sPyNNaker

sPyNNaker enables simulation of spiking neural networks (SNNs) defined in PyNN on the SpiNNaker neuromorphic platform for real-time execution and performance analysis.

Key Features:

• Real-Time SNN Execution: Supports real-time execution of SNNs using an event-based operating system with time-driven neuron state updates and pipelined spike-event processing.
• Preprocessing and Neuron/Synapse Models: Provides preprocessing support and implementations of various neuron and synapse models to simulate diverse neural dynamics.
• Performance Profiling: Offers detailed profiling of software–hardware interactions to analyze how different network configurations impact simulation efficiency.
• System Performance Metrics: Targets approximately 10,000 synaptic events per millisecond with observed performance within a factor of two and highlights strong dependence on SNN topology.
• Cost Model for Connectivity and Partitioning: Implements a cost model to characterize how network connectivity and partitioning affect simulation performance and to assist performance estimation.

Scientific Applications:

• Neuromorphic SNN experimentation: Enables exploration of SNN dynamics and real-time neural interactions on the SpiNNaker neuromorphic hardware.
• Performance analysis and optimization: Facilitates analysis of the impact of topology, connectivity, and partitioning on throughput using profiling and the cost model.
• Computational neuroscience research: Supports studies of brain-like computations and neural processing mechanisms using biologically motivated neuron and synapse models.
• Biologically inspired AI development: Supports development of AI applications that mimic spiking network behavior via real-time neuromorphic simulation.

Methodology:

Integration of PyNN-defined SNNs with the SpiNNaker neuromorphic architecture using an event-based operating system to perform time-driven neuron state updates and pipelined spike processing, together with detailed performance profiling and a cost model to characterize effects of connectivity, partitioning, and SNN topology on throughput.