Eliyan Unveils Optical Chiplet Platform for AI Data Center Links
Eliyan has launched its NuLight platform, featuring the Nexus and Oscar chiplets, which use 2D microVCSEL technology to connect AI compute resources over optical links. The design aims to overcome copper bandwidth and distance limits by using many lower-speed NRZ optical lanes, targeting roughly 50 meters of reach. The company presented the architecture at OCP Summit 2026, emphasizing scalability and lane repair for reliability.
Eliyan’s NuLight offering pairs two chiplets: Nexus, a 1.6T reverse gearbox built on TSMC’s 3nm process, and Oscar, an electrical interface chip linking high-speed electrical signals to microVCSEL and photodiode arrays. Nexus uses 212G PAM4 SerDes and converts each high-speed electrical lane into four slower NRZ optical lanes, aiming for roughly 50 meters of reach.
The design includes telemetry with FEC monitoring, termination, regeneration, IEEE auto-negotiation, and host link training. Oscar supplies laser-driver and TIA connections, plus crosstalk cancellation and thermal management. Spare optical lanes can take over if an active lane degrades, supporting uninterrupted operation for AI systems that scale up or scale in.
If optical chiplets like NuLight mature, AI data center operators, chip designers, cloud providers, and ultimately users of AI services could benefit from longer-reach, lower-power links and more resilient large-scale systems. Improved lane repair may reduce downtime, while optical complexity and manufacturing demands could shape costs and supply chains. Broader access to capable AI infrastructure may depend on whether these components scale economically, so societal effects remain uncertain and uneven.