[25p-E202-1~15] 21.1 Joint Session K "Wide bandgap oxide semiconductor materials and devices"

▼ [25p-E202-6] Fermi Level Prediction of Solution-processed Ultra-wide Bandgap Amorphous Gallium Oxide via Supervised Machine Learning Models

〇(D)Diki Purnawati¹, Paul Rossener Regonia¹, Juan Paolo Bermundo¹, Kazushi Ikeda¹, Yukiharu Uraoka¹ (1.Nara Institute of Science and Technology)

Keywords:a-Ga2Ox, Ultra-wide bandgap, Machine learning

Ultra-wide bandgap amorphous gallium oxide (a-Ga₂O_x) has been attracting great attention as an excellent candidate for Internet of Things (IoT) applications. In fabricating a-Ga₂O_x-based device, the amount of carrier concentration is critical since it tends to shift the Fermi level (E_F) closer to the conduction band minimum (CBM) to improve a-Ga₂O_x conductivity and exhibit better device performance. However, experimental optimization of E_F is quite complex and time consuming. Therefore, machine learning (ML) approach was proposed to accelerate the optimization. Random Forest Regression (RFR) exhibits the best model performance with RMSE of 0.15, MAE of 0.12 and R² of 75% which is ~10× > than previous reported model. This result implied that RFR has 75% model accuracy on predicting E_F±0.12 eV. The lowest E_F of 1.38 eV was successfully obtained with the best parameters of 90 nm-thick film deposited at 425°C under 4% H₂ gas-flowed which is near to our experimental result. This result shows that ML models can be used to predict E_F of UWB a-Ga₂O_x film given an input of fabrication parameters and also identify optimized fabrication parameters to achieve the optimized E_F.