Inc. Magazine
AI is costly today, but new research in Science Advances shows optical computing could cut energy use and heat. Led by Xingjie Ni at Penn State University, a light based prototype ran AI faster and cheaper, pointing to more affordable AI in the future.
How an ‘Infinity Mirror’ Device Could Help Solve AI’s Biggest Problem
Author: Chris Morris
Artificial intelligence, as it stands today, isn’t cheap. The leaders in the field are spending tens of billions of dollars on chips, data centers and more, running up steep losses so far. That has raised questions of sustainability. When these companies shift into profit mode, will AI be something that companies and consumers can afford—or want to pay the expected high prices for?
Like any new technology, of course, the price of generating AI is expected to drop in years to come, but a new paper, published earlier this month in the Science Advances journal, suggests that optical computing could be one way to reduce the energy costs and heat production of today’s AI.
Optical computing isn’t a widely-known term. Rather than using transistors on silicon chips to do its computations, an optical computer uses light. It’s a technology that sounds like something from science fiction, but is, in fact, very real. Researchers have been working on it for years.
Optical devices are still years away from commercial use, but scientists have been exploring the concept since the 1960s. In theory, they can operate faster than modern computers, since light moves faster than the electricity in circuits. And, since they rely on photons, they use substantially less energy. (They also don’t generate heat signatures on par with today’s data centers, so could have environmental benefits.)
Photons, the atomic building blocks of light, are the key to the speed gains, since they don’t normally interact with each other. That means multiple light signals can pass through the same system simultaneously, letting optical computers process large data sets at a rate today’s computers cannot.
The research, led by Xingjie Ni, associate professor of electrical engineering at the Penn State School of Electrical Engineering and Computer Science, saw a team develop a prototype device that reduces the energy cost of AI computation. The prototype uses an “infinity mirror” setup (two parallel mirrors with lights sandwiched between them) that loops “tiny optical elements, encoding data directly into the beams of light.” That data is captured by a microscopic camera.
The result is AI models that ran faster and used less energy. If the computational-intense parts of AI could be done with these devices, the team concluded, companies “could offer the same capabilities for less overhead cost, which translates to cheaper, more sustainable AI services for consumers.”
Perhaps even more encouraging, the device that Ni and his team built didn’t rely on expensive, high-end processors and rare earth materials. It was created with everyday items.
“The core of our system is built from widely available components — like what’s used in everyday LCD displays and LED lights — rather than exotic materials or high-power lasers,” Ni said in a Q&A with Penn State University. “By arranging these familiar elements in a multi-pass loop, we can produce the energy AI needs, while remaining incredibly compact and efficient.”
Ni says he doesn’t expect optical computing to replace electronic computing, but says it could accelerate it. In the scenario he envisions, conventional electronics would handle things like memory and flexibility, while optical components would handle high-volume computations.
“We’re working to shrink the setup into a compact unit that can plug into real computing platforms, so the optical part does more of the work,” he said. “If this technology matures into something that can plug into today’s platforms, we could power AI models with smaller, faster and more sustainable hardware.”
Optical computing isn’t imminent, though. The first devices are likely to be hybrid machines, which could start to go online in the next five years or so. And Nvidia’s Jensen Huang has said the technology is not yet reliable enough to use its flagship GPUs.
But as the prototype demonstrates, there’s potential in the new form of computing that could not only accelerate the growth of AI, but do so in a more affordable way.
Credits: TCA, LLC.