Molecular dynamics simulations using machine learning potential for a-Si:H/c-Si interface: Effects of oxygen and hydrogen on interfacial defect states

Molecular dynamics simulations of a-Si:H/c-Si models with and without an oxygen layer at the interface were performed using a machine learning potential (MLP) that was efficiently trained using an on-the-fly scheme with ab initio molecular dynamics. The relaxation processes up to 1 ns at 500 and 700 K were simulated using MLP, and snapshots were evaluated using ab initio calculations to examine the in-gap states that could significantly affect the solar cell performance. The results showed that oxygen atoms passivated surface dangling bonds on c-Si, but simultaneously generated strain-induced in-gap states at the Si-O/a-Si interface. The hydrogen atoms suppressed the recrystallization of a-Si, distributed in a-Si particularly at the Si-O/a-Si interface because of the repulsive potential of the Si-O layer and contributed to the reduction of the in-gap states. Our results support experimental observation where optimization of the a-Si:H/O/c-Si interface could improve the performance of solar cells.