READY4COPY//Magnetic whirlpool device trained to recognize handwritten numbers shows promise for low-energy computing

Researchers at RIKEN have developed a computing device that utilizes tiny magnetic swirls, known as skyrmions, to recognize handwritten numbers in a breakthrough in neuromorphic computing that spotlights the potential of magnetic whirlpools for creating low-energy computing systems inspired by the brain.

According to a news release from RIKEN, a Japanese research institution founded in 1917 and known for it research in a wide range of areas, the device which is based on the reservoir computing model, draws upon the short-term memory and energy saving abilities of skyrmions, which can run on low energies.

“Another advantage of using skyrmions is energy saving, because skyrmions can be controlled using very small current densities,” Tomoyuki Yokouchi of the RIKEN Center for Emergent Matter Science, said in a post on the company’s website.

RIKEN’s news release detailed how by encoding information into magnetic fields and analyzing the output voltage signals, the device can attain an accuracy of 95% in identifying handwritten digits, and the researchers hope to build upon its performance and ultimately make way for motion tracking and speech recognition tasks.

Yokouchi and the rest of the team of researchers have developed a device that is created from the reservoir computing model, and streamlines short-term memory into its workings, according to the website. Those magnetic patterns offer inherent memory, which is drawn from their behavior and structure, which banks information from past exposure to magnetic fields. The skyrmions, according to the post, show low-energy consumption and are key to developing energy efficient computing.

"Another advantage of using skyrmions is energy saving, because skyrmions can be controlled using very small current densities,” Yokouchi said.

The researchers, according to the RIKEN website, developed a neuromorphic device that includes a series of bars coated with a platinum-cobalt-iridium film capable of hosting micrometer-sized skyrmions. The data is input into the device by encoding information into a magnetic field that is then applied to the skyrmions, which creates a voltage, which relies upon the number of skyrmions used in the unit.

The team relied upon a dataset of more than 13,000 images of handwritten digits from 0 to 9 to train the unit, and those images were transferred into magnetic signals, with the device then being honed to make sure the output voltage signals match the correct digits. Later testing included an additional 5,000 images and found a recognition accuracy of nearly 95%, according to the website, topping other neuromorphic devices.

The work is worth highlighting, Yokouchi noted in the news release.

“Our work indicates that energy-saving neuromorphic computing can be realized using skyrmions,” he said in the RIKEN news release.

Moreover, the website noted the researchers can now focus on building up the capabilities of the unit by using electrical current for input over the magnetic field, an improvement that could result in increased performance and additional energy savings.

Yokouchi, according to the RIKEN website, said this could lead to the development of applications that include motion tracking and speech recognition and may open the potential uses of the magnetic whirlpool device beyond use for recognizing handwritten digits, widening the use of low-energy computing systems inspired by the capabilities of the human brain.

The researchers, the RIKEN website said, see skyrmions and the memory-effect impact they possess, open a range of possibilities to drive the development of neuromorphic units that are energy efficient, and as the technology advances and the exploration of magnetic whirlpools could lead to the creation of computing systems that lead to the development of computing systems that are effective and efficient.