Fig. 3: Construction of neuro-atoms and numerical simulation of TDNN.
From: Holographic multiplexing metasurface with twisted diffractive neural network
a Schematic diagram of a neuro-atom, consisting of an SiO2 substrate and polymer nanopillars. Polarization conversion and geometric phase modulation can be achieved by varying the local rotation angle \(\Theta\). b Polarization conversion efficiency and variation as the width and height of nanopillars. To achieve highest efficiency, we select the parameter set at the red star, ensuring a high polarization conversion rate for the incident wave. c Average correlation coefficient of the stored images and training time with changes in the number of storage images. As the number increases, the training time gradually rises, while the storage correlation coefficient remains at a high level. d Video storage of a blooming and lightning. Storage examples for 36 and 40 frames are demonstrated, each of which corresponds to the storage rotation intervals of 10° and 9°, respectively.
