Ghent researchers develop revolutionary computer chip


Scientists at the University of Ghent have developed a new optical chip that takes its inspiration from the human brain

The new chip's neural network is made of 16 silicon neurons

“Faster and more efficient”

The infamous Moore’s law states that computing power doubles every two years. But these days, computer chips have become so small and complex that hardware engineers are starting to run up against the limits of the nanoworld.

One of these is the problem with "the gap", or the core of a transistor, where electrons jump across. The smaller the transistor, the narrower the gap, and below a certain width, the electrons are no longer controllable. This renders the chip useless.

To overcome this so-called “nano barrier”, scientists are now exploring a number of different strategies. One is to replace transistors by a neural network of tiny silicon nodes or “neurons” that communicate with each other through light pulses. The result is a so-called optical, or photonic, chip.

The major advantage of this revolutionary chip is that it opens the door to high-speed computing. Moreover, a chip like this uses much less energy than a standard one. This, in turn, means it will require much less cooling, an important breakthrough considering that the ventilator is often the first part to break down in computers and laptops.

Scientists from the University of Ghent (UGent) have now created a prototype of a photonic chip implementing such a neural network – a first. Although the network is tiny – it counts only 16 nodes -  the chip represents an important first step in the move toward a different approach to computation, according to Peter Bienstman from the Photonic Research Group at UGent.

“The chip takes inspiration from the way our brain works to perform certain computations, both faster and more power-efficient compared to electronic chips,” he says.

The researchers have successfully demonstrated that their invention – although still a modest prototype – can perform multiple tasks. The photonic chip can, for example, identify certain bit patterns in a data stream, and it can perform basic elements of speech recognition.

The challenge now is to extend the neural network so that the chip can learn to perform more complex tasks.

Photo by P Bienstman