Validation of kill-rate and yield model for clustering visual defects in semiconductor manufacturing

This paper validates the kill rate for different types of visual defects and resulting yield by using large amount of real inspection data in semiconductor manufacturing. Visual defects are assumed to be distributed in cluster. The results of this estimation model is slightly different from real inspection data. Therefore, this model is feasible to estimate the kill rate and the yield of mature product. However, since the corresponding manufacturing process for a new product may often result in multiple visual defects on a die. That can be verified by incorporating real inspection data of new product and this estimation model as well.