Explicit and refined QUBO modeling of anomalously reflective metasurfaces for annealing-assisted design optimization

Abstract

This paper presents an annealing-assisted optimization framework for anomalously reflective metasurface design based on quadratic unconstrained binary optimization (QUBO) modeling. First, an explicit QUBO formulation is derived under the physical optics (PO) approximation, enabling large-scale combinatorial optimization of metasurface elements without introducing a periodicity assumption. To overcome the limitations of the PO approximation, the QUBO model is further refined through iterative learning using full-wave electromagnetic (EM) simulation results within an annealing-assisted optimization loop. The proposed EM-QUBO framework progressively updates the QUBO coefficients based on accurate EM responses, enabling efficient optimization with a limited number of iterative EM simulations. Numerical results demonstrate significant improvements in the anomalous reflection performance compared with conventional Bayesian optimization and gradient-based methods. Furthermore, the performance of the designed metasurface reflectors for single-band 140 GHz and dual-band 140 GHz/160 GHz operations is experimentally demonstrated. Communication experiments using 64-QAM signals verify high-quality non-line-of-sight wireless links enabled by anomalous reflection, demonstrating the effectiveness of the proposed framework for practical terahertz wireless communication applications.