Topological Insulators Boost Ultralow-Power Neuromorphic Spintronics: Advancing Handwritten Digit Recognition with...

Area of Science:

• Spintronics
• Artificial Intelligence
• Materials Science

Background:

• Neuromorphic spintronics devices offer advantages for artificial intelligence but are limited by low spin-orbit torque (SOT) efficiency in conventional materials.
• Achieving ultralow power consumption is crucial for advancing high-performance AI systems.

Purpose of the Study:

• To demonstrate low-power artificial synapse and neuron devices with enhanced SOT efficiency.
• To explore the potential of topological insulators for efficient neuromorphic computing.

Main Methods:

• Utilized (BiSb)2Te3, a topological insulator, to achieve high SOT efficiency (θSH = 1.11).
• Fabricated artificial synapse and neuron devices capable of long-term potentiation/depression and excitatory/inhibitory postsynaptic potential processes.
• Implemented an artificial neural network using the developed devices.

Main Results:

• Demonstrated artificial synapse and neuron devices with an ultralow activation current density of 1.8 × 10^5 A/cm^2, 1-2 orders of magnitude lower than conventional systems.
• Achieved simultaneous long-term potentiation/depression and excitatory/inhibitory postsynaptic potential processes.
• The artificial neural network achieved 92.8% accuracy in handwritten digit recognition.

Conclusions:

• Topological insulators like (BiSb)2Te3 exhibit exceptional SOT efficiency, enabling ultralow-power neuromorphic spintronics.
• The developed devices and artificial neural network show significant promise for the future of energy-efficient AI.
• This work paves the way for practical, low-power neuromorphic computing applications.