DT-AIM: Digital-Twin-Accelerated Inverse Metasurface Optimization

This project aims to revolutionize the way advanced wireless and sensing technologies are developed by introducing a new digital-twin framework for designing large, reconfigurable metasurfaces. Metasurfaces are engineered 2D structures that can manipulate electromagnetic waves, enabling innovations such as high-speed wireless communications, high-resolution radar imaging, and high-efficiency wireless power transfer. Traditionally, designing these complex surfaces is slow and costly, relying on trial-and-error simulations and manual adjustments. The proposed framework promises to dramatically reduce design time and risk by creating a virtual replica (i.e., a digital twin) that can accurately predict and optimize the performance of these metasurfaces before they are built. This leap in design efficiency will accelerate the deployment of next-generation technologies like 6G wireless networks, intelligent reflecting surfaces for smart environments, advanced automotive and drone radar, and safer wireless power systems. By making the design process faster, more reliable, and accessible, the project stands to benefit society and the nation by enabling smarter infrastructure, improved public safety, and new educational opportunities. The involvement of industry partners and the commitment to open-source tools and diverse student training further ensure that the benefits will be widely shared and have lasting impacts. The project sets out to develop the Digital-Twin-Accelerated Inverse Metasurface (DT-AIM) framework, which integrates advanced electromagnetic simulation, reduced-order modeling, machine learning, and hardware-in-the-loop validation. The goals are to unify the design parameters of metasurfaces, from the microscopic geometry of individual elements to the macroscopic configuration of entire arrays, into a single physics-informed digital-twin framework. The approach involves three main themes: (1) building a closed-loop data framework that captures all relevant design and performance metrics, (2) constructing a digital twin using physics-informed variational autoencoders, Gaussian-process surrogates, and Bayesian active learning for efficient forward and inverse designs, and (3) fabricating and testing real metasurface tiles to validate and refine the digital twin. This methodology is expected to achieve at least a tenfold reduction in design time and a 90% decrease in the need for resource-demanding full-wave simulations, while maintaining high accuracy in key performance metrics. The project will produce datasets, codes, and models, advancing the state-of-the-art knowledge in electromagnetic design, machine learning, and digital-twin technology. The innovations, such as interpretable surrogate models, multi-fidelity data fusion, and hardware-calibrated digital twins, will be applicable not only to metasurfaces but also to antennas, photonics, acoustics, and power-beaming systems, setting new standards for rapid, reliable, and scalable wave-based engineering designs. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.