半導體製程控制技術為目前產學界著重之領域,本論文以批次製程控 制(Run By Run Process Control;RBR製程控制)作為研究之基礎,將 通用迴歸神經網路(GRNN)應用在批次製程控制上,GRNN演算法裡,因子與 因子間的關係不事先被假定,而是由歷史資料讓因子本身去學習彼此之間 的關係,從得到的結果來應用於特定問題的預測與控制。本研究利用GRNN 具有逼近因子間函數的能力來判斷製程的穩定狀態,並導入一新的方法於 RBR製程控制。本論文中,利用模擬的方法,於製程中加入漂移的因子, 並從模擬得知GRNN可以在少量樣本的情況下,快速學習製程變動的趨勢, 並判斷製程是否需要調整,提早偵測出異常。
In this thesis, we propose a new method to monitor the run by run (RBR) process control by general regression neural network (GRNN). In The GRNN, it’s not necessary to assume the functional form between factors, the relationship between variables is formed through learning historical data.. The result of GRNN can be used to predict or manage in many areas. The GRNN RBR process control model is established, the performance of the established model is studied by simulation. Simulation results show that GRNN can learn the actual trend of the process rapidly even in few data, then a manager can take proper strategy promptly to lower the risk.