Scientists Just Automated Designing Light-Based Computers
Light-based computers are closer than ever.
A close-up of light moving through the silicon interconnect. Image: Stanford University
Your computer ferries data around using electrons and conductive wires; for now, anyway. The age of electronics could soon end, maybe even within the next decade, thanks to an algorithm that can automatically design components that use light instead of electrons.
Currently, sending electrons around a computer using conductive wires consumes up to an estimated 80 percent of a computer's processing power. Basically, wires are a bottleneck when it comes to processing speed. Sending data with light, similar to fiber optics but at a micro-scale, would allow computers to process much more data, and much faster.
Although much promising work has been done in this area, a real and constant concern is the cost of building light-based components. Electronic circuits for consumer devices are usually designed quickly and automatically using software, but in the experimental arena of light-based computers, not so much. In 2013, scientists at Stanford University designed an algorithm that did just this.
Now, they've actually built components using the program, and they work. The results of their study were published today in Nature Photonics.
"Traditionally, optoelectronics circuits have been designed by brute force design of individual components and connecting them one by one," Jelena Vuckovic, one of the scientists who led the work, told me in an email. "Clearly, this traditional approach is not sustainable as the complexity of optical circuits increases, and an equivalent of the electronics circuit design algorithms in optics was needed—which is what we achieved with our algorithm."
"We make hundreds of such circuits on a silicon chip in each fabrication run"
Instead of having to design an optical interconnect by hand, which requires an immense amount of technical knowledge, the researchers' algorithm only asks the user to set their desired performance values and device-specific design requirements. The researchers call this an "inverse design" approach. The result is a tiny piece of swiss cheese-looking silicon, 20 of which can fit inside the width of a human hair, that can refract specific wavelengths of light carrying data inside a computer.
Since discovering that the results of their algorithmic design approach actually worked, the Stanford researchers have been busy pumping out components.
"We make hundreds of such [optical] circuits on a silicon chip in each fabrication run," Vuckovic wrote. "But the full integration into a computing platform would have to be done in industry—university nanofabrication facilities are not as state of the art as the industrial [ones] are, which is needed for full systems integration."
Industrial outfits have already begun looking at light-based computer chips, and this month IBM unveiled a computer chip that uses light to transmit data, although with components much larger and unwieldy than those designed by the algorithm built by Vuckovic and her colleagues.
With the ability to algorithmically churn out optical circuits on an industrial scale, super-fast, light-based computers are closer to reality than ever before.