The problem
Simulating a million neurons on conventional hardware takes a rack of machines and kilowatts of power; a brain does far more on the energy of a dim light bulb.
The innovation
A mixed analogue-digital multichip system in which silicon transistors operating in their analogue regime emulate the physics of neurons and synapses directly, rather than computing equations about them.
How it worked
Sixteen custom chips share a tree-routed communication fabric that carries spikes as events, so power is spent only when something happens.
Why it mattered
Roughly a million neurons and billions of synaptic connections simulated in real time at a few watts.
The legacy
Neuromorphic architecture is now pursued by major chip manufacturers for low-power sensing and inference at the edge.
Sources
Neurogrid: A Mixed-Analog-Digital Multichip System for Large-Scale Neural Simulations (Proceedings of the IEEE, 2014)
ScholarshipBrains in Silicon laboratory records
Institutional record
Follow the invention
The West Area computers become programmers
Dorothy Vaughan taught herself FORTRAN and machine operation, then taught her staff, converting a pool of hand calculators into the programming staff of the new Analysis and Computation Division.
The precision wire resistor
A wire-wound resistance element designed so its value held under thermal and mechanical stress, and could be manufactured cheaply and consistently.
Foil-electret microphone
A polymer film holding a permanent electrostatic charge, removing the need for an external bias voltage.