ABSTRACT
Surface defects generated during semiconductor wafers processing are among the main challenges in micro- and nanofabrication. The wafers are typically scanned using optical microscopy and then the images are inspected by human experts. That tends to be a quite slow and tiring process. The development of a reliable machine vision-based system for correct identification and classification of wafer defect types for replacement of manual inspection is a challenging task, due to the variety of possible defects. In this work we developed a machine vision system for the inspection of semiconductor wafers and detection of surface defects. The system integrates an optical scanning microscopy system and an AI algorithm based on the Mask RCNN architecture. The system was trained using a dataset of microscopic images of wafers with Micro Electro-Mechanical Systems (MEMS), silicon photonics and superconductor devices at different fabrication stages including surface defects. The achieved accuracy and detection speed makes the system promising for cleanroom applications.
