AUTOMATIC VISUAL INSPECTION OF IC PACKAGES

Abstract

With the advances in IC packaging techniques, there is a strong need for computer vision based inspection systems to be able to efficiently detect various defects on IC packages. This thesis focuses on some unresolved issues on the automatic inspection of IC packages and presents computer vision based techniques that were developed to perform automatic IC orientation checks, and to inspect for certain defects on IC packages that are becoming prevalent. The proposed algorithm for automatic IC orientation checks involves several procedures including IC boundary detection using histogram analysis and edge detection, IC package alignment using line-fitting methods, image segmentation using adaptive double thresholding, and notch/dimple identification using region-based analysis. The algorithm for the detection of the die extrusion defect on IC packages, which is gaining attention as the packages become thinner, is proposed using an optimal linear filter for wide linear features. The automatic surface inspection of IC packages including mold defect inspection and marking checks is also discussed. A region-growing algorithm based on pixel clustering and adaptive neighbourhood processing is proposed to segment the regions of low contrast and/or blurred borders. The segmented regions are then isolated for marking checks and mold defect inspection using the mathematical morphology operations. Our experiments have shown that the proposed algorithms work successfully with the IC samples, and features of interests are rapidly identified at appropriate resolutions.