Role of hydrogen-doping for compensating oxygen-defect in non-stoichiometric amorphous In2O3−x: Modeling with a machine-learning potential

Transparent amorphous oxide semiconductors (TAOSs) are essential materials and ushering in information and communications technologies. The performance of TAOS depends on the microstructures relating to the defects and dopants. Density functional theory (DFT) is a powerful tool to understand the structure–property relationship relating to electronic state; however, the computation of DFT is expensive, which often hinders appropriate structural modeling of amorphous materials. This study, thus, applied machine-learning potential (MLP) to reproduce the DFT level of accuracy with enhanced efficiency, to model amorphous In2O3 (a-In2O3), instead of expensive molecular dynamics (MD) simulations with DFT. MLP-MD could reproduce a-In2O3 structure closer to the experimental data in comparison with DFT-MD and classical MD simulations with an analytical force field. Using the relatively large models obtained by the MLP-MD simulations, it was unraveled that the anionic hydrogen atoms bonding to indium atoms attract electrons instead of the missing oxygen and remedy the optical transparency of the oxygen deficient a-In2O3. The preferential formation of metal–H bonding through the reaction of oxygen vacancy was demonstrated as analogous to InGaZnOx thin films [Joonho et al., Appl. Phys. Lett. 110, 232105 (2017)]. The present simulation suggests that the same mechanism works in a-In2O3, and our finding on the structure–property relationship is informative to clarify the factors affecting the optical transparency of In-based TAOS thin films.