隨著半導體產業在全球供應鏈中的地位日益重要，但晶圓製造技術未來將遭遇物理限制瓶頸，先進封裝技術被視為成為延續摩爾定律的可行解決方案而逐漸被重視。在晶片封裝產業中，雖已經廣泛應用自動光學檢測(AOI)技術於晶圓表面瑕疵檢測，但仍然依賴人工進行瑕疵分類，這在品質一致性、效率以及成本上存在諸多挑戰。鑒於卷積神經網路(CNN)和遷移學習技術在圖像處理與瑕疵分類中的卓越表現，本研究旨在評估並開發一種基於遷移學習的自動光學檢測瑕疵分類模型，並將其應用於晶圓級封裝產品的重佈線層(RDL)製程中。
本研究使用了InceptionV3、ResNet50和DenseNet161三種預訓練模型，並結合Logistic Regression、KNN、SVM和XGBoost等常見分類器進行模型的構建與測試。實驗結果表明，DenseNet161結合Logistic Regression及SVM分類器的混合模型在準確性、穩定性和資源需求達到了適當的平衡，平均分類準確率達89.33%，F1分數達達89.24，且訓練時間大幅減少。本研究為半導體封裝產業提供一種低成本、高效能的技術方案，並有助於提升生產效率和產品品質，也為自動光學檢測系統的改進提供了實踐依據。同時，未來研究可在此基礎上進一步優化模型性能，並探索新型深度學習模型在該領域中的應用可能。

As the semiconductor industry becomes increasingly crucial in the global supply chain, advanced packaging technology has gradually emerged as a viable solution to extend Moore's Law. While automatic optical inspection (AOI) technology has been widely applied in the packaging industry, defect classification still relies heavily on manual processes, leading to challenges in consistency, efficiency, and cost. Given the exceptional performance of Convolutional Neural Networks (CNN) and transfer learning techniques in image processing, this study aims to develop and evaluate a transfer learning-based defect classification model for AOI systems, specifically applied to the redistribution layer (RDL) process in wafer-level packaging products.
This research employs three pre-trained models—InceptionV3, ResNet50, and DenseNet161—combined with common classifiers such as Logistic Regression, KNN, SVM, and XGBoost to construct and test the models. The experimental results indicate that the DenseNet161 model combined with Logistic Regression and SVM classifiers achieves an optimal balance in terms of accuracy, stability, and resource requirements, with an average classification accuracy of 89.33% and an F1 score of 0.8924, along with a significant reduction in training time. This study provides a low-cost, high-efficiency solution for the semiconductor packaging industry, contributing to improved production efficiency and product quality, and offers practical insights for enhancing AOI systems. Future research could further optimize the model's performance and explore the potential applications of novel deep learning models in this field.

中文文獻
[1] 經濟部產業發展署智慧電子產業計畫推動辦公室. "半導體產業現況." https://www.sipo.org.tw/industry-overview/industry-state-quo/semiconductor-industry-state-quo.html (accessed.
[2] 經濟部統計處. "國內生產毛額(名目金額)." https://dmz26.moea.gov.tw/GA/common/Common.aspx?code=A&no=6 (accessed.
英文文獻
[3] G. E. Moore, "Cramming More Components Onto Integrated Circuits," (in Chinese), Proceedings of the IEEE, vol. 86, no. 1, pp. 82-85, 1998, doi: 10.1109/jproc.1998.658762.
[4] S.-L. MA and T.-Q. ZHANG, "Research progress in hybrid bonding technology for advanced packaging applications[J]," (in Chinese), Microelectronics & Computer, vol. 40, pp. 22-42., 2023, doi: 10.19304/J.ISSN1000-7180.2023.0677.
[5] Z. Zhengkai, "ResearchprogressofRDL-firstprocessinadvanced packaging[J]," (in Chinese), Electronics & Packaging, vol. 23, no. 10, p. 100205, 2023, doi: 10.16257/j.cnki.1681-1070.2023.0134.
[6] X. Tong, Z. Yu, X. Tian, H. Ge, and X. Wang, "Improving accuracy of automatic optical inspection with machine learning," (in Chinese), Frontiers of Computer Science, vol. 16, no. 1, 2021, doi: 10.1007/s11704-021-0244-9.
[7] K. Pfab, R. Eichler, A. Mallandur, and M. Rothering, "Data of automated optical inspection of surface-mounted technology electronic production," (in Chinese), Data Brief, vol. 53, p. 110110, Apr 2024, doi: 10.1016/j.dib.2024.110110.
[8] C.-H. Huang et al., "Automated Optical Inspection Method for Light-Emitting Diode Defect Detection Using Unsupervised Generative Adversarial Neural Network," (in Chinese), Crystals, vol. 11, no. 9, 2021, doi: 10.3390/cryst11091048.
[9] B. Ali and M. Marshall, "Automated Optical Inspection (AOI) for FOPLP with Simultaneous Die Placement Metrology," (in Chinese), International Symposium on Microelectronics, vol. 2019, no. 1, pp. 000203-000210, 2019, doi: 10.4071/2380-4505-2019.1.000203.
[10] Gao Huang, Danlu Chen, Tianhong Li, Felix Wu, Laurens van der Maaten, and K. Weinberger, "Multi-Scale Dense Networks for Resource Efficient Image Classification," (in Chinese), ICLR 2018 conference, 2018, doi: ar5iv.org/abs/1703.09844.
[11] S. O. Chishti, S. Riaz, M. BilalZaib, and M. Nauman, "<Self-Driving Cars Using CNN and Q-Learning.pdf>," (in Chinese), IEEE Access, 30 December 2018, doi: 10.1109/INMIC.2018.8595684.
[12] S. Grigorescu, B. Trasnea, T. Cocias, and G. Macesanu, "A survey of deep learning techniques for autonomous driving," (in Chinese), Journal of Field Robotics, vol. 37, no. 3, pp. 362-386, 2019, doi: 10.1002/rob.21918.
[13] H. Liu, C. Wu, and H. Wang, "Real time object detection using LiDAR and camera fusion for autonomous driving," (in Chinese), Sci Rep, vol. 13, no. 1, p. 8056, May 17 2023, doi: 10.1038/s41598-023-35170-z.
[14] D. Komura and S. Ishikawa, "Machine Learning Methods for Histopathological Image Analysis," (in Chinese), Comput Struct Biotechnol J, vol. 16, pp. 34-42, 2018, doi: 10.1016/j.csbj.2018.01.001.
[15] K. Fukushima, "Neocognitron: a self organizing neural network model for a mechanism of pattern recognition unaffected by shift in position," (in Chinese), Biol Cybern, vol. 36, no. 4, pp. 193-202, 1980, doi: 10.1007/BF00344251.
[16] Y. LeCun et al., "Backpropagation Applied to Handwritten Zip Code Recognition," (in Chinese), Neural Computation, vol. 1, no. 4, pp. 541-551, 1989, doi: 10.1162/neco.1989.1.4.541.
[17] C. Szegedy et al., "Going deeper with convolutions," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 1-9.
[18] S. Ioffe and C. Szegedy, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," in International conference on machine learning, 2015: pmlr, pp. 448-456.
[19] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, "Rethinking the inception architecture for computer vision," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2818-2826.
[20] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, "Densely connected convolutional networks," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 4700-4708.
[21] Q. Yang and X. Wu, "10 challenging problems in data mining research," (in Chinese), International Journal of Information Technology & Decision Making, vol. 5, no. 04, pp. 597-604, 2006.
[22] S. Thrun, "Is learning the n-th thing any easier than learning the first?," (in Chinese), Advances in neural information processing systems, vol. 8, 1995.
[23] L. Pratt and B. Jennings, "A survey of transfer between connectionist networks," (in Chinese), Connection Science, vol. 8, no. 2, pp. 163-184, 1996.
[24] R. Caruana, "Multitask learning," (in Chinese), Machine learning, vol. 28, pp. 41-75, 1997.
[25] S. J. Pan and Q. Yang, "A survey on transfer learning," (in Chinese), IEEE Transactions on knowledge and data engineering, vol. 22, no. 10, pp. 1345-1359, 2009.
[26] H. Yang, S. Mei, K. Song, B. Tao, and Z. Yin, "Transfer-Learning-Based Online Mura Defect Classification," (in Chinese), IEEE Transactions on Semiconductor Manufacturing, vol. 31, no. 1, pp. 116-123, 2018, doi: 10.1109/tsm.2017.2777499.
[27] M. K. Ferguson, A. Ronay, Y. T. Lee, and K. H. Law, "Detection and Segmentation of Manufacturing Defects with Convolutional Neural Networks and Transfer Learning," (in Chinese), Smart Sustain Manuf Syst, vol. 2, 2018, doi: 10.1520/SSMS20180033.
[28] K. Imoto, T. Nakai, T. Ike, K. Haruki, and Y. Sato, "A CNN-Based Transfer Learning Method for Defect Classification in Semiconductor Manufacturing," (in Chinese), IEEE Transactions on Semiconductor Manufacturing, vol. 32, no. 4, pp. 455-459, 2019, doi: 10.1109/tsm.2019.2941752.
[29] J. Liu, F. Guo, H. Gao, M. Li, Y. Zhang, and H. Zhou, "Defect detection of injection molding products on small datasets using transfer learning," (in Chinese), Journal of Manufacturing Processes, vol. 70, pp. 400-413, 2021, doi: 10.1016/j.jmapro.2021.08.034.
[30] J. A. Mat Jizat, A. P. Abdul Majeed, A. F. Ab. Nasir, Z. Taha, E. Yuen, and S. X. Lim, "Evaluation of the transfer learning models in wafer defects classification," in Recent Trends in Mechatronics Towards Industry 4.0: Selected Articles from iM3F 2020, Malaysia, 2022: Springer, pp. 873-881.
[31] K. H. Ang et al., "Deep Learning-Based Silicon Wafer Defect Classification: A Performance Comparison of Pretrained Networks," in International Conference on Mechatronics and Intelligent Robotics, 2023: Springer, pp. 129-139.
[32] A. K. Abdelmoula, "Bank credit risk analysis with k-nearest-neighbor classifier: Case of Tunisian banks," (in Chinese), Accounting and Management Information Systems, vol. 14, no. 1, p. 79, 2015.
[33] R. Kumar, S. M. Naik, V. D. Naik, S. Shiralli, V. Sunil, and M. Husain, "Predicting clicks: CTR estimation of advertisements using logistic regression classifier," in 2015 IEEE international advance computing conference (IACC), 2015: IEEE, pp. 1134-1138.
[34] A. Prabhat and V. Khullar, "Sentiment classification on big data using Naïve Bayes and logistic regression," in 2017 International Conference on Computer Communication and Informatics (ICCCI), 2017: IEEE, pp. 1-5.
[35] T. Pranckevičius and V. Marcinkevičius, "Comparison of naive bayes, random forest, decision tree, support vector machines, and logistic regression classifiers for text reviews classification," (in Chinese), Baltic Journal of Modern Computing, vol. 5, no. 2, p. 221, 2017.
[36] S. S. Dhaliwal, A.-A. Nahid, and R. Abbas, "Effective intrusion detection system using XGBoost," (in Chinese), Information, vol. 9, no. 7, p. 149, 2018.
[37] N. Manju, B. Harish, and V. Prajwal, "Ensemble feature selection and classification of internet traffic using XGBoost classifier," (in Chinese), International Journal of Computer Network and Information Security, vol. 11, no. 7, p. 37, 2019.
[38] Y. Wang and X. S. Ni, "A XGBoost risk model via feature selection and Bayesian hyper-parameter optimization," (in Chinese), arXiv preprint arXiv:1901.08433, 2019.
[39] I. Hanif, "Implementing extreme gradient boosting (xgboost) classifier to improve customer churn prediction," ed, 2020.
[40] K. Budholiya, S. K. Shrivastava, and V. Sharma, "An optimized XGBoost based diagnostic system for effective prediction of heart disease," (in Chinese), Journal of King Saud University-Computer and Information Sciences, vol. 34, no. 7, pp. 4514-4523, 2022.
[41] S. Marianingsih and F. Utaminingrum, "Comparison of support vector machine classifier and Naïve Bayes classifier on road surface type classification," in 2018 International Conference on Sustainable Information Engineering and Technology (SIET), 2018: IEEE, pp. 48-53.
[42] K. Mittal, G. Aggarwal, and P. Mahajan, "Performance study of K-nearest neighbor classifier and K-means clustering for predicting the diagnostic accuracy," (in Chinese), International Journal of Information Technology, vol. 11, pp. 535-540, 2019.
[43] Y. Narayan, "Comparative analysis of SVM and Naive Bayes classifier for the SEMG signal classification," (in Chinese), Materials Today: Proceedings, vol. 37, pp. 3241-3245, 2021.
[44] M. Amjad, I. Ahmad, M. Ahmad, P. Wróblewski, P. Kamiński, and U. Amjad, "Prediction of pile bearing capacity using XGBoost algorithm: modeling and performance evaluation," (in Chinese), Applied Sciences, vol. 12, no. 4, p. 2126, 2022.
[45] S. Zhang, Y. Yuan, Z. Yao, X. Wang, and Z. Lei, "Improvement of the performance of models for predicting coronary artery disease based on XGBoost algorithm and feature processing technology," (in Chinese), Electronics, vol. 11, no. 3, p. 315, 2022.
[46] Y. Yuan-Fu, "A deep learning model for identification of defect patterns in semiconductor wafer map," in 2019 30th Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC), 2019: IEEE, pp. 1-6.
[47] S. Kumaresan, K. S. J. Aultrin, S. S. Kumar, and M. D. Anand, "Transfer Learning With CNN for Classification of Weld Defect," (in Chinese), IEEE Access, vol. 9, pp. 95097-95108, 2021, doi: 10.1109/access.2021.3093487.
[48] J. A. M. Jizat, A. P. A. Majeed, A. F. A. Nasir, Z. Taha, and E. Yuen, "Evaluation of the machine learning classifier in wafer defects classification," (in Chinese), ICT Express, vol. 7, no. 4, pp. 535-539, 2021.
[49] L. S. Xuen et al., "The classification of wafer defects: a support vector machine with different DenseNet transfer learning models evaluation," in International conference on robot intelligence technology and applications, 2022: Springer, pp. 304-309.
[50] S. K. Behera, S. P. Dash, R. Amat, and P. K. Sethy, "Wafer defect identification with optimal hyper-parameter tuning of support vector machine using the deep feature of ResNet 101," (in Chinese), International Journal of System Assurance Engineering and Management, vol. 15, no. 3, pp. 1294-1304, 2024.