Neural Network-Driven Molecular Insights into Alkaline Wet Etching of GaN: Toward Atomistic Precision in Nanostructure Fabrication

Abstract

We present large-scale molecular dynamics (MD) simulations based on a machine-learning interatomic potential to investigate the wet etching behavior of various GaN facets in alkaline solution-a process critical to the fabrication of nitride-based semiconductor devices. A Behler-Parrinello-type neural network potential (NNP) was developed by training on extensive DFT datasets and iteratively refined to capture chemical reactions between GaN and KOH. To simulate the wet etching of GaN, we perform NNP-MD simulations using the temperature-accelerated dynamics approach, which accurately reproduces the experimentally observed structural modification of a GaN nanorod during alkaline etching. The etching simulations reveal surface-specific morphological evolutions: pyramidal etch pits emerge on the $-c$ plane, while truncated pyramidal pits form on the $+c$ surface. The non-polar m and a surfaces exhibit lateral etch progression, maintaining planar morphologies. Analysis of MD trajectories identifies key surface reactions governing the etching mechanisms. To gain deeper insights into the etching kinetics, we conduct enhanced-sampling MD simulations and construct free-energy profiles for Ga dissolution, a process that critically influences the overall etching rate. The $-c$, $a$, and $m$ planes exhibit moderate activation barriers, indicating the feasibility of alkaline wet etching. In contrast, the $+c$ surface displays a significantly higher barrier, illustrating its strong resistance to alkaline etching. Additionally, we show that Ga-O-Ga bridges can form on etched surfaces, potentially serving as carrier traps. By providing a detailed atomistic understanding of GaN wet etching, this work offers valuable guidance for surface engineering in GaN-based device fabrication.