Researchers from the CNRS and the Physics Laboratory at ENS-PSL have developed a prototype ionic neuron with the same transmission properties as a neuron. This innovative research, published in Science, could pave the way for the implementation of simple learning algorithms, and thus serve as the basis for the electronic memories of tomorrow.

Brain-inspired electronics have been booming for a few years. But the brain’s overall efficiency is unbeatable, as Lydéric Bocquet, CNRS research director at the Physics Laboratory and professor at ENS-PSL, explains in an interview with the establishment:

“For an energy consumption equivalent to two bananas a day, the human brain is capable of performing a large number of complex tasks, whereas artificial intelligence can only perform these at the cost of energy consumption tens of thousands of times higher.”

While the origin of this incredible performance is still poorly known to scientists, the prototype artificial ionic neuron produced by Lydéric Bocquet’s team and his students could provide new answers as to the origin of the brain’s performance and open up prospects for the development of electronic memory.

Nanofluidics for artificial intelligence

The researchers want to create electronic systems that are as energy-efficient as the human brain, systems that will use ions rather than electrons as information carriers. Lydéric Bocquet details:

“Considerable progress has been made in the field of fluid transport at nanoscales. However, the artificial systems developed so far remain far from the impressive machinery that exists in Nature: ultra-selective channels, ion or proton pumps, channels that open or block under certain stimuli, etc. These are true “ionic machines”. These are real ‘ionic machines’ that far surpass anything we know how to make artificially, both in terms of complexity and efficiency of the functions performed.

For him, nanofluidics opens new perspectives for artificial intelligence:

“By using the new behaviors of fluids at nanoscales, we can mimic some of these functions that usually take place in the brain and imagine basic bricks to artificially build these famous ionic machines.”

He and his team are working on ultrathin channels a few nanometers or even angstroms thick, and exploring ion transport in two-dimensional structures, in this case a single layer of water molecules made of graphene nanoslots.

“Two-dimensional physics is always a source of singular, even bizarre behavior… and that’s indeed what we observed for ion transport in these sheets: bizarre.”

Experiments with unexpected results

Since 2014, Lydéric Bocquet has been the head of the microMégas team at the ENS Physics Laboratory, which he co-founded with Alessandro Siria, a CNRS researcher, and joined in 2019 by Antoine Niguès, an ENS research engineer. Scientists there develop experiments and theories to understand how fluids and ions flow, in channels whose size can reach only a few molecules. This is the field of nanofluidics, a rapidly expanding area. He explains:

“At these infinitesimal scales, new and sometimes exotic properties emerge and we need to rethink the way fluids flow, to propose a new framework.”

Together with his team, they faced many challenges:

“First and foremost, we had to invent a completely new experimental toolbox to be able to build such systems with such tiny dimensions, which we did by making some sort of “legos” at the nanoscales, which use in particular all the new nanomaterials that have emerged in the last ten years or so, nanotubes, graphene…”

We also had to develop suitable measuring instruments and new techniques to measure how fluids flow in such channels.

This has led to the discovery of unexpected phenomena, such as the fact that water flows almost frictionlessly in carbon nanotubes, especially the small ones – a singular quantum coupling between water and carbon. Lydéric Bocquet believes that:

“fluids at nanoscales are therefore proving to be much stranger than we could have anticipated, and nanofluidics is a teeming research topic, which opens up multiple very fundamental questions”.

Research that tries to mimic nature

First, the team developed theoretical predictions: under the action of an electric field, the ions from this layer of water assemble into elongated coils and then develop a memory of stimuli received in the past. This “memory” leads to a property known as the “memristor effect”, i.e. “memory resistance”. In neural systems, these memristors play a central role. In biological systems, this memory is built up through ion channels that open and close to modulate the current. Lydéric Bocquet summarizes:

“In the artificial systems that interest us, the origin is very different, but the result is very similar. So we have to find tricks to do as well as Nature.

In a second step, the researchers were able to show that these artificial ionic systems are capable of reproducing the physical mechanism of the emission of action potentials from neurons, and thus of transmitting information.

“Of course, in this constant back and forth between theory and experiments, we very quickly went hunting for memristors in our nano-slit systems… and we finally found them!”

An artificial reproduction of the behavior of a synapse, which by regulating the link between two neurons, allows learning, was possible. Lydéric Bocquet tempers:

“These artificial ionic machines are still primitive, and far from the fantastic architecture of the brain. But if making an artificial ‘ionic computer’ is still a dream, these advances open up a new path and will also allow us to better understand the role of ions as information vectors.”

And when asked what the next step is:

“Implementing simple learning algorithms, which can serve as the basis for tomorrow’s electronic memories.”

Start-ups that care about the environment

The results of this research can be applied in connection with the environmental transition. For example, some materials can convert differences in salinity into electrical energy very efficiently, which is what the start-up Sweetch Energy is developing with osmotic energy in collaboration with the microMégas team and is now working on the industrialization of the process. For Lydéric Bocquet:

“This is a major breakthrough, there is considerable hope towards this completely renewable and non-intermittent energy source.”

The new fundamental properties revealed at the nanoscale will make it possible to “develop breakthrough innovations, for example in the field of membranes or nanotechnologies”. With his colleague Alessandro Siria, co-founder of microMegas, he has since created two other start-ups, Hummink, which does nanoscale printing for the semiconductor industry, and Altr, which develops membranes to reduce the alcohol content in drinks.

According to Lydéric Bocquet:

“Nanofluidics is at the confluence of various fields. It is the frontier where the continuum of fluid mechanics meets the atomic nature of matter, or even its quantum nature

.
The research proposes a new framework for fluid dynamics at small scales. One of the main axes concerns the interface between fluid dynamics and the quantum world, as Lydéric Bocquet explains:

“Our recent work on quantum friction opens up multiple avenues for modulating flows by quantum effects, including the possibility of a form of “quantum engineering” of flows in certain materials that we are currently exploring. Bio-inspiration is also a completely open field, with the aim of developing artificial channels whose properties reproduce – thanks to emerging properties – their biological counterparts: developing ion pumps, stimulable channels, all aiming to develop an “ion-tronics”.

Nanofluidics is also of interest in the field of membranes, with new materials and concepts for separation, remediation and desalination: membranes based on graphene, graphene oxide, etc. For example, desalination membranes are very useful in the field of agriculture. Graphene, a material that is highly resistant to the aggression of chemical agents, is also particularly interesting for the textile industry. For Lydéric Bocquet, nanofluidics is therefore an exciting universe where everything remains to be discovered.

Reference:
Modeling of emergent memory and voltage spiking in ionic transport through angstrom-scale slits. Paul Robin, Nikita Kavokine and Lyderic Bocquet. Science, August 6, 2021. DOI: 10.1126/science. abf7923

Translated from Un neurone artificiel pourrait servir au développement de la mémoire électronique de demain