Credit: Shutterstock
Researchers harness light waves to perform AI calculations instantly without electronics
Artificial intelligence is hungry for power and speed, but researchers at Aalto University may have found an elegant solution hiding in plain sight. They’ve developed a way to use light itself as a computer processor, performing complex AI calculations in a single flash instead of the millions of electronic steps today’s chips require.
The breakthrough centers on something called tensor operations, the mathematical backbone of modern AI. Think of these as solving a multidimensional puzzle where everything must happen at once. Current computer chips tackle these problems step by step, burning through electricity and time. The Finnish research team found a way to let light do all the work simultaneously.
Their system works by encoding information directly into light waves, specifically into properties called amplitude and phase. As different wavelengths of light pass through optical materials, they naturally interact in ways that complete the mathematical operations needed for AI tasks like image recognition or language processing. No electronic switches flipping on and off, no sequential processing, just pure optical physics doing the heavy lifting.
Dr. Yufeng Zhang, who led the research, compares it to sorting packages. Traditional computing inspects each package individually through multiple machines. Their optical method merges everything together, connecting inputs to outputs through “optical hooks” that process everything in one pass.
What makes this especially promising is its passivity. The system doesn’t need constant electronic control during computation, which dramatically reduces energy consumption. Professor Zhipei Sun notes the technique works on nearly any optical platform and could eventually be built into photonic chips.
The timing couldn’t be better. As AI models grow larger and more demanding, they’re pushing conventional hardware to its limits. This light-based approach could break through current bottlenecks in speed, power usage, and scalability.
The technology is still in early development, with researchers estimating it could reach commercial hardware within three to five years. Challenges remain in manufacturing photonic chips at scale and integrating them with existing systems, but the potential payoff is enormous.
