β-Ga2O3 is a wide-bandgap semiconductor of significant technological importance for electronics, but its low thermal conductivity is an impeding factor for its applications. In this work, an interatomic potential is developed for β-Ga2O3 based on a deep neural network model to predict the thermal conductivity and phonon transport properties. Our potential is trained by the ab initio energy surface and atomic forces, which reproduces phonon dispersion in good agreement with first-principles calculations. We are able to use molecular dynamics (MD) simulations to predict the anisotropic thermal conductivity of β-Ga2O3 with this potential, and the calculated thermal conductivity values agree well with experimental results from 200 to 500 K. Green–Kubo modal analysis is performed to quantify the contributions of different phonon modes to the thermal transport, showing that optical phonon modes play a critical role in the thermal transport. This work provides a high-fidelity machine learning-based potential for MD simulation of β-Ga2O3 and serves as a good example of exploring thermal transport physics of complex semiconductor materials.

中文翻译：

---

用于研究 β-Ga2O3 热导率的深度神经网络原子间势

β-Ga2O3 是一种对电子技术具有重要技术意义的宽带隙半导体，但其低热导率是其应用的阻碍因素。在这项工作中，基于深度神经网络模型开发了 β-Ga2O3 的原子间势，以预测热导率和声子传输特性。我们的潜力是由 ab initio 能量表面和原子力训练的，它再现了与第一性原理计算非常一致的声子色散。我们能够使用分子动力学 (MD) 模拟来预测具有这种潜力的 β-Ga2O3 的各向异性热导率，并且计算出的热导率值与 200 至 500 K 的实验结果非常吻合。进行 Green-Kubo 模态分析以量化不同声子模式对热传输的贡献，表明光学声子模式在热传输中起着关键作用。这项工作为 β-Ga2O3 的 MD 模拟提供了基于机器学习的高保真潜力，并作为探索复杂半导体材料热传输物理的一个很好的例子。