Most computer hardware available nowadays relies on magnetic memory devices which employ magnetic states – the direction microscopic magnets are pointing towards – to store and access information.
To meet today’s computing needs, however, more advanced techniques are becoming mandatory. One way of accomplishing that is by building systems based on exotic magnetic states – such as a point where three south poles meet – which function in a similar fashion to our brains and other complex natural systems.
Computing applications inspired by neural mechanisms – called ‘neural networks’ – are extremely powerful, software-based solutions capable of learning and highly efficient problem-solving.
One of the key drawbacks of neural networks, though, is the need to run them on conventional hardware, thereby limiting the breadth of potential applications and processing power.
Now, researchers from Imperial College London (ICL) have developed a technique for encoding magnetic data in any pattern, using a tiny magnetic probe, called a magnetic force microscope.
The new technique allows an array of magnetic nano-wires to function as a physical neural network, which could prove significantly more effective than using their software-based counterparts.
“With this new writing method, we open up research in ‘training’ these magnetic nano-wires to solve useful problems. If successful, this will bring hardware neural networks a step closer to reality,” said Dr Jack Gartside from the Department of Physics at the ICL, and first author on the study.
In addition to its benefits in computing, the new technique could also prove useful in studying the fundamental principles of complex systems.
A paper where the research team demonstrates the capabilities of their method by writing patterns that have never been seen before has been published in the journal Nature Nanotechnology on the 20 November.