Hierarchical multi-resonator metamaterial absorber with graphene integration for ultra-broadband solar energy harvesting and machine-learning-assisted performance prediction

This study presents a hierarchically engineered multi-resonator metamaterial absorber for ultra-broadband solar energy harvesting across the 0.2–3.0 μm spectral range, covering the ultraviolet, visible, near-infrared, and far-infrared regions. The proposed structure integrates a titanium rectangular patch, tungsten square resonator, dual Fe₂O₃ circular inclusions, monolayer graphene sheet, chromium ring resonator, and a SiO₂/Ti substrate stack, all composed of CMOS-compatible, noble-metal-free materials suitable for scalable fabrication. The synergistic interaction of plasmonic and magnetic resonances enables efficient impedance matching and broadband light confinement, resulting in a peak absorptance of 99.94%, an average absorptance of 98.13% over the operating spectrum, an AM1.5 G solar-weighted absorptance of 97.84%, and a 90% absorption bandwidth of 2800 nm. Numerical investigations performed using the finite element method (FEM) in COMSOL Multiphysics were validated through Transfer Matrix Method (TMM), Temporal Coupled-Mode Theory (TCMT), and equivalent LC-resonance modeling, all showing excellent agreement with the simulated results. Comprehensive parametric and power-dissipation analyses reveal that the broadband response originates from cooperative excitation of multiple resonant mechanisms, with the tungsten resonator primarily governing spectral tuning and the chromium ring enhancing bandwidth through surface plasmon polariton coupling. Fabrication tolerance studies confirm robust operation, maintaining average absorptance above 96.9% under realistic dimensional deviations, while polarization-insensitive behavior and angular stability are preserved up to 60° incidence for both TE and TM polarizations. To further accelerate device optimization, a Bayesian Gaussian Process Regression surrogate model was developed for predicting absorptance under geometric and angular variations, achieving mean R² values of 0.9574 and 0.8999, respectively, with RMSE below 0.034. Compared with full-wave FEM simulations, the surrogate model provides an approximate computational speed-up of 3.15 × 10⁵, enabling rapid optimization and uncertainty quantification. These results demonstrate the potential of the proposed absorber as a high-performance, fabrication-tolerant platform for solar energy harvesting and thermophotovoltaic applications.