Bio-inspiration. When crickets inspire a CEA researcher to design memories on silicon

Did you know that we still have a lot to learn about insects, even in the field of electronics? This is the bet of Elisa Vianello, researcher at the CEA and coordinator of the CEA-Leti Edge AI program in artificial intelligence, who has been working for several years in the field of memory computing.

It has just received a grant of 3 million euros from the European Research Council (ERC), with the aim of developing new memory devices on silicon at the nanometric scale, inspired by the nervous system of insects.

With, among the objectives she has set for herself on this five-year research project (which will include four doctoral students and a post-doctoral fellow), the development of the very first smart chip, associated with a local neural module capable of processing sensory data in real time.

Because for the researchers, the whole challenge was summed up here: after having developed powerful algorithms aimed at enabling machines to learn through experience and with their environment, it was still necessary to develop memories, which today can be summed up in very small electronic devices (nanosystems), which must be powerful in the processing of their data, while being less and less energy-consuming.

This is where the study of crickets, and in particular of some of their characteristics, made sense, according to Elisa Vianello: because she discovered that the different functions of the insect’s nervous system closely resembled the functions carried out by the memories required for on-board intelligence projects.

Take inspiration from biology to reduce the consumption of these systems

“The study of crickets is not new, and for this I approached Professor Jérôme Casas of the University of Tours (also a member of the CEA/LETI/CARNOT chair in bio-inspired technology – Grenoble, editor’s note ), which had already developed bio-inspired micro-electromechanical systems (MEMS) sensors. Having myself worked in the field of artificial intelligence, with the challenge of manufacturing chips that reduce their own energy consumption, I wanted to draw inspiration from biology to see how it was possible to reduce the consumption of these systems when ‘we have to run AI algorithms », explains the researcher.

Because insects, and in particular crickets, would offer the scientific world new avenues to explore: “ we already knew that insects have a limited number of neurons and synapses, which makes them easier to study than the human brain for example, opening up the possibility of mapping them one by one “.

But it is above all the fact that these animals actually have not one, but several brains that particularly interested Elisa Vianello:

Crickets have a secondary brain in their abdomen, which allows them to process information received from sensors located on their tails and hairs. These capture the airflows and decide in particular locally if they should escape or not, without sending the information to the central brain. It is a very smart and energy efficient system “, underlines Elisa Vianello.

Because by processing certain information at the local level, such an organization inspired by insects could also make it possible to reduce the quantity of flows sent to the central server, by transmitting only a few more targeted and reduced packets. ‘Cause so farthese learning functions based on artificial intelligence are not done directly on the chips, since it is a very energy-intensive phase, and must necessarily be sent back to the cloud.

Bio-inspiration with crickets also opens a way to process the “noise” generated during data transmission. “ One of the objectives of our work will be precisely to make the comparison between the variability of biological noise, and that which can be encountered in devices resulting from nanotechnology”.

Work which, while still at the very early stage of research, should however pave the way for different types of applications in consumer robotics, implantable chips for medical diagnosis and portable electronics.

Applications in medical diagnosis and autonomous navigation

Among the first possible applications, Elisa Vianello mentions, for example, the field of autonomous navigation, and in particular drones and shuttles “ who must interact with their environment and manage continuous information flows, coming from several sensors, while making decisions and reducing their energy consumption in order to be able to process this information locally, without going through a cloud”.

Another longer-term application would be that of medical devices, and in particular devices that need to record signals from the heart in order to detect potential abnormalities. ” This is an area where there is also a lot of noisy information, which depends a lot on the patient and where it is very important to process the information locally, for privacy reasons”.

His research project should also allow him to overcome a technological obstacle by changing his approach: “On the memory side, we are going to work on developing a new technology by taking the best of the different types of existing memories, in order to combine them to obtain certain properties. The idea being not to develop things separately but on the contrary to maintain a strong coupling between memory and computing capacities, which must be developed together and in parallel “, explains Elisa Vianello.