Deep open-set recognition for silicon wafer production monitoring

The chips contained in any electronic device are manufactured over circular silicon wafers, which monitored by inspection machines at different production stages. Inspection detect and locate defect within the wafer return a Wafer Defect Map (WDM), i.e., list of coordinates where defects lie, can be considered huge, sparse, binary image. In normal conditions, wafers exhibit small number randomly distributed defects, while grouped specific patterns might indicate known or novel categories failures line. Needless to say, primary concern semiconductor industries is identify these intervene as soon possible restore conditions. Here we address WDM monitoring an open-set recognition problem, aim classify promptly patterns. particular, propose comprehensive pipeline for based on Submanifold Sparse Convolutional Network, deep architecture designed process sparse data arbitrary resolution, trained classes. To novelties, define outlier detector Gaussian Mixture Model fitted latent representation classifier. Our experiments real dataset WDMs show that directly processing full-resolution Convolutions yields superior classification performance classes than traditional Neural Networks, require preliminary binning reduce size images representing WDMs. Moreover, our solution outperforms state-of-the-art solutions novelty detection.