本研究探討自動光學外觀檢查機（AOFI）在晶圓測試印刷電路板（Wafer Test Board, WTB）檢測中的應用與優勢。隨著半導體產業技術的快速發展，傳統人工檢測方法因其高耗時與不穩定性，已難以滿足高精度與高效率的生產需求。本研究針對現行人工檢測與AOFI系統進行比較，透過實驗法、模擬法與比較法，評估檢測速度、準確性、誤檢率與漏檢率等指標。 實驗結果顯示，AOFI系統可大幅提升檢測效率，將檢測時間由人工檢測的40~50分鐘縮短至5分鐘，同時確保0%漏檢率，避免不良品流入後續製程。雖然AOFI系統存在較高的誤檢率，但透過參數優化與機器學習演算法調整，可有效降低誤報率，進一步提升檢測精準度。此外，本研究驗證AOFI系統在應對晶圓測試板技術與精密線路板檢測時的適應性，並證明其對提升晶圓測試印刷電路板品質管理的可行性。 綜合而言，本研究結果顯示，自動光學外觀檢測技術能顯著提高PCB製造品質、縮短生產週期，並降低人工檢測的不確定性。未來研究可進一步探討深度學習與AI影像分析技術在AOFI檢測中的應用，以提升智能化檢測能力並減少誤檢率，確保半導體製程的高品質與高效率。;This study explores the application and advantages of Automated Optical Final Inspection (AOFI) in the inspection of Wafer Test Boards (WTB). With the rapid advancement of semiconductor technology, traditional manual inspection methods have become increasingly inadequate in meeting the demands for high precision and efficiency due to their time-consuming nature and inconsistent results. This research compares manual inspection with AOFI systems using experimental, simulation, and comparative approaches, evaluating key performance indicators such as inspection speed, accuracy, false detection rate, and missed detection rate. The results indicate that AOFI systems significantly enhance inspection efficiency, reducing inspection time from 40–50 minutes to just 5 minutes while maintaining a 0% missed detection rate, thereby preventing defective products from entering subsequent processes. Although AOFI systems initially exhibit a relatively high false detection rate, this can be effectively reduced through parameter optimization and machine learning algorithm adjustments. Furthermore, this study verifies the adaptability of AOFI systems in detecting advanced Wafer Test Board structures, demonstrating their value in improving quality management. Overall, the findings confirm that AOFI technology enhances PCB manufacturing quality, shortens production cycles, and reduces the uncertainties associated with manual inspection. Future research may explore the integration of deep learning and AI-based image analysis to further improve detection accuracy and enable intelligent defect classification.
