Polarization-Modulated Elliptical Plasmonic Synapses for Optical Neural Network Applications
Chengyan Zhong, Xiaobo She, Xiang Wang, Yu Liu · 2025
This work presents a plasmonic aluminum (Al) patch synapse with an elliptical geometry modulated by polarization angle for an optical neural network. Compared to conventional rectangular structures, elliptical patches offer superior fabrication advantages with reduced processing deviations and enhanced structural symmetry. The distribution and propagation characteristics of the electromagnetic field as they vary with polarization angle are investigated using finitedifference time-domain simulation. As the polarization angle increases from 0° to 90°, the synaptic weights are quantified within a range of 0.15 to 0.9. We capitalize on these polarization-dependent transmission characteristics to develop an optical computing platform tailored for Fashion-MNIST image classification tasks, achieving a remarkable classification accuracy of 86.8%. This approach paves the way for novel integration of plasmonic synaptic elements in optical neural network applications.