Efficient deep learning approach for fault detection in the semiconductor industry
Résumé
The semiconductor industry is a very cost sensitive industry and yield iskey to profitability. The ability to analyse and detect the faulty parts atseveral manufacturing steps is also very important to ensure the qualityof the delivered integrated circuits. Several factors as alignments, shiftsor masks rotations can lead to errors during the front-end step (waferfabrication), or others causes such as fingerprints, scratches and stainscan cause cosmetic damage during the back-end step (silicon packaging).Therefore, an automatic visual inspection is required to ensure that the partsare free of any defects. In this chapter, we focus specifically on classifyingwafer maps according to predefined defaults. We propose a platform whichaims at making the classification process more energy efficient, by means ofthe interconnection of two hardware parts. The first one, the microprocessorSTM32MP1, is responsible for image pre-processing and for offloadinginference to a dedicated hardware accelerator. The second one, the hardwareaccelerator, is implemented in a Xilinx Zybo Z7-20 FPGA and uses aquantized neural network model. Preliminary results show that, for this lowthroughput applications that has a limited number of classes, the solutionpresented in this article can classify in real-time with accuracy above 80%using limited resources.
