A Neuroevolution Potential for Gallium Oxide: Accurate and Efficient Modeling of Polymorphism and Swift Heavy-Ion Irradiation

TL;DR

This work addresses the challenge of accurately modeling the polymorphic and nonequilibrium behavior of Ga2O3 by developing a neuroevolution potential with an energy-dependent training weighting. The NEP-based MLIP, implemented in GPUMD, outperforms the state-of-the-art tabGAP in energy, force, and virial predictions while delivering high computational throughput. Augmenting the training data with gamma phase and beta heating–cooling pathways improves extrapolation and yields gamma phase accuracy near 3 meV/atom, enabling reliable irradiation simulations. Applying the augmented NEP to swift heavy-ion irradiation of beta-Ga2O3, the authors reproduce core–shell ion tracks and track diameters consistent with experiments, demonstrating the method’s potential for large-scale, device-relevant atomistic modeling.

Abstract

Gallium oxide (Ga2O3) is a wide-bandgap semiconductor with promising applications in high-power and high-frequency electronics. However, its complex polymorphic nature poses substantial challenges for fundamental studies, particularly in understanding phase-transformation behaviors under nonequilibrium conditions. Here, we develop a robust, accurate, and computationally efficient machine-learning interatomic potential (MLIP) for Ga2O3 based on the neuroevolution potential (NEP) framework combined with an energy-dependent weighting strategy. The resulting NEP potential demonstrates clear advantages over the state-of-the-art tabGAP potential with respect to both accuracy and computational efficiency. Furthermore, we introduce a physically process-oriented sampling strategy to systematically augment the training dataset, thereby enhancing the MLIP performance for targeted physical phenomena. As a representative application, a dedicated NEP potential is constructed for swift heavy-ion (SHI) irradiation simulations of \b{eta}-Ga2O3. The simulated results are in quantitative agreement with experimental observations and provide a consistent physical explanation for the reported experimental discrepancies regarding phase transformations in the ion track of \b{eta}-Ga2O3.

Figures (9)

• Figure 1: Detailed comparison of the prediction accuracy of the NEP and tabGAP models. Blue and red dots represent NEP and tabGAP results, respectively. Panels (a), (c), and (e) show the predicted energies, forces, and virials of the GAP-dataset configurations with average energies ranging from $-5.9$ to $-0.9$ eV/atom. Panels (b), (d), and (f) present the corresponding results for a narrower energy window from $-5.9$ to $-5.4$ eV/atom.
• Figure 2: Comparison of the computational throughput between NEP and tabGAP for systems of different sizes. The throughput of NEP is measured on a single NVIDIA H100 GPU, while that of tabGAP is measured on eight CPU nodes, each equipped with two 24-core Intel Xeon Gold 6240R processors.
• Figure 3: Comparison of the MAE of (a) energy, (b) force, and (c) virial predictions by NEP and tabGAP for different config types in the GAP dataset.
• Figure 4: (a) Illustration of the energy-dependent weighting strategy, where the black curve represents the rescaling factor as a function of the average potential energy. All samples in the GAP dataset are shown as dots colored according to their configuration types; their positions along the horizontal axis indicate their average potential energies. (b) Comparison of the energy-prediction performance of tabGAP, NEP with energy-dependent weighting, and NEP with configuration-type weighting over different energy ranges.
• Figure 5: Energy-volume curves for the $\alpha$, $\beta$, $\delta$, $\gamma$, and $\kappa$ phases predicted by NEP (solid circles) and tabGAP (solid squares), compared with DFT reference data (open circles). Different phases are color-coded as indicated.