Multiplexed supercell metasurface design and optimization with tandem residual networks

Abstract

Complex nanophotonic structures hold the potential to deliver exquisitely tailored optical responses for a range of applications. Metal–insulator–metal (MIM) metasurfaces arranged in supercells, for instance, can be tailored by geometry and material choice to exhibit a variety of absorption properties and resonant wavelengths. With this flexibility, however, comes a vast space of design possibilities that classical design paradigms struggle to effectively navigate. To overcome this challenge, here, we demonstrate a tandem residual network approach to efficiently generate multiplexed supercells through inverse design. By using a training dataset with several thousand full-wave electromagnetic simulations in a design space of over three trillion possible designs, the deep learning model can accurately generate a wide range of complex supercell designs given a spectral target. Beyond inverse design, the presented approach can also be used to explore the structure–property relationships of broadband absorption and emission in such supercell configurations. Thus, this study demonstrates the feasibility of high-dimensional supercell inverse design with deep neural networks, which is applicable to complex nanophotonic structures composed of multiple subunit elements that exhibit coupling.

A Generative Machine Learning-Based Approach for Inverse Design of Multilayer Metasurfaces

“…Inspired by the revolution that data-driven machine-learning methods have made in material informatics such as discovery of new quantum materials, pharmaceuticals, and other compounds [15], deep machine learning [16]- [21] and statistical learning [22] methods can help to build surrogate models that can provide fast predictions of the properties of each unit cell. Moreover, several machine-learning techniques have been proposed to tackle the challenges of the second step [23]- [33]. Some of these proposed methods deal with the inverse design of a uniform EMMS where the impact of mutual coupling is less of an issue [23]- [29], while the rest optimize over a simple solution space that is composed of only one scatterer shape [30]- [33].…”

“…Moreover, several machine-learning techniques have been proposed to tackle the challenges of the second step [23]- [33]. Some of these proposed methods deal with the inverse design of a uniform EMMS where the impact of mutual coupling is less of an issue [23]- [29], while the rest optimize over a simple solution space that is composed of only one scatterer shape [30]- [33]. It is worth noting that dielectric optical EMMSs can be designed using global optimization methods due to the analytical relation between the scatterers' properties and the EMMS's scattering parameters [34].…”

Generalized multi-channel scheme for secure image encryption

“…Advances in the study of optical metamaterials 1 , 2 have introduced new possibilities for encoding information in optical waves, in both the spatial 3 – 6 and spectral 7 – 12 domain. At infrared wavelengths, metamaterials have been designed to encode images in their spatial emission profiles 13 , 14 .…”

Physics-augmented deep learning for high-speed electromagnetic simulation and optimization

“…To address this bottleneck, deep neural networks serving as high speed surrogate Maxwell simulators have emerged as promising algorithms that can operate orders-of-magnitude faster than conventional Maxwell simulators [21][22][23][24][25][26]. Many initial attempts to use neural networks in this manner were based on the training of fully connected deep neural networks, which could accurately model the spectral response of photonic structures described by a handful of geometric parameters [27][28][29][30][31].…”