New spin-toque diodes show higher sensitivity while consuming less power

Microwave detecting, energy harvesting, and neuromorphic computing are just a few potential applications of spin-torque diodes based on magnetic tunnel junctions. While quantum and semiconductor devices have generally been used in these areas, spin-torque diodes have the advantages of a simple structure, nanoscale size, and a wide tunable frequency range.

Modern spin-torque diodes typically require a high current density which increases power consumption and the risk of device breakdown. Zhang et al. developed a device that exhibits high sensitivity while requiring low power consumption.

“By optimizing device structure, we demonstrate a highly sensitive spin-torque diode with ultra-low current density based on magnetic tunnel junctions,” said Like Zhang.

The researchers deposited a magnetic multilayer stack, containing Co₂₀Fe₆₀B₂₀ at a thickness of 1.72 nanometers, MgO at 0.9 nanometers, and Co₄₀Fe₄₀B₂₀ at 3.0 nanometers, onto a thermally oxidized silicon substrate. Electron-beam lithography and ion milling techniques patterned the stack in ellipse-shaped pillars.

The results showed that in the absence of a magnetic field, sensitivities exceeded 3,785 volts per watt at current densities 10 times less than previous works. Applying small, out-of-plane magnetic fields enhanced sensitivity to levels exceeding 20,000 volts per watt.

“In addition, we constructed an artificial neural network with spin-torque devices to perform the recognition of handwritten digits in the Mixed National Institute of Standards and Technology database, obtaining an accuracy of up to 94.92%,” said Zhang.

Future work involving spin-torque diodes will continue to focus on low power consumption in the areas of communications and neuromorphic computing.

Source: “Ultralow-current density spin-torque diodes with high sensitivity,” by Like Zhang, Huayao Tu, Yanxiang Luo, Kexin Zeng, Xu Tao, Dong Zhao, Bin Fang, and Zhongming Zeng, Applied Physics Letters (2023). The article can be accessed at https://doi.org/10.1063/5.0141113 .