隨著高效能運算對資料傳輸速率的要求日益嚴苛，共同封裝光學（Co-Packaged Optics， CPO）技術已成為半導體先進封裝的發展核心。然而CPO光纖陣列單元（FAU）的點膠與固化製程涉及高精密與異質材料整合，極易受環境與製程參數波動影響導致量產初期良率低迷。本研究旨在解析製程變異來源、建構多因子統計模型，並推導最佳化參數以建立標準作業程序（SOP）與品質管制計畫。研究首先運用品質管制工具，界定出「內部氣泡」與「表面分層」為兩大關鍵缺陷。隨後導入田口實驗設計法（DOE），採用L9(3^4)直交表針對環境溫度、相對濕度、UV曝光能量強度與UV曝光時間等四項關鍵因子進行多水準實驗。研究中結合光學顯微鏡之數位影像量化與望小特性訊號雜音比（S/N比），並透過變異數分析（ANOVA）精準拆解系統變異。相對濕度為誘發氣泡的絕對因子，而UV曝光能量強度則是導致介面分層的關鍵。經由主效應與等高線操作視窗分析推導出最佳製程參數組合。經產線大樣本確認實驗實證此最佳化參數不僅具備跨批次（Lot-to-lot）之穩健性，更成功將製程綜合良率大幅提升。本研究有效排除了傳統試誤法的盲點，為前瞻矽光子封裝奠定了科學化的量產管制基準。;Background and Objectives: Co-Packaged Optics (CPO) is critical for high-speed computing but faces severe yield bottlenecks during Fiber Array Unit (FAU) packaging due to high sensitivity to process variations. This study aims to identify variance sources， optimize parameters， and establish Standard Operating Procedures (SOPs) for robust mass production.
Methods: "Internal voids" and "interfacial delamination" were identified as primary defects. A Taguchi L9(3^4) orthogonal array was utilized to evaluate four control factors: temperature， relative humidity (RH)， UV exposure intensity， and UV exposure time. Data were analyzed using digital image quantification， signal-to-noise (S/N) ratios， and Analysis of Variance (ANOVA).
Results: ANOVA revealed that RH is the dominant factor for void formation， while UV intensity primarily drives delamination. The optimal parameter combination was determined and Confirmation experiments verified cross-batch robustness， successfully improving the overall yield from 5% to over 40%.
Conclusions: This study replaces traditional trial-and-error approaches with a scientifically validated framework， establishing a highly stable mass production baseline for advanced silicon photonics packaging.
