碳化矽是一種具有優異電子特性和高熱穩定性的半導體材料，4H導電型碳化矽基板作為高功率半導體元件製造中的最常用也最重要的材料，其中晶體缺陷對對後續元件的品質有很大的影響，輕則元件效率降低重則元件失效，所以提高對晶體缺陷的辨識精度為重中之重的任務，而KOH蝕刻控制不易，缺陷大小變化程度很大，本研究使用Faster RCNN來對4H-N型碳化矽半導體的晶體缺陷進行自適應辨識，使用Faster RCNN卷積神經網絡配合預訓練模型ResNet50作為訓練晶體缺陷辨識的模型架構，使用三種KOH蝕刻的持溫時間來揭露碳化矽晶體缺陷，將其區分為大、正常和小三種尺寸大小的缺陷，並使用G公司的S機器來對碳化矽晶圓進行缺陷檢測掃描蒐集缺陷照片並做為訓練資料。
接著對缺陷照片上的缺陷類型做分類並對其做標記，使用卷積神經網絡學習後，對十組碳化矽晶體缺陷照片進行缺陷辨識測試，辨識後再使用人工檢查補償錯誤對其進行漏抓率、誤判率和正確率的計算，針對辨識結果進行分析並更改標記策略和調整資料庫缺陷照片，觀察模型權重測試後的正確率變化，最後再和G公司的S機器的辨識準確率做比較，以補足超出S機器檢測範圍之正確率的缺失。

Silicon carbide is a semiconductor material with excellent electronic properties and high thermal stability. Among various materials used in the manufacturing of high-power semiconductor devices, 4H-N type silicon carbide substrates are the most commonly used and critical.
Crystal defects in these substrates significantly impact the quality of subsequent semiconductor components, leading to decreased efficiency or even device failure.
Thus, enhancing the precision of crystal defect recognition is of paramount importance.
KOH etching is not easy to control, and the defect size varies greatly.
In this study, Faster RCNN (Region-based Convolutional Neural Networks) is used for adaptive recognition of crystal defects in 4H-N type silicon carbide semiconductors.
Faster RCNN, a deep learning approach, is combined with the pre-trained ResNet50 model to construct the defect recognition framework.
To reveal crystal defects, three different durations of KOH (Potassium Hydroxide) etching are employed, categorizing the defects into three size classes: large, normal, and small.
And to build the recognition model, defect images are collected by scanning the silicon carbide wafers using G Company's S machine, which serves as the defect inspection system.
Subsequently, the defect types in the images are classified and annotated and employ a convolutional neural network for learning.
The model is tested on ten sets of silicon carbide crystal defect images, and any errors are manually checked and compensated.
Key performance metrics, such as recall rate, false positive rate, and accuracy, are calculated to evaluate the recognition results.
Through analysis of the recognition outcomes, the annotation strategy and refine the database of defect images are adjusted as needed. Additionally, the changes in accuracy by testing the model's weights are observed. Finally, the recognition accuracy of our Faster RCNN-based approach with that of G Company's S machine to address any deficiencies in recognition accuracy beyond the S machine's detection range is compared.
This research integrates machine learning and deep learning techniques, and its outcomes are expected to contribute significantly to defect identification and quality control in the manufacturing of high-power semiconductor devices using 4H-n type silicon carbide substrates.

[1] Chen, PC., Miao, WC., Ahmed, T. et al. Defect Inspection Techniques in SiC. Nanoscale Res Lett 17, 30 (2022). https://doi.org/10.1186/s11671-022-03672-w
[2] G.A. Bootsma, W.F. Knippenberg, G. Verspui, Phase transformations, habit changes and crystal growth in SiC, Journal of Crystal Growth,
Volume 8, Issue 4, 1971, Pages 341-353 ISSN 0022-0248
[3] T. Kimoto, “Bulk and epitaxial growth of silicon carbide,” Progress in Crystal Growth and Characterization of Materials, vol. 62, no. 2, pp. 329–351, Jun. 2016, doi: 10.1016/J.PCRYSGROW.2016.04.018.
[4] 趙培勛, “導輪磨耗於線鋸切割影響研究”, 國立臺灣科技大學機械工程系碩士學位論文, 2011
[5] 李承遠。「碳化矽基板碳包裹物衍生磊晶缺陷研究」。碩士論文，國立臺灣科技大學機械工程系，2023。https://hdl.handle.net/11296/tdwe6j。
[6] Lasertec Co., “Lasertec SiC Wafer Inspection and Review System SICA88.”
[7] He, Kaiming, XiangyuZhang, ShaoqingRen, andJianSun. 2016. “Deep Residual Learning for Image Recognition.” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition 2016-Decem: 770–78.
[8] D. Erhan, C. Szegedy, A. Toshev, D. Anguelov, "Scalable object detection using deep neural networks.", Proceedings of the IEEE conference on computer vision and pattern recognition(CVPR), 2014, 2147-2154.
[9] S. Ren, K. He, R. Girshick, J. Sun, "Faster r-cnn: Towards real-time object detection with region proposal networks.", Proceedings of the 28th International Conference on Neural Information Processing Systems, Vol.1, 2015, 91–99
[10] K. He, X. Zhang, S. Ren, J. Sun, "Spatial pyramid pooling in deep convolutional networks for visual recognition.", IEEE transactions on pattern analysis and machine intelligence, Vol.37, No.9, 2015, 1904-1916.
[11] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436-44, May 28, 2015.
[12] R. Jain, R. Kasturi, and B. G. Schunck, Machine vision: McGraw-hill New York, 1995.
[13] D. G. Lowe, "Object recognition from local scale-invariant features." pp. 1150-1157.
[14] P. F. Felzenszwalb, and D. P. Huttenlocher, “Pictorial structures for object recognition,” International Journal of Computer Vision, vol. 61, no. 1, pp. 55-79, Jan, 2005.
[15] R. Girshick, J. Donahue, T. Darrell, and J. Malik, "Rich feature hierarchies for accurate object detection and semantic segmentation." pp. 580-587.
[16] R. Girshick, "Fast r-cnn." pp. 1440-1448.
[17] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” IEEE Trans Pattern Anal Mach Intell, vol. 39, no. 6, pp. 1137-1149, Jun, 2017.
[18] H. Jiang, and E. Learned-Miller, "Face detection with the faster R-CNN." pp. 650-657.
[19] C. M. Bishop, Neural networks for pattern recognition: Oxford university press, 1995.
[20] Violinleechan. (2019). Non-Maximum Suppression Algorithm（Non-Maximum Suppression，NMS）. Available at：https://www.twblogs.net/a/5c856cedbd9eee35fc140c18
[21] C. Papageorgiou, M. Oren, and T. Poggio, “A general framework for object detection,” Proceedings of the IEEE International Conference on Computer Vision, 1998.
[22] S. Ren, K. He, R. Girshick, and J. Sun,, Faster R-CNN: Towards real-time object detection with region proposal networks, in NIPS, 2015.
[23] R. Girshick, “Fast R-CNN.” Proceedings of the IEEE International Conference on Computer Vision, 1440-1448, 2015.
[24] Zhang, K., Zuo, W., Chen, Y., Meng, D., & Zhang, L. (2017). Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising. IEEE Transactions on Image Processing, 26(7), 3142-3155.