Photonic Spiking Convolutional Neural Networks for High-Speed Image Processing *Note: Sub-titles are not captured in Xplore and should not be used

Abstract

A multi-layer photonic spiking convolutional neural network is presented, based on miniaturized VCSELs, targeting image classification. Spatial information is encoded using a time multiplexing scheme providing a radical reduction of physical neuron-count. Hardware-friendliness is enhanced by exploiting unsupervised spike-timing dependent plasticity.