A multilayer WSe₂–graphene plasmonic architecture for polarization-insensitive and machine learning–optimized glucose detection in the terahertz regime

The development of advanced biosensing technologies has been prompted by the increasing global diabetes prevalence and the necessity for non-invasive glucose monitoring in real-time. These technologies surpass the limitations of classic enzymatic and electrochemical methods. This study presents a glucose biosensor utilizing a multilayer plasmonic architecture integrated with tungsten diselenide (WSe₂) and graphene hybrid coatings. The designed sensor features a central circular resonator (2.1 μm in diameter) surrounded by square resonators (1.8 μm) coated with WSe₂ and enclosed by graphene-coated rectangular resonators (12 μm × 0.5 μm) on a glass substrate. COMSOL Multiphysics simulations indicate that the proposed design achieves a sensitivity of 1000 GHz/RIU throughout the refractive index range of 1.335–1.347 RIU, beyond the capabilities of most existing glucose detection techniques. The sensor demonstrates its peak response at 0.195 THz, where the electric field is highly localized at the resonator-dielectric interfaces, thereby amplifying analyte interaction. Subsequent optimization using a K-nearest regressor (KNR) machine learning model confirms both angular resilience and predictive reliability, resulting in coefficient of determination (R 2 ) values above 0.999. The integration of 2D materials (WSe₂ and graphene) provides tunable optical properties through chemical potential modulation, while the polarization-insensitive design ensures measurement reliability across varying operational conditions. All reported results are obtained through numerical simulation using COMSOL Multiphysics, providing a theoretical foundation for future experimental realization.