AUTOMATED DEFECT INSPECTION ALGORITHM FOR SEMICONDUCTOR-PACKAGED CHIPS

Detecting product quality defects through image recognition technology is one of the key technologies of intelligent manufacturing and an important step for enterprises to construct a smart factory. The internal wire bonding of a chip easily receives interference and produces defects in the capsulation step of the semiconductor enterprise. Companies need to pick out defective chips to prevent them from entering the market. A traditional method is to use human visual inspection, which may lead to low efficiency and high labor cost. Hence, this study intends to use a machine vision detection method based on image processing technology. The objectives are to identify the defect of a chip and replace the workers with human manual detection work. This study proposes two algorithms to solve such problems. The template matching algorithm (TMA) determines whether the chip is defective based on the standard template. Meanwhile, the neighborhood comparison algorithm (NCA), which is implemented by Halcon software, calculates the similarity of the neighbor chips to judge the target chip's defects. A German semiconductor company has provided enough samples to support our research. Experimental results show that the two algorithms are effective in the defect detection of specific products. The advantage of the TMA lies in its processing speed, but the applicability and accuracy of NCA are excellent. The algorithm proposed in this study can be used for enterprises through integration into the actual detection process.