Enhancing the effectiveness of output projection in wafer fabrication using an Industry 4.0 and XAI approach

Industry 4.0 and deep learning methods have been widely used for output projection in wafer fabrication. Although such applications appeared to be effective, they were difficult to understand and/or improve. Therefore, in this study, an Industry 4.0 and explainable artificial intelligence (XAI) approach is proposed to enhance the interpretability of a deep learning application for output projection. First, a deep neural network (DNN) is constructed to estimate the output times of jobs based on real-time job, machine, and factory information collected using Industry 4.0 technologies. Then, the DNN is approximated by a random forest (RF) to generate interpretable decision rules. SHAP analysis is also performed to generate local explanations. The proposed methodology has been applied to a real case from a wafer fab. Through the synergy of Industry 4.0 and deep learning, the proposed methodology achieved a satisfactory output projection performance. The MAE was only about two wafer lots. Additionally, user communication was also facilitated since the output projection mechanism using deep learning has been explained using XAI techniques.