電漿輔助化學氣相沉積設備之射頻阻抗匹配網路的模型建立與控制器設計

Abstract

在半導體製程中，使用射頻功率產生電漿來製造晶圓的技術中，功率的傳輸是一個重要的製程問題，不穩定的功率傳輸將會造成電漿源的不穩定，使得製程無法達成所需的目標，而阻抗匹配網路就扮演著功率傳輸的橋樑，可降低反射功率的產生，增加進入腔體中的傳輸功率，並且提供腔體穩定的操作功率。 本論文將利用化學氣相沉積設備中的阻抗匹配網路，應用神經網路技術建立匹配網路之阻抗模型，並與傳統的建模方式做ㄧ比較，說明神經網路在滯留阻抗處理上的優異性，其次再分析阻抗匹配網路中射頻感測器的信號關係，並賦予其數學模型，最後應用Neuro-Sliding控制技術實現阻抗匹配網路的控制，並與原有的控制技術作一比較，說明新的控制器效能，與射頻感測器信號處理對阻抗匹配網路控制的重要影響。

In semiconductor processing, the Radio Frequency (RF) power source is used to create plasma for wafer fabrication. Power transmission plays a very important role in wafer processing. Unstable power transmission will lead to unstable plasma, and wafer processing can not achieve it's goal. The impedance matching network behaves like a bridge for power transmission. It reduces the reflective power from the chamber, and increases the transmission power into the chamber, and provides a stable operating power. In this thesis, we use an impedance matching network in a plasma enhanced chemical vapor deposition (PECVD) equipment. First, we use neural network approach to establish the impedance model of the matching network, and the result will be compared with the traditional method to explain its elegance in dealing with stray impedance. Next, we analyze the relationship of input/output signals of RF sensor board in the matching network, and create mathematical model accordingly. Finally, we apply Neuro-Sliding mode control theory to control impedance matching network, and compare its result with the existing controller. The performance of new controller design will be discussed, and the importance of RF sensor signal processing to the control of impedance matching network will be high lighted in the conclusion portion of this research.