In MANNGA, we seek to explore and challenge the limits of spin-based devices and their energy efficiency. This will be achieved by combining two inherently energy-efficient technology paradigms: (i) magnonics (using spin waves – low energy magnetic excitations – to process signals and data) and (ii) neuromorphic computing (using large-scale integrated systems and analog circuits to solve data-driven problems in a brain-like manner). Going well beyond existing paradigms, we will use nanoscale chiral magnonic resonators (CMRs) as building blocks of artificial neural networks. The power of the networks will be demonstrated by creating magnonic versions of field programmable gate arrays, reservoir computers, and recurrent neural networks. The ultimate efficiency of the devices will be achieved by (a) maximising their magnetic nonlinearity (via spin wave power focusing within chiral magnonic resonators of minimal intrinsic loss); (b) using epitaxial yttrium iron garnet (YIG), which has the lowest known magnetic damping allowed by physics, for thin film magnonic media and resonators; and (c) using wireless delivery of power (minimising Ohmic loss in interconnects). Sensitive to the resonators’ micromagnetic states, such artificial neural networks will be conveniently programmable and trainable within existing paradigms of magnetic data storage. The latter includes magnetic random-access memory (MRAM), which is already compatible with CMOS, while compatibility with other technology paradigms of spintronics will also be sought, explored, and exploited.

Read more