An Ensembled Anomaly Detector for Wafer Fault Detection

Area of Science:

• Semiconductor Manufacturing
• Data Science
• Anomaly Detection

Background:

• Semiconductor wafer production involves multiple critical phases, including fault detection.
• Yield loss in manufacturing is often caused by undetected anomalies, necessitating robust root-cause analysis.
• Traditional fault detection relies on tracking parameters like pressure and voltage, but faults often arise from complex interactions within control limits.

Purpose of the Study:

• To develop an advanced anomaly detection system for semiconductor manufacturing.
• To address the challenge of identifying subtle faults caused by combinations of parameters.
• To improve the accuracy and efficiency of fault detection to minimize yield loss.

Main Methods:

• Proposed an ensembled anomaly detector combining univariate and multivariate analyses.
• Investigated three distinct ensemble balancing strategies.
• Conducted experiments on two real-world semiconductor industry datasets.

Main Results:

• The ensembled anomaly detector demonstrated promising performance in detecting anomalies.
• Achieved high recall rates, indicating effective identification of true anomalies.
• Significantly reduced the number of false alarms compared to traditional methods.

Conclusions:

• The proposed ensembled anomaly detection approach is effective for semiconductor manufacturing.
• This method offers a significant improvement in identifying yield detractors.
• The findings suggest a valuable tool for enhancing quality control and reducing production losses.