sPyNNaker: A PyNN Implementation for SpiNNaker

Introduction

sPyNNaker is an implementation of the PyNN language for SpiNNaker that lets you build spiking neural networks and simulate them in real-time or close to real-time, depending on network complexity and system configuration.

This is the documentation of the part of sPyNNaker, implemented in C, that is actually deployed onto SpiNNaker to execute. The Python parts of the sPyNNaker package are responsible for the user interface, data preparation, execution planning, and runtime control.

Note that this code depends on SpiNNFrontEndCommon for many support functions.

Neuron Simulation Implementation

The major part of this code is the implementation of the neuron simulator itself, which is comprised of several pieces.

• A main entry point, c_main.c
• The core neuron implementation, neuron.c
• The core synapse implementation, synapses.c
• The main spike processing system, spike_processing.c
• The master population table, used to decode a spike and identify which neurons it is to be delivered to, population_table.h
• The neuron models, neuron_model.h, and subclasses
• The synaptic input types, input_type.h, and subclasses
• The synaptic shaping, synapse_types.h, and subclasses
• The neural threshold model, threshold_type.h, and subclasses
• The plasticity models, both for STDP (synapse_dynamics_stdp_mad_impl.c) and structural plasticity (topographic_map_impl.c)
• The additional model inputs, such as calcium ion modulation (additional_input_ca2_adaptive_impl.h)

Support Binaries

These enable the neuron simulation to work, or to work more efficiently.

• Bitfield Expander, which builds inter-core message masks on the machine. See bit_field_expander.c
• Delay Extension, which handles messages that need to be delayed a long time before delivery. See delay_extension.c
• Robot Motor Control, which translates spiking rates of SpiNNaker messages into activation levels for an motor device (part of integrating with external peripherals). See robot_motor_control.c
• Poisson Spike Source, which injects random spikes (using a Poisson distribution) into the system. See spike_source_poisson.c
• Synapse Expander, which efficiently constructs synaptic connectivity data on machine from statistical descriptions. (Note that literal descriptions of connectivity data need to be uploaded directly, which can be a slow process by comparison to generating the data in parallel on SpiNNaker.) See synapse_expander.c
• Delay Expander, which efficiently constructs synaptic delay data on machine. This is a reduced versino of the synapse expander that is designed to support delay extensions. See delay_expander.c

Preferred Citation

Rhodes, O., Bogdan, P., Brenninkmeijer, C., Davidson, S., Fellows, D., Gait, A., Lester, D., Mikaitis, M., Plana, L., Rowley, A., Stokes, A., Furber, S. sPyNNaker: A Software Package for Running PyNN Simulations on SpiNNaker. Frontiers in Neuroscience, vol 12, pp 816, 2018. DOI: 10.3389/fnins.2018.00816