Novel Feature Extraction Enhances YOLO on FOCoS RDL Recognition

The precisions of redistribution layers (RDLs) process and equipment have become an important and challengeable issue due to the design requirements of miniaturization of components and high density placement on fan out chip on substrate (FOCoS) of integrated circuit (IC) package. The inspection equipment for the automated optical inspection (AOI) in IC package line also have become essential tools to recognition the defects such as open, short, contamination...etc. The high false detection rate of AOI will affect the production yield rate and overall equipment effectiveness. In order to solve this problem, the novel edge feature extraction enhances you only look once (YOLO) algorithm based on machine learning approach in this research. After processing over ten thousand huge database of heterogeneous integration IC package images, the new feature extraction method has shown a high sensitivity image detection capability. The results help artificial intelligence (AI) for making the management of image judgment more accurate.