sPyNNaker: A Software Package for Running PyNN Simulations on SpiNNaker

Area of Science:

• Computational Neuroscience
• Neuromorphic Engineering

Background:

• Spiking neural networks (SNNs) are computational models inspired by biological neural systems.
• The SpiNNaker platform is a large-scale neuromorphic system designed for SNN simulations.

Purpose of the Study:

• To present sPyNNaker 4.0.0, the latest software for simulating PyNN-defined SNNs on SpiNNaker.
• To detail the operations enabling real-time SNN execution and analyze system performance.
• To introduce a cost model for predicting SNN simulation performance based on network characteristics.

Main Methods:

• Implementation of an event-based operating system for efficient neuron state updates and spike processing.
• Profiling of simulation performance for a simple SNN to understand software-hardware interaction.
• Development of a cost model correlating SNN topology, partitioning, and simulation performance.

Main Results:

• sPyNNaker 4.0.0 supports real-time SNN execution on SpiNNaker hardware.
• Achieved system performance is within a factor of 2 of the 10,000 synaptic events/ms target.
• SNN topology significantly influences performance, as captured by the developed cost model.

Conclusions:

• sPyNNaker 4.0.0 advances real-time SNN simulation capabilities on SpiNNaker.
• The cost model provides valuable insights for optimizing SNN design and predicting performance.
• The findings underscore the ongoing potential of SpiNNaker hardware for large-scale neural simulations.