Spiking Convolution Engine for Spiking Convolution Neural Networks

A promising alternative in artificial vision tasks that considerably reduces computational cost and power is neuromorphic event- based processing. In this context, we employed a multiconvolution event- based system on an FPGA (Field-Programmable Gate Array), inspired by the Leaky Integrate-and-Fire (LIF) neuron, to demonstrate its viability in implementing a Spiking Convolutional Neural Network (SCNN). In this work, we demonstrate that the convolution layers of the LeNet-5 network, trained with MNIST, can be implemented in a spiking manner and discuss the necessary modifications to the architecture to offer this solution in real-time.