就半導體產業而言,生產良率之高低會影響產品的成本以及企業之競爭力,因此良率的改善已成為各競爭廠商的重要課題。在半導體製造過程中,往往都會蒐集每個製程步驟的詳細晶圓生產資料並將這些資料儲存於資料庫中,以便進行製程監控、故障分析與製造管理,但往往因為半導體的製程複雜,而且影響的變因眾多且通常具有相互關係,工程師很難以迅速從龐大的資料中找到可能會導致製程異常的原因以及可能隱藏的有效資訊。 本研究乃是利用決策樹方法來建構半導體資料探勘架構,並希望尋找出可能造成製程變異的原因以做為製程工程師及領域專家解決問題的參考依據,並且運用貝氏網路來說明這些製程因素之間的相互影響程度及機率關係,希望藉由變異的原因探討,進而能來提升半導體製程的良率,本研究並以某半導體廠之案例為實證,以檢驗本研究的可行性。
In the semiconductor manufacturing, the process yield plays an important role. A high yield can bring profit to the industry and also can proof the company have the good competition, thus how to achieve high and stable yield is a principal task for the business. During the process in semiconductor manufacturing, each operation step will collect the process and engineer data automatically or manually in order to monitor process situation and defeat diagnosis, and for manufacturing management. But it’s difficulty for engineer to find out the critical factor timely and rapidly in massive data because the semiconductor manufacturing process is very complex and has many interactive influence parameters. This research is tried to using the decision tree approach to build the data mining architecture of semiconductor and using it to find out the relation between input parameter and outputting yield, and in order to dig out valuable information. Besides, in this paper proposed the Bayesian network, because Bayesian network model has both a causal and probabilistic semantics, it is an ideal representation for combining prior knowledge and data go to control the parameter value of the products, in order to improve the process yield. An empirical study was conducted in a fab and the results showed the practical viability of this approach。