. 1. Summary diagram of the second-generation Gallium Oxide machine learning database model independently developed in this study. The database covers (i, ii) all experimentally known and theoretically predicted high-precision models of Gallium Oxide crystal phases, and is universally compatible with data models of (iii) amorphous/high-temperature molten and (iv) high-energy/discrete Gallium Oxide systems, successfully meeting both high-precision and general-purpose molecular dynamics simulations.

      The final published potential function has the advantages of high precision, universality and high computational efficiency, and can effectively simulate the structural evolution of Gallium Oxide materials in 100,000 to million atomic systems. Corresponding molecular dynamics simulations successfully revealed the complex dynamic process of relatively independent gallium and oxygen sublattice during recrystallization at the solid-liquid interface of the β-phase Gallium Oxide (FIG. 2). The subsequent potential function can be applied to the study of phase transformation by high-energy ion beam irradiation, lattice heat transport, surface gas epitaxy growth and other important technical fields related to Gallium Oxide materials.

FIG. 2. Molecular dynamics simulation of interfacial recrystallization in solid-liquid phase transformation using machine learning potential functions. A unique low mobility oxygen atom at the interface was successfully revealed. This group of oxygen atoms corresponds to the rapidly formed face-centered cubic sublattice oxygen atoms at the defect region of the interface.

      The University of Southern Science and Technology is the first unit of the paper. Mengyuan Hua, assistant professor in the Department of Electronic and Electrical Engineering, and Junlei Zhao, research assistant professor, are the corresponding authors. Junlei Zhao is the first author of the paper. The work was supported by the National Natural Science Foundation of China, the Basic and Applied Basic Research Foundation of Guangdong Province, and the Basic Research Project of Shenzhen City.

Paper Link：https://doi.org/10.1038/s41524-023-01117-1