Objective: Magnetic skyrmions in synthetic antiferromagnetic (SAF) systems have emerged as highly promising information carriers for next-generation non-volatile memory and neuromorphic computing paradigms, owing to their nanoscale dimensions, suppression of the skyrmion Hall effect (SkHE), and high topological stability. While voltage-controlled magnetic anisotropy (VCMA) offers an ultralow-power approach for skyrmion manipulation, existing research predominantly focuses on the global VCMA modulation of the free layer. The dynamic evolution of skyrmions driven by localized, size-dependent VCMA effects remains largely unexplored. To address this gap, this study systematically investigates the dynamic evolution—specifically the generation and annihilation mechanisms—of skyrmions under localized, size-dependent VCMA modulation, and subsequently proposes a highly efficient neuromorphic computing architecture.
Methods: Micromagnetic simulations were rigorously conducted using the OOMMF framework on a cylindrical SAF magnetic tunnel junction (MTJ) model. The free layer features a Top FM/NM/Bottom FM structure (e.g., CoFeB/Ru/CoFeB) with a fixed diameter of 100 nm, whereas the top electrode diameter (D) is adjustable from 50 nm to 100 nm. By applying vertical bias voltages, localized electric fields induce VCMA effects at the MgO/Top FM interface, dynamically lowering or raising the local perpendicular magnetic anisotropy constant (K_u) to precisely drive skyrmion generation and annihilation.
Results: The simulation results reveal a profound size-dependent effect during localized VCMA modulation. As the electrode diameter (D) decreases, the threshold voltage required to induce skyrmions increases, correspondingly broadening the effective K_u modulation window. Remarkably, under identical VCMA driving conditions, a deterministic polarity reversal of the generated skyrmions is observed. Specifically, skyrmions generated under smaller electrodes (D≤82 nm) exhibit a downward polarity, whereas those under larger electrodes (D≥90 nm) exhibit an upward polarity. Comprehensive energy analyses indicate that this bifurcation fundamentally stems from the energetic competition between the exchange energy (E_ex) and the Dzyaloshinskii-Moriya interaction (DMI) energy (E_dmi) during the intermediate nucleation phase. For larger sizes, the system minimizes E_ex, leading to an edge-canted magnetic state; conversely, for smaller sizes, the system maintains a low E_dmi, preserving the initial vertical edge spins. As E_dmi further decreases, these distinct intermediate states deterministically induce skyrmions of opposite polarities.
Leveraging these unique size-dependent physical characteristics, we construct a hierarchical, VCMA-controlled skyrmion neuromorphic computing system. In this architecture, D=80 nm MTJ devices are configured as a 4×4 crossbar array to provide highly stable, non-volatile binary synaptic weight storage. Concurrently, D=90 nm devices utilize a distinct four-state cyclic magnetic switching mechanism to intrinsically emulate biological leaky integrate-and-fire (LIF) neurons. This design accurately replicates resting, integration, firing, and essential refractory periods at the device level, effectively eliminating the need for complex external CMOS circuitry. Algorithm-level validation using a spiking neural network (SNN) on the MNIST dataset achieved a robust classification accuracy of approximately 92%.
Conclusions: This paper systematically investigates the impact of localized voltage-controlled magnetic anisotropy (VCMA) modulation on the generation and annihilation of skyrmions in synthetic antiferromagnetic (SAF) magnetic tunnel junctions. Our results indicate that as the electrode size decreases, the threshold voltage required for skyrmion generation increases, accompanied by an expanded modulation range of the magnetic anisotropy (K_u). Additionally, we reveal a size-dependent skyrmion polarity reversal phenomenon driven by energy competition. Practically, small-size modulation facilitates the generation of skyrmions for constructing binary memory cells, whereas large-size modulation enables controllable multi-state cyclic switching to realize artificial neuron functions. Leveraging these size-dependent characteristics, we propose a neuromorphic computing system featuring high stability, high integration density, and ultra-low power consumption.