應用類神經網路減少TFT-LCD產品測試項目之研究

Abstract

台灣是國際上TFT-LCD (Thin-Film Transistor Liquid-Crystal Display)，薄膜電晶體液晶顯示器) 主要的製造和生產基地。由於TFT-LCD在生產設備和儀器的投資金額相當龐大，因此如何降低製造的時間，維持一定的品質，縮短產品的生命週期和加速新產品的開發，比競爭對手更早推出產品在市場上，是企業生存的必要條件。在TFT-LCD的生產過程中，檢驗和測試是一個大瓶頸。 本研究透過運用類神經網路的方法減少TFT-LCD在生產製造上的測試項目，以期能降低測試的時間和設備的投資，並希望當利用較減少測試項目之檢驗的結果和使用原有測試項目所檢驗出的結果要非常近接近。此外，在減少測試項目所做的TFT-LCD面板的品質分類結果，也希望能和原有測試項目所做的TFT-LCD面板的品質分類結果非常近接近。本研究以一實際案例來說明所提類神經網路的方法的有效性，並與傳統的統計迴歸方法進行比較。

Taiwan is the major TFT-LCD (Thin-Film Transistor Liquid-Crystal Display) manufacture and engineering base in the world. Owing to quite huge amount of expense in design and production facilities, the necessary survival criteria for this industry based on how to shorten the manufacturing time, maintain the high quality, cut the product lifecycle and expedite the newer model design on current market, have become critical issue.

This research aims to deduct the TFT-LCD test items on manufacturing process via applying Neural Network approach. We expect to reduce the test time and the facility investment, and hopelly to get the same or less inaccuracy test results between reduced test items and original test items. Moreover, the TFT-LCD quality classified results by reduced test items are approximately the same as them by keeping the original test items. This research uses a real example to demonstrate the validity of Neural Network approach and also does the comparison to the traditional statistic regression analysis.