Elliptical skyrmions are a noncircular variant of magnetic skyrmions whose dynamics are exceptionally rich and complex owing to shape anisotropy, and they hold strong promise for next-generation spintronic devices. Our theoretical analysis and micromagnetic simulations show that, under spin-transfer torque driving, both the translational velocity and the skyrmion Hall angle of elliptical skyrmions vary anisotropically as the angle between the ellipse major axis and the current direction is changed. $\varphi$ is defined as the angle between the major axis of the ellipse and the current; $\varphi$ does not determine the shape of the magnetic structure, the magnetic moment distribution, or the magnetic moment tilt angle. Remarkably, the dependence of the skyrmion Hall angle ${\theta }_{\mathrm{SkHE}}$ on $\varphi$ follows a sigmoid-type nonlinear activation function. Exploiting this intrinsic nonlinearity, we fitted ${\theta }_{\mathrm{SkHE}}\left(\varphi \right)$ to a sigmoid nonlinear activation function and used the fitted activation to train a three-layer, fully connected neural network. The training accuracies for both the activation function and the network weights exceeded 90%, and the software simulation achieved up to 97% accuracy. This paper theoretically proposes a sigmoid activation method for spintronic neural network computing.

Close

PrevNext

View figure in article

• Magnetic texture
• Magnetism
• Magnetization dynamics
• Micromagnetism
• Spintronics