Computers that use light instead of circuits to run calculations may sound like a plot point from a Star Trek episode, but researchers have been working on this novel approach to computing for years.
They’re called optical computers, and labs around the world have been exploring how they might be useful in everyday life.
On Wednesday, a team of researchers from Penn State published a paper in the journal Science Advances that examines how optical computing could reduce the power consumption of artificial intelligence systems.
Xingjie Ni, an engineering professor at Penn State and one of the paper’s authors, told that the work is a proof of concept for how optical computing could benefit the fast-growing AI industry in the future.
“Sometimes progress comes from rethinking familiar physics with a new goal,” Ni said. “By revisiting classic ideas in optics through the lens of modern AI challenges, we can open up practical new directions for faster, greener computing hardware.”
Powering AI
As AI is increasingly adopted for work and home use, the issue of AI’s energy costs is pertinent. So much computing power is required to run AI products and services like ChatGPT, and a lot of energy is consumed in the process.
You may live in or near a town where a tech company is planning to build a data center, or your monthly utility bill could increase due to higher demand on the local power grid.
The International Energy Agency estimates that data centers accounted for about 1.5% of global energy consumption in 2024 and that this figure increased 12% per year in the five years prior. The IEA also estimates that data center energy use could double by 2030.
That’s why using an alternative computational method to reduce the power AI consumes is an attractive prospect.
Light speed
Optical computers — computers that use light instead of electricity — still mostly exist in the tech industry’s moonshot category, where they’re years away from commercial use. They’ve been a concept since the 1960s, with the roots of optical information processing stretching back much further.
True optical computers have mostly been relegated to research laboratories. But optical data transfer, which quickly transmits data via pulses of light, is used today in some large data centers and for ground-to-plane transmissions.
Still, using optical computing in artificial intelligence is an emerging field of study. There are real challenges in getting light to cooperate so it can perform the functions required by neural networks, which is a subset of AI used in products like today’s chatbots.
Essentially, light naturally moves in a straight line. To build a computer that can process data, you need an optical system that produces nonlinear functions. For optical computers to do this, they often require other materials that can be hard to manufacture and consume a lot of power.
“True optical nonlinearity is typically weak and hard to access — it often requires high-power lasers or specialized materials, which adds complexity and can undermine the energy-efficiency advantage of optics,” Ni said. “Our approach avoids those requirements while still delivering performance that is comparable to nonlinear digital networks.”
Infinity mirror
The researchers at Penn State found an interesting solution that could help optical computers perform nonlinear functions better suited to the kind of data processing AI needs.
The prototype the team built uses an “infinity mirror” setup that loops “tiny optical elements, encoding data directly into the beams of light,” creating a nonlinear relationship over time. Then, the light patterns are captured with a microscopic camera.
“The key takeaway is that a carefully designed optical structure can produce the nonlinear input–output behavior AI needs without relying on strong nonlinear materials or high-power lasers,” Ni said. “By letting light ‘reverberate’ through the system, we generate this nonlinear mapping while keeping the hardware simple, low power, and fast.”
The (above) figure shows how light is focused into a tiny processing unit, allowing vast strings of computational information to be transferred without the use of energy-intensive circuitry. The other figure (below) illustrates how the team’s process works conceptually. Light input is repeatedly reflected through lenses and other optical devices, encoded with complex strings of information, and finally focused into a camera that provides a simplified output.
It’s an interesting concept, but turning the prototype into a system with real-world applications will take a lot more time, work and money.
From the lab to the data center
Ni acknowledges that we’re still years away from AI optical computers.
“A realistic timeline to reach an industry-facing prototype and early demonstrations is about two to five years, depending on the level of investment and the target application,” he said.
Nonetheless, it’s a hot topic in the computing world. Francesca Parmigiani, principal research manager at Microsoft Research, told that optical chips could one day work alongside traditional GPUs to help AI systems perform specific tasks.
“Optical computing has the potential to efficiently perform vastly more operations in parallel and at significantly higher speeds than conventional digital hardware,” Parmigiani said. “This can translate into substantial gains in energy efficiency and reductions in latency for workloads.”
The traditional computers we use for AI are not being replaced by optical computers anytime soon. But in a few years, it’s possible that optical computers could be integrated into AI systems to work with regular computers.
“The goal is a hybrid approach: Electronics still handle general-purpose computing, memory and control, while optics can accelerate specific high-volume computations that dominate AI’s time and energy cost,” Ni said.
