Active metasurfaces have recently attracted more attention since they can make the light manipulation be versatile and real-time. Metasurfaces-based holography possesses the advantages of high spatial resolution and enormous information capacity for applications in optical displays and encryption. In this work, a tunable polarization multiplexing holographic metasurface controlled by an external magnetic field is proposed. The elaborately designed nanoantennas are arranged on the magneto-optical intermediate layer, which is placed on the metallic reflecting layer. Since the non-diagonal elements of the dielectric tensor of the magneto-optical material become non-zero values once the external magnetic field is applied, the differential absorption for the left and right circularly polarized light can be generated. Meanwhile, the amplitude and phase can be flexibly modulated by changing the sizes of the nanoantennas. Based on this, the dynamic multichannel holographic display of metasurface in the linear and circular polarization channels is realized via magnetic control, and it can provide enhanced security for optical information storage. This work paves the way for the realization of magnetically controllable phase modulation, which is promising in dynamic wavefront control and optical information encryption.

Recently, deep learning has been used to establish the nonlinear and nonintuitive mapping between physical structures and electromagnetic responses of meta-atoms for higher computational efficiency. However, to obtain sufficiently accurate predictions, the conventional deep-learning-based method consumes excessive time to collect the data set, thus hindering its wide application in this interdisciplinary field. We introduce a spectral transfer-learning-based metasurface design method to achieve excellent performance on a small data set with only 1000 samples in the target waveband by utilizing open-source data from another spectral range. We demonstrate three transfer strategies and experimentally quantify their performance, among which the “frozen-none” robustly improves the prediction accuracy by ∼26 % compared to direct learning. We propose to use a complex-valued deep neural network during the training process to further improve the spectral predicting precision by ∼30 % compared to its real-valued counterparts. We design several typical teraherz metadevices by employing a hybrid inverse model consolidating this trained target network and a global optimization algorithm. The simulated results successfully validate the capability of our approach. Our work provides a universal methodology for efficient and accurate metasurface design in arbitrary wavebands, which will pave the way toward the automated and mass production of metasurfaces.

Optical metasurfaces, which consist of subwavelength scale meta-atoms, represent a novel platform to manipulate the polarization and phase of light. The optical performance of metasurfaces heavily relies on the quality of nanofabrication. Retrieving the Jones matrix of an imperfect metasurface optical element is highly desirable. We show that this can be realized by decomposing the generalized Jones matrix of a meta-atom into two parallel ones, which correspond to the ideal matrix and a phase retardation. To experimentally verify this concept, we designed and fabricated metasurface polarizers, which consist of geometric phase-controlled dielectric meta-atoms. By scanning the polarization states of the incident and transmitted light, we are able to extract the coefficients of the two parallel matrices of a metasurface polarizer. Based on the results of the Jones matrix decomposition, we also demonstrated polarization image encryption and spin-selective optical holography. The proposed Jones matrix retrieval protocol may have important applications in computational imaging, optical computing, optical communications, and so on.