A magnetic synapse: multilevel spin-torque memristor with perpendicular anisotropy

Area of Science:

• Materials Science and Engineering
• Nanotechnology
• Neuroscience and Neuromorphic Computing

Background:

• Memristors, as non-volatile nano-resistors with tunable resistance, are crucial for multilevel non-volatile memories and artificial nano-synapses.
• Developing memristive neuromorphic hardware requires technologies compatible with mainstream non-volatile memory candidates.

Purpose of the Study:

• To demonstrate the first experimental multilevel memristor compatible with spin-torque magnetic random-access memories (MRAM).
• To explore the potential of this device for analog magnetic neural computation and neuromorphic hardware.

Main Methods:

• Utilized perpendicularly magnetized magnetic tunnel junctions where resistive switching is driven by spin-torque-induced magnetic domain wall displacement.
• Engineered device geometry to optimize spin-torque efficiency and minimize energy consumption at low current densities.

Main Results:

• Achieved a multilevel memristor exhibiting a large number of intermediate resistance states, suitable for neural computation.
• Demonstrated low current densities for domain wall displacement, indicating reduced energy cost.
• Established compatibility with spin-torque MRAM technology.

Conclusions:

• The developed spin-torque memristor offers a viable pathway for creating energy-efficient, large-scale analog magnetic neural computation.
• Device geometry engineering is key to optimizing performance and energy efficiency for neuromorphic applications.
• This work paves the way for spin-torque-based analog magnetic neural computing.