MEMOS is an interdisciplinary project at the interface between Spintronics, Neuroscience and Microelectronics. Its goal is to implement an Associative Memory, a bio-inspired concept that operates differently from classical Random Access Memories, by exploiting the complex properties of magnetic nano-devices. Associative memories are memories with cognitive abilities, allowing information to be reconstructed from noisy or incomplete data. Because their operations are performed in parallel, they are fast and consume little energy. They are inspired by the brain, for which it is thought that associative memory operations are performed via the synchronization of neural assemblies. Indeed neurons can in some contexts be abstracted as weakly coupled nonlinear oscillators, and it has been shown theoretically that arrays of oscillators can perform associative memory operations. In that case, data storage and retrieval is obtained by modifying the coupling between oscillators. Although several implementations have been proposed, experimental demonstrations are extremely scarce. One reason is that the performance of these memories depends on the number of oscillators in the network. Spin Torque Nano-Oscillators (STNOs), which are nano-scale, non-linear, tunable oscillators that can couple and synchronize, therefore appear as extremely promising candidates as building blocks for large scale, fast, low power associative memory processors. Our main research objectives are to build a lab-scale demonstrator of an associative memory based on STNOs, and to demonstrate its potential for scaling to large networks through multi-physics and multi-scale simulations. In order to achieve its ambitious goals, the interdisciplinary MEMOS consortium, by uniting internationally recognized academic (3) and industrial (1) partners, gathers the wide spectrum of necessary expertise: bio-inspired computing approaches (UMPhy-CNRS, IEF), synchronization and coupling of STNOs (experiments: UMPhy-CNRS, CEA-Spec & theory: IEF, UMPhy-CNRS), spintronics/CMOS system co-integration and design (IEF), large scale dynamic simulations (IEF, Thales-TRT) and evaluation of benefit for applications (IEF, Thales-TRT).
Unité Mixte de Physique CNRS/Thales (Laboratoire public)
L'auteur de ce résumé est le coordinateur du projet, qui est responsable du contenu de ce résumé. L'ANR décline par conséquent toute responsabilité quant à son contenu.
CEA/DSM/IRAMIS/Service de Physique de l'Etat Condensé
Université Paris-Sud / Institut d'Electronique Fondamentale
Unité Mixte de Physique CNRS/Thales
Aide de l'ANR 769 428 euros
Début et durée du projet scientifique : septembre 2014 - 48 Mois