Strain-mediated multistate skyrmion for neuron devices.

Area of Science:

• Spintronics
• Neuromorphic Computing
• Artificial Intelligence

Background:

• Magnetic skyrmions offer potential for neuromorphic computing due to stability and nanoscale size.
• Current-driven skyrmion motion in artificial neurons faces challenges in energy efficiency and integration density for deep networks.

Purpose of the Study:

• To present a compact and energy-efficient skyrmion-based artificial neuron.
• To mimic the Integrate-and-Fire (IF) function of biological neurons using voltage manipulation.

Main Methods:

• Fabrication of a skyrmionic neuron device using ferromagnetic/heavy metal/ferroelectric layers.
• Utilization of strain-mediated voltage control for skyrmion state manipulation.
• Implementation of a spiking neural network (SNN) with the proposed devices.

Main Results:

• The skyrmionic neuron successfully mimics the Integrate-and-Fire (IF) function.
• A spiking neural network (SNN) achieved 95.08% accuracy on the MNIST dataset.
• The device demonstrated low power consumption of approximately 46.8 fJ per epoch per neuron.

Conclusions:

• The developed skyrmionic neuron offers a novel approach for energy-efficient and high-density neuromorphic computing.
• Voltage-controlled skyrmion manipulation is a viable strategy for artificial neuron design.
• This technology paves the way for more powerful and efficient artificial intelligence systems.