Logo image
Surface Plasmon Polariton Neural Network for Broadband Computing
會議論文

Surface Plasmon Polariton Neural Network for Broadband Computing

Komal Gupta, Anand Hegde, Wei-Hsiang Shen, Meng-Lin LiChen-Bin Huang
Nanophotonics XI (Auditorium Schweitzer, 12/04/2026–16/04/2026)
15/04/2026

摘要

Plasmonic and spoof-plasmon neural networks enable ultrafast subwavelength optical computing. We experimentally demonstrate a Surface Plasmon Polariton (SPP)-based neural network operating at 1550 nm that performs high-speed, high-density, all-optical inference. Plasmonic waveguides implement analog weighted summation via controlled SPP interference, while adjustable neuron weights are programmed through input polarization, eliminating electronic control. Nonlinear activation emerges from localized field enhancement, providing sub-picosecond, activation-like responses within nanometric volumes. We realize photonic neural primitives including multi-neuron fan-in and dual-input/dual-output processing, and furthermore we map a two-neuron input through a two-neuron hidden layer to a single-neuron output to achieve cascaded architecture. Polarization-resolved measurements confirm weight control, strong confinement, and nonlinear behavior. This architecture delivers a fivefold footprint reduction and defines a scalable route to hybrid plasmonic/dielectric neuromorphic processors.

相關連結

指標

1 檢視次數

詳細資料

Logo image