本研究針對半導體微影製程中的旋轉塗佈技術，進行光阻厚度均勻性之最佳化分析。為有效提升膜厚均勻性並降低實驗成本，採用均勻設計法作為實驗設計工具，分別運用傳統的好格子點法與近代新興的元啟發式法(粒子群演算法與模擬退火法)，針對塗佈量、塗佈轉速、主轉速時間及主轉速等四項關鍵參數進行系統性規劃。透過多元迴歸分析建立平均厚度、標準差與各因子間的預測模型，並以判定係數與F檢定驗證模型的解釋力與顯著性。進一步運用迴歸模型，找出最佳化的製程參數組合，以提升膜厚均勻性與製程穩定性。經實驗結果顯示，好格子點法得出的厚度標準差為91.2 Å，元啟發式法得出的厚度標準差為95.1Å，兩種均勻設計法得出的標準差皆小於原始實驗配置的結果，同時厚度規格也達到規範要求，顯示導入均勻設計法能有效提升光阻厚度的均勻性，且迴歸模型具備高度的預測能力，並展現高度的可靠性。;This study aimed to optimize the uniformity of photoresist thickness in the spin coating process of semiconductor photolithography. To effectively enhance film uniformity and reduce experimental costs, a uniform design method was employed as the experimental design tool. Both the traditional Good Lattice Point method and modern metaheuristic algorithms (Particle Swarm Optimization and Simulated Annealing) were systematically applied to four critical process parameters, namely coating volume, coating speed, main spin duration, and main spin speed. Predictive models correlating average thickness and standard deviation with these factors were developed using multiple regression analysis, with the models’ explanatory power and significance validated through the coefficient of determination and F-test. The regression models were further utilized to identify optimal process parameter combinations, thereby improving both film uniformity and process stability. Experimental results demonstrated that the standard deviation of thickness achieved by the Good Lattice Point method was 91.2 Å, while that obtained via metaheuristic algorithms was 95.1 Å. Both uniform design methods produced standard deviations lower than those obtained from the original experimental setup. Additionally, the film thickness specifications met the required standards. This indicates that introducing uniform design methods can effectively improve the uniformity of photoresist thickness, and the regression model demonstrates high predictive capability and reliability.
