Independent Bifocal Metalens Design Based on Deep Learning Algebra

We present a single-layer, transmissive metalens for focusing the orthogonal polarization components of incident light to two independent focal spots without affecting the initial orthogonal polarization states. To produce the required phase profile, the cross-shaped nanorod, as the unit cell of the metalens, is developed to support arbitrary combinations of two independent phase shifts (0-2 pi) of transverse electric and transverse magnetic polarized light. A deep neural network is trained to generate the design parameters of each unit cell efficiently and accurately. The metalens is simulated with finite-difference time-domain method and good agreements are observed comparing with theoretical prediction. This work provides a new solution to multi-foci metalens and polarization beam splitter and may find potential applications in deep sensing, holography, information encryption and display.

Automated summary

An innovative single-layer, transmissive metalens has been developed for focusing orthogonal polarization components of incident light to two independent focal spots, while maintaining the initial polarization states. By utilizing a deep neural network for efficient and accurate design parameter generation, this metalens represents a new solution for multi-foci metalens and polarization beam splitter applications. Good agreement between simulation and theoretical predictions indicates the potential suitability of this technology for deep sensing, holography, information encryption, and display purposes.