應用改良式之基因遺傳演算法於化學氣相沈積製程之最佳化

Abstract

本論文研究探討以一「改良式之基因遺傳演算法則」應用於CVD磊晶 沈積製程之最佳化。基因遺傳演算法為一基於生物演化概念而發展成之最 佳化搜尋法則，在此，我們整合一些策略如：人口菁英政策及順位揀選法 ，配合多點交配法以增加搜尋速率；可自我調適之運算子機率以避免過早 收斂、及解決內部參數設定的問題；結合混合式之基因運算子、移民運算 子及具啟發式之適應函數型態以加強其微調能力，來改善傳統基因法則的 種種缺點。我們除了應用「改良式之基因遺傳演算法則」於搜尋可解析函 數以證明改良效果之外；並應用於有外界干擾雜訊下的化學氣相沈積製程 之搜尋，成功且有效率地搜尋到一組製程之最佳解，使晶元沉積厚度變異 量在此製造過程中降到最低，以提高整體製程良率。相較於其他已提出針 對同一製程之搜尋方式而言，本論文不僅能有更佳的搜尋結果之外，也對 今後製程最佳化的方式，提供了另一新的思維方向。 A vertical chemical vapor deposition process (CVD) optimization method using modified geneticalgorithms (MGA) has been proposed. Genetic algorithms (GA) are a computational optimization paradigm modeled after biological evolution concept. Strategies such as: elitist with ranking selection reproduction scheme and multiple points crossover are used to raise the search efficiency of the traditional GA. Self-adjusted operator probability not only helps to avoid premature but also define parameters automatically. Moreover, we integrate hybrid genetic operator, immigration operator, and heuristic fitness function to enhance its local fine tuning ability. In order to prove the improvement results, we initially optimize several highly nonlinear functions with MGA, then, with a well-defined fitness function, the optimization procedure has been successfully applied to the CVD process with various noise level. Through the optimal solution, we obtained the thickness in deposition layers which is more uniformly distributed over the wafers. These results demonstrate the superiority of the proposed optimization solution in comparison with other existing optimization algorithms.