- 學位論文
半導體製程虛擬量測技術之研究
Virtual Metrology Technique for Semiconductor Manufacturing
摘要
中文摘要 我國半導體是以生產製造為主體，在全球的激烈競爭下，嚴格的製程控制已經成為不可或缺的要求，要如何有效的降低生產成本、增加設備效能、減少製程風險、提昇製程良率，為現今晶圓廠與面板廠提升其國際競爭力最重要的因素。目前的晶圓廠和面板廠大部分仍是採用統計製程管制技術來監控重要的製程參數，並以例行測機、機台預防保養等方法提高製程穩定性，改善製程良率。然而在測機或後續檢驗與量測的結果無法立即產生，故這種監控方式對於無預警式的系統良率損失而言，當製程中發生異常狀況時，就有可能造成大量數目的基板報廢，將會嚴重地影響到生產成本與產能。 由於近幾年資訊科技與網際網路技術的蓬勃發展，於是產生了網路式診斷與網路式預防保養的觀念，透過網際網路與資訊技術來達到虛擬量測、自我診斷、預測性維修與等之目標。藉由此類網路式診斷技術，將可快速地診斷與修復設備、大幅縮短平均修復時間，以及降低設備的維修服務成本。而虛擬量測的概念便是以網路式預兆偵測之技術，結合先進半導體技術，來達成品質預測之目標。 本篇論文以半導體技術結合虛擬量測的概念，利用小波分解來濾除干擾雜訊，再搭配類神經網路與模擬控制，分別進行趨勢預測以及配方調整效果之預測，並整合這兩方面對於響應的影響來做Metrology的預測，藉此達到改善製程品質，提昇設備整體效能的目標。使用在虛擬量測的系統架構下，結合趨勢狀態與配方響應的預測結果，在模擬資料的部分，絕對誤差在25以下且誤差率在1.5%以下，而在預測實際的製程資料，絕對誤差約在0.05左右且誤差率也能維持在3.84%左右，整個虛擬量測系統的預測結果的準確性令人滿意。
並列摘要
Abstract The semiconductor of our country regards manufacturing as the subject. Under global keen competition, strict process control already become the indispensable demand. It is most important factor of international competitiveness for Panel factory and Liquid Crystal Display(LCD) factory how does it want effective reducing production cost, increasing equipment efficiency, reducing process risk, increasing process yield. Present most of Panel factory and LCD factory use control technology for statistics process control technique to monitor important process parameters. However the result can’t be measure right away, this monitor maybe cause a large number of base plate loss if it happen process unusual. It will influence the production cost and yield seriously. Because of the flourishing development of information technology and internet network technology in recent years, produced the idea about e-Diagnostics and e-Maintenance. Pretext of internet network and information technology to achieve the goal of virtual metrology, diagnose-self and predict maintain. E-Diagnostics technology can diagnose and repair the equipment, shorten repair time and reduce service cost. Idea of virtual metrology is combine e-Diagnostics and advanced process control to achieve the goal of quality predict. The theory combines advanced process control technology and idea of virtual metrology to filter noise by wavelet theory, and using fuzzy neural networks predict trend and recipe influence. Pretext of predict metrology to improve the quality of process, promote equipment efficiency. Absolute error is under 25 and absolute error rate is under 1.5% in simulate data and absolute error is about 0.05 and absolute error rate is about 3.84% in actual data.