Building a Spiking Neural Network Model of the Basal Ganglia on SpiNNaker

Abstract

We present a biologically inspired and scalable model of the basal ganglia (BG) simulated on the spiking neural network architecture (SpiNNaker) machine, a biologically inspired low-power hardware platform allowing parallel, asynchronous computing. Our BG model consists of six cell populations, where the neuro-computational unit is a conductance-based Izhikevich spiking neuron; the number of neurons in each population is proportional to that reported in anatomical literature. This model is treated as a single-channel of action-selection in the BG, and is scaled-up to three channels with lateral crosschannel connections. When tested with two competing inputs, this three-channel model demonstrates action-selection behavior. The SpiNNaker-based model is mapped exactly on to SpineML running on a conventional computer; both model responses show functional and qualitative similarity, thus validating the usability of SpiNNaker for simulating biologically plausible networks. Furthermore, the SpiNNaker-based model simulates in real time for time-steps ≥ 1 ms; power dissipated during model execution is ≈ 1.8 W.