Meander geometry based skyrmionic synapse for image classification

The neuromorphic computing paradigm, inspired by the biological nervous system, aims to replicate brain-like functions, composed of numerous artificial neurons interconnected through an even greater number of artificial synapses. Magnetic skyrmions are promising for neuromorphic computing due to their nanometric size, inherent topological stability, energy efficiency, and remarkably low depinning current density. However, the inherent stochastic nature of skyrmions can hinder the precise manipulation and control of skyrmions for synaptic applications, which is crucial for reliable computing. Hence, we present a meander geometry for a skyrmion-based artificial synapse device. The geometry allows the control of the spatial distribution and dynamics of skyrmions. Control over skyrmionic behavior enables the emulation of essential synaptic functions, such as synaptic weight modulation. Providing the external stimuli, the synaptic weights of the proposed device can be adjusted to replicate the process of potentiation and depression observed in biological synapses. Finally, the synaptic device characteristics are employed to simulate a multilayer perceptron (MLP) neural network architecture for the modified national institute of standards and technology (MNIST) dataset achieving a recognition accuracy of ∼ 95 %. Furthermore, the accuracy of ∼ 85 % is achieved with VGG8 architecture for Canadian institute for advanced research (CIFAR-10) dataset classification with proposed synaptic device weights. This work provides an innovative perspective regarding combining meandered step geometries and skyrmion technology to develop effective and adaptive synapses for neuromorphic computing.