Researchers at the University of Oxford's Department of Physics have made a discovery that could pave the way for the next big breakthrough in AI.
The team has successfully created “hurricane-like” magnetic swirls in hematite, the main component of oxide, that could potentially power energy-efficient brain-like processors running at hundreds of Gigahertz.
These magnetic eddies, capable of moving at astonishing speeds of up to kilometers per second, are touted as potential information carriers for the next generation of ultra-fast, green computing platforms.
100-1000 times faster
The study, published in Nature Materials, overcame the challenge of producing these swirls in materials that were previously incompatible with silicon, a major obstacle in their practical application.
The research was led by Dr Hariom Jani from the University of Oxford, in collaboration with the National University of Singapore and Swiss Light Source. Dr. Jani believes that silicon-based computing, with its high energy inefficiency, is not suitable for the next generation of computing applications, such as large-scale AI and autonomous devices.
The solution, he suggests, lies in harnessing magnetic eddies in a special class of materials called antiferromagnets, which work 100 to 1,000 times faster than modern devices.
“Today, silicon transistors use charges to perform calculations, which can dissipate as soon as the power is turned off. Therefore, they tend to consume a lot of power and are inefficient. In addition, conventional computers maintain data storage separated from data processing. “The information between them consumes a lot of time and energy, which creates a performance bottleneck,” Dr. Jani told us.
“On the other hand, the hurricane-like eddies in antiferromagnetic materials are 'shielded' nanoscale spin structures that are intrinsically stable thanks to their unique coiling. They can remain stable even in the absence of electrical pulses. In addition, they also harbor rich and ultra” Fast dynamics that can be leveraged to perform unconventional information processing, where memory and logical functions can be brought together. “In the future, these platforms could be used to perform brain-like computing to create efficient and fast AI hardware.”
The team achieved their breakthrough by fabricating ultrathin crystalline membranes of hematite, on a glass template coated with a special “sacrificial layer.” The layer was dissolved in water, separating the hematite from the crystalline base, which was then transferred to silicon and various other platforms.
The researchers also developed a new imaging technique that uses linearly polarized X-rays to visualize the nanoscale magnetic patterns within these membranes. The technique revealed that independent layers can host a robust family of magnetic eddies, opening the door to ultrafast information processing.
The team is now developing prototype devices to exploit the dynamics of these superfast eddies. Dr. Jani concluded: “Eventually, these devices could be integrated into new types of computers that function more like the human brain; we are very excited about what will come next.”
