Nanoconstriction spin-Hall nano-oscillators (NC-SHNOs) have shown promise for using spintronics to expand non-conventional computing, particularly in neuromorphic systems that mimic neural networks. However, most work in spintronic computing, remains theoretical, with many of the technical details about the performance and effects of different materials in question.
Researchers simulated the micromagnetic properties of NC-SHNO arrays focusing on how synchronization achieves computational performance. Capriata and Malm constructed model 2x2 and 4x4 arrays of micromagnetic oscillators separated by up to 300 nm while virtually inserting grains in the free magnet layer of the devices. They studied the impact of these changes on a variety of oscillator properties, such as stable current operation range, synchronized oscillator frequencies, and collective output power.
The results shed light on the intrinsic working principles of these simple models of synapses and how they reach fully synchronized states for neuromorphic computing, said Capriata.
“The inclusion of the presence of local variations in material properties directly into the micromagnetic simulations is a technique developed at KTH,” he said. “Moreover, we evaluated extensively a plethora of different array configurations and we gave hints at how larger arrays might behave.”
For higher fidelity, the group supplied their model with real layer characteristics from spintronic devices using an atomic force microscope.
They found that SHNO arrays are mostly stable and continue to show potential for neuromorphic computing. Their work, for example, revealed that de-coupling oscillators along rows or columns can boost power output, likely from more favorable phase shifts between the oscillators.
The group looks to verify their findings with microwave measurements and use their realistic simulation technique on materials and spintronic memory devices.
Source: “Grain structure influence on synchronized two-dimensional spin-Hall nano-oscillators,” by Corrado Carlo Maria Capriata and Bengt Gunnar Malm, AIP Advances (2023). The article can be accessed at https://doi.org/10.1063/5.0147668.