Design and Optimization of a Rose-Inspired Plasmonic Filter using Machine Learning

Bio-inspired plasmonic structures offer a powerful route toward compact and high-performance photonic components by ena-bling strong confinement of surface plasmon polaritons (SPPs) at deeply subwavelength scales. In this research, a tunable mul-tichannel bandpass filter using surface plasmon polaritons (SPPs) is introduced and numerically studied on a metal-insulator-metal (MIM) waveguide substrate. Consisting of a central circular resonator, two larger main waveguides, and six smaller sat-ellite resonators, the proposed structure exploits a unique rose-like geometry. High Q-factor (Q-factor) multiple resonance modes can be achieved due to this blend. Nevertheless, time-consuming numerical methods such as the Finite-Difference Time-Domain (FDTD) approach are employed in the conventional design and optimization of such intricate structures. Re-cently, a machine learning framework was utilized to considerably accelerate the design procedure. Through training on a vast dataset from FDTD runs, a Random Forest surrogate model was developed and trained to predict the filter spectral response from its geometrical parameters with very high accuracy. FDTD simulation results reveal that the proposed rose-like structure produces several sharp transmission peaks of high Q-factor, up to ۱۲۷.۲. Moreover, the machine learning system possesses a high coefficient of determination (R² up to ۰.۹۸۶) and excellent predictive ability and can replicate the transmission spectrum within a fraction of the time it takes for conventional runs. Through its powerful volume-based optimization tool and inverse design tool, the proposed mixed-method approach brings within grasp possible future applications of photonic integrated circuits (PICs), high-sensitivity biosensors, and wavelength-division multiplexing (WDM) systems.

کلیدواژه ها:

Plasmonic Filter ، MIM Waveguide (Metal-Insulator-Metal) ، Circular Resonator ، machine learning ، Quality Factor