Machine-Learned Fermi Level Prediction of Solution-Processed Ultrawide-Bandgap Amorphous Gallium Oxide (a-Ga₂O_x)

The Fermi level (EF) relative position to the conduction band minimum is a crucial consideration for controlling electrical conductivity and semiconductor device performance in thin-film transistors, sensors, and photodetectors. Experiment complexity and expensive material resources for predicting EF via an experimental approach render a machine learning (ML) approach to be more appropriate. This work presents ML-assisted EF prediction of solution-processed ultrawide-bandgap (UWB) amorphous gallium oxide (a-Ga2Ox). Three regression models─kernel ridge regression, support vector regression, and random forest regression─were trained with experimental features including the film thickness, baking temperature, and gas environment during solution deposition of the a-Ga2Ox film. The results show that ML models can be used to predict EF of the UWB a-Ga2Ox film and also identify optimized fabrication parameters to achieve the optimized EF. Moreover, the ML approach can significantly accelerate the fabrication of semiconducting UWB a-Ga2Ox-based material for future device applications. This work is a big step toward rapid and cost-effective optimization methods for developing UWB a-Ga2Ox-based devices.