Computing in Time Using Magnetic Excitations

Avik Ghosh, University of Virginia

June 18, 2021

The size scaling of magnetic memory devices is constrained by their read-write energies, latency and retention. These constraints arise from fundamental limits on drivability and reliability, as well as material limitations. Computing in the time domain has the potential to significantly reduce the energy cost of computation. For low-barrier soft magnets, probabilistic computing has been proposed as a way to quickly scan the available phase space for various search and optimization algorithms. We will argue that a variant of these soft magnets can be used to construct an analog stochastic neuron, which may be well suited for analog information processing such as reservoir computing for temporal inferencing and event based imaging. On the opposite end are high barrier non-volatile magnets, where scalability maybe be accomplished while retaining adequate storage barrier by engineering solitonic excitations such as domain walls, as well as topological skyrmions with added spatial symmetry breaking. We will discuss the underlying materials physics that controls the size, speed and lifetime of isolated metastable skyrmions, ways to deal with their Magnus force, and their potential use in temporal information processing, such as a native memory for certain classes of timing based logic.