Spintronics for neuromorphic computers
Computers, and artificial intelligence algorithms in particular, use a lot of energy - considerably more than our brains do for similar tasks. Now researchers are using a quantum mechanical property of electrons to design novel devices that function similarly to the human mind.
Artificial intelligence (AI) has become an indispensable part of our everyday life: The algorithms sort emails, are useful language assistants, analyze the content of texts or images - and sometimes even better than a human being. For example, in 2017, DeepMind's AlphaGo program defeated Korean Go world champion Ke Jie. Experts had reckoned with this performance in ten years at the earliest.
However, to achieve such impressive results, the algorithms consume a great deal of energy, far more than the human brain doing equivalent tasks. For example, a speech recognition program like Bert from Google requires about 1000 kilowatt hours of electricity for a training session; such an amount of energy would last the human brain for six years!
Modern artificial intelligence algorithms are often based on deep neural networks, the structure of which is partly based on that of the visual cortex. So why is their capacity so different from our mind? The problem is the hardware on which the programs run. The structure of conventional computers and graphics cards differs significantly from that of the brain.
Therefore, many researchers and companies are working on new approaches to develop more efficient devices. So-called spintronics, which uses the quantum mechanical properties of electrons, offers a promising possibility…