Verification of a neuromorphic computing network simulator using experimental traffic data.

Area of Science:

• Computational neuroscience
• Hardware simulation
• Neuromorphic engineering

Background:

• Simulations are crucial for hardware design, but require accurate results.
• A previously developed simulator estimates network load and latency in neuromorphic computing (NC) systems.
• This simulator is particularly effective for large-scale, heterogeneous neural networks (NNs).

Purpose of the Study:

• To validate the accuracy of the NC network simulator.
• To compare simulated network loads with experimental data from a SpiNNaker board.
• To confirm the simulator's representativeness for biological plausible heterogeneous NNs.

Main Methods:

• The study utilized a previously developed network simulator for NC systems.
• Experimental data was gathered from a SpiNNaker board running various NN configurations.
• Simulated network loads were compared against measured network loads from the SpiNNaker board.

Main Results:

• Simulated network loads closely matched measured loads on the SpiNNaker board.
• Observed differences were minor and within the margin of error due to statistical variations in NN generation.
• The simulator demonstrated high accuracy for heterogeneous neural network configurations.

Conclusions:

• The network simulator provides representative results for biological plausible heterogeneous NNs.
• The validated simulator can be reliably used for more complex neuromorphic network analyses.
• This work confirms the simulator's value in exploring the design space of NC hardware systems.