In-memory computing (neuromorphic spintronics)

The framework of this post-doctoral position is the EIC Pathfinder MultiSpin.AI project. The latter is a radically novel approach for neuromorphic computing in which we are developing a multi-level nano synapse based on M2TJ (multi-level magnetic tunnel junction) cells. In contrast to binary MTJs, i.e., with two states, the M2TJs use multi-level (2n) or n-ary state cells. Those levels can thus be encoded as an n-bit word which improves energy efficiency, processing speed for fast AI inference, and reduces neuromorphic chips footprint, especially at the edge, at least n times. A paradigm shift of this magnitude requires various ambitious technological advancements, whose feasibility is supported, by some of our recent publications. Building on the above, MultiSpin.AI will achieve a technological breakthrough for spintronic neuromorphic computing by developing a disruptive AI co-processor since it will combine massively parallel local in-memory processing of data, and high energy efficiency that are 10x that of the most advanced neuromorphic devices. This will facilitate efficient and sustainable AI inference, especially for edge computing, enabling the acceleration of the development of AI solutions for complex grand challenges in various fields.