在本篇論文中，我們討論在兩個部分，一個是晶圓圖的多變量分析，另一個是模型分析以及實驗結果。在本篇研究中，我們先透過預備知識了解良率、 B-score及相似度的參數性質，接著，在晶圓圖的多變量分析中，我們運用良率、B-score以及相似度這三個變數來進行主成份分析最後再利用K-medoids分群法進行分群來表示多變量分析之成果。 再來我們利用晶圓圖的多變量分析的步驟，去對自製的合成特殊晶圓圖做實驗，在合成的晶圓中，我們使用 系統性 錯誤樣態的 晶圓圖作分析，其錯誤樣態含有Center、Donut、Scratch、Edge-Ring、Edge-Loc以及Loc我們將每個錯誤樣態的晶圓圖分別製成不同良率、不同大小的晶圓圖進行討論。 最後我們將所實驗的步驟應用到真實晶圓，此真實晶圓為台積電提供的 WM-811K晶圓資料庫所標記錯誤樣態的172K拿來使用，進而討論在真實晶圓中的應用情形。;In this paper, we discuss two parts. First, we focus multivariate analysis of wafer maps. Second part is model analysis and the results of experiment. In this research, we should know yield, B-score and similarity according to previous research. Then, we use above three variables to do multivariate analysis. Last but not least, we use K-medoids cluster to grouping the data what we get in previous analysis. Second part, we use first part steps to test synthetic wafer maps. We use synthetic wafer maps which are symptomatic, including Center, Donut, Scratch, Edge-Ring, Edge-Loc and Loc. Each failure type wafer maps are produced in different yields and die sizes respectively. Finally, we apply our experiment steps to real world wafer maps which are produced by TSMC WM-811K database which is contained labeled wafer map lots in 172K. We discuss the situations in real world wafer maps.
