- 學位論文
應用虛擬量測技術於物理氣相沉積製程之面電阻均一性預測
Prediction of Surface Resistance Uniformity of PVD Process Based on Virtual Metrology Technique
摘要
中文摘要 TFT LCD 產業生產的薄膜電晶體液晶顯示器，由於具有輕巧、攜帶方便不佔空間之特性，取代傳統映像管是必然的現象。正所謂適者生存不適者淘汰的原則，在全球的各大廠的激烈競爭中，要如何使旗下的商品能在市埸上佔有一席之地，除了外在的商品行銷，於內在首要的就是有嚴格品質管制。目前的面板廠大部分仍是採用統計控制 ( Statistical Process Control，SPC ) 的方法來監控製程上的重要參數，然而這些警訊幾乎只能在產品品質發生異常後發出警告，並無法預測出即將生產的下一批產品是否可能會發生異常，若於生產中的例行性量測監控中發現品質異常的情形 ( 超出製程的管制界限 ) ，則在下一批貨生產之前予以修正，製作出符合品質要求之產品。由於產品的測試結果並不一定緊接著在製程結束之後，若量測機台無法在下一批產品之前確認產品有無異常的情況，就有可能造成下一批產品的良率損失，甚至有可能造成大量數目的成品因製程因素造成報廢，嚴重地影響到公司成本與產能。 虛擬量測便是一種預兆偵測之技術，可視為目前業界所用 SPC之進化系統，用以結合 TFT LCD 技術上，來達成品質預測之目標。本篇論文以 TFT LCD 技術，結合虛擬量測的應用，利用小波分解來濾除干擾雜訊、模糊類神經來分別進行產品品質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衡量後有效性的預測，藉以達到有效改善製程品質，提昇設備整體生產效能的目標。
並列摘要
Abstract TFT-LCD display has replaced the traditional CRT because of its advantages, such as lightweight, portability, and small volume. However, under the extreme competition, TFT-LCD plants have to not only market but also promote their product quality in order to dominate the market share. Currently, statistical process control (SPC) is a major method adopted by most of fabs to monitor the important process parameters. Once an abnormal operation occurs, SPC system will issue an alarm. Unfortunately, it cannot predict the happening of any abnormal operations. Engineers must also rely on routine metrology to inspect defect products which are out of specifications (beyond process control limits). Usually, the product measurements are not performed after the fabrication process. The lot of product often stands in a queue for metrology. In this situation, several lots of defect products might be not inspected when abnormal operation occurs. It results in yield loss and, moreover, producing many scraps. Virtual metrology is a technology for predictive detection. It can be regarded as the evolution system of SPC to achieve the quality prediction for TFT-LCD manufacturing. In this study, the developed virtual metrology technique is applied to the prediction of uniformity of surface resistance in the PVD process. It integrates the wavelet transform for predicting the process trend and the fuzzy neural network for estimating the process outcome due to recipe tuning. The virtual metrology technique can effectively predict the uniformity of surface resistance in the PVD process. It promotes the process quality and the overall equipment efficiency.