Multiband circular dichroism and data-driven prediction in a VO2/graphene chiral terahertz metasurface.

Dynamic tuning of chiral terahertz metasurfaces depends on the coupling between the electromagnetic properties of functional materials and localized resonant modes. However, as the number of tunable materials and structural degrees of freedom increases, the circular dichroism (CD) response becomes highly nonlinear. Traditional full-wave parameter scanning is therefore inefficient for both mechanism analysis and rapid performance evaluation. To address this issue, we design a tunable VO2/graphene chiral terahertz metasurface and further introduce a data-driven method for rapid prediction of the complete CD spectrum. The structure uses the electromagnetic parameter changes caused by the insulator-to-metal transition of VO2 and the tunable conductivity of graphene to control the coupling between circularly polarized waves and the resonant units, as well as the selective absorption. This produces a tunable triple-band CD response in the terahertz range. The results show pronounced CD resonances near 2.01, 3.12, and 3.40 THz. Changes in the VO2 phase state and graphene chemical potential further tune the response intensity of different resonant modes, indicating a complex nonlinear relationship between material states, structural parameters, and the CD spectrum. To efficiently obtain complete CD spectra for a large number of parameter combinations, we construct a multilayer perceptron (MLP) model with the structural azimuth angle and VO2 conductivity as inputs and the complete CD spectrum as the output. The model learns the nonlinear mapping between the tuning parameters and spectral response, achieving a coefficient of determination of 0.9972 on the test set. Beyond conventional simulation-based analysis, this approach integrates physical response analysis with data-driven spectral prediction. It maintains high prediction accuracy while significantly improving spectral acquisition efficiency. This provides a new approach for rapid performance evaluation and parameter optimization of tunable chiral terahertz devices.