標題: | 資料探勘技術在晶圓針測誤宰分析之應用 Applying Data Mining Techniques to the Overkill Analysis of Wafer Testing |
作者: | 丁世杰 劉敦仁 理學院應用科技學程 |
關鍵字: | 晶圓針測;誤宰;資料探勘;Wafer Probe;Overkill;Data Mining |
公開日期: | 2012 |
摘要: | 半導體晶圓針測(Wafer Probing)為半導體晶圓製造完成後,對產品良率進行驗證的重要程序,其目的在於找出晶圓上晶粒的瑕疵;主要以探針的方式與每顆晶粒上的銲墊接觸,輸入電流或信號,並接收所反饋出來的數值,以判斷是否為良品。然而,由於測試時可能因為測試機台、針測機或探針卡的穩定性不足、操作人員的參數設定動作不確實,而造成將良品的晶粒誤判為壞品的誤宰(Overkill)現象出現;此現象會影響產品的良率,事後還必須進行重測程序,而使生產效率降低。對半導體晶圓測試產業來說,誤宰現象除了造成測試成本的浪費,也嚴重影響客戶的信任度。
本研究透過與產業領域專家的溝通討論後,建立多項與誤宰現象有相關的偵測演算法,並經由資料探勘分類技術 (Classification),針對這些偵測演算法所計算結果,結合實際的晶圓針測資料為實驗數據進行分析,對這些偵測演算法的準確度進行確認,以用來協助業界在實際生產時參考,確認是否為誤宰的決策輔助。
由於造成誤宰的根本原因,包含了設備、針測卡損耗、人員設定問題,甚至是這些狀況的組合。本實驗也從實際的針測資料中,透過運用資料探勘技術的關聯規則 (Association Rule),找出與晶圓誤宰有關的可能原因,做為製程改善的參考。 Wafer probing is a critical process employed to measure the yield of wafer fabrication. The major object of wafer probing is to find the defect dice on the wafer. It inputs electrical current and signals through the probe needles which contact the pad of each dice and receives the outputs to determine whether or not the dice are good. However, the probing result could be affected by the stability of tester, prober, probe card or the setting actions of operators. Overkill situations happen if good dice are misjudged as bad dice caused by one or more of the above factors causing an abnormal yield and requiring re-probe actions which diminish production performance and the trust of customers. In this study, we talked to the specialists of this industry in order to build some overkill-related detection methods. The aim was to implement these detection methods based on the real wafer probing data from one of Taiwan’s testing facilities. We used classification technologies of data mining on those results generated by these detection methods to determine the correctness of these methods, to ascertain if they could be implemented in the data analysis system as a real-time alarm to handle the overkill issue. The root causes of overkill situations are not easy to find. They include: stability of tester, usage of probe card, setting actions by operators or a combination of these factors. We used association rule technology of data mining to find the possible causes of overkill situation, to serve as reference for improving actions. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#GT079973610 http://hdl.handle.net/11536/50904 |
Appears in Collections: | Thesis |
Files in This Item:
If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.