Abstract
CMOS image sensor includes color filter and microlens process, which is used to manufacture cameras and phone lens. In color filter and image sensor manufacturing company’s manufacturing process, it may cause various defect types and defect rate in sensing or non-sensing area. To improve product’s yield and find causes of defect type, we should repair the tools in time and reduce the rework rate. Now it almost uses engineers’ experience for trouble shooting. Try and error method is not quick enough and may cause errors because of less of experience. This research is aim for constructing a data mining framework of color filter and microlens to help engineers detecting causes of defect types. By using defect types’ data in fab, we could combine Chi-square test for independence, Cramer’s V correlation coefficient and divide training data set of Association Rules to build model. Using the correct rate of testing data set to select suitable model and setting threshold of three indexes：support, confidence and lift to screen useful rules before executing evaluation.