Performance Analysis of Rare-earth Doped Oxide Thin-Film Transistors Using Neural Network Method

Abstract The work analyzes the key impact factors on the performances of rare-earth element doped oxide thin film transistors (TFTs), which are potentially used for high performance displays, by comparatively using a Bayesian Neural Network (BNN) method and Artificial Neural Network (ANN) method based on published and self-experimental data which was exhaustively collected. Both BNN and ANN methods can effectively identify the primary impact factors among rare-earth element type, doping concentration, thin film thickness, channel length and width, which are key factors to determine the TFTs performances. Comparisons between the ANN and BNN methods, the BNN approach offers more reliable and robust predictions on the dataset. Accordingly, the efficient neural network models tailored to the data features were accurately established. A key outcome from the BNN models is the relative importance ranking of the influence factors and relationship between the carrier mobility and element type, concentration as well. To the TFT mobility, rare-earth element concentration is the most critical factor, suggesting lower concentration exhibited higher mobility, followed by the rare-earth element type. To the sub-threshold swing performance of TFTs, the rare-earth element type is the most significant influence factor, suggesting higher valence rare-earth is superior to lower valence one, followed by the element concentration. The results are basically consistent with experimental tendency. These insights could effectively guide the design of oxide semiconductor materials and TFT device structure, to achieve high-performance (high mobility and high stability) oxide TFT devices for displays.