Multiplexing the aperture of a metasurface: inverse design via deep-learning-forward genetic algorithm

Abstract

The modern design of advanced functional materials or surfaces has increasingly undergone miniaturization and integration, so as to implement multiple functions using the same aperture. Metasurfaces, as an emerging kind of artificial functional surface, have provided unprecedented freedom in manipulating electomagnetic (EM) waves upon two-dimensional surfaces. It is desirable that the aperture of metasurfaces can be multiplexed with a number of functions for EM waves. Based on previous research, an artificial neural network-forward design can achieve an inverse design. In this paper, we propose an inverse design method of multiplexing metasurface apertures. A deep learning network (DLN) is trained to serve as a forward model for a genetic algorithm (GA), that is, a deep-leaning-forward genetic algorithm (DLF-GA). With the DLF-GA model, the phase of meta-atoms can be predicted for orthogonally linearly polarized waves simultaneously. The DLN was trained by a data set consisting of 70 000 samples with an accuracy of 95% on the test data. To demonstrate the competency of this method, we demonstrate the design of a multiplexed metasurface that can achieve focusing and diffuse scattering for polarization. The metasurface, which consists of 24 × 24 meta-atoms, can be generated monolithically with the input phase profile. A prototype was fabricated and measured. The predicted, simulated, and measured results are well consistent and underscore the validity of this inverse design method. This method provides an efficient and accurate method for the fast design of multiplexed metasurfaces and will find applications in microwave engineering such as in satellite communications.

Designing Metasurfaces for Efficient Solar Energy Conversion

“…Deep learning based on multilayer artificial neural networks (Figure a) has recently attracted significant attention from the photonics’ community . Deep learning that brings substantial acceleration capability and forth a feasible avenue for global optimization, has been introduced in metasurface design problems, including multilayer perceptrons, , convolutional neural networks, generative adversarial networks, and variational autoencoders. − The method allows combination with other optimization techniques such as genetic algorithms, , topology optimization, , or adjoint optimization , to enable high-performance, large-scale metasurface designs.…”

Inverse design of structured materials for broadband sound absorption

“…the SIMP method [42], level-set method [43], or the ESO method [44] could also be adopted for the present problem, which have advantages in computational efficiency. A good data-driven method, such as deep learning [45][46][47] or machine learning [48,49], would also be optional for this work. However, considering the specific objective function and broad operating frequency range, we choose GA to obtain various structures, without relying on any predetermined initial design or gradient information.…”

Intelligent metaphotonics empowered by machine learning

“…Chiral metasurfaces are also among the typical applications of the DL design procedures [60,[84][85][86][87][88]. Phase-amplitude engineering supported by DL algorithms helps to overcome chromatism [89], achieve tunable beam-steering [62], and multiplex an aperture [90] of metasurfaces. AI-assisted design enables development and improvement of broadband achromatic [91,92], bifocal [58], thermally tunable [93], and RGB (red, green, blue) [94] metalenses.…”