結合遞迴最小平方動態模型調變及最小變異控制器進行PECVD機台先進製程控制

Abstract

本論文利用實驗設計法找出電漿輔助化學氣相沉積(Plasma Enhanced Chemical Vapor Deposition，PECVD)機台輸入參數(製程配方，recipes)、內部參數及輸出參數(晶圓良率變數：膜厚與均勻度)間之關係，利用主成份分析法將PECVD機台內部參數(Plasma Index，Straatum sensor取得之機台內部參數)加以轉換成相互間獨立之新的參數(主成)，設定這些參數的容許範圍，可做到即時錯誤偵測的目的。利用實驗設計法(Design of experiment，DOE)找出輸出輸入間之模型，從機台給定之輸入參數值與量測機台得到之輸出參數值，以遞迴最小平方法(Recursive Least Square，RLS)，做模型係數的動態調變，可使所找出來的模型更能適應機台不確定性的變異，最後加入非線性最小變異控制器(Minimum Variance Controller，MVC)，使機台的輸入參數可以在改變量最小的情況下，讓輸出達到所期望的目標值，可改善MEWMA(Multivariate EWMA)控制器無法解決之輸出輸入間的非線性關係，達到提升PECVD製程良率的目的。

In this thesis, the relationships among input recipes, Plasma Index (measured from Straatum sensor) and output variables of PECVD equipment are found by utilizing DOE method. The Plasma Index is projected into another space to reduce the dimensions by using Principal Component Analysis (PCA), and get some principal components which are independent to each others. Set some constraints in PCA score plots, and use PCA score plots to find if there is any fault or not (Fault Detection). Further, the nonlinear models between input recipes and output variables (I/O) are found by using DOE method. Finally, use RLS dynamical model tuning method and MVC to tune the coefficients of the nonlinear models and make the output variables in a good situation. In this thesis, we find a solution to solve the nonlinear relationships between I/O and promote the quality of PECVD process.