透過您的圖書館登入
IP:3.17.154.171
  • 學位論文

利用線性迴歸偵測電漿化學氣相沉積製程變異

The Detection of Process Variation in PECVD Using Linear Regression Method

指導教授 : 章明

摘要


TFT-LCD是目前平面顯示器(Flat Panel Display,FPD)產業最廣泛應用的技術,也是目前商業化最多的顯示器面板製程,在產業互相競爭壓力下,生產線上的製造流程順暢、生產良率提升及報廢率降低等,均直接影響各家面板製造廠生產績效,本研究重點即在探討如何即時偵測電漿氣相沉積製程變異,以達減少報廢及提昇良率之效。 本文提出以線性迴歸分析(Linear Regression Analysis)之迴歸模型(Regression Model)偵測電漿化學氣相沉積(Plasma Enhance Chemical Vapor Deposition,PECVD)沉膜變異。我們選擇氣相沉積設備中的V-peak電壓做為此次偵測訊號,以基本統計方式找出V-peak訊號特徵值,並以訊號特徵值作為迴歸分析預測變數x ,而迴歸分析應變數y代表沉膜電漿狀態,沉膜正常訊號設定0,異常訊號設定1。 本研究使用V-peak 100筆正常訊號資料與100筆異常訊號資料建立迴歸模型,迴歸模型建立完成後,我們先進行Re-Call處理驗證,把200筆的訓練訊號代回迴歸模型,驗證結果顯示y值設定0.5為偵測訊號最佳值,大於0.5為異常圖形,小於0.5為正常圖形。驗證完畢後再以此迴歸模型進行偵測製程變異,本研究另行收集了V-peak正常訊號400筆及異常訊號400筆,混合代入迴歸線性關係式,驗證結果顯示偵測正確率高達97%。

並列摘要


TFT-LCD is the technology to be used the most in the display industry, also very famous in the market of display manufacture process. In the high competitive market, fluently manufacture process, good quality production rate increasing, and lower defect production rate are the mission of every panel manufacturer. Firstable, to reduce defect production rate and increase good quality rate is testing production equipment on time and stop production at the right timing. This paper will use linear regression analysis model to detect the membrane film variation of plasma enhanced chemistry vapor deposition . We decided to have the V-peak voltage which in the PECVD to be the detect signal. In the regression analysis independent variable is x and dependent variable is y: the normal deposition signal is 0 and abnormal signal is 1. The regression model of the research is based on 100 normal signals and 100 abnormal signals, to detect the variation of manufacture process. When finish the model, we do the Re-Call verification which put those 200 signals back to the regression model. The result of testing: when y is 0.5 which is the best detect signal, if bigger than 0.5 which is abnormal value, and smaller than 0.5 will be the normal value. The research is also collecting 400 normal V-peak signals and 400 abnormal V-peak signals to put into the linear regression analysis, the accuracy rate is 97%.

參考文獻


的比較”,台灣科技大學工業管理系碩士論文, 2005.
【3】 楊金聲,“利用類神經網路與線性迴歸進行成本預測之研
pp.83-89, 2006.
【7】 曾健逢,“大量資料集之線性迴歸模式的估計問題,”國立
【8】 李佳勳,“以簡單線性迴歸分析實作掌紋辨識系統,”大同

延伸閱讀