image: Ilustration of artificial neural networks for AI./ Pedro David García (ICMM/CSIC). view more
Credit: Pedro David García (ICMM/CSIC)
The Institute of Materials Science of Madrid (ICMM), CSIC, is leading an international project that will create artificial neural networks that will be used in integrated photonic circuits. They will do so with less energy consumption, an essential advance to promote the development of artificial intelligence. The Neuropic project (Nano electro-optomechanical programmable integrated circuits) has received 3 million euros through Pathfinder grants from the European Union.
Artificial intelligence has long ceased to be a matter of science fiction. However, what for many years has been a dream has proven to have a dangerous component for the environment: solving the algorithms with which they work involves an unaffordable energy expenditure in the current context. That is why research into energy efficient projects is especially essential.
"Clearly we have a problem with energy: we are using algorithms more and more, but we are moving towards a future in which the abundance of energy that we have experienced in recent years is going to end and we need to find much more efficient ways to process information. That is the objective of the project", says Pedro David García, a CSIC researcher at the Institute of Materials Science in Madrid (ICMM) and coordinator of the project.
The teams, from four European countries, will imitate in electronic devices how the human brain works through a complex system of interconnected neurons. It is what is known as neuromorphic computing, an alternative way of processing information different from that of standard computers or mobile phones. They will also do so with a new and sustainable approach: "We propose using thermal energy from the system to create this type of artificial neurons," says García.
"We propose using the thermal energy that these structures already have at room temperature to create the functionalities of the neuronal system," explains the researcher, adding: "As they are very small systems, they vibrate at room temperature. These vibrations are generally harmful to other types of of applications, but we intend to take advantage of them and exploit them in our network of neurons.
All of this, moreover, will be carried out with silicon, a very abundant material in nature and whose associated technologies are highly developed, which also implies less damage to the environment.
An international nanotechnological project
Neuropic involves "some of the world's leading researchers and companies in nanotechnology, photonics, advanced manufacturing, and complex systems in a risky and highly interdisciplinary research project with the ambition of founding a new paradigm of programmable silicon photonics," says the researcher. . From his point of view, "nanotechnology is now ready to begin exploring programmable photonic chips scaled to thousands of programmable nodes."
The project, which will officially start on March 1 and has an estimated duration of four years, has the participation of two Universities (Technical University of Denmark and University of Cork), two Spanish Research Institutes (ICN2 and ICMM) and two small -medium-sized companies (Beamfox and Fincotec) from four different EU countries (Denmark, Ireland, Germany and Spain). While the Spanish centers will work more on the design and characterization of these complex photonic systems, it will be in Denmark where the first prototypes will be manufactured (which will be studied again in Spain). The 'partners' from Ireland and Germany, for their part, will collaborate with the DTU in the development and manufacture of the final prototype of the system. The researcher Cefe López (PC Group) is also participating in the project from the ICMM.
"We have some idea of what a neuron is for us, that is our proposal, but we have to manufacture it and characterize its properties in the laboratory," García emphasizes. The final objective will be to create a complex network of these neurons connected to each other and check how they react to the information introduced by photons to, in this way, "demonstrate that this network can solve Artificial Intelligence algorithms", points out García, a researcher with extensive experience in complex photonic systems.
"I find it very interesting to create artificial neural networks to test questions about complex systems and emerging processes," says García, who concludes: "This will translate into a way of solving all those algorithms that no longer depend on us, but energetically much more efficient. That's what these kinds of projects are for, to find much more sustainable ways of solving problems."