Machine learning-optimized terahertz ultra-wideband tunable metamaterial absorber

Abstract

Ultra-wideband absorbers are essential devices capable of efficiently absorbing electromagnetic waves over a broad frequency range, with extensive applications in radar detection, wireless communication, and stealth technology. Their primary advantage lies in the ability to simultaneously cover both low and high-frequency absorption bands, thereby significantly enhancing stealth performance and anti-interference capabilities. However, the design of ultra-wideband absorbers still faces two major technical challenges: first, achieving stable absorption performance across an ultra-wide frequency range; and second, further improving absorption efficiency while maintaining broadband stability to meet the demands of various application scenarios. In this study, we propose a terahertz metamaterial absorber based on a three-layer composite structure incorporating patterned graphene sheets. This structure enables dynamic tunability between absorption and reflection states. To optimize the absorption performance, an innovative machine learning-based optimization strategy is introduced. Firstly, forwarding prediction is employed to quantify the optimization weights of different structural parameters, allowing for the selection of key tunable parameters. Subsequently, inverse prediction is utilized to determine the optimal structural configuration based on the target absorption performance. As a result, the proposed design achieves an absorption rate exceeding 90 % within the 2.28–4.68 THz frequency range, demonstrating significant improvements in absorption efficiency and tunability.