Theoretical simulations reveal a path forward for antiferromagnetic artificial neurons

Numerous approaches have tried to create artificial neuromorphic computing hardware that mimics the computational function of the human brain. This includes current-driven auto-oscillators based on antiferromagnetic materials proposed for spintronics-based artificial neurons. New work explores one theoretical framework that might lead to antiferromagnetic neurons capable of outpacing their biological counterparts.

Sulymenko et al. reported results from numerical simulations and analytical calculations that demonstrate an antiferromagnetic neuron with clock speeds capable of reaching tens of gigahertz, making the design a promising candidate as a base element in neuromorphic computing. The artificial neurons would use spin Hall oscillators working at subcritical currents to perform a variety of tasks of biological neurons, but at millions of times faster speed.

Spin Hall oscillators based on antiferromagnetic materials and working in subcritical regimes generate picosecond-long spikelike pulses when an external current stimulus reaches a certain threshold. Larger stimuli make these devices capable of emitting a burst of pulses similar to the millisecond-long behavior of biological neurons.

The authors’ simulations revealed that their artificial neurons would use these ultrashort spikelike pulses to reach processing speeds of up to 50 GHz. When combined in units of five neurons or fewer, the neurons can form dynamic memory loops with a variable clock frequency, or can perform the logic functions of OR, AND, MAJORITY, and Q-gates — or a “full adder” circuit. This leads to a belief that substantially fewer artificial neurons could achieve similar signal processing results.

The authors hope their work stokes interest among experimentalists to build the device that would one day lead to fully neuromorphic computers capable of performing tasks, such as image recognition or classification.

Source: “Ultra-fast logic devices using artificial ‘neurons’ based on antiferromagnetic pulse generators,” by Olga Sulymenko, Oleksandr Prokopenko, Ivan Lisenkov, Johan Åkerman, Vasyl Tyberkevych, Andrei N. Slavin, and Roman Khymyn, Journal of Applied Physics (2018). The article can be accessed at https://doi.org/10.1063/1.5042348 .