Investigation into the pragmatism of phenomenological spiking neurons for hardware implementation on FPGAs

Spiking neurons (SNs) are biologicallyplausible neuron models that offer new informationprocessing paradigms for neuroengineers. It is expectedthat artificial representation of these neurons willenhance the link between biological and artificialsystems. The complexity of spiking neuron models withlow level abstraction makes them unsuitable for largescale implementations, limiting network scalability. Thishas led to the development of simpler, phenomenologicalspike models, such as the Leaky Integrate and Fire model.However, no clear guidelines exist to help select whichphenomenological model to implement. The aim of thispaper is to reduce this ambiguity, through a systematiccomparative performance evaluation. An evolutionarystrategy for the supervised training of networks to twoformal models is used to solve computational benchmarkproblems in software. The models are then designed,simulated and implemented onto a Field ProgrammableGate Array (FPGA) through a novel hardware designflow. It is envisaged that this information will helpneuroengineers in future hardware implementationdecisions.