Die Chipping Measurement Using Computer Vision for Chip-to-Wafer Hybrid Bonding Application

In the semiconductor industry, specifically in the context of chip-to-wafer (C2W) hybrid bonding, dicing is a critical process of cutting a semiconductor wafer into individual chips for bonding. However, the current methods used in this dicing process often do not prioritize achieving smooth edges on the chips leaving behind rugged edges. These imperfections can negatively impact the overall quality and reliability of the bonded structures, resulting in defects like void and delamination in the subsequent steps. In recent years, the industry has increasingly embraced vision-based measuring approaches due to their ability to provide high measurement efficiency and accuracy in estimating the deviation of trimmed edge from the smooth edge. A novel vision-based algorithm is proposed to precisely measure this deviation. Advanced computer vision and machine learning analysis are applied to detect and quantify the chipping deviation. Furthermore, the proposed approach effectively measures the deviation with minimum mean square error. By leveraging the advantages of vision-based measuring, it becomes possible to undertake the post process measures to improve the quality of the resulting chips.