Power-efficient simulation of detailed cortical microcircuits on SpiNNaker

Area of Science:

• Computational neuroscience
• Neuroscience
• Computer architecture

Background:

• Computer simulations are crucial for understanding brain function.
• Existing models of cortical circuits are limited by slow speeds and high power consumption.
• The intricate structure of the cortex has been mapped, providing data for simulations.

Purpose of the Study:

• To introduce the SpiNNaker computer architecture for efficient brain simulations.
• To demonstrate real-time simulation of a large cortical circuit.
• To assess the power efficiency of the SpiNNaker hardware for neural modeling.

Main Methods:

• Utilized four SpiNNaker chips to simulate a cortical circuit.
• Modeled ten thousand spiking neurons and four million synapses in real-time.
• Measured energy consumption per neuron and per postsynaptic potential.

Main Results:

• Achieved an energy consumption of 100 nJ/neuron/ms and 43 nJ/post-synaptic potential.
• Demonstrated real-time simulation capabilities on a large-scale neural network.
• The SpiNNaker hardware exhibited the lowest reported energy consumption for digital computer simulations.

Conclusions:

• SpiNNaker offers a power-efficient solution for large-scale brain simulations.
• This architecture approaches fast, power-feasible, and scientifically useful simulations of cortical areas.
• The findings pave the way for more extensive and practical computational neuroscience research.