What is the advantage of a neuromorphic processor?

Currently, processors have all their cores active at any time. Despite a use that is often close to idle (no activity). So many users do not need all cores to navigate, play, edit files, picture, or videos. And so many cores are for the most of the time useless, they are called the dark cores.

A neuromorphic processor is very similar to GPUs. In the case of Intel Loihi (which will have required six years of development), there are more than 130,000 neurons and 130 million synapses that connect them. As a reminder, a fly has 100,000 neurons; a human being has 100 billion. However, a human brain requires less energy than a conventional bulb. We are still a long way from it with Loihi, but the 14 nm engraving already makes it possible to create “clusters” of neurons (128 clusters for now). Each cluster contains 1024 neurons. Each neuron communicates with other neurons via synapses; it is inside these synapses that the learning engines are found. The synapses of a neuron will activate other neurons if necessary, ie if the information has changed enough between two neurons. Otherwise the neurons nearby will remain inactive.

Ultimately, the energy consumption will be much lower than with a conventional processor. And it is one of the goals sought besides the quest for performance.

Intel will provide universities Artificial Intelligence microprocessors to see how to optimize their system and especially to teach neurons how to work with different tasks.

Lo’ihi is the name of the volcano of the island of Hawaii. A young volcano in full construction.