Vision-based defect detection and quality control for semiconductor industry

Effectively detecting defects in production components is a crucial element in many industrial segments. Considering the semiconductor industry’s high-quality standards, the capability of automating this process is a necessity to keep up with the worldwide demand of semiconductor products. This thesis, developed during a six-month internship in the R&D department of Applied Materials Italia, focuses on i) the development of a vision-based defect detection and quality control system for components of equipment for manufacturing of semiconductors and ii) the creation of Deep Learning models for defect detection in flat-optics elements. We describe the hardware choices made to build the machine vision system, and the models and algorithms used for defect detection and quality control. Finally, we introduce an algorithm for the extraction of pre-aligned contours with sub-pixel accuracy, used to perform measurements on the edges of through-holes and flat-optics elements.