Thursday, May 2025
09:40 AM - 10:00 AM
Room: LL21EF
Session: Artificial Intelligence for Active Matrix Devices
Prediction of Electrical Properties in LnIZO Thin-Film Transistors Based on Machine-Learning Solutions
Description:
High-mobility oxide thin-film transistors (TFTs) are promising candidates for high-end display applications. Traditional TFT process development involves performing design of experiments (DOEs) based on empirical process parameters, which can lead to significant experimental costs. In this study, we developed a predictive model for the electrical properties of high-mobility LnIZO TFTs using machine learning (ML) techniques, based on actual DOE data. The proposed model accurately forecasts the electrical performance of devices across various process conditions. Among the ML models and architectures evaluated, a stacking-based ML model demonstrates the highest predictive accuracy for electrical parameters. To improve interpretability and practical implementation, SHAP-based explainable AI (XAI) analysis was conducted to assess the influence of process parameters, and the model was encapsulated into a program. This approach significantly reduces the number of required DOEs and improves the efficiency of process development for high-mobility TFTs.