Machine learning-assisted performance prediction of ZnMSnO-based thin-film transistors

Amorphous oxide semiconductors (AOS) are highly promising for optoelectronic devices due to their exceptional electrical properties and optical transparency. However, a significant barrier to their development is the lack of a comprehensive materials database, which hinders systematic studies and the discovery of emergent AOS materials. This study addresses this gap by constructing a dedicated database of zinc-tin-based doped oxide semiconductors (Zn–M–Sn–O) and their thin-film transistor (TFT) performance parameters, compiled from an extensive review of existing literature. Our research aims to perform a systematic analysis of the correlations between key material properties and device performance. We summarize and analyze existing Zn–M–Sn–O based optoelectronic devices, extracting key features such as material compositions, processing parameters, and performance metrics. These features are then used to construct feature vectors. By applying a machine learning algorithm to this dataset, we establish a performance prediction model for Zn–M–Sn–O TFTs. This machine learning-assisted approach allows us to efficiently screen materials and predict the optimal M element for high-performance devices. This methodology significantly accelerates the discovery and development of advanced AOS materials, paving the way for next-generation optoelectronic technologies.