Defective Wafer Detection Using Sensed Numerical Features

One of the fundamental processes in semiconductor manufacturing is slicing, which means cutting an ingot into many wafers. During the slicing process, it is possible to produce defective wafers. Unfortunately, the inspection to identify defective wafers is time-consuming and difficult. To solve this problem, we build a system, which uses sensors to collect features (e.g., temperature, thickness, pattern on wafer surface, etc.) during the slicing process to detect if the wafers are defective in the manufacturing process. Two different models, the GRU neural network and XGBoost, are implemented in the proposed system. After fine-tuning both models, experimental results based on real dataset indicate that the GRU neural network outperforms XGBoost for wafer defective detection in both the prediction accuracy and model training time.