Fig. 12
From: Intelligent designs in nanophotonics: from optimization towards inverse creation
(a) ANN-assisted designs of vectorial field for the optical holography. Left: a floating display of a 3D kangaroo image. The vectorial hologram incorporates a phase hologram and a 2D vector field. Right: an amplified view stressing an arbitrary 3D vectorial field originated from the ANN model. (b) TO-developed microstructured fibers for third-harmonic generation. Left: a general sketch depicting fiber geometries. Right: dispersion relations (solid line) and radiative lifetimes Q (dashed line) verses propagation constant k. The graph indicates that the optimized kopt achieves frequency matching condition and leads to a large nonlinearity. Top corner: fiber cross-sections at power density of ω1 (red frame) and ω3 (blue frame), respectively. (c) ANN-aided optical information retrieval from digit-encoded nanostructures. Left: a sketch of an individual geometry storing a 9-bit (e.g., ‘101000001’) data. The digits are encoded in tagged (labels of ‘i’ to ‘ix’) silicon blocks (entity: ‘1’; vacuum: ‘0’) and the added L-shaped sidewall allows to distinguish symmetric arrangements in case of polarized excitations. Right: selected examples illustrating ANN training data (lower panel) produced from a numerical expansion of the experimental spectra (upper panel). The legend of ‘X’ and ‘Y’ denotes the applied light polarizations, i.e., X for vertical direction and Y for horizontal orientation. Figures reproduced with permission: (a) Ref. [136], Copyright 2020 AAAS; (b) Ref. [137], Copyright 2018 OSA; (c) Ref. [138], Copyright 2019 Nature Publishing Group