Machine Learning Strategy for Optimizing Multiple Electrical Characteristics in Dual-Layer Oxide Thin Film Transistors

A machine learning (ML) strategy is suggested to optimize dual-layer oxide thin film transistor (TFTs) performance. In this study, Bayesian optimization (BO), an algorithm recognized for its efficiency in optimizing material design, is applied to guide the design of a channel layer composed of IZO and IGZO. The sputtering fabrication process, which has attracted attention as an oxide semiconductor channel layer deposition method, is fine-tuned using ML to enhance multiple electrical characteristics of transistors: field-effect mobility, threshold voltage, and subthreshold swing. Using BO, the sputtering conditions─plasma power, pressure, and gas ratio, which intricately influence device performance─were modified using 19 data sets of 84 scenarios. It reveals that the modulated process conditions improve field-effect mobility up to 46.7 cm²V^-1s^-1, achieving more than double the performance of conventional IGZO TFTs. Furthermore, it was observed that threshold voltage is optimized to zero voltage, and the subthreshold swing is considerably improved, contributing to reduced power consumption. This study demonstrates that leveraging ML to optimize TFTs design not only accelerates the design process but also improves device performance dramatically. Overall, this ML strategy manages complex correlations among process parameters, properties, and performance and sets a precedent for the expeditious optimization of semiconductor devices.