Failure Detection and Remaining Life Estimation for Ion Mill Etching Process Through Deep-Learning Based Multimodal Data Fusion

Abstract

Abstract Ion mill etching (IME) is an advanced process technology that uses ion-beam sources to remove materials by atomic sandblasting in order to reveal a specific pattern on the substrate. The high-precision capability of IME requires stringent process monitoring and control of the flowcool system which is critical to prevent thermal deformation of temperature-sensitive materials from the heat generated by the ion beam irradiation. This study focuses on a multi-sensor data analytics to monitor the IME process condition—enabling diagnostics and prediction of three main failure mechanisms of the IME flowcool system. A generalizable framework of engineering-based data-driven failure diagnostics and prediction are developed using random forest-based classification and a deep long short-term memory (LSTM) based method. The proposed framework and methods are demonstrated and validated in an IME process using multi-sensor data collected from multiple run-to-failure cycles. Three different failure modes related to the flowcool system are detected and identified in real time, and the time to the next failure is accurately predicted. The proposed method provides a systematic and generalizable approach for process monitoring and early prediction of failures by using heterogeneous sensor measurements and operational data under various operating conditions and settings.