Defect detection on optoelectronical devices to assist decision making: A real industry 4.0 case study

Abstract

This paper presents an innovative approach, based on industry 4.0 concepts, for monitoring the life cycle of optoelectronical devices, by adopting image processing and deep learning techniques regarding defect detection. The proposed system comprises defect detection and categorization during the front-end part of the optoelectronic device production process, providing a two-stage approach; the first is the actual defect identification on individual components at the wafer level, while the second is the pre-classification of these components based on the recognized defects. The system provides two image-based defect detection pipelines. One using low resolution grating images of the wafer, and the other using high resolution surface scan images acquired with a microscope. To automate the entire process, a communication middleware called Higher Level Communication Middleware (HLCM) is used for orchestrating the information between the processing steps. At the last step of the process, a Decision Support System (DSS) collects all information, processes it and labels it with additional defect type categories, in order to provide recommendations to the optoelectronical engineer. The proposed solution has been implemented on a real industrial use-case in laser manufacturing. Analysis shows that chips validated through the proposed process have a probability to lase at a specific frequency six times higher than the fully rejected ones.