Data and Code

In semiconductor manufacturing, accurate and quick anomaly detection is critical, yet numerous sensors with tight specification limits often generate high false alarms. This case study introduces the Sync Ratio, a new method for grouping sensor data based on the synchronization of rare data points in time. Unlike traditional methods such as pearson correlation, our approach captures local sensor behavior and genuine physical relationships without prior domain knowledge. Applied to data from a dry etcher, this method demonstrated superior performance in distinguishing true from false alarm and outperforming benchmark algorithms. Despite limitations in handling temporal delays and highly controlled sensors, our findings indicate that this method is an intuitive and robust tool to enhance process quality in semiconductor manufacturing.