本研究針對超穎介面中，因為幾何結構與材料數據匱乏而導致設計困難，提出了解決方法以提升設計效率。為此我們採用了條件深度卷積生成對抗網路，通過學習光譜數據來訓練模型，實現了使用光譜特性即可預測幾何結構的能力。為了提高運算效率，我們使用了嚴格耦合波展開法，結合GPU加速運算，大幅提高光學響應計算的速度。我們構建了多種幾何結構與材料特性的資料庫，涵蓋不同波長下的穿透光譜資訊，其中包括x偏振光、y偏振光、左旋圓偏振光與右旋圓偏振光入射時的穿透光譜。通過嚴格耦合波展開法與條件深度卷積生成對抗網路結合，我們在設計階段能夠快速生成並驗證符合需求的奈米結構，大幅減少了傳統設計所需的計算時間與資源。這項基於人工智慧的逆向設計方法為超穎介面的設計提供了一種高效、精確的替代方案，並有望在光學元件的設計與開發中發揮重要作用。隨著AI模型的進一步完善，我們期望該方法能推動光學元件設計領域的發展，為未來的科學研究與工業應用提供強有力的支持。

This thesis proposes a solution to the scarcity of geometric structures and material data while designing metasurface, so that the design efficiency can be improved. We employed a conditional deep convolutional generative adversarial network (cDCGAN), training the model by learning spectral data to enable the prediction of geometric structures based on spectral characteristics. To improve computational efficiency, we utilized the rigorous coupled-wave analysis (RCWA) method combined with GPU-accelerated computing, significantly increasing the speed of optical response calculations. We constructed a database of various geometric structures and material properties, encompassing transmission spectral information across different wavelengths, including transmission spectrum under incident x-polarized light, y-polarized light, left-handed circularly polarized light, and right-handed circularly polarized light. By integrating the RCWA with the cDCGAN, we can rapidly generate and validate nanoscale structures that satisfy design requirements during the design phase, greatly reducing the computational time and resources required by traditional methods. This artificial intelligencebased inverse design approach provides an efficient and precise alternative for designing metasurface and is expected to play a significant role in the design and development of optical components. With further advancements in AI models, we anticipate that this method will drive progress in the field of optical component design, offering robust support for future scientific research and industrial applications.

[1] Y. D. Shah, J. Grant, D. Hao, M. Kenney, V. Pusino, and D. R. Cumming, "Ultra-narrow line width polarization-insensitive filter using a symmetry-breaking selective plasmonic metasurface," Acs Photonics, vol. 5, no. 2, pp. 663-669, 2018.
[2] V. E. Babicheva and A. B. Evlyukhin, "Resonant lattice Kerker effect in metasurfaces with electric and magnetic optical responses," Laser & Photonics Reviews, vol. 11, no. 6, p. 1700132, 2017.
[3] T. Ellenbogen, K. Seo, and K. B. Crozier, "Chromatic plasmonic polarizers for active visible color filtering and polarimetry," Nano letters, vol. 12, no. 2, pp. 1026-1031, 2012.
[4] C.Y. Yu, Q.C. Zeng, C.J. Yu, C.Y. Han, and C.M. Wang, "Scattering analysis and efficiency optimization of dielectric Pancharatnam–Berry-Phase metasurfaces," Nanomaterials, vol. 11, no. 3, p. 586, 2021.
[5] C. Yeung, R. Tsai, B. Pham, B. King, Y. Kawagoe, D. Ho, J. Liang, M. W. Knight, and A. P. Raman, "Global inverse design across multiple photonic structure classes using generative deep learning," Advanced Optical Materials, vol. 9, no. 20, p. 2100548, 2021.
[6] C. C. Nadell, B. Huang, J. M. Malof, and W. J. Padilla, "Deep learning for accelerated all-dielectric metasurface design," Optics express, vol. 27, no. 20, pp. 27523-27535, 2019.
[7] R. Zhu, J. Wang, X. Fu, X. Liu, T. Liu, Z. Chu, Y. Han, T. Qiu, S. Sui, and S. Qu,"Deep-learning-empowered holographic metasurface with simultaneously customized phase and amplitude," ACS Applied Materials & Interfaces, vol. 14, no. 42, pp. 48303-48310, 2022.
[8] P. La-aiddee, P. Sangwongngam, L. Wuttisittikulkij, and P. Vanichchanunt, "A Generative Adversarial Network-Based Approach for Reflective-Metasurface Unit-Cell Synthesis in mmWave Bands." 2024 International Technical Conference on Circuits/Systems, Computers, and Communications (ITC-CSCC). IEEE, 2024.
[9] R. Yong, "L1, L2 Regularization原理與L1 Regularization的稀疏性," Medium, Aug. 12, 2021. [Online]. Available:https://roger010620.medium.com/l1-l2-regularization-%E5%8E%9F%E7%90%86-5beb1b68b955
[10] E. Eunice, "Day 15 模型正則化方法(1)：L1 和 L2 正則化," iT 邦幫忙, 2024. [Online]. Available:https://ithelp.ithome.com.tw/m/articles/10359655
[11] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, "Dropout: a simple way to prevent neural networks from overfitting," The journal of machine learning research, vol. 15, no. 1, pp. 1929-1958, 2014.
[12] R. Lu, "Learning Model: Gradient Descent - learning rate 優化器介紹與選擇," Medium, Mar. 25, 2019. [Online]. Available:https://medium.com/ai%E5%8F%8D%E6%96%97%E5%9F%8E/learning-model-gradient-descent-%E5%84%AA%E5%8C%96%E5%99%A8%E4%BB%8B%E7%B4%B9%E8%88%87%E9%81%B8%E6%93%87-8d8e49554e18
[13] J.-Y. Shiau, "為什麼 Adam 常常打不過 SGD？癥結點與改善方案," Medium, Apr. 26, 2020. [Online]. Available:https://medium.com/ai-blog-tw/deep-learning-%E7%82%BA%E4%BB%80%E9%BA%BCadam%E5%B8%B8%E5%B8%B8%E6%89%93%E4%B8%8D%E9%81%8Esgd-%E7%99%A5%E7%B5%90%E9%BB%9E%E8%88%87%E6%94%B9%E5%96%84%E6%96%B9%E6%A1%88-fd514176f805
[14] A. Bouguettaya, A. Kechıda, and A. M. Taberkıt, "A survey on lightweight CNN-based object detection algorithms for platforms with limited computational resources," International Journal of Informatics and Applied Mathematics, vol. 2, no. 2, pp. 28-44, 2019.
[15] V. Dumoulin and F. Visin, "A guide to convolution arithmetic for deep learning," arXiv preprint arXiv:1603.07285, 2016.
[16] A. Zafar, M. Aamir, N. Mohd Nawi, A. Arshad, S. Riaz, A. Alruban, A. K. Dutta, and S. Almotairi, "A comparison of pooling methods for convolutional neural networks," Applied Sciences, vol. 12, no. 17, p. 8643, 2022.
[17] Ann, "Activation Functions — Sigmoid & ReLU & tanh & LeakyReLU & ELU," Medium, Aug. 9, 2021. [Online]. Available:https://medium.com/@adea820616/activation-functions-sigmoid-relutahn-20e3ae726ae
[18] S. Ioffe, "Batch normalization: Accelerating deep network training by reducing internal covariate shift," arXiv preprint arXiv:1502.03167, 2015.
[19] N. Bjorck, C. P. Gomes, B. Selman, and K. Q. Weinberger, "Understanding batch normalization," Advances in neural information processing systems, vol. 31, 2018.
[20] S. Durr, Y. Mroueh, Y. Tu, and S. Wang, "Effective dynamics of generative adversarial networks," Physical Review X, vol. 13, no. 4, p. 041004, 2023.
[21] H. Borse, "CNN — quick learn," Medium: Analytics Vidhya, May 9, 2020. [Online].Available:medium.com/analytics-vidhya/cnn-quick-learn-12dced578b01
[22] Z. Li, F. Liu, W. Yang, S. Peng, and J. Zhou, "A survey of convolutional neural networks: analysis, applications, and prospects," IEEE transactions on neural networks and learning systems, vol. 33, no. 12, pp. 6999-7019, 2021.
[23] S. Yan, "Understanding LSTM networks," Online). Accessed on August, vol. 11, 2015.
[24] Brandon, "How Recurrent Neural Networks and Long Short-Term Memory Work," Example Website, Jun. 27, 2017. [Online]. Available: https://e2eml.school/how_rnns_lstm_work.html
[25] C.-H. Liu, "深度學習 Paper 系列 (03): Generative Adversarial Networks," Medium, Apr. 15, 2023. [Online]. Available:https://tomohiroliu22.medium.com/%E6%B7%B1%E5%BA%A6%E5%AD%B8%E7%BF%92paper%E7%B3%BB%E5%88%97-03-generative-adversarial-networks-a64e89387b62
[26] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, "Attention is all you need," Advances in Neural Information Processing Systems, 2017.
[27] ENLITECH, "AM1.5 ". [Online]. Available: https://enlitechsy.com/solar-simulator01/fbclid=IwY2xjawHmMPJleHRuA2FlbQIxMAABHXEdvsCRe1BRWGBDS45HaYEVxxtb7onoDP5Hdmr8wyoZWAiIAsR80C4bgg_aem_WQuiz9UIIJ7xHt6jdWXwIw
[28] Thorlabs, SLS201L - Stabilized Fiber-Coupled Light Source w/ Universal Power Adapter, 360 - 2600 nm, 1/4"-20 Taps, Thorlabs. [Online]. Available: https://www.thorlabs.com