Artificial Intelligence with Microwave Spintronic Nanodevices – SPIN-IA

SPIN-IA is an interdisciplinary project which objective is to develop the foundations for an hybrid CMOS/spintronics chip accelerating Artificial Intelligence. Recently, UMPhy and C2N, with the participation from Thales, have obtained pioneer results showing that magnetic tunnel junctions can achieve pattern recognition with remarkable performance. A small neural network of four junctions can learn how to recognize vowels pronounced by different speakers with success rates above 85%. In these lab-bench demonstrations, the magnetic tunnel junctions emulate neurons and communicate with the microwave signals they emit. This type of wireless communication can allow building large scale, densely interconnected neural networks, especially since magnetic tunnel junctions are nanodevices and can be fabricated by billions in CMOS circuits. In the Astrid SPIN-IA project, we want to step up from the lab-bench proof-of-concept towards a packaged prototype, to demonstrate that these systems can achieve complex tasks like image recognition and to design the RF circuits that will interface the different layers of neuron-junctions in deep networks. Towards this goal, we will make all the necessary scientific and technical advances: physics of coupled junctions networks (UMPhy and C2N), system-level design and dedicated algorithms (C2N, Thales and UMPhy) and the design of compatible RF CMOS circuits (IMS and Thales). At the end of this two years project, the outcomes of SPIN-IA will allow building a chip densely co-integrating tunnel junctions and transistors towards an embedded artificial intelligence, fast, compact, with low energy consumption, capable to process in real time important data for civilian and military applications.