Today, the hugegest challenge in AI is not the chip itself but how chips connect to one another. As AI models receive larger and more complex, shifting data between processors, accelerators, and memory has become the main obstacle.
CamGraPhIC, a company founded in 2018 at the University of Cambridge’s Graphene Centre by Professor Andrea Ferrari and Dr Marco Romagnoli, has worked on this problem for seven years.
This week, it received €211 million in Italian state aid, approved by the European Commission, to bring its graphene-based optical interconnect technology to industest. This is one of the largest public investments ever built in an Italian deep-tech startup.
The company creates graphene-based optical transceivers that utilize light, not electrons, to sconclude and receive data. In these devices, graphene replaces the silicon utilized in standard photonic interconnects.
CamGraPhIC declares its transceivers utilize 80% less energy than traditional silicon versions, offer higher bandwidth and lower latency, and operate reliably across a wide temperature range without requiring complex cooling. The first utilize for this technology is the GPU-to-HBM (high-bandwidth memory) connection, where data bottlenecks are a huge problem in AI systems.
Its direct competitors include Ayar Labs and Lightmatter. What creates the graphene approach different is its claimed energy savings over silicon and its possible fit with current semiconductor factories. If CamGraPhIC can reveal it can produce at scale applying existing factories, the cost savings could be large.
The €211 million grant is meant to test this ability. A pilot facility near Milan, set to open in 2028, will reveal whether the technology works with commercial semiconductor and photonics factories for production.
Leave a Reply