How they differ from ordinary chips
A conventional processor keeps memory and computing apart and moves data between them constantly, which costs energy. Neuromorphic designs place memory next to many simple, neuron-like units that communicate with spikes, and they work on events. If nothing changes in the input, almost nothing is computed.
Why that is interesting
Event-driven operation can use a tiny fraction of the energy of a conventional chip for some tasks, such as always-on sensing, detecting a gesture or a sound, or tracking motion with an event camera. That is attractive for sensors that must run for years on a small battery.
Where it stands
Research chips from large companies and universities have shown the idea works, but the field lacks mature programming tools, and spiking networks are harder to train than standard ones. Expect niche uses in low-power sensing first, not a replacement for graphics processors.
Where it is used
- Always-on sensing Detecting a keyword, a gesture or motion on very little power.
- Robotics Fast, reflex-like responses from event cameras.
- Research Studying brain-like learning in hardware.
Neuromorphic chips compared with CPUs and GPUs
CPUs handle general instructions. GPUs do huge numbers of calculations in parallel and dominate today’s AI. Neuromorphic chips compute with sparse events and keep memory next to the computing, aiming for very low energy on suitable tasks.
They are not a replacement for GPUs, and most of today’s AI software has to be rebuilt to run on them.
Key terms
- Spiking neural network
- A network whose units communicate with brief pulses.
- Event-driven
- Computing only when an input changes.
- In-memory computing
- Processing data where it is stored, to cut the energy spent moving it.
- Event camera
- A sensor whose pixels report only changes in brightness.
Common questions
Are neuromorphic chips the same as brains?
- No. They borrow ideas such as spikes and local memory, but are far simpler than biological brains.
Will they replace GPUs?
- Unlikely soon. GPUs excel at the large models in wide use today. Neuromorphic chips target low-power, event-driven tasks.
What is an event camera?
- A camera whose pixels report only changes in brightness, producing a stream of events instead of full frames. It pairs naturally with neuromorphic processing.
Sources and further reading
Independent pages we checked while writing this guide. They are not Reinnder products, and Reinnder is not affiliated with them.
Reinnder’s angle
How Reinnder looks at chips that think like neurons
Reinnder Labs lists neuromorphic chips as a research interest, in the context of very low-power sensing.
This guide explains the technology in general terms. It is not advice, and it does not describe a Reinnder product on sale.