Project description
Neural networks with strongly interacting particles
Exciton-polaritons are hybrid quasiparticles made of light (photons) and matter (excitons), and they inherit properties of both of them: they can propagate as fast as photons while displaying much larger interactions. The demonstration of exciton-polaritons at room temperature has paved the way for practical applications. Funded by the European Innovation Council, the PolArt project intends to use exciton-polaritons to realise artificial neural networks implemented into hardware rather than software. Exciton-polaritons promise faster processing speeds and lower energy consumption, contributing to more energy-efficient devices exploitable for recognition of images, sounds and genome-wide patterns of biomarkers.
Objective
Exciton-polaritons, hybrid light-matter particles, have recently come into the spotlight for their peculiar properties (sizable interaction, small mass, long coherence, etc.) leading to spectacular effects such as phase transitions, superfluidity, bistability, ultra-efficient fourwave-mixing, and quantum blockade. On the other hand, polaritons have also been proposed for different kinds of devices (including optical switches, transistors, low threshold lasers and simulators), with beautiful experiments showing proofs-of-principle. However, it is only recently that polaritons have been operating efficiently at room temperature, giving the promise of a real technological impact in the future. In a recent work, made by some of the theoretical and experimental partners of this proposal, we could demonstrate that such hybrid state of matter, when used for realising artificial neural networks, shows extremely interesting performances in terms of speed and success rate.
Given the strong interest in the realisation of hardware-based (not simulated) artificial neural networks, the goal of PolArt is to demonstrate a new way to build artificial intelligence-dedicated circuits using polariton neural networks as optical accelerators.
Thanks to this new concept device, complex applications related to neural-like processing, will be efficiently implemented, therefore enabling neuromorphic computation to be done in small devices that cannot rely on remote, large bandwidth connection. This proposal benefits from the contribution of several complementary partners coming from many different research areas (material science, physics, optics, chemistry, genetics) and industrial participants that assure the interdisciplinarity and technological oriented target.
Fields of science (EuroSciVoc)
CORDIS classifies projects with EuroSciVoc, a multilingual taxonomy of fields of science, through a semi-automatic process based on NLP techniques.
CORDIS classifies projects with EuroSciVoc, a multilingual taxonomy of fields of science, through a semi-automatic process based on NLP techniques.
- natural sciencesbiological sciencesgenetics
- natural sciencesphysical sciencesopticslaser physics
- natural sciencescomputer and information sciencesartificial intelligencecomputational intelligence
You need to log in or register to use this function
Keywords
Programme(s)
- HORIZON.3.1 - The European Innovation Council (EIC) Main Programme
Funding Scheme
HORIZON-EIC - HORIZON EIC GrantsCoordinator
00-927 Warszawa
Poland