Abstract
For CMOS image sensor (CIS) manufacturing, defect reduction is a key taskforce for quality assurance and yield enhancement. Indeed, automatic optical inspection (AOI) is the critical equipment for defect inspection. Although AOI can capture possible defect images with high throughput and low manual labor, it cannot identify defect types for troubleshooting purpose. In particular, the advanced AOI equipment can provide a high resolution defect image of a whole wafer for overall judgments. This study aims to develop a hybrid data mining approach for defect classification for whole wafer images based on the result of classifier. The proposed approach consists of two stages similarity matching to rearrange the order of features from different images of CMOS. This concept could not only reduce the false alarm rate but enhance the correct rate. An empirical study was conducted with a leading CIS manufacturing company in Taiwan to estimate the validity and the results also demonstrated the practical value of the proposed approach.