A Spike-based Cellular-Neural Network Architecture for Spatiotemporal filtering

Abstract

The foundation and architecture for a spike-based, neuromorphic cellular neural network is presented. Spike information from an event-based, dynamic vision sensor is processed asynchronously by the architecture in parallel. An array of $N^{2}$ processing elements (PEs) with eight neighbor clique is the primitive unit of the processor. Spatiotemporal filtering of spike data is accomplised via mixed-signed, embedded morphological processing using a simplicial piecewise linear approximation. Preliminary simulation and modeling on data acquired from event-based sensors show a clear pathway towards the realization of the architecture in hardware.