In earlier studies, the research team had developed a reservoir computer based on Brownian motion of magnetic skyrmions - small but stable magnetic domains, which move in a magnetic material due to thermal fluctuations. They successfully showed that simple logic functions can be achieved. However, this technique had not yet been applied to data processing of real-world data.
Traditional reservoir computing methods often have a gap between real-world signal timescales and the fast intrinsic dynamics of magnetic systems. The team used stochastic, low-energy dynamics of skyrmions to process Doppler radar gesture data in real time, achieving classification without the need for energy-intensive preprocessing.
Time-aligned reservoir computing
In their latest study, the researchers revealed that skyrmions can effectively manage complex, real-world data, such as hand gestures, achieving accuracy comparable to contemporary neural networks but at a fraction of the energy cost. This means that Brownian reservoir computing technology not only performs logic functions but can also be effectively used for gesture recognition, with much better energy efficiency. "We are now able to process gesture data with a hardware approach that equals or even exceeds other methods, all while using significantly less power,” researcher Johan Mentink says.
This innovative technique can be applied for various applications requiring real-time data processing. By modifying device layouts or using multiple skyrmions, researchers can enhance complexity and improve performance even further.
Energy-efficient computing
These results pave the way to developments in low-power computing applicable to everything from mobile devices to large-scale data centers. “An advantage of our proof of concept is that it is very easy to adapt to the timescale of incoming real data. Therefore, we think it can be applied to many applications without the need of additional external data processing equipment", Mentink says.
Note
This article, entitled 'Gesture recognition with Brownian reservoir computing using skyrmions' , has been selected as highlight. https://www.nature.com/collections/bjiiabbacg
