應用資料探勘技術控制PI製程之薄膜捲曲

Abstract

對於PI(polyimide)薄膜生產公司而言，捲曲(Curl)是一個衡量品質的重要指標。台灣某個案公司在PI薄膜品質上一直存在著捲曲品質不夠好的問題，不過因為製造程序的複雜，加上龐大的製程資料，很難在短時間內找出良率改善規則，所以希望本研究發展一個分析架構可以快速找出影響捲曲品質的關鍵製程及改善方案。近年來有許多學者提出，應用資料探勘的技術在製程良率改善的問題上面，都獲得不錯的效果。本研究使用個案公司所提供的製程資料，利用資料探勘技術中分類規則的1R及C4.5為主軸建立此分析架構，並藉由實際的行動方案檢驗本分析架構的結果。從結果顯示此分析架構可以快速找出關鍵製程參數並實際提出改善規則，該PI薄膜生產公司之製程良率也大幅從43.36提升到85%，較以往增加約40%。

Curl is an unfavorable quality attribute in polyimide film (PI) production. In practice, it is hard to reduce the curl issues for PI films because the production process includes a few hundreds of process parameters. This research combined two data mining algorithms, 1R and C4.5, for identifying the critical process parameters for reducing the curl issues. Three critical process parameters as well as their appropriate ranges of values were identified by the proposed approach. An empirical experiment reveals that the process yield for reducing the PI curl issues could be improved from 43.36% to 85%.