Design by neural network of concentric multilayered cylindrical metamaterials
Design by neural network of concentric multilayered cylindrical metamaterials
复制标题
DOI：
10.35848/1882-0786/ab7cf1
复制
发表时间：
2020-03
影响因子：
2.3
通讯作者：
Naoto Akashi;M. Toma;K. Kajikawa
中科院分区：
文献类型：
--
作者：
Naoto Akashi;M. Toma;K. Kajikawa
人工神经网络深度学习了超材料的光响应，可以在不求解麦克斯韦方程组的情况下预测给定超材料的光谱。由于计算复杂度低，预测速度极快。本文报道了两种同心多层圆柱体超材料的反设计，使用训练好的神经网络模型，并讨论了预测的准确性。我们还预测了一种性能优于变换光学的隐形条件，作为神经网络在超材料设计中的进一步应用。
Artificial neural networks (NNs) that have deeply learned the optical responses from metamaterials can predict the optical spectra from a given metamaterial without solving Maxwell’s equations. Prediction is extremely fast because of the low computational complexity. We report here two inverse designs of concentric multilayered cylinder metamaterials, using trained NN models, and discuss the accuracy of the prediction. We also predict a cloaking condition for invisibility having performance better than that derived by transformation optics, as a further application of NNs to metamaterial design.