建构半导体制造过程产品异常资料挖矿技术跟研究.docVIP

PAGE 建構半導體製造過程產品異常資料挖礦技術之研究 梁蕙姿 簡禎富 彭金堂 國立清華大學工業工程與工程管理研究所 國立清華大學工業工程與工程管理研究所 元培科學技術學院經營管理研究所 摘 要 為因應半導體製造過程中產品製程技術的快速變化、複雜產品組合及生產週期時間長的生產環境，如何在自動化系統輔助的大量生產之下，在製造過程中提供工程人員可確認正確的製程參數，以及發現產品異常時可立即針對問題的特徵進行問題真因追蹤與分析，針對取得的資訊進行合適的決策並將可能已受影響的產品列表管理，以減少產品的良率損失，已成為重要的產業應用與學術研究議題。 本研究目的係研究資料挖礦技術中關聯規則的Apriori 演算法，結合半導體領域知識，將製造過程中追蹤已發生問題的解決方式，進行關聯資料的搜尋並予以模式化，並利用決策樹歸納法進行可疑原因區別的建議；本研究並建構此半導體製造過程中的資料挖礦應用系統雛型，而以目前半導體製造中自動化系統所蒐集的資料結合此方法而建構一半導體製造過程中的資料挖礦應用系統為此研究的實作驗證，以完整地檢驗資料挖礦的方法與步驟進行大量資料的篩選、推演與模式建構等過程；結論歸納本研究結果與貢獻，並探討未來研究方向。 關鍵字：資料挖礦、決策分析、半導體製造管理、關聯規則 The Study of Data Mining Techniques for Analyzing Semiconductor Manufacturing Product Abnormal Hui-Tzu Liang Chen-Fu Chien Jin-Tang Peng Department of Industrial Engineering and Engineering Management, National Department of Industrial Engineering and Engineering Management, National Graduate Institute of Yuanpei Institute of Science and Technology ABSTRACT The environment within the semiconductor industry is one that is made up of rapidly changing technologies, complicated product groups, extensive production hours. The following points have become crucial issues for industrial application and academic research within the semiconductor industry: methods by which accurate processing parameters may be provided for engineers to confirm to, within an automation system during mass production; methods by which abnormal wafers may be instantly tracked and analyzed, based on the problems/defects of each individual case; methods by which obtained data may be appropriately processed, by which affected products may be managed through a report chart, and by which yield loss of products may be lessened. This study will focus on the Apriori algorithm used in data mining technologies, and will incorporate working knowledge of the semiconductor industry, to search for, and to develop the solution used to track existing problems within the fabrication process. The decision tree analysis method will also be used to categorize and dif