應用類神經網路預測電漿化學氣 相沉積膜厚與參數調整建議系統

A Suggestive Tuning System Using Neural Network Theory for the Prediction of the Membrane Thickness in PECVD Process

陳宥佐(Kobebryant Chen)

指導教授 ： 章明

中原大學/工學院/機械工程學系/碩士(2006年)

https://doi.org/10.6840/CYCU.2006.00483

摘要

摘要
本研究主要是利用薄膜電晶體液晶顯示器（TFT-LCD）前端陣列（Array）製程中，電漿化學氣相沉積機台的參數與沉膜厚度建立類神經網路模型，再以模糊規則萃取的方法從神經網路中擷取輸入與輸出間的定量模糊規則。分析的資料係由LCD廠中array部門之PECVD機台取得，參數包括沉膜製程中機台內的各種氣體流量、溫度、壓力、傳遞功率等。資料在整理過後，以機台參數與沉膜厚度分別當作輸入與輸出，共蒐集700筆資料，取其中500筆資料做為訓練樣本，再利用倒傳遞（Back-Propagation）類神經網路進行500次訓練，訓練完畢後的類神經網路模型在以200筆資料作為測試樣本檢測，其平均誤差為9.09 ％，證實能夠有效模擬與預測PECVD製程中之沉膜狀況。由訓練完成的神經網路模型，再利用模糊規則萃取演算法，自類神經網路中擷取出輸入（氣體等參數）與輸出（膜厚等）間的定量模糊規則（fuzzy rules），將這些定量規則予以檢查、解釋後，建立出一套模糊專家系統，做為工程師更改配方時的參考依據。

關鍵字

沉膜厚度 ； 類神經網路 ； 模糊規則萃取 ； 模糊專家系統

並列摘要

ABSTRACT
The principal object of this research is to develop a neural network model by using the parameters of the plasma enhanced chemical vapor deposition（PECVD）and the membrane thickness, then extracting the ration fuzzy rules between input and output from neural network with fuzzy rule extraction algorithm. The data of analysis was collected from the PECVD machines in the TFT-Array department, the machine parameters including the rate of all gases, temperature, pressure and bias of the chamber in membrane deposition process, etc. After arraying the data, use the parameters of the machine as input, and membrane thickness as output, the neural network model was established with back-propagation network(BPN) 500 times with 500 training data. The model was tested with 200 test data and the error rate was ％ in average, that prove this model can simulate and predict the situation of membrane deposition. The extraction rules were created from the trained neural network with the fuzzy rule extraction algorithm to describe the knowledge with fuzzy logic rules. After checking ,explaining and integration the rules into the knowledge base, then construct a fuzzy expert system enable to provide a reference to the engineers for the work of recipe adjustment.

並列關鍵字

Deposition Membrane Thickness ； Fuzzy Expert System ； Fuzzy Rule Extraction ； Neural Network

參考文獻

[3] 張瑞文,“電漿化學氣相沉積膜厚預測類神經網路模型與規則庫

[1] 盧慶儒, “技術洞察,”Digi Times, 2000.

[8] B. Kim and C. Lee.“Monitoring plasma impedance match characteristics in a multiple inductively coupled plasma process control,”J Vac. Sci. Technol. A, Vol.18, No.1, pp58-62, 2000.

[12] D. E. Rumelhart, G. E. Hinton ,and R. J.Williams,“Learning Internal Representation by Error Propagation in Parallel Distributed Processing,”Vol.1, pp318-362, 1986.

[13] G. Bologna, “A Study on Rule Extraction form Neural Networks Applied to Medical Databases,”pp119-260, 2000.