頁籤選單縮合
題 名 | 晶圓良率損失資料分群模式之研究=A Data Clustering Model for Wafer Yield Loss in Semiconductor Manufacturing |
---|---|
作 者 | 劉淑範; 陳飛龍; | 書刊名 | 工業工程學刊 |
卷 期 | 21:4 2004.07[民93.07] |
頁 次 | 頁328-338 |
分類號 | 494.568 |
關鍵詞 | 良率損失; 工程資料; 良率特徵值; 分群品質指標值; 修正後模糊分群; Low-yield; Engineering data; Yield characteristic value; Clustering quality index; M-FCM; |
語 文 | 中文(Chinese) |
中文摘要 | 半導體廠由以往的6吋、8吋、至今12吋之規模,使得國內晶圓代工製造將成為全球最具競爭力的產業,如何降低代工成本、縮短代工交期、提升代工製造的良率,都成為面對全世界競爭的關鍵。由於任何良率的損失都會造成成本的提升,因此各半導體廠莫不汲於藉由各種分析手法,針對生產過程進行嚴格之監控,以達到良率提升(Yield Enhancement)之最終目的。一般而言,當晶圓良率損失時,良率工程師會利用所收集之大量工程資料(Engineering Data)來進行分析,這些工程資料包括缺陷資料、晶圓圖樣資料、製程 / 機台參數、WAT參數等,然而傳統的工程資料分析方式對於尋找製程問題之效率有限,且花費太多的時間。基於此,本研究提出一晶圓良率損失資料分群模式,將整片晶圓製程與測試之相關工程資料分析其良率特徵值,並以本研究所提之修正式模糊分群演算法(Modified-Fuzzy C Means; M-FCM),利用分群品質指標值(Ψ)從所有良率損失晶圓之特徵向量中求得最佳分群品質解。本研究以某半導體廠實際資料進行驗證後發現,此分析模式能有效輔助工程師透過相同分群中的晶圓特徵資料,快速的找出原因並瞭解各資料間的關連性,使其能在較短時間內做出更正確的決策。 |
英文摘要 | For the past few years, semiconductor manufacturing has emerged as one of the most important industries in Taiwan. The complex manufacturing processes, expensive raw materials and machines, and the near particle free environment have made semiconductor manufacturing a high-cost industry. If low yield occurs in certain lots, the yield loss will result in high manufacturing cost. For this reason, all the semiconductor manufacturing companies have devoted huge efforts to analyze the yield related data in the hope for reducing the occurrence of process variations and achieving the goal of yield enhancement and cost reduction. However, many of the different yield enhancement methods used by the semiconductor manufacturing companies are in lack of integration and therefore are inefficient. Engineers from different departments usually analyze large amount of engineering data. They also spend lots of time tracing the possible variations by analyzing the defect maps, wafer bin maps, process parameters, machines, and WAT parameters with simple statistical analysis techniques and tools. In this situation, the efficiency and effectiveness of locating the root causes are strongly associated with the experiences of the engineers. Based on the above consideration, this research intends to construct a model to analyze the characteristic values of low-yield wafer in semiconductor manufacturing. Firstly, the characteristic values of low-yield wafers are retrieved. Secondly, the proposed modified fuzzy c means (M-FCM) algorithm along with clustering quality index (ψ) is applied to cluster the characteristic values of low-yield wafers and therefore an optimal solution is obtained. With the proposed model, the possible root causes can be identified more easily and decisions can be made to correct manufacturing problems more efficiently. |
本系統中英文摘要資訊取自各篇刊載內容。