As semiconductor manufacturing processes become more and more complex along with increase in global semiconductor product demand, semiconductor manufacturing equipments get more vulnerable to unexpected failures. As a result, predictive maintenance (PdM) rises as a key technology, enabling timely maintenance of equipments, which otherwise might fail.
Approach
In our approach, AI algorithm learns the normal pattern of sensor data from semiconductor manufacturing equipments and detects possible anomaly patterns during manufacturing process. It also estimates the remaining lifetime of equipment (aka, the time-to-failure (TTF)) and identifies the cause of detected anomalies.
Results
With true positive rate over 90 %, it is possible to predict anomalies 12 to 24 hours prior to their occurrence. Production losses due to downtime can be minimized by facilitating preventive measures against equipment failures.
Related Use Cases
Automobile
Anomaly detection
Robot Arm Anomaly Detection
Using an unsupervised learning algorithm, it detects those anomalies before they occur in the equipment.