WaferSight

WaferSight by Team MiniTehnicus is an end-to-end semiconductor yield intelligence platform designed to catch wafer defects and process drift before they result in costly scrapped silicon. Built and accelerated using the AMD AI Notebook, WaferSight classifies spatial defect patterns—such as Center, Edge-Ring, Edge-Loc, and Scratch—with high precision across single wafers and 25-wafer production lots. Rather than acting as a black-box classifier, WaferSight provides full explainability through Grad-CAM spatial heatmaps and cosine similarity search over CNN embeddings to retrieve the top five matching historical wafers. When recurring defects trigger a statistical drift alarm (at least three of the last five wafers), an integrated AI Agent cross-references fabrication troubleshooting manuals to deliver immediate root-cause diagnostics. Finally, WaferSight translates AI predictions into real-time economic decisions—comparing expected scrap costs against line downtime to automatically recommend whether engineers should Hold or Continue production.