Phase-to-pattern inverse design paradigm for fast realization of functional metasurfaces via transfer learning

Abstract

Metasurfaces have provided unprecedented freedom for manipulating electromagnetic waves. In metasurface design, massive meta-atoms have to be optimized to produce the desired phase profiles, which is time-consuming and sometimes prohibitive. In this paper, we propose a fast accurate inverse method of designing functional metasurfaces based on transfer learning, which can generate metasurface patterns monolithically from input phase profiles for specific functions. A transfer learning network based on GoogLeNet-Inception-V3 can predict the phases of 28×8 meta-atoms with an accuracy of around 90%. This method is validated via functional metasurface design using the trained network. Metasurface patterns are generated monolithically for achieving two typical functionals, 2D focusing and abnormal reflection. Both simulation and experiment verify the high design accuracy. This method provides an inverse design paradigm for fast functional metasurface design, and can be readily used to establish a meta-atom library with full phase span.

Intelligent metaphotonics empowered by machine learning

“…As an alternative, a free-form design approach can be used. In this case, the DL algorithm process is based not on sets of parameters, but rather employs pixelated images of the unit cells [55]. Such techniques extend the expressivity of this approach, significantly extending the range of possible cell geometries.…”

“…The metastructures consist of a patterned MoS 2 layer with thickness 30 nm and refractive index n = 4.3+0.034i. The ANN [54,55,[58][59][60]. Representation of metasurface elements can be done in several ways.…”

“…(a) General scheme of the inverse design implemented for metasurfaces. (Adapted from Refs [54,55,[58][59][60]…”

CaSee: A lightning transfer-learning model directly used to discriminate cancer/normal cells from scRNA-seq

“…These emerging deep learning analysis methods provide us with a novel idea in scRNA-seq data mining. At the algorithm technology level, the strength of deep learning lies not only in its ability to achieve accurate sample classification based on big data, but also in its excellent "transfer learning" ability [17][18][19][20] to learn in a specific data space, and then apply the "knowledge" in subspace or similar data space 17,18 . At the hardware equipment level, the increasingly mature GUP hardware technology greatly shortens the time of deep learning training and prediction and greatly improves the efficiency of analysis.…”

Switchable Multifunctional Meta-Surface Composed by Dielectric-Metal Hybrid Antenna Array Architecture

“…There is also a meta-surface design method for directivity improvement, and in Reference [7], an effective technique was come up with for radiation patterns' improvement of the Fabry-Perot cavity antenna by improving the near-field phase distribution, that led to a 5.6 dBi increase in the peak directivity of the antenna. Additionally, based on the operation method for changing the structure and arrangement of meta-atoms [8], lots of meta-surfaces with different functions have been proposed and designed, such as polarization manipulation [9][10][11], abnormal reflection [12,13], focusing lens [14][15][16], and absorber [17][18][19], etc. Although these meta-surfaces have been designed well, the function of these proposed meta-surfaces is mostly single, which is not suitable for the high-integration electromagnetic system and the complex optical integration system; hence, multifunctional meta-surfaces are proposed and have gradually become a research hotspot.…”