From stochasticity to functionality: harnessing magnetic domain walls for probabilistic and neuromorphic computing

Domain walls (DWs) in magnetic nanowires have been of intense interest due to proposals to use them to represent data in logic and memory devices. However, these have been challenging to realise because DWs behaviour is highly stochastic, making conventional digital devices unreliable. Here, we show how embracing DW stochasticity as a functional feature can facilitate novel computational devices. We first present results showing how integrating tuneable stochastic DW pinning into DW logic networks allows “stochastic computing”, where numbers are represented by random bit streams and individual logic gates perform complex mathematical operations. We then go on to demonstrate how DW stochasticity can be used to facilitate neuromorphic devices: (a) a neural network where the probabilities of DW propagation through nanowires perform the roles of synaptic weights and (b) a reservoir computing platform based on the emergent dynamics of DWs within an extended nanowire ensemble.