← All Terms

Neuromorphic Computing

Chips and systems that copy the brain's structure, processing information as sparse spikes that fire only when something changes, with memory placed next to the computation.

Implementation

A conventional processor works to a clock, fetching instructions and moving data on every tick whether or not anything has changed. A neuromorphic chip uses artificial neurons that stay idle until enough input arrives, then send a short spike to the neurons they connect to. Intel’s Loihi 2, IBM’s TrueNorth and NorthPole, and BrainChip’s Akida are the best-known examples. Paired with event-based sensors, which report only changes in light or sound, they can run AI on milliwatts.

The low power is what puts this AI inside cameras, wearables, vehicles and medical devices, outside the data centre. Some neuromorphic chips can also learn on the device after deployment, so the model in the field can drift from the model that was tested.

For security and governance, the shift matters more than the speed. Behaviour depends on the timing of inputs, the chip is within physical reach, and the usual scanners and patches do not reach it. Threat models need to cover light, power and the learning process as well as code.

Read more on this
Security

Neuromorphic AI Moves the Attack Surface Into the Device

Brain-inspired neuromorphic chips run AI as bursts of events inside sensors, cameras and controllers, and some of them keep learning after they ship. That changes cybersecurity in three ways: attackers can work through light, timing and power as well as code; the usual scanners and patches do not reach the chip; and the device you approved may not be the device in the field a year later. Maria Singh and I set out what changes, what the research already shows, and how to govern AI that adapts inside physical devices.