Anomaly detection of interference lithography system

Abstract

We are developing an interference exposure system. This system can expose high fine periodic patterns without shading mask. To improve the operational stability and maintainability of the system, we are studying anomaly detection and failure prediction by using operation data. This report describes an initial study of anomaly detection in the stage operation of the system. As an anomaly detection method, we employed a positive learning that detects anomalies by comparing the output of the device with the output of the model. Since both inputs and outputs are time series data, we used a seq2seq model with LSTM. The predicted values accurately simulated the behavior of a normal device. In addition, although the training data unintentionally contained anomalous data, the obtained model learned only normal behavior without learning anomalous behavior. In future works, we will consider extending the training data, and processing calculations for the difference between model predictions and actual measured data.