Neuro-inspired information processing: a book by Alain Cappy
An overview of research on bio-inspired systems :
from electronic components to the architecture of neurons, it's just a short step ...
Interview with Alain Cappy, former Director of IEMN and IRCICA.
Edited by Karine Perrot, IEMN Publishing, Communication and Multimedia Department
Since the brain is reputed to be an extraordinary machine, can we draw inspiration from it to designing intelligent, energy-efficient systems ?
This is the question, as fundamental as it is pragmatic, that Alain Cappy asked himself when he switched his research focus from electronic components to neural circuits, and which he attempts to answer in his book Neuro-inspired information processing(September 2020).
This ground-breaking book sets out to assess summary of work on neuro-inspired systems for information processing. It presents the essential concepts of electronic information processing and the operating principles of the cerebral cortex, and deals with the properties and networks of living and artificial neurons and synapses. The energy issues is the common thread running through the targets students and young researchers wishing to focus on this subject.
The researcher's path began with a two-fold observation. Firstly, the multiplication of terminals, linked to the activity of smartphones and computers, is leading to a rapid growth in electricity consumption equivalent today to 15 % of world consumption. Secondly, current technology has reached its physical limitswith the end of the Moore's Law which predicted that the power of microprocessors would double every eighteen months. With the addition of the energy constraint, the architecture of the information processing system had to be completely overhauled, and Alain Cappy's conversion was an obvious choice:
"For a long time, I had been working on ultra-fast devices that consumed a lot of power and could not be significantly reduced. It was while I was looking for more energy-efficient architectures that I got the idea of looking at information processing in the brain, and therefore at the properties of neurons. I was very surprised to find that the physics of how neurons work was similar to that of the components (field effect transistors) that I had always worked on. We can even draw a formal analogy between the insulating membrane that surrounds the neuron and the insulator that forms the basis of the transistor effect in semiconductor devices".
Integrated circuit emulating artificial neurons
Moving into familiar territory, and surrounding himself with a team of researchers, he quickly got down to the nitty-gritty. But we had to go beyond the biological model. technology isn't about copying, it's about trying to do things differently and better ».
That's how the team made artificial neurons which are energetically 1000 times more efficient than a living neuron, i.e. they use around 1000 times less energy to create the same electrical impulse. The reason is simple: an electronic neuron is 1000 times smaller than a living neuron, so the ratio of sizes determines the ratio of energies.
Alain Cappy opened two lines of research in the field of bioinspired architectures, which is now a major area of research at IEMN and IRCICA. One, with François Danneville and Christophe Loyez, aims to be inspired by the architecture of the brainto create intelligent and efficient vision systems. The other way, to therapeutic aimwith Virginie Hoel, is working to make living neurons interact with artificial neurons to form implants, in order to treat neurodegenerative diseases. The start up Axoruscreated in 2019 and dedicated to treatment of AMD (age-related macular degeneration), aims to promote this research.
https://onlinelibrary.wiley.com/doi/book/10.1002/9781119721802
