本研究之主要目的為建立一套應用於薄膜電晶體液晶顯示器( TFT-LCD )前端陣列( Array )製程中,判斷電漿化學氣相沉積( Plasma Enhanced Chemical Vapor Deposition,PECVD )機台製程參數與該道流程沉膜厚度之類神經網路模型,再將此神經網路之知識萃取為布林規則形式,以供建立專家系統知識庫之用。在資料的收集上,係自LCD廠中之array部門的PECVD機臺取得,包括了該道沉膜製程中,使用氣體種類、流量、壓力、腔室( chamber )溫度等,配合影像檢測部門對於該流程成品的檢查結果,分別以此兩項數據為輸入與輸出,採用倒傳遞( Back-Propagation )的學習法則來建立神經網路系統模型,總共蒐集了92筆資料作為訓練樣本,進行500次的訓練,訓練完畢的類神經網路模型經30筆測試樣本的檢測,其平均誤差值為±1.62%,證實能夠有效預測PECVD製程之沉膜品質狀況。由此神經網路模型,再以規則萃取演算法擷取出其所學習到的知識,由於這些知識能以布林邏輯的規則方式呈現,將這些定性規則予以檢查、解釋後,彙整成專家系統知識庫形式,可作為系統膜厚預測、診斷與預警上的依據,當機台參數的漂移可能令膜厚超出規格時,即可提出警示,並提供工程師執行參數修正的參考方案。
The principal object of this research is to develop a neural network model, which can simulate the plasma enhanced chemical vapor deposition (PECVD) process in TFT-Array procedure. Then the Boolean logic rules were extracted from the trained network model to establish a knowledge base of the expert system. Here the input data of neural network was collected from the process parameters of PECVD machines in the TFT-Array department, which included the flow rate of all gases, pressure and temperature of the chamber, etc. Furthermore, the inspecting values from the image inspection department were adopted as the output. The neural network model was constructed with back-propagation network (BPN) algorithm and trained 500 times with 92 training data. The model was tested with 30 test data and the error rate was ±1.62% in average. The extraction rules were obtained from the trained neural network with the rule extraction algorithm to describe the knowledge in the network system with logic rules. After checking, explaining and integrating the rules into the knowledge base, the rules can then be the basics of membrane thickness prediction and alarm diagnosis in PECVD system. When the parameters are abnormal, to make the membrane thickness out of control limit, an alert message can be offered and also the knowledge base can provide a reference to the engineers for the work of recipe adjustments.