在半導體業界中，晶圓針測 (Circuit Probe) 屬於生產過程中的後段測試，其測試結果包含了晶圓良率的好壞及追溯異常製程與設備所需之重要資訊，並且產生視覺化的錯誤圖樣分佈於晶圓上，此資訊為製程錯誤診斷及機台故障檢視提供很多有用的線索。因此，了解晶圓針測圖樣所代表的意義及造成錯誤圖樣的原因，在工程資料關聯性分析中是件非常重要的工作。本論文應用影像處理中霍夫轉換等技術來偵測直線及圓形等故障圖樣特徵，用以判別晶圓針測圖中是否含有直線，中心聚集，環形，區域性及邊緣失效等常見的故障圖樣。我們的目標是建立一個全自動化的晶圓針測故障圖樣分類及監控系統，提供後續的工程資料分析及減少人力資源的消耗。為了評估這些特徵所建構出來的模型是否準確，我們使用了J48決策樹與類神經網路資料探勘分類演算法並評估其演算法之效益。

The spatial pattern of failures in a Circuit Probe (CP) wafer map can provide some useful information on which step of the process and which equipment is responsible for the production of failures. Thus, it is crucial to understand root causes of failures and problems related for each characteristic pattern on semiconductor CP wafer maps. In this paper, we applied several selected features to detect CP wafer map representing in scratch, center, ring, region, and edge characteristic spatial patterns. In order to evaluate the advantage of our approach created by selected features, decision tree and neural network classification algorithms have been used. Our goal is create a fully automatic classification procedure to screen signatures for further engineering analysis without human intervention on CP wafer maps.