3D Random Magnetic Nanowire Network as Potential Neuromorphic Computing Element

Interconnected magnetic nanowire (NW) networks can be the building blocks of 3-dimensional (3D) information storage and integrated neuromorphic computing platform. In this regard, we have previously shown discrete propagation of magnetic states driven by magnetic field and current in quasi-periodic 3D Co networks [1, 2]. In this study, we explore the possibility of utilizing random magnetic NW networks as neuromorphic computing elements. The NWs are synthesized via electrodeposition and sintered to form an interconnected network. Next, arrays of electrode pairs are fabricated to connect varying number of interconnected NWs. When magnetoresistance (MR) is measured between an electrode pair, multiple discrete jumps are observed due to domain wall pinning/depinning at the intersections, exhibiting step-by-step switching of the network. Each electrode pair shows unique MR feature as the number and geometry of NW interconnects vary between them. If this domain wall pinning/depinning is further controlled by applying current pulses of varying magnitudes, pulse-widths, or repetition numbers, synaptic weights could be diversely programmed by assigning different electrode pairs as inputs and outputs. This demonstration is a step towards integrating these 3D magnetic structures into artificial neural networks that can perform tasks such as pattern recognition.

[1] Burks et al, Nano Lett, 21, 716 (2021)

[2] Bhattacharya et al, submitted