Automated diagnosis of anomalies via sensor-step data outlier detection: An application in semiconductors

No Thumbnail Available
Authors
Al Iqbal, MD Ridwan
Vargas, Andrés
Erickson, John S.
Bennett, Kristin P.
Issue Date
2018-08-20
Type
Article
Language
Keywords
Research Projects
Organizational Units
Journal Issue
Alternative Title
Abstract
The diagnosis of potential mechanical defects and sources of variation impacting wafer quality is an essential task in semiconductor manufacturing. During manufacturing, semiconductor fabrication tools are monitored by sensors that capture critical environmental information including temperature, power, and pressure. Our goal is to create an online and unsupervised anomaly detection system able to detect potential mechanical faults and other significant anomalies by monitoring sensor data from semiconductor process tools and then automatically generating a comprehensive report to help define putative causes of these anomalies for process engineers. Our framework represents the raw multivariate time series data as a tensor of sensor-step pairs. Exploiting information at recipe step granularity enables the framework to be easily deployed to new tools and recipes. The system finds instances that significantly differ from a normal model of the sensor-step pairs. A moving window Principal Component Analysis (PCA) model is used to capture the normal model, and departures from this model indicate potential anomalies. The system is modular, so different anomaly detection algorithms can be used. The dynamically-generated reports produced by the system contain lists of influential sensor-step pairs and also include visualizations and statistical tests explaining and validating the detected anomalies. The generated report includes the processed data and sample analysis code process engineers need to expedite further exploration of the anomalies. The anomaly detection and diagnosis system has been demonstrated by experimenting on wafer trace data generated by three different process nodes or types, with data from 13 different chambers. Our experimental results show that the detected sensor-step anomalies are indeed of engineering significance.
Description
Full Citation
Automated diagnosis of anomalies via sensor-step data outlier detection: An application in semiconductors MR Al Iqbal, A Vargas, JS Erickson, KP Bennett - SIGKDD ODD v5. 0 Workshop, 2018
Publisher
ACM
Journal
Volume
Issue
PubMed ID
DOI
ISSN
EISSN