Researchers at the University of Pennsylvania have developed a new computer chip that uses light instead of electricity. This could improve the training of artificial intelligence (AI) models by improving the speed of data transfer and, more efficiently, reducing the amount of electricity consumed.
Humanity is building the exascale supercomputers today that can carry out a quintillion computations per second. While the scale of the computation may have increased, computing technology is still working on the principles that were first used in the 1960s.
Researchers have been working on developing computing systems based on quantum mechanics, too, but these computers are at least a few years from becoming widely available if not more. The recent explosion of AI models in technology has resulted in a demand for computers that can process large sets of information. The inefficient computing systems, though, result in high consumption of energy.
Chips that use light, not electricity
A team led by Nader Enghata, a professor at the School of Engineering and Applied Science at the University of Pennsylvania, has designed a silicon-photonic (SiPh) chip that can perform mathematical computations using light. The team turned to light as it is the fastest means of transferring data known to humanity. However, using widely abundant silicon ensures the technology can be scaled quickly.
The researchers aimed to design a chip that can perform vector-matrix multiplications. A common mathematical computation, the system is widely used in developing and functioning neural networks that are critical when developing architecture to power AI models being developed today.
Since the chip is silicon, the researchers could have reinvented the fabrication process completely. Instead, they reduced the height of the chip in specific regions to control how light propagates inside.
A wider chip allows light to scatter through the chip. Still, by controlling the height variations, the research team ensured that the light inside the chip traveled only in a straight line.
Unhackable chip ready for deployment
The researchers approach a commercial foundry to fabricate their SiPh chips. Since the foundry could only design chips at sizes currently existing in the market, the chip design had to be modified accordingly and is ready for deployment immediately.
Representational stock image showing that the chip cannot be hacked
Firooz Aflatouni, an associate professor in Electrical and Systems Engineering at the University, was also involved in the research, said that their chips could replace the graphics processing units (GPUs) that companies use to train and classify their AI models. The associate professor recommends that the SiPh platform can serve as an add-on to existing infrastructure being used by AI companies.
RECOMMENDED ARTICLES
In addition to performing computations faster and with less electricity consumption, SiPh chips can also address data privacy concerns. Since the chip can perform multiple computations in parallel, there is no need to store the information in a working memory while the computations are performed. “No one can hack into a non-existing memory to access your information,” added Aflatouni in a press release.
The research findings were published today in the journal Nature Photonics.
0COMMENT
NEWSLETTER
The Blueprint Daily
Stay up-to-date on engineering, tech, space, and science news with The Blueprint.
Ameya Paleja Ameya is a science writer based in Hyderabad, India. A Molecular Biologist at heart, he traded the micropipette to write about science during the pandemic and does not want to go back. He likes to write about genetics, microbes, technology, and public policy.
