隨著摩爾定律，集成電路上可放置的晶體管數目隨著製程技術精進而倍增，半導體產業快速發展，晶片也從28奈米，逐漸濃縮到7奈米甚至3奈米，晶片尺寸的縮小導致熱密度急遽上升，熱管理問題變成各產業不可輕視的一環，「熱遮罩」即是管理廢熱的方法之一。熱遮罩具有引導及屏蔽熱流，不僅能使得內部保護區不受外部溫度場干擾，也能達到內部熱不會影響到外部溫度場的隱身效果。熱遮罩發展起源是轉換理論，轉換理論結合熱學超材料擁有非均質和各項異性的特性，但熱超材料製程過程繁雜取得不易，慢慢演進雙層理論，雙層理論僅使用兩種均質且均勻的自然材料，進行內外層設計即可達到熱隱形效果，但雙層理論內層遮罩必須假設內層為完美絕熱材料即熱導率必須為零，且不同熱遮罩尺寸所需自然材料取得成本不一，在實際運用上有所困難，因此本研究提出利用機器學習設計雙層熱遮罩，文中討論在多個不同幾何結構的雙層熱遮罩，使其達到熱隱身效果，遮罩不但內層不受限於雙層理論之完美絕熱條件，也能利用常見自然材料製程成本降低，實際工程能較易實現。本文研究使用機器學習中深度學習來設計在不同自然材料雙層熱遮罩中分別該有的層厚，機器學習在計算上大大減少了時間成本，也能在應用上更加實用且直觀的設計熱遮罩該有的層厚，最終也透過數值模擬方法驗證這種方法的準確性。本文末也透過舉例了幾個範例說明該方法在多種不同幾何形狀且無解析解的情形下皆能準確地預測出熱遮罩對應各層層厚，未來在各產業與工程應用只要有需要遮罩來管理熱或是隱身效果，此方法都能快速且準確地計算出所需的參數。

Thermal manipulation has garnered significant attention in scientific research owing to its vast potential for novel applications, including cloaking, illusion creation, and sensing. However, thermal cloak has been extensively studied by many scholars. Among them, bilayer theory uses the assumption that the inner layer is an adiabatic material with zero thermal conductivity. After setting the geometric size, the thermal conductivity of the outer layer is predicted, but the predicted thermal conductivity will produce corresponding to natural materials, which do not exist in nature, so thermal metamaterials are needed to satisfy them. In order to break through the above problems, this research innovatively proposes a machine learning based thermal cloak that uses known and common natural materials as the material for the thermal mask to optimize the thickness of the double layer to achieve the effect of thermal invisibility. By establishing a deep learning neural network, using the geometric size and background heat flux as a database, the relationship between the geometric size of each layer and thermal stealth performance can be intelligently explored.
In the end, thermal cloak using different natural materials can successfully predict the thickness of the outer layer of the thermal cloak and guide the heat flow to bypass the invisible object. The external heat will not affect the internal area, and the internal heat will not affect the peripheral background. In the case of circles and ellipses with analytical solutions, the error between the predicted value of deep learning and the analytical solution is less than 0.5%. Even the geometry that achieves thermal cloaking is perfectly predicted in the rounded rectangle thermal cloak for which there is no analytical solution. The results of this paper also verify the accuracy of this method through numerical simulation. In the future, as long as there is a need for cloaks to manage heat or stealth effects in various industries and engineering applications, this method can quickly and accurately calculate the required parameters.

[1] Guenneau, S., Amra, C., and Veynante, D., Transformation thermodynamics: cloaking and concentrating heat flux. Optics Express. 20(7), 8207-8218,(2012).
[2] Han, T., Bai, X., Gao, D., Thong, J.T., Li, B., and Qiu, C.-W., Experimental demonstration of a bilayer thermal cloak. Physical review letters. 112(5), 054302,(2014).
[3] Yang, F.-Y., Hung, F.-S., Yeung, W.-S., and Yang, R.-J., Optimization method for practical design of planar arbitrary-geometry thermal cloaks using natural materials. Physical Review Applied. 15(2), 024010,(2021).
[4] Leonhardt, U., Optical conformal mapping. Science. 312(5781), 1777-1780,(2006).
[5] Pendry, J.B., Schurig, D., and Smith, D.R., Controlling electromagnetic fields. Science. 312(5781), 1780-1782,(2006).
[6] Cummer, S.A. and Schurig, D., One path to acoustic cloaking. New journal of physics. 9(3), 45,(2007).
[7] Raza, M., Liu, Y., Lee, E.H., and Ma, Y., Transformation thermodynamics and heat cloaking: a review. Journal of Optics. 18(4), 044002,(2016).
[8] Fan, C., Gao, Y., and Huang, J., Shaped graded materials with an apparent negative thermal conductivity. Applied Physics Letters. 92(25), 251907,(2008).
[9] Chen, T., Weng, C.-N., and Chen, J.-S., Cloak for curvilinearly anisotropic media in conduction. Applied Physics Letters. 93(11), 114103,(2008).
[10] Li, T.-H., Zhu, D.-L., Mao, F.-C., Huang, M., Yang, J.-J., and Li, S.-B., Design of diamond-shaped transient thermal cloaks with homogeneous isotropic materials. Frontiers of Physics. 11, 1-7,(2016).
[11] Farhat, M., Chen, P.-Y., Bagci, H., Amra, C., Guenneau, S., and Alu, A., Thermal invisibility based on scattering cancellation and mantle cloaking. Scientific reports. 5(1), 1-9,(2015).
[12] Alekseev, G. and Levin, V. An optimization method for the problems of thermal cloaking of material bodies. in Doklady Physics. 2016. Springer.
[13] Yeung, W.-S. and Yang, R.-J., Introduction to Thermal Cloaking: Theory and Analysis in Conduction and Convection. 2022: Springer.
[14] Peurifoy, J., Shen, Y., Jing, L., Yang, Y., Cano-Renteria, F., DeLacy, B.G., Joannopoulos, J.D., Tegmark, M., and Soljačić, M., Nanophotonic particle simulation and inverse design using artificial neural networks. Science advances. 4(6), eaar4206,(2018).
[15] Liu, D., Tan, Y., Khoram, E., and Yu, Z., Training deep neural networks for the inverse design of nanophotonic structures. Acs Photonics. 5(4), 1365-1369,(2018).
[16] Liu, B., Xu, L., and Huang, J., Thermal transparency with periodic particle distribution: A machine learning approach. Journal of Applied Physics. 129(6), 065101,(2021).
[17] Lu, H., Yu, Y., Jain, A., Ang, Y.S., and Ong, W.-L., Deep learning techniques elucidate and modify the shape factor to extend the effective medium theory beyond its original formulation. International Journal of Heat and Mass Transfer. 184, 122305,(2022).
[18] Ji, Q., Chen, X., Liang, J., Fang, G., Laude, V., Arepolage, T., Euphrasie, S., Martinez, J.A.I., Guenneau, S., and Kadic, M., Deep learning based design of thermal metadevices. International Journal of Heat and Mass Transfer. 196, 123149,(2022).
[19] Ji, Q., Qi, Y., Liu, C., Meng, S., Liang, J., Kadic, M., and Fang, G., Design of thermal cloaks with isotropic materials based on machine learning. International Journal of Heat and Mass Transfer. 189, 122716,(2022).
[20] Dede, E.M., Schmalenberg, P., Nomura, T., and Ishigaki, M., Design of anisotropic thermal conductivity in multilayer printed circuit boards. IEEE Transactions on Components, Packaging and Manufacturing Technology. 5(12), 1763-1774,(2015).
[21] Dede, E.M., Zhou, F., Schmalenberg, P., and Nomura, T., Thermal metamaterials for heat flow control in electronics. Journal of Electronic Packaging. 140(1),(2018).
[22] Jin, P., Yang, S., Xu, L., Dai, G., Huang, J., and Ouyang, X., Particle swarm optimization for realizing bilayer thermal sensors with bulk isotropic materials. International Journal of Heat and Mass Transfer. 172, 121177,(2021).
[23] Wang, W., Ai, Q., Shuai, Y., and Tan, H., Topology optimization of bilayer thermal scattering cloak based on CMA-ES. International Journal of Heat and Mass Transfer. 206, 123959,(2023).
[24] Yeung, W.-S., Yang, F.-Y., and Yang, R.-J., Analysis of a trilayer thermal cloak. AIP Advances. 10(5), 055102,(2020).
[25] Han, T., Yang, P., Li, Y., Lei, D., Li, B., Hippalgaonkar, K., and Qiu, C.W., Full‐parameter omnidirectional thermal metadevices of anisotropic geometry. Advanced Materials. 30(49), 1804019,(2018).
[26] Hebb, D.O., The organization of behavior: A neuropsychological theory. 2005: Psychology press.
[27] McCulloch, W.S. and Pitts, W., A logical calculus of the ideas immanent in nervous activity. The bulletin of mathematical biophysics. 5, 115-133,(1943).
[28] Rosenblatt, F., The perceptron: a probabilistic model for information storage and organization in the brain. Psychological review. 65(6), 386,(1958).
[29] Rumelhart, D.E., Hinton, G.E., and Williams, R.J., Learning representations by back-propagating errors. nature. 323(6088), 533-536,(1986).
[30] Hornik, K., Stinchcombe, M., and White, H., Multilayer feedforward networks are universal approximators. Neural networks. 2(5), 359-366,(1989).
[31] Nair, V. and Hinton, G.E. Rectified linear units improve restricted boltzmann machines. in Proceedings of the 27th international conference on machine learning (ICML-10). 2010.
[32] Hinton, G.E., Osindero, S., and Teh, Y.-W., A fast learning algorithm for deep belief nets. Neural computation. 18(7), 1527-1554,(2006).
[33] Qian, N., On the momentum term in gradient descent learning algorithms. Neural networks. 12(1), 145-151,(1999).
[34] Duchi, J., Hazan, E., and Singer, Y., Adaptive subgradient methods for online learning and stochastic optimization. Journal of machine learning research. 12(7),(2011).
[35] Tieleman, T. and Hinton, G., Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural networks for machine learning. 4(2), 26-31,(2012).
[36] Kingma, D.P. and Ba, J., Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.(2014).
[37] Multiphysics, C., Introduction to comsol multiphysics®. COMSOL Multiphysics, Burlington, MA, accessed Feb. 9(2018), 32,(1998).
[38] Van Rossum, G. and Drake Jr, F.L., Python tutorial. Vol. 620. 1995: Centrum voor Wiskunde en Informatica Amsterdam, The Netherlands.