Gradual magnetization switching via domain nucleation driven by spin–orbit torque

Abstract

Gradual magnetization switching driven by spin–orbit torque (SOT) is preferred for neuromorphic computing in a spintronic manner. Here we have applied focused ion beam to selectively illuminate patterned regions in a Pt/Co/MgO strip with perpendicular magnetic anisotropy, soften the illuminated areas, and realize the gradual switching by a SOT-driven nucleation process. It is found that a large in-plane field is helpful to reduce the nucleation barrier, increase the number of nucleated domains and intermediate states during the switching progress, and finally flatten the switching curve. We proposed a phenomenological model for descripting the current dependence of magnetization and the dependence of the number of nucleation domains on the applied current and magnetic field. This study may promote the birth of SOT devices applicable in neuromorphic computing applications.

Spin–Orbit Torque‐Induced Domain Nucleation for Neuromorphic Computing

“…In the following sections, we used our L1 1 ‐CuPt/CoPt bilayer as an artificial neuron to build and train a deep learning network. In a previous study, [ 26 ] both theoretical calculations and experimental observations show that the number of nucleated domains exhibits a normal distribution with the magnitude of SOT. We illustrate in Figure a that this mechanism naturally transforms our bilayer device to a domain nucleation‐based sigmoidal neuron.…”

“…Second, domain nucleation can be understood from an energy perspective. The switched magnetization ( m ) per SOT (τ) is proportional to the probability of switching () and the proportion of unswitched area ( 1−m ), i.e., , [ 26 ] where E n , k B , and T are the switching energy barrier, Boltzmann constant, and the temperature, respectively. E n has a spatial distribution across the sample, which follows the normal distribution statistically.…”

“…[ 21,24,25 ] Microscopically, the magnetization is switched via either the coherent mode or the incoherent modes, where the latter are further broken down as domain wall propagation and domain nucleation. [ 26 ] The incoherent modes are particularly valuable for neuromorphic computing, since they not only lower the switching threshold current thus improve the energy efficiency, but also provide stable intermediate states that effectively transform to memristive plasticity. Spintronic devices based on domain wall propagation have been utilized to mimic synapses [ 25 ] and spiking neurons, [ 4 ] or to demonstrate spike‐timing‐dependent plasticity.…”

In‐Memory Mathematical Operations with Spin‐Orbit Torque Devices

“…The mechanism of magnetic field sensing can be understood that the I EN causes the demagnetization state of device, and thus a domain nucleation‐dominated magnetization reversal is very sensitive to the collinear magnetic field, either direction or magnitude, caused by SOT. [ 24,25 ] It was proved by the magneto‐optical Kerr effect (MOKE) microscopy investigation. The MOKE images (Figure 1c) depict that the proportion of − M z domains (shown in black) grows in a dispersed manner when scanning H x from +40 to −40 Oe, together with I x = 30 mA, after initializing the magnetization to the saturated state under +200 Oe.…”

“…1b). The mechanism of magnetic field sensing can be understood that the IEN causes the demagnetization state of device, and thus a domain nucleation-dominated magnetization reversal is very sensitive to the collinear magnetic field, either direction or magnitude, caused by SOT 24,25 . It was proved by the magnetooptical Kerr effect (MOKE) microscopy investigation.…”