Menu

Neuromorphic computing

Our main research focus: computing that works only when something changes, for motion and sensing hardware that is quiet most of the time. Neural-network studies run alongside.

See it working

A live model: an event camera feeding 32 spiking neurons that find a moving spot. Stop the spot and watch the work fall away. How it works, and how it compares with the brain

1 · Event cameraPixels report only changes: blue ON, black OFF. Ring: the spot. Line: where the network thinks it is.
2 · Spiking neurons32 neurons tuned to position, last 0.6 s. Tap a row to watch that neuron.
Neuron 17: membrane potentialdashed: firing threshold
0events per second from the sensor
0synaptic operations per second, event-driven
12.3 Mmultiply-accumulates per second, clocked at 1 kHz
0.0×less work when only changes are processed
0.0 pxdecoding error, rms over the last second

Simulation time runs at a quarter of real time so single spikes are visible. Operation counts are per second of simulated time. Set the spot still: the sensor goes quiet and the event-driven work drops towards zero, while the clocked processor keeps reading every pixel. Turn on noise: the work rises and decoding gets worse, which is why real sensors filter noise events.

How we work

Each study asks one question about a problem our own hardware designs raised, compares our approach with the method an engineer would use today, and reports the numbers as they came out. The simulations are plain code with fixed random seeds, so every figure repeats exactly.

Simulation is where a study starts, not where it ends. The next step for each is a measurement on real hardware, and we will publish those results here, whichever way they go.

Results generated 2026-11-18, from 6 minutes of compute.