A new type of computer can use the behaviour of magnetic particles to recognise handwritten digits. If made smaller and faster it should be able to process information in an energy-efficient way.
Magnetic materials can be thought of as grids of tiny bar magnets, with each bar magnet representing a particle. One type, called a skyrmion, is like the bar magnets swirling around. Researchers have long thought that skyrmions could be used to build energy-saving electronic devices because it doesn鈥檛 take a lot of energy to create and control them.
Now, at the University of Tokyo and his colleagues have built a skyrmion computer.
Advertisement
The computer consists of thin, microscopic, rectangular blocks made of layers of platinum, cobalt and iridium containing skyrmions. These are then connected together in such a way as to mimic an artificial neural network, an algorithm that imitates neurons and synapses in the brain.
To give the computer a task, the researchers change the magnetic field, which, in turn, changes the number, shape and arrangement of the skyrmions. This changes the electric voltage across each block, which the researchers can then translate as the computer鈥檚 answer.
The team used a standard test of neural network computing 鈥� recognising handwritten digits 鈥� to see how the computer would perform. The researchers converted numbers ranging from 0 to 9 into variations in a magnetic field and fed them into the computer. Across 5000 different tests, the device recognised handwritten digits about 95 per cent of the time. This is around 5 per cent better than other non-traditional computers that use the same neural network structure.
The design of the new device means it may not have to use a lot of energy, says at the University of Gothenburg in Sweden. Skyrmion computers could be manufactured cheaply with technology that already exists for making hard drives, though the current design would have to be made smaller and faster to have an advantage over conventional computers, he says.
Science Advances
