Inverse Design of an end-fire Broadband Hybrid Plasmonic Leaky-Wave Antenna based on the CNN approach

This paper presents an inverse-design methodology for developing a broadband hybrid plasmonic leaky-wave antenna (LWA) featuring efficient end-fire radiation across the 900–2500 nm wavelength range. The proposed antenna integrates a hybrid plasmonic waveguide consisting of a 200 nm silver (Ag) layer, a thin SiO₂ guiding film, and a dielectric substrate, achieving strong optical field confinement while minimizing propagation losses. Periodically arranged subwavelength slits along the waveguide enable the controlled conversion of guided plasmonic modes into leaky radiation, resulting in highly directive end-fire emission.A convolutional neural network (CNN)-based inverse design framework is employed to optimize the antenna’s geometrical parameters. The CNN is trained on electromagnetic simulation data correlating radiation patterns with structural features, enabling rapid and accurate prediction of optimal slot dimensions and spacing with a high regression accuracy (R² >0.93).Simulation results confirm that the optimized antenna exhibits broadband operation with stable radiation efficiency and a peak gain of 8.2 dBi at 1 µm, while maintaining gain levels above 6 dBi throughout most of the operational band. The integration of hybrid plasmonic principles with deep-learning-based inverse design demonstrates a powerful and efficient pathway for the automated development of compact, low-loss, and high-performance optical antennas suitable for photonic integrated circuits and nanoscale communication systems.