Physics

Energy-efficient computing using magnetic vortexes

Voltage is used to move skyrmions on a triangular thin film. From the skyrmion’s movement, the type of hand gesture the system has detected can be interpreted. Credit: Grischa Beneke / JGU

Researchers at Johannes Gutenberg University Mainz (JGU) have now succeeded in enhancing the Brownian motion reservoir computing framework by recording hand gestures, transferring them to a system and using skyrmions to detect individual gestures.

“We were impressed by how well our hardware approach and concept worked, outperforming energy-intensive software solutions employing neural networks,” said Grischa Beneke, a member of Professor Matthias Kraui’s research group at the JGU Institute of Physics.

Beneke, working with other experimental and theoretical physicists, has demonstrated that Brownian motion reservoir computing can recognize simple hand gestures with relatively high accuracy.

The research is published in the journal Nature Communications.

Reservoir computing requires no training effort and reduces energy consumption.

Reservoir computing systems are similar to artificial neural networks. Their advantage is that they don’t require extensive training, reducing overall energy consumption. “You only need to train a simple output mechanism to map the results,” Beneke explains.

The exact computational process remains unknown and the details are not important. The system can be likened to a pond into which stones have been thrown, producing complex wave patterns on the surface. Just as the waves imply the number and position of the stones that were thrown, the system’s output mechanism provides information about the original input.

In the paper, the researchers describe how they use two radar sensors from Infineon Technologies to record simple hand gestures, such as swiping left and right, with range-Doppler radar. The radar data is converted into a corresponding voltage and fed into a reservoir, which in this case consists of a multi-layered thin-film stack of different materials, formed into a triangle with contacts at each corner.

Two of the contacts supply a voltage, which causes the skyrmion to move within the triangle. “It reacts to the supplied signal and detects complex movements,” explains Grischa Beneke. “From this movement of the skyrmion we can infer the movements recorded by the radar system.”

Skyrmions are chiral magnetic vortices that are believed to have great potential for use as information carriers in unconventional computing devices and innovative data storage devices.

“Skyrmions are truly amazing. Although we initially only considered them as a candidate for data storage, they also have great potential for computing applications in combination with sensor systems,” emphasized Professor Matthias Kraui, who leads this research field at JGU.

Comparing the results obtained using Brownian reservoir computing with those recorded using a software-based approach, we find that the accuracy of gesture recognition is comparable or even better than that of Brownian reservoir computing.The advantage of combining the concepts of reservoir computing and Brownian computing is that local differences in magnetic properties have less impact on the skyrmion’s response, allowing the skyrmion to perform random movements freely.

This means that skyrmions can be moved with very low currents, in contrast to conventional reactions, representing a significant improvement in energy efficiency compared to software-based approaches.

Because the data collected by the Doppler radar and the intrinsic dynamics of the reservoir operate on similar time scales, the sensor data can be fed directly into the reservoir, and the time scale of the system can be adapted to solve a variety of other problems.

“We found that radar data for a range of hand gestures can be detected in the hardware reservoir with fidelity equal to or greater than state-of-the-art software-based neural network approaches,” the researchers concluded.

According to Beneke, further improvements should be possible regarding the readout process, which currently uses a magneto-optical Kerr effect (MOKE) microscope. Employing a magnetic tunnel junction instead could potentially reduce the overall size of the system. The signal provided by a magnetic tunnel junction has already been emulated to demonstrate the capacity of the reservoir.

Further information: Grischa Beneke et al., “Gesture recognition via Brownian motion reservoir computing with geometrically constrained skyrmion dynamics,” Nature Communications (2024). DOI: 10.1038/s41467-024-52345-y

Courtesy of Johannes Gutenberg University Mainz

Source: Energy-Efficient Computing with Magnetic Vortices (September 16, 2024) Retrieved September 17, 2024 from https://phys.org/news/2024-09-energy-magnetic.html

This document is subject to copyright. It may not be reproduced without written permission, except for fair dealing for the purposes of personal study or research. The content is provided for informational purposes only.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button