Brain-based computers compute faster and more energy-efficiently

from the magazine

Computers are increasingly able to recognize patterns. Yet the human brain often works even faster, more efficiently, and more economically. By processing and storing data in the same place as in the brain, chips can use up to a thousand times less energy, the researchers believe.

The computer was once conceived as a great calculator to relieve people. Since the advent of artificial intelligence (AI), the role of chips has expanded significantly. Thanks to AI, a smartphone can recognize speech and the software ‘sees’ that there is, for example, a dog in the fine picture you just took.

Lots of power

To work properly, an AI algorithm must first be trained. Before recognizing the cute dog in the picture, it went through tens of thousands of pictures to learn what such an animal looks like. It takes some time, but more importantly: it costs a lot of energy – almost as much as a household spends in two weeks.

Even when training is over, image recognition still uses a lot of power. This is because in today’s computers, processors and memory are physically separated from each other. Information must be constantly back and forth: a slower and energy-inefficient process.

Network of nerve cells

Anyone who’s ever seen one captcha have personally experienced the limits of conventional computers’, says Professor Hans Hilgenkamp, ​​one of the directors of the Center for Brain-Inspired Nano Systems (BRAINS) at the University of Twente (UT). Such a captcha is used to prevent unwanted crash and consists of letters or numbers that are distorted in such a way that the computer can no longer recognize them.

For the human brain, on the other hand, this is a piece of cake, thanks to a communicating network of nerve cells (neurons) in our brain. Each neuron opens at several (axon) ends and transmits information to other neurons via the synapses – the connecting pieces between neurons.

Powerful chip

Here, processing and memory are intertwined, saving a lot of energy and speeding up the process. ‘A powerful chip that, like the brain, can process and store data in the same place, opens up a new world of possibilities,’ says Hilgenkamp. A chip that works according to these brain principles would use a thousand times less energy for certain processes than current computer chips.

In Memory Computing is the new concept where memory and processing somehow coexist, just like in the brain.
Illustration: IBM Research

Inspired by the brain

Researchers around the world at universities and in companies are working on such neuromorphic chips, ie chips that are inspired by how the brain works. This is not new software that will allow the traditional computer to perform increasingly complex actions, but materials and designs that act as the brain. The focus is on enabling multiple processes at the same time and on combining memory and information processing.

The American chip manufacturer Intel makes artificial neuron chips from silicon. These so-called Loihi chips contain about 130,000 neurons and are already capable of ‘smelling’ chemicals in odor samples. By connecting 768 chips, Intel has already built a large neural network that acts as a faster and more energy-efficient computer for certain AI processes.

Plastic as the brain

As the entire chip industry is built on silicon, these chips are suitable for large-scale production. However, they cannot be retrained. ‘The neural networks are engraved in silicon and are therefore not plastic as in the brain’, says Yoeri van de Burgt, associate professor at Eindhoven University of Technology (TU / e). His research team also works with neuromorphic chips. ‘Because the brain is plastic, new networks can be constantly created or existing networks strengthened: that’s how you learn,’ says Van de Burgt. ‘Chips made from materials that can do this too are also capable of learning new tasks.’

READ MORE ABOUT THE BRAIN COMPUTERS?

The full story of brain-based computers can be found in the April issue of De Ingenieur. Buy the digital version for € 7.50, or buy – with a significant discount of 25% – a digital annual subscription of 12 issues for € 69.

Opening photo A ‘chip’ with organic artificial synapses with electrolyte (water with salts) on top. Photo: Eveline van Doremaele

If you found this article interesting, you can subscribe to our weekly newsletter for free.

Print this article

Leave a Comment