本篇論文為藉由實驗室先前研究晶圓圖特性分界，針對其分界所產生出的新參數：B-score，並利用原先有的參數NBD(瑕疵晶粒總數)、NCL(瑕疵晶粒連續線總數)進行隨機性的分析，而一般晶圓圖是由隨機性和系統性錯誤所組成，我們將分界進行更嚴謹的分析，強化分界之準確性，並增加測試效率。 首先，我們先使用隨機數產生的合成晶圓，將原本1.96倍的標準差(95%信賴區間)增加至2.58倍(99%信賴區間)及3.89倍(99.99%信賴區間)，在全域的分析之下，加強B-Score隨機性之驗證。其後再以正規化NBD做各區段的常態性分析，觀察同一區段所產生的B-Score之分布在我們所找的幾個臨界點上是否符合，驗證B-score為一個標準分數。 最後，利用自製的合成特殊晶圓圖，除了原本分類的錯誤型態，也加入了便於我們觀察較於特殊的晶圓圖，觀察其與B-score之關係，並了解B-score的特性。也將我們利用B-score所找出來結果較為不準確的實際晶圓圖，利用減少一些較分散的瑕疵晶粒或瑕疵晶粒群，得知相較於人工判別這些皆會影響我們隨機性分析之結果。 ;In this thesis, a boundary of wafer map characteristics is used by previous research to generate a new parameter: B-score. We use this parameter to verify randomness by the original parameters: NBD (Number of Bad Die) and NCL (Number of Contiguous Line). The general wafer map is composed of random and systematic errors. We will strictly verify the boundary to enhance its accuracy and increase test efficiency. First, we generate synthetic wafers by random number. Increasing the original standard deviation of 1.96 times (95% confidence interval) to 2.58 times (99% confidence interval) and 3.89 times (99.99% confidence interval), we use full-range analysis to enhance the verification of randomness. And then verify the normality of normalized NCL. Observed whether distribution of B-score meets to our critical points. By these steps, verified B-score is a standard score. At last, we derive synthetic special patterns to the original classification of failure types. we are also added some special wafer maps which facilitated our observation. To observe the relation with B-score, and know more about its behaviors. We also use B-score to judge some inaccurate real wafer. By removing some single dies or clusters, we can know more about these single dies or clusters will affect the result of randomness test.
