AI Driven Defect Map Recognition

This paper describes an AI-driven automated system that leverages Convolutional Neural networks (CNNs) for real-time wafer map defect recognition. The proposed framework integrates data preparation, wafer map image generation, data augmentation, and a CNN-based classification model to identify systematic defect patterns such as line, streak and ring. An interactive user interface (UI) enables engineers to search for similar defect signatures across historical production data, adjust similarity thresholds, and compute hit rates for anomaly detection. InceptionResNetV2 model that achieved 94.5% accuracy on testing data is utilized as the main framework for this study. The system’s scalability in linking defects to assignable causes such as polishing pad hardening or uneven cleaning, helps to enhance process monitoring and yield management. This work bridges the gap between machine learning and practical industrial application by providing a solution for semiconductor manufacturing challenges.