Automatic defect classification, usually shortened to ADC, is the use of software to sort defect candidates from inspection or review tools into meaningful categories. In semiconductor manufacturing, those categories might include particles, scratches, bridges, voids, residues, pattern nuisance, or "unknown" cases that need expert review.
Why It Matters
ADC matters because fabs generate far more defect candidates than engineers can review manually. If every optical or e-beam signal has to be checked by a person, review queues grow, nuisance sites consume attention, and real yield excursions can take longer to confirm. ADC helps teams reserve human effort for the cases where judgment is most valuable.
Why It Matters In AI
AI makes ADC more useful because modern models can learn from inspection imagery, wafer-map context, process metadata, and feedback from prior reviews. In practice, ADC often overlaps with computer vision, anomaly detection, active learning, virtual metrology, and advanced process control. The strongest systems do not just assign labels. They also preserve uncertainty, route unknowns, and support faster process response.
What To Keep In Mind
Strong ADC is not the same as fully autonomous inspection. The model still needs class updates, drift checks, and human review for new or ambiguous defect types. Good systems therefore treat "unknown" as a valid outcome and keep enough feedback in the loop to expand the defect library safely over time.
Related Yenra articles: Semiconductor Defect Detection, Microtechnology and Nanotechnology Design, Micro-Fabrication Process Control, Digital Twin Modeling in Manufacturing, and Industrial Robotics.
Related concepts: Computer Vision, Anomaly Detection, Active Learning, Virtual Metrology, Advanced Process Control (APC), Nanofabrication, and Model Monitoring.