Scientists can solve mathematical equations by shining light through a so-called meta-surface of cleverly designed silicon nanostructures. This is faster and more energy efficient than calculating with electronics.
Technological developments, such as artificial intelligence and image recognition, are becoming increasingly complex, but they also require more and more energy. The electrical currents that shoot through electronic circuits encounter resistance and energy is lost as heat.
Analogous calculations with light can provide comfort. It generates less heat and is therefore more energy efficient. Furthermore, it calculates much faster than electronics. Light travels at almost the speed of light through the nanostructures that make the calculations possible. This makes it suitable for applications where speed and efficiency are important, such as image recognition in self-driving cars.
Brain training apps lack scientific support Many apps claim to make their users smarter. Not only are these brain training apps pretty boring, their effect on our performance is…
The specialty optical computing, where physicists investigate ways to calculate with light, is therefore on the rise. Researchers from AMOLF, in collaboration with the University of Pennsylvania and the City University of New York, have now taken a new step, they write in Nature. They show that with light and a cleverly designed nanostructure, a metamaterial, you can perform complex mathematical calculations called matrix inversions.
Metamaterials are artificially developed structures that behave like materials with properties not found in natural materials. Thus, for example, they can refract light differently than is possible with natural materials.
The research builds on previous work where part of this research group used the technique to recognize the edges of objects in images. Detecting edges of buildings and people, for example, is an important part of image recognition.
For this, they developed a meta-surface, which consists of silicon structures a few ten-thousandths of a millimeter in size, on a transparent surface. When the light from an image shines through it, it spreads and bounces around this surface in such a way that an image appears on the other side with only the edges of the objects in the image.
This edge detection method is faster and more efficient than a computer, says physicist Albert Polman from AMOLF. ‘In the meta-surface, the light from the image is immediately processed. With a computer, more steps would be necessary.’ The image must first be converted into bits with a camera that a computer can process. Then the machine does the calculation to find out the edges, and only then do you have the result.
After the successful edge detection, the researchers looked at whether they could apply the technique to more complex mathematical calculations called matrix inversions. These calculations are widely used in science, engineering and economics, for example in aircraft and robot operating systems or in navigation systems and computer animation.
For this purpose, the researchers developed a system consisting of a mirror and a meta-surface with a thin silicon structure. The mirror continues to reflect the light back to the surface. This is necessary because steps must be repeated in these calculations. The answer to the equation can be read from the light rays that this produces.
The results show that you can solve matrix inversions with this technique in less than a trillionth of a second. This is faster than is possible with most computing methods.
More meta surfaces
Even today, each computational task requires its own custom-designed meta-surface. The technique is therefore particularly suitable for taking over specific tasks from a computer. To make the method more flexible, the researchers are working on a new structure where they can electrically change the properties of the metamaterial. You can then electrically program which calculation takes place.
‘The next step is for the light itself to adjust the metamaterial so that it can perform the desired computation with that light,’ says Polman. ‘It’s the dream application, but it’s still a long way off.’