在本篇論文中，我們分析台積電WM-172K晶圓資料庫所有LOT內含晶圓之間的相似性。我們使用相似性公式來計算相似性數值，該數值反映兩片晶圓的相似程度。接著，我們定義相似性門檻，也就是判斷是否相似的參考數值。 在建立相似性門檻時，我們考慮良率和晶片尺寸兩個因素，建立兩個新的相似性門檻，加上過去的一個門檻，我們使用這些門檻會找到LOT中不同的相似晶圓集合。此外，我們還定義了絕對相似，即在同一批次內找出互相完全相似的一組晶圓圖像。 最後討論相似性門檻在真實晶圓圖上分析應用的部分，選擇實驗的對象為某公司提供的真實晶圓資料集，我們透過相似性門檻和chamber表格檢測，找出每一批次中具有相似的特徵及關連，快速判斷chamber效應存在與否以及發生位置，提早找出異常機台， 達到提高生產效率及良率。 通過相似性的分析，我們可以判別晶圓間的相似程度，如此可以提供協助判斷晶圓發生的錯誤樣態與問題。 ;In this paper, we analyze the similarities between wafers within the TSMC WM-172K wafer database. We employ a similarity formula to calculate similarity values, which reflect the degree of similarity between two wafers. Next, we establish a similarity threshold, which serves as a reference value for determining whether wafers are similar.
When establishing the similarity threshold, we consider two factors: yield and chip size. We create two new similarity thresholds in addition to the previous one. Using these thresholds, we identify sets of dissimilar wafers within a lot. Furthermore, we define absolute similarity, meaning finding a group of wafers that are identical within the same batch.
Finally, we discuss the practical application of similarity thresholds on actual wafer images. We select the real wafer data set provided by a certain company as the object of the experiment. We use similarity threshold and chamber table to identify batches with similar characteristics and relationships, rapidly determining the presence and location of chamber effects, and early detecting anomalous machines. This leads to improved production efficiency and yield.
Through the analysis of similarity, we can assess the degree of similarity between wafers, aiding in the identification of error patterns and issues that occur within wafers.
