Voltage-Controlled Skyrmion Breathing Mode for Reservoir Computing

Reservoir computing (RC), a neuromorphic computing paradigm enabling universal approximation, faces challenges in structural complexity and energy efficiency, particularly in hardware implementations. Specifically, spintronic reservoir has attracted great interest due to its nanoscale integration, high durability, and physical mapping of nonlinear dynamics. Here, we propose a skyrmion-based physical reservoir with energy-efficiency based on simulation methods, leveraging the oscillatory relaxation dynamics triggered by the voltage for sub-femtojoule (0.15 fJ) nonlinear transformations. The skyrmion breathing mode—driven by voltage-tunable perpendicular magnetic anisotropy—exhibits robust nonlinear oscillations, emulating the dynamic behavior of reservoir neurons to capture high-dimensional temporal features. Moreover, a rectified linear unit (ReLU) neuron and high linear synapse based on this skyrmion device enhance classification performance in fully connected layers. By constructing an all-spin RC network, we demonstrate effective pattern recognition, achieving 91.7% accuracy on the Modified National Institute of Standards and Technology (MNIST) dataset. Our work offers a pathway toward high-density, energy-efficient RC by voltage-controlled nonlinear dynamics.