Semiconductor Research Corporation
Circuit Anomaly Detection System
Built outlier detection algorithms on 250+ datasets for analog and mixed-signal circuits, improving anomaly detection accuracy by 12%.
March 1, 2024
mlpythonbackendresearch
01
Context
Summer internship at Semiconductor Research Corporation in Chennai, funded by Texas Instruments. The project focused on improving quality assurance for analog and mixed-signal circuits through automated anomaly detection.
02
What I Built
Implemented multiple outlier detection algorithms including One-class SVM, K-means, and DBSCAN on 250+ datasets, each containing 100-200 features. Used clustering and predictive modeling techniques to assess performance for high-sensitivity circuits.
03
Key Decisions
1Compared multiple algorithms (One-class SVM, K-means, DBSCAN) for best fit
2Designed feature engineering pipeline for high-dimensional circuit data
3Implemented clustering for pattern recognition in circuit behavior
4Built predictive models for performance assessment
04
Challenges
→Handling high-dimensional feature spaces (100-200 features per dataset)
→Processing 250+ diverse datasets with varying characteristics
→Balancing detection accuracy with false positive rates
05
Outcomes
✓12% improvement in anomaly detection accuracy
✓Accuracy score increased from 80% to 87%
✓Results applied to Texas Instruments circuit quality assurance
06
Tech Stack
PythonScikit-learnPandasNumPyMatplotlib