Energy-efficient multipath ring network for heterogeneous clustered neuronal arrays

Abstract

Simulating large spiking neural networks (SNN) with a high level of realism in a field programmable gate array (FPGA) requires efficient network architectures that satisfy both resource and interconnect constraints, as well as changes in traffic patterns due to learning processes. Based on a clustered SNN simulator concept, in this thesis, an energyefficient multipath ring network topology is presented for the neuronto-neuron communication. It is compared in terms of its mathematical properties with other common network topology graphs after which the traffic distributions across it and a two dimensional torus network are estimated and contrasted. As a final characterization step, the energy-delay product of the multipath topology is estimated and compared with other low power architectures. In addition, a simplified