Efficient multi-objective optimization for switchable digital coding metasurface absorber empowered by machine learning

Multi-functional metasurface absorbers have emerged as highly promising platforms in the fields of wireless security, radar communication, and biosensing. Nevertheless, the traditional trial-and-error design methods for these absorbers rely extensively on prior experience and are exceedingly time-consuming. Although machine learning algorithms have been recently utilized to improve the design of multifunctional metasurface absorbers, the existing methods are mainly limited to single-objective tasks and often result in suboptimal efficiency. To address these limitations, we propose a universal and efficient multi-objective optimization framework. This framework employs the machine learning Non-dominated Sorting Genetic Algorithm II (NSGA-II) to expedite the design of a four-channel dual-layer multi-functional digital coding metasurface absorber. The device to be optimized incorporates nine programmable encoding modes, with seamless transitions among these modes realized through four digital logic switches. The Results indicate that the proposed framework can effectively address both homogeneous and heterogeneous multi-objective optimization problems, achieving rapid convergence within merely 10 generations, which represents a significant improvement compared to conventional genetic algorithms. Owing to the independently tunable dual-layer architecture, the optimized device exhibits a reduced full width at half maximum (FWHM). In addition, the optimized device exhibits excellent free-space impedance matching, simplicity in fabrication, robust angular tolerance, and polarization insensitivity across all operating modes. The proposed efficient multi-objective optimization framework has the potential for the design of achromatic metalenses, sensors, and detectors.