Large-scale data-driven inverse EM design based on metacircuit-embedded surface

Abstract

Abstract General-purpose electromagnetic model is central to intelligent electromagnetic applications. However, its development is limited by severe scarcity of training data, arising from the high computational cost of full-wave simulations and reliance on specialized topologies that yield single EM function. To overcome this challenge, we introduce a metacircuit-embedded surface (MCES) framework, consisting of a fixed structural topology and flexible metacircuits that can be realized through lumped components. By shifting design focus from full-wave simulation to circuit-level modeling, over six million samples can be easily generated with laptop computation resources. Subsequently, an MCES-based forward and inverse AI design method was developed, providing rich design capability in frequency, amplitude, phase and polarization domains. Finally, three kinds of classical metasurfaces were designed and fabricated, showing superior performance compared to reported advanced designs. Overall, the proposed MCES-based framework greatly raises the design efficiency and training scale, paving the way for future general-purpose electromagnetic model.