Abstract
Purpose－CMOS image sensor (CIS) manufacturers usually rely on domain knowledge for troubleshooting, yet it may not be effective nor efficient due to lack of experience and increasing complexity. Focusing on realistic needs, this study aims to develop a framework based on data mining and big data analytics for analyzing defect patterns in color filter and microlens to enhance yield.Design/methodology/approach － The proposed framework integrates data collection and preprocessing of defect patterns, univariate correlation analysis (by Chi-square independence test and Cramer’s V correlation coefficient), and association rules (generated by Apriori algorithm via splitting dataset into training and testing repeatedly). The derived rules were evaluated by the indices of support, confidence, and lift. Findings－An empirical study was conducted in a leading CIS manufacturing company in Taiwan to validate the proposed approach. The results have shown thepractical viability of the proposed approach to effectively and efficiently resolve the present problem. Research limitations/implications－This study focused only on three data types(including process tools, recipes, and process time) to generate rules to support troubleshooting. Future research could incorporate inline data such as metrology and equipment parameters to construct the model for further investigation. Practical implications－The color filter and microlens processes are critical for CMOS image sensor manufacturing. It is crucial to identify possible root causes ofdefect patterns to troubleshoot and enhance product yield and reduce the loss of reworks in short time. The proposed framework provides a systematic approach to identify theroot causes of specific defect patterns effectively and efficiently. Originality/value－This study firstly employ data mining and big data analyticsfor troubleshooting and yield enhancement of CIS manufacturing and developed an effective solution.