A Squeeze-and-Excitation and Transformer Based Model for Remaining Useful Life Prediction in Ion Mill Etching Process

In semiconductor manufacturing, ion mill etching (IME) is an emerging technique that uses ion-beam sources to remove materials to a certain depth, which requires a high degree of stability and precision. Accurate prediction of the remaining useful life (RUL), which provides a significant tool for fault diagnosis, is essential for maintenance scheduling and health management in IME process. However, RUL prediction is a challenging task, as the distribution of effective information is uneven and the degradation trend is unclear in the process. Recently, data-driven methods are widely applied in RUL prediction, due to their ability in extracting key features from the condition monitoring data without prior knowledge. Most deep learning based models treat the data collected from different sensors independently, without considering the dependencies across the sensor channels. To solve the above problems, this study proposes a squeeze-and-excitation and Transformer based model for RUL prediction. The squeeze-and-excitation module enables the re-calibration of feature weights to make more efficient use of important features and suppress useless features. The improved Transformer network is applied to have a stronger ability for temporal feature extraction. Experiments are conducted to evaluate the effectiveness of the proposed method, and the results show that the performance of our model is better than other competitive methods.