多響應系統之製程配方最佳化

Process Recipe Optimization in Multiple Response System

江孟龍(Meng-Lung Chiang)

指導教授 ： 張永鵬

中原大學/工學院/機械工程研究所/碩士(2002年)

https://doi.org/10.6840/cycu200200486

摘要

在市場需求瞬息萬變的現代社會裡，新技術、新產品導致製程必須不斷更新。為了縮短開發時程以提升企業競爭力，快速而有效的尋求最佳製程配方已成為各界關注的焦點。在實際的製程中，產品通常包含數個規格條件（如：尺寸、抗壓性），在此多響應系統的問題中，各個輸出的目標值之間可能相互衝突，因此必須同時考慮每個響應來求解最佳配方。本研究係利用經由DOE輸入-輸出對資料訓練的RBF類神經網路，配合渴望函數以尋找多響應系統中，符合輸出參數條件的最佳配方之智慧型製程配方最佳化技術。其主要研究內容分為兩個部份：
一、函數近似部份：
第一部份利用以DOE法取得的輸入-輸出對來訓練RBF類神經網路，建立製程模型的輸入輸出關係。模擬結果顯示，RBF類神經網路能夠快速的達到函數近似，且具有良好的準確性。

二、製程配方最佳化部份：
由第一部份所得之結果，再利用渴望函數將多響應系統求解最佳化的問題簡化為求解總體渴望函數的最大值。

本研究在製程配方最佳化上，具有下列之優點：
（一） 利用RBF類神經網路可以快速建立系統模型，可針對個別機台狀態之差異做配方最佳化。
（二） 當機台累積的實驗數據增加時，可利用新的實驗數據將類神經網路做修正，以獲得更準確的製程模型和最佳配方；並建立製程知識庫。

關鍵字

最佳化 ； 多響應系統 ； 配方 ； 製程 ； 製程參數

並列摘要

Now a rapid change about the market, new techniques and new products resulted in new process. In order to reduce the time of develop the new process and promote the enterprises competitiveness, how to find the best process recipe effective is the focal of all circles.
In actual process, good product include server specifications, and these specifications relative to the recipes maybe conflict. Therefore we should consider all responds at the same time. This article is use DOE data to train RBF neural network, immediately to find the optimum recipe in multiple response system.
There are two topics in this study. In the first part, we use DOE data to train RBF neural network to construct process model. The result presented RBF neural network can approximate function effective. In the second part, we use the desirability function to solve the optimization problem in multiple response system. This study has two superiorities:
(1) Use RBF neural network can construct process model quickly and accurately, so we can find the optimal recipe in individual machine.
(2) When we get more experimental data, we can amendment neural network to obtain process model and optimal recipe more accurately.

並列關鍵字

optimization ； process ； multiple response ； recipe

參考文獻

6.W.S.McCulloch, and W.Pitts, “A logical calculus of the ideas immanent in nervous activity”, Bulletin of Mathematical Biophysics, vol.5, pp.115-133, (1943).

7.F.Rosenblatt, “The perceptron : A probabilistic model for information storage and organization in the brain”, Phychological Review, vol.65, pp.386-408, (1958).

8.D.E.Rumelhart, and J.L.McClelland, “Parallel Distributed Processing”, Cambridge, MA:MIT Press, vol.1, (1986).

10.J.Moody and C.Darken, “Fast learning in networks of locally-tuned processing units”, Neural Compute, vol.1, pp.281-294, (1989).

13.D.Pantić, S.Milenković, T.Trajković, V.Litovski and N.Stojadinović, “Inverse Modeling of Semiconductor Manufacturing Process”, 20th International Conference on Microeletronics, Vol.1, pp.321-326, Sep. (1995).