3D networks of artificial neurons for new generation of computers – Journal en direct – Le journal de la recherche et du transfert de l’Arc Jurassien

Photo Sandrine Quarroz, FEMTO-ST Institute

Digital computers are unquestionably champions in performance and computing speed. But the human brain remains the master of solving complex problems, which requires the acquisition of prior knowledge. This faculty of learning comes in the technique of machine learningwhich by giving a computer the possibility ofintegrate a large amount of informationlays the foundations of artificial intelligence.

Directly inspired by the functioning of the brain, artificial neural networks are a tool to complete this learning. But their development potential comes up against a technical limit: the electronic systems on which the operation of the classical computer is based are in two dimensions, a configuration which only admits the connection of a few tens of artificial neurons, where it would take millions in the ideal.

CNRS research fellow at the FEMTO-ST Institute, Daniel Brunner works at the development of neural networks according to a 3D optical architecture , which allows the implementation of artificial neurons and their connections. “A network of neurons and connections is printed on the geometrically available physical surface, then this network extends in 3D like the branches of a tree.

This deployment takes place thanks to matrices of lasers that create new neurons and new connections in an entirely optical process,” explains Daniel Brunner. The process, based on the mathematical principle of nonlinear dynamics, is inspired by the human brain as much for its 3D architecture as for its neural connection system.

Neuromorphic calculators using artificial neural networkspromise an impressive change of scale in the computer field. According to Daniel Brunner, they will one day capable of producing 20 billion results per second, an incredible leap if we compare this figure to the 100,000 results/s already provided by a conventional computer for the purposes of artificial intelligence. And to stay within the equivalences, the operation of this new generation computer would require 10 to 1,000 times less energy . “There is no doubt for me that the 3D neuromorphic calculator will see the light of day; it will be hybrid, combining the advantages of electronics for signal transformation and opticsfor the transmission of information”, assures the researcher.

An ERC grant and a CNRS bronze medal

Developed within the framework of the INSPIRE project, this research has now earned Daniel Brunner the award of a remarkable ERC grant. Consolidator Grants1 2021, for an amount of 2 million euros, which will support the continuation of its work over five years; the researcher also obtains a magnificent bronze medal from the CNRS 2022.

1 ERC Consolidator Grants is a European funding aimed at researchers who obtained their thesis 7 to 12 years previously. In France, 29 projects were rewarded in this context for the year 2021.

We would like to thank the writer of this write-up for this amazing material

3D networks of artificial neurons for new generation of computers – Journal en direct – Le journal de la recherche et du transfert de l’Arc Jurassien

You can view our social media pages here and other pages related to them here.https://www.ai-magazine.com/related-pages/