應用視覺及資料探勘技術於TFT-LCD陣列電路工程線中瑕疵辨識之研究-源/汲電極光罩瑕疵自動分類系統開發

Development of an Automatic Source and Drain Electrode Mask Defect Classification System for the Inline Inspection of TFT-LCD Array Engineering

林思賢(Szu-Hsien Lin)

指導教授 ： 劉益宏

中原大學/工學院/機械工程學系/碩士(2006年)

https://doi.org/10.6840/cycu200600375

摘要

就目前的面板製程而言，所有的製程流程都在無塵室內進行，廠區也將外在環境的影響降至最低，惟多數的面板在製程時仍會因為異物的掉落或是基台故障等因素，導致產品出現缺陷。例如面板上出現持續不變色的亮點或亮線，輕者可以用雷射進行修補，重則必須報廢，這對面板廠商而言即是成本的消耗，故提高面板的良率，以降低生產成本並增加公司獲利，對面板廠商而言是最重要的目標。由於影響面板良率的主因為面板的製程方面，因此若能在面板製程中的陣列電路工程及時發現錯誤並加以補救或是停止，即可提升面板的良率。
為了增加良率，多數的公司都設有檢查部門，主要是當製程進行到某一個階段時進行人工的瑕疵分類工作，以防止較大的錯誤繼續發生。但是為了降低人為誤判的影響以及增加處理的速度，以達到全檢的目標，將整個檢查系統自動化是必須的。
本論文的目標即是發展一個自動化瑕疵分類系統。藉由融合各種數位影像處理技巧、統計紋理特徵抽取、資料探勘以及類神經網路辨識等方法，針對檢查機台所拍攝的瑕疵影像進行自動即時分類，以利產品良率的提升。本研究主要是針對檢查機台於陣列電路工程中、第三道光罩（源/汲電極光罩）微影製程後所拍攝到的瑕疵影像進行分析與辨識，發展了一個『源/汲電極光罩中微影製程瑕疵影像辨識系統』。此系統可以對九種常見的瑕疵影像（GE殘、SE殘、SD殘、源/汲電極短路、異物、塗佈異常、顯影異常、刮傷和GE圖案突出體）進行分類，讓瑕疵分類達到自動化、即時化，以上的瑕疵影像皆由華映面板廠商提供。
經由最後的實驗測試結果可知，我們輸入了886張含有瑕疵的實際照片，『源/汲電極光罩中微影製程瑕疵影像辨識系統』的辨識率可達96％以上，表示整套系統針對以上的九種瑕疵影像是可以精準的分類出來，防止下一批面板經過微影製程時發生同樣的瑕疵，對於面板製程中的檢測及良率提升是有相當大的幫助。此外，系統平均辨識一張瑕疵影像不需要3秒鐘，代表系統能快速的分辨瑕疵。

關鍵字

陣列電路工程 ； 自動化瑕疵分類系統 ； 類神經網路 ； 源/汲電極光罩

並列摘要

For panel companies today, reducing their costs and increasing company revenues by raising the yield rate of panels is one of their most important goals. One of the major factors hugely affecting the quality of panels is the process of making panels. Though all of panel-making processes are practiced in dust-free rooms, and all factories have tried their best to reduce the influences of outside factors to the least degree, defects do occur in the process, either because of the falling-down of particles, or because of equipment failure. As a result, companies make defected products. For example, some of the defects are bright spots or bright lines that do not change color for a long time. Less serious defects could be repaired by laser; more serious ones make the product being abandoned. This is an unnecessary waste of resources for a company. As a result, if mistakes could be found in time in the process of array engineering and get fixed or prevented, the yield rate of panels would thus be increased, and the cost decreased.
Most companies set up quality control departments to increase the yield rate of their products. They are meant to perform manual defect classification when the process is practiced to a certain stage, intending to call a stop before major mistakes occur. However, in order to reduce the influences of human factors, to accelerate the speed of processing, and to achieve the goal of a full inspection, an automatic inspection system is in great need.
The purpose of this study is to provide an automatic defect recognition system. In this study, we consult theories of digital image processing techniques, statistic textured feature extraction, data mining, and neural network. We want our system to automatically classify defect images shot by inspection machines, with the intension of increasing the yield rate of products. We focus on the analysis and recognition of defect images shot by inspection machines in array engineering, during the lithography process in the third mask, and devised a “Defect Recognition System for the Lithography Process Inspection in the SD(Source and Drain Electrode)-Mask” in the study. This system is able to automatically classify nine common defect images, providing a real-time automatic defect classification. The above-mentioned defect images are all offered by a panel company.
The experimental results show that, among the 886 defect pictures offered by a listed panel company in Taiwan, the “Defect Recognition System for the Lithography Process Inspection in the SD-Mask” achieves a recognition rate higher than 96%. This means that our system is able to classify the nine defect images above promptly and accurately, to prevent the same defects occur in panels-to-come during the lithography process, and is capable of increasing the accuracy of inspection and the yield rate in panel processing. Moreover the developed system is also to classify one defect image within 3 second, which means that the goal of high-speed defect inspection is achieved.

並列關鍵字

automatic inspection system ； Source and Drain Electrode ； array engineering ； neural network

參考文獻

[1] R. Jain, R. Kasturi and B. G. Schunck, Machine Vision, McGraw-Hill, New York, 1995.

[2] W. S. Kim, D. M. Kwak, Y. C. Song, D. H. Choi, and K.H. Park, “Detection of spot-type defects on liquid crystal display modules,” Key Engineering Materials, pp. 808-813, 2004.

[3] C. J. Lu and D. M. Tsai, “Automatic defect inspection for LCDs using singular value decomposition,” International Journal of Advanced Manufacturing Technology, vol. 25, pp. 53-61, 2005.

[6] K. Nakashima, “Hybrid inspection system for LCD color. filter panels,” Proceedings IMTC/94. Conf. 10th Anniversary, vol. 2, 10-12, pp. 689 -692, May 1994.

[7] N. Otsu, “A threshold selection method from gray-level histograms,” IEEE Trans. System Man Cybernet, vol. 9, no. 1, pp. 62-69, 1979.