整合資料探勘技術與即時回饋控制演算法於半導體製程改善之研究

Abstract

隨著半導體製程從微米進入深次微米時代，在晶圓品質要求上面臨更多技術問題挑戰，為了因應製程不斷改進，減少晶圓報廢及重工(Rework)，藉由電腦整合製造系統來控制製程參數因應而生。由收集得來的製程資料，與SPC（Statistic Process Control）結合,運用資料探勘（Data Mining）來分析歸類，以即時迴授方法，使每批產品在最佳製程參數下進行生產，以求達成產能及良率提升，進而降低生產成本，創造企業利潤。 本論文針對半導體晶圓廠為例，將Data Mining架構於CIM與MES之間，以Association rule 和 K-means Clustering演算法，再加上即時回饋分析法，將每個可能會影響到生產之環境變數參數，萃取及歸類分析，動態地消除及補償不同機台與產品間之差異，使各種機台在生產不同產品上各具有彈性。 研究結果顯示可改善單一產品製程良率，而Rework次數亦可降低，運用自動化配合最佳參數生產，使整個製程能力更為提升，使企業建立更穩定及更彈性的競爭優勢。 提出的系統架構可應用於傳統製造業在製程上的分析，對提升產品良率的同時，亦可降低生產成本。利用標準流程控制，結合了SPC與過去製程資料，以即時回饋方式(Feedback Control)來預測下一批最佳製程的參數，亦可增進產品競爭優勢。

Along with the movement of semiconductor process from micron into deep sub-micron era, there are more technical challenges on the wafer quality requirement. In order to deal with continuous process improvement and reduce wafer rejection and rework, the use of computer integrated manufacturing system to control process parameters thus arises. The collected process data is combined with SPC（Statistic Process Control）, Data Mining technology is used to analyze and induce, in-time feedback method is used to let each lot of product manufactured under optimum process parameters so that production capability and yield rate can be enhanced and the goal of low cost manufacturing can be achieved. This article aims at studying semiconductor foundry, Data Mining architecture is built in between CIM and MES, Association rule and K-means Clustering algorithm are used together with in-time feedback control analysis to extract, induce and analyze each parameter which might affect production environment, delete and compensate dynamically the difference among different machines and products, therefore, each machine will have more flexibility in manufacturing different product. The study result shows that the process yield rate of a single product can be improved, through the use of automatic and optimum parameter manufacturing, the whole process capability will be enhanced,enterprise can then build a more stable and flexible competitiveness. The system architecture proposed can be applied in the process analysis of traditional manufacturing industry, it not only enhance product yield rate but also reduce manufacturing cost. Use standard process control together with SPC and the past process data, plus Real-time Feedback Control, we can predict the optimum process parameters for the next lot in order to enhance product competitiveness.