EU H2020 NEURONN: 2D Oscillatory Neural Networks For Energy Efficient Neuromorphic Computing

Stefania Carapezzi (1) , Gabriele Boschetto (1) , Corentin Delacour (1) , Madeleine Abernot (1) , Thierry Gil (1) , Aida Todri-Sanial (1)

1 SmartIES - Smart Integrated Electronic Systems

In this paper, we showcase a leading-edge implementation of oscillatory neural networks (ONNs) using beyond Complementary-Metal-Oxide-Semiconductor devices based on vanadium dioxide to mimick neurons, and 2D molybdenum disulfide memristors to emulate synapses. We explore the ONN technology through simulations from materials to devices up to circuits. We show that ONNs naturally behave like associative memories and can be used for pattern recognition, a task to be exploited in edge devices. Finally, we develop a reconfigurable digital ONN-on-FPGA to assess ONN functionality in real world applications.

Mots clés: Oscillatory Neural Network (ONN); Neuromorphic computing; Beyond-CMOS device; vanadium dioxide (VO2); Molybdenum disulfide (MoS2); Memristor; pattern recognition; FPGA

Domaines: Intelligence artificielle [cs.AI]; Modélisation et simulation; Technologies Émergeantes [cs.ET]; Matériaux; Micro et nanotechnologies/Microélectronique

Stefania Carapezzi, Gabriele Boschetto, Corentin Delacour, Madeleine Abernot, Thierry Gil, et al.. EU H2020 NEURONN: 2D Oscillatory Neural Networks For Energy Efficient Neuromorphic Computing. 15ème Colloque National du GDR SoC², Jun 2021, Rennes, France. , 2021. ⟨lirmm-03270397⟩