In November 2018, the SpiNNaker machine at the University of Manchester reached one million processor cores, becoming one of the largest neuromorphic computers ever built. SpiNNaker, short for Spiking Neural Network Architecture, was designed by a team led by Steve Furber, one of the original designers of the ARM processor that powers most of the world’s smartphones.
SpiNNaker took a different route to brain-like computing than custom chips such as TrueNorth or Loihi. Instead of designing exotic silicon neurons, it wired together a million small, conventional ARM cores - the same low-power processors used in phones - with a custom communication fabric tuned to do one thing well: deliver the enormous number of tiny “spike” messages that neurons send each other. Each core simulates a few hundred neurons, and the whole system runs in biological real time, modeling neural activity at roughly the speed of a real brain.
The architecture’s purpose was scientific. By simulating large networks of spiking neurons fast and cheaply, SpiNNaker let computational neuroscientists test theories about how brains process information, and let engineers explore brain-inspired control for robots. It was a flagship piece of the European Union’s Human Brain Project.
SpiNNaker embodies a recurring idea in the neuroscience-and-AI overlap: that the brain’s secret may lie less in any single neuron than in massive parallelism and sparse spike-based communication. A successor, SpiNNaker 2, pushed the design further with millions more cores aimed at both brain simulation and energy-efficient AI.