Abstract: The rapid growth of the semiconductor industry continues to push for advances in manufacturing quality assurance capability. Classical human-based visual inspection is becoming increasingly impractical because of high inspection cost and rapid manufacturing throughput, and thus machine vision systems for surface defect detection are becoming common solutions. To facilitate the design and evaluation of such systems, wide-ranging research into computer vision solutions for surface defect detection in semiconductor manufacture is presented. This comprises an overview of semiconductor fabrication processes and defect types, consideration of essential concepts in computer vision, examination of common detection approaches, and the creation of two openly available image datasets. These datasets, designed specifically for semiconductor contexts, are carefully annotated and then employed to assess classical image processing techniques and machine learning detectors.

The general nature of the discussion makes it relevant across numerous manufacturing domains, with emphasis on fundamental computer vision concepts of relevance to any defect detection task. Special focus is given to the preparation of image datasets, and particular attention is placed on design considerations essential for real-world deployment in high-throughput manufacturing environments.

Keywords: Semiconductor fabrication; defect detection; computer vision; image processing; machine learning; dataset; Automated Optical Inspection (AOI); Wafer Defect Detection; Deep Learning-Based Visual Inspection; Semiconductor Manufacturing Yield Analysis; Convolutional Neural Networks (CNNs) for Inspection; Inline Process Monitoring; Submicron Defect Classification; Image-Based Fault Diagnosis; Smart Manufacturing (Industry 4.0); Real-Time Quality Control Systems.

Downloads: | DOI: 10.17148/IJIREEICE.2022.101222