運用資料探勘技術於LED產業製程良率

Abstract

由於市場的競爭激烈，高良率成為LED產業所追求的目標之一，對於LED產業而言，良率越高表示其產品的單位成本越低，產品在市場上的競爭力代表越高，並可大幅提昇公司的產品利潤。從近幾年來，由於產品良率偏低而無法在市場上取得優勢的案例層出不窮，想要在這個競爭激烈的環境下生存，如何提昇良率是一個非常重要的課題。 本研究提出以資料探勘的關聯性規則-Apriori演算法及類神經網路—霍普菲爾坦克網路方法去實作製程良率改善系統以找出LED製造流程良率缺陷原因及相關對應的最佳生產機台的組合，期望能由建置產品製程良率問題之資料倉儲，並透過資料探勘的技術，分析關於產品在製造過程中所遇到影響良率的關鍵因子及相關機台組合，進而得到快速而又可信度高的結果，作為後續良率問題處理的主要決策依據。 本研究結果在問題分析效益上，可以找出生產機台組合造成產品良率不佳原因。並可提升生產排程機台組合的製程良率，或調整機台參數設定，提供相關人員做決策，進而達到提升良率的目的。比較兩種演算法的分析結果，霍普菲爾—坦克類神經網路在平均良率改善較Apriori演算法為佳。在降低產品生產週期上，霍普菲爾—坦克類神經網路在縮短產品生產週期較Apriori演算法佳。 所提出的系統架構也可應用於相關製造業在製程機台組合上的分析，除了提升產品良率，亦可降低生產成本，提高訂單的達交率，協助提升企業競爭力。

Due to keen competition of marketing, high yield is the target of the Led industry. In Regard to the Led industry, the higher process yield to mean the lower product cost, and the product competion is more excellent. And it promotes the product profit. Recently, the cases that lower process yield make inferior competion emerge unceasingly. So it is very important to promote the process yield in the keen competition business activities. This paper presented the assocation rule-Apriori algorithm and Hopfield-Tank neural network for data mining to find the best association of process equipments and the defect code of LED process flow. To this end, we construct the datamart to analyze the process yield and find the root causes that influence the process yield and equipment assocation. By this method, we could find the best solution, and using the result to enhance the performance of process efficiently. Benefits to use the data mining system that implement by Apriori algorithm and Hopfield-Tank neural network, we use this system to find the worse equipment association to avoid these problem occurrence. By the above method to provide the process decision-making to promote the process yield. To compare Apriori algorithm and Hopfield-Tank neural network, we have the summary as the following: To promote the yield, Hopfield-Tank neural network improves better than Apriori algorithm. To reduce product cycle time, Hopfield-Tank neural network improves more than Apriori algorithm. We could apply this methodology to not only on LED industries but also on another manufacturing industries in the problem of equipment association. It could help us to find the factors in various parameters that affect process. It could enhance process yield accuracy, reduce production costs and promote the product delivery and strength the enterprise competitiveness.