Please use this identifier to cite or link to this item:
標題: Development of Wafer Bin Map Multiple Pattern Recognition Model-Using Genetic Algorithm
作者: 顏懷先
Huai-Hsien Yen
關鍵字: Bin Map
Pattern Recognition
Genetic Algorithm
引用: [1] 李偉傑,「半導體之工程資料與診斷系統」,碩士論文,國立清華大學工業工程與工程管理學系研究所,1996。 [2] 林寅智,「以工程資料為基礎之半導體故障分析系統」,碩士論文,國立清華大學工業工程與工程管理學系研究所,1998。 [3] 林景堂,「晶圓圖像辨識」,碩士論文,國立台灣大學資訊工程學系研究所,1998。 [4] 簡禎富、林鼎浩、劉巧雯、彭誠湧、徐紹鐘、黃佳琪,「建構晶圓圖分類之資料挖礦方法及其實證研究」,Journal of the Chinese Institute of Industrial Engineers, Vol. 19, No. 2, pp. 23-38,2002。 [5] 莊達人,VLSI製造技術,高力書局,1997 [6] 劉淑範,「以工程資料為基礎之晶圓良率特徵值模式架構與診斷分析」,博士論文,國立清華大學工業工程與工程管理學系研究所,2004。 [7] 鐘安聖,「以小波轉換與類神經網路方法建構半導體晶圓圖樣辨識系統」,碩士論文,國立清華大學工業工程與工程管理學系研究所,2007。 [8] 魏連均,「應用類神經網路建構晶圓圖故障圖樣辨識模式」,碩士論文,國立清華大學工業工程與工程管理學系研究所,2006。 [9] Nag, P. K., Maly, W and Jocobs, H. J., 'Simulation of Yield/Cost Learning Curves with Y4,' IEEE Transactions on Semiconductor Manufacturing, Vol. 10, No.2, pp.256-266, 1997. [10] Mirza, A. I., O'Donoghue, G., Drake, A. W. and Graves, S. C., 'Spatial Yield odeling for Semiconductor Wafers,' IEEE/SEMI Advance Semiconductor Manufacturing Conference, pp. 276-281, 1995. [11] Cox, I. and Reynolds, A., 'Manufacturing Digital's Alpha AXP: Rapid Applications Development to Assure Critical Success Through Quality,' SAS Software Solutions for The Semiconductor Industry, pp. 32-37, 1996. [12] Kaempf, U., 'The Binominal Test: A Simple Tools to Identify Process Problems,' IEEE Transactions on Semiconductor Manufacturing, Vol. 8, No. 2, pp. 160-166,1995. [13] Hansen, M. H., Nair, V. N. and Friedman D. J., 'Monitoring Wafer Map Data From Integrated Circuit Fabrication Processes for Spatially Clustered Defects,' Technometrics, Vol. 39, No. 3, pp. 241-253, 1997. [14] Hu, M., Visual pattern recognition by moment invariants. IRE Transactions of Information Theory. v8. 179-187.
摘要: The complexity of semiconductor manufacturing has increased a lot in recent years. The semiconductor manufacturing consists of more than 400 process steps. In order to ensure product quality, equipment parameters and measurement result of wafers are collected and save into database of automation systems during production, and many tests are needed to be performed afterwards. Engineers need to check all relative process parameters if the test result is abnormal and then find out the root cause according to their expertise and experience. However, the process data are too enormous to analyze effectively for engineers. Therefore, how to analyze huge amounts of data effectively and covert into valuable information is a very important topic for semiconductor industry. Fortunately, different abnormal excursion may form different patterns on wafer bin map, engineers often use wafer bin map to find out the root cause of abnormal excursion. Currently most semiconductor fabs still analyze wafer bin map manually, a few fabs use pattern recognition system in wafer bin map but these systems assume that a wafer bin map only has a pattern. Actually a wafer bin map may consists of serval independent patterns of different regions in wafer map. This research uses supervised machine learning to implement classification of bin map patterns. Extract features of each patterns of training data which has been pre-processed, then calculate weight of each features via genetic algorithm, and then use Weighted Euclidean Distance to calculate similarity of testing data and each patterns. This research use real fab data to build up a recognition system and validate the effectiveness. The experimental results demonstrate that the proposed methodology can not only recognize bin map patterns by extracting features, but also effectively identify multiple bin map patterns on the wafer by recognition rate of 90%
近年來半導體製程日益繁複精密,製造過程包含了三、四百道製程程序,為了確保產品品質,在生產過程中會進行機台參數的收集與晶圓量測的動作,生產結束後還會進行多項測試。測試結果如有異常,工程師就需要調閱該批產品的相關製程參數,根據工程師的專業知識與經驗判斷出造成異常的真因。然而這些工程資料太過龐雜,工程師往往難以有效率的加以分析,因此如何有效的分析大量的工程資料,進而轉化成有價值的資訊,成為各大半導體廠的重要課題。 所幸實務上不同的異常事故發生時可能會在晶圓圖上形成不同的故障圖樣,因此工程師常利用晶圓圖來追查事故原因。目前半導體廠對晶圓圖的分析大多仍採人工辨識,少數晶圓廠雖有自動化的晶圓辨識系統,但皆假設一張晶圓圖只有一個故障圖樣,但實務上一張晶圓圖可能依所在區域不同,而同時存在多個故障圖樣,故障圖樣間彼此獨立,互不相關。本研究以監督式的機器學習手法將故障圖樣分類,將晶圓資料經過資料前處裡之後,提取圖樣的特徵值,以基因演算法得出各特徵值的權重值,利用加權的歐式距離計算出晶圓資料與各故障圖樣的相似度,進而達成辨識的效果。 本研究以實際的晶圓廠資料以建立系統並驗證成效,實證分析結果顯示本研究手法可成功辨識多重晶圓故障圖樣,辨識率達到90%。
其他識別: U0005-2601201516102300
文章公開時間: 10000-01-01
Appears in Collections:資訊管理學系



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.