Magnetic skyrmion artificial synapse for neuromorphic computing (1907.00957v2)

Abstract: Since the experimental discovery of magnetic skyrmions achieved one decade ago, there have been significant efforts to bring the virtual particles into all-electrical fully functional devices, inspired by their fascinating physical and topological properties suitable for future low-power electronics. Here, we experimentally demonstrate such a device: electrically-operating skyrmion-based artificial synaptic device designed for neuromorphic computing. We present that controlled current-induced creation, motion, detection and deletion of skyrmions in ferrimagnetic multilayers can be harnessed in a single device at room temperature to imitate the behaviors of biological synapses. Using simulations, we demonstrate that such skyrmion-based synapses could be used to perform neuromorphic pattern-recognition computing using handwritten recognition data set, reaching to the accuracy of ~89 percents, comparable to the software-based training accuracy of ~94 percents. Chip-level simulation then highlights the potential of skyrmion synapse compared to existing technologies. Our findings experimentally illustrate the basic concepts of skyrmion-based fully functional electronic devices while providing a new building block in the emerging field of spintronics-based bio-inspired computing.

• The paper demonstrates a novel approach using magnetic skyrmions as artificial synapses for neuromorphic computing.
• It employs Pt/GdFeCo/MgO multilayers and STXM to electrically create, manipulate, detect, and annihilate skyrmions at room temperature.
• Neuromorphic simulations on MNIST digits reached ~89% accuracy, highlighting the potential for energy-efficient, scalable AI hardware.

Magnetic Skyrmion Artificial Synapse for Neuromorphic Computing

The paper presents an in-depth exploration of leveraging magnetic skyrmions as artificial synapses in neuromorphic computing frameworks. The authors have demonstrated a room-temperature, electrically-operable skyrmion-based synaptic device, envisioned as a cornerstone for bio-inspired computing systems. This research capitalizes on the unique properties of skyrmions—topologically nontrivial spin textures capable of efficient data storage and processing.

The experimental focus is on ferrimagnetic multilayers, particularly Pt/GdFeCo/MgO, known for their ability to stabilize skyrmions at room temperature owing to sizeable Dzyaloshinskii-Moriya interactions. The authors used scanning transmission X-ray microscopy (STXM) to visualize and measure skyrmion dynamics, successfully demonstrating the electrical creation, manipulation, detection, and annihilation of skyrmions. The skyrmion synapse shows potential for analog memory applications, matching the performance characteristics of biological synapses.

A critical aspect of the paper is the neuromorphic pattern recognition simulation. The implementation of skyrmion-based synapses in a feedforward neural network to process the MNIST handwritten digits dataset resulted in an accuracy of ~89%, which is competitive with software-based models that achieve ~94%. Although there are deviations from ideal synapse behaviors, such as the non-uniform skyrmion size distribution and limited resistance states, the paper demonstrates promising directions for future optimizations in device engineering and material science.

From a theoretical standpoint, exploiting the topological stability and particle-like behavior of skyrmions offers a significant advantage over traditional domain-wall based synaptic devices, which suffered from high threshold currents and stochastic domain-wall pinning issues. The ability of skyrmions to exhibit rigid-body properties negates these issues, facilitating their accumulation in nanoscale areas without interaction with topographic defects.

Practical implications for this research include the development of highly energy-efficient synapses for neuromorphic computing that mimic the massively parallel processing capabilities of biological systems. The device architecture proposed—rooted in spintronics—holds promise for scalable, low-power, and versatile computing elements fundamental to advancing artificial intelligence research.

Looking forward, the research anticipates extended advancements in skyrmion technology, including reduced skyrmion size for increased synapse density and exploration of materials with higher tunable magnetic properties. Potential exists for integrating these technologies into existing ANN frameworks, further reducing energy consumption and improving computational efficiencies.

Significantly, this work opens pathways for integrating skyrmion-based synapses into broader computational architectures, potentially influencing the design of future computing systems. Future challenges will likely focus on improving the on/off ratios through advancements in magnetic tunnel junctions (MTJs) and optimizing skyrmion stabilization under varied environmental conditions. As such, the paper contributes a foundational understanding and practical approach to spintronics-based, bio-inspired computing technologies.