Polytechnique team finds way to cut AI data center power
A new organic material could reduce electricity consumption of AI systems by improving how photonic chips process light.
The day's top stories, food & events — every morning at 7. Unsubscribe anytime.
Researchers at Polytechnique Montreal have identified an organic material that could significantly reduce the power consumption of artificial intelligence systems — a breakthrough that comes as the UN warns data center electricity use is set to double by 2030.
The team, led by Professor Stéphane Kéna-Cohen, created a thin organic layer that can be integrated directly onto existing silicon photonic chips without redesigning their architecture. Instead of constantly converting electrical signals to light and back again — a process that happens multiple times for each AI query — the new material allows light to be processed directly on the chip.
"The strategy since 2005 has been to add more transistors and enlarge chips," Kéna-Cohen explained. "But the larger chips become, the more difficult communication between different parts becomes. The solution is to use light to communicate between chips — what we call photonic chips."
The breakthrough works by allowing light beams to interact as they pass through the material, enabling functions like amplification and modulation directly on the chip. This means a high-power signal can amplify a weaker one for long-distance transmission without extra conversion steps.
Because the material can be added at the end of the manufacturing process, existing chip makers can adopt it without overhauling their facilities. The discovery addresses a real problem: every question asked to an AI system like ChatGPT requires multiple round trips between components, each one requiring signal conversion that consumes energy.
Montreal's research community continues to punch above its weight on the problems that matter.