AI-augmented surface Plasmon Resonance (SPR) fiber sensor simulation in COMSOL v6.2 for ultra-sensitive biosensing applications

Surface Plasmon Resonance (SPR) fiber sensors are suited for biosensing and chemical assays because they detect minute changes in refractive index. We proposed a hybrid system in which COMSOL Multiphysics simulations couple with a machine learning environment for enhancing the analysis’s precision and flexibility of the SPR fiber sensor. We have simulated the electromagnetic response of a plasmonic grating structure at different wavelengths, angles of incidence, and analyte refractive index. It was observed that resonance excitation occurs only under TM polarization and varies with the angle with a linear redshift as refractive index increase. Obtained sensitivity was around 1000 nm/RIU and confirmed the feasibility of the design for label-free detection. The AI methods have been introduced to circumvent the limitations of standard post-processing. Neural regression was used for strong and automatic peak recognition in noisy spectra, and support vector machines were used for reliable categorization of refractive index bins that matched biological samples. This AI-enhanced pipeline makes real-time sensing more accurate, cuts down on the need for human intervention, and lets you intelligently sort analytes. The suggested method is a step toward the next generation of plasmonic biosensors that might be used in many areas, such as healthcare diagnostics, environmental monitoring, and chemical detection. This is because TM polarization provides the necessary electric field component perpendicular to the metal-dielectric interface to excite surface plasmons, which is crucial for the dual mode sensor analysis. This value represents a significantly higher sensitivity compared to conventional grating-based sensors, which typically exhibit sensitivities in the range of 500–800 nm/RIU. Further sensitivity improvements are strongly limited by structural sensing parameters such as the thickness of the plasmonic layer and the grating periodicity. The neural regression model improves peak detection in noisy SPR spectra by learning complex, non-linear representations of the entire spectral signature.