Eliyan Introduced Optical Chiplet Interconnect Platform
The new NuLight platform enables 12.8 Tbps connections across 50 meters, aiming to break copper-based distance limits in AI data centers.
Updated on Oct. 14, 2026 in Semiconductors
Live Poll
Do you believe new hardware innovations will effectively solve the scaling challenges of modern AI infrastructure?
Eliyan has announced its NuLight 2D microVCSEL platform and begun sampling its associated Nexus and Oscar chiplets. The architecture is currently in the sampling phase and aims to solve scaling bottlenecks in AI infrastructure.
Why it matters
The platform addresses the physical reach and bandwidth limitations of copper-based connections between server racks in large-scale AI clusters. By moving to optical interconnects, the design seeks to support higher density while maintaining the reliability of traditional wiring.
The Nexus gearbox chiplet, built on a TSMC 3nm process, converts 212G PAM4 host links into 4 x 53G NRZ streams. The platform scales from a 1.6 Tbps OSFP slice to a 12.8 Tbps XPO configuration.
The players
Eliyan
A semiconductor startup developing chiplet-based interconnect technology for high-performance computing and AI clusters.
TSMC
The world's largest contract semiconductor manufacturer, providing the 3nm fabrication process used for the Nexus chiplet.
The details
The system utilizes microVCSEL — microscopic vertical-cavity surface-emitting lasers — and photodiode arrays to facilitate high-speed optical data transmission. A reverse gearbox, which changes signal formats to ensure compatibility, splits high-bandwidth host signals into manageable streams. The architecture includes an on-the-fly optical lane repair feature that automatically redirects traffic to spare lanes if a physical link fails, ensuring continuous operation.
Timeline
October 14, 2026: Eliyan introduced the NuLight chiplet platform.
Week of October 14, 2026: The technology was unveiled at the OCP Summit 2026.
The Tech Race
Eliyan is positioning the NuLight platform against legacy copper cabling that limits how far apart server racks can be placed within a data center. By moving the optical conversion to the chiplet level, the company is competing directly with proprietary optical module solutions currently dominating the AI infrastructure market.
Data center architects and AI hardware developers can begin evaluating the Nexus and Oscar chiplets now as they enter the sampling phase. This technology will eventually enable designers to build larger, more modular server clusters that were previously constrained by the physical distance limitations of copper.
The takeaway
The industry is moving toward optical integration at the chiplet level to overcome the physical reach barriers of current AI hardware clusters. Watch for future versions of the Oscar chiplet, which are expected to interface directly with host ASICs to further simplify the optical-electrical transition.
Further reading
For more on how new architectures are transforming data center throughput, visit our Semiconductors section.
Source note: This article includes information reported by The Manila times.
Live Poll
Do you believe new hardware innovations will effectively solve the scaling challenges of modern AI infrastructure?
