Marvell Unveils 1.6T Silicon Photonics Light Engine for AI Networks at OFC 2025

Marvell Technology has introduced its 1.6T silicon photonics light engine at OFC 2025, aimed at transforming rack-scale AI interconnects. The new device, part of Marvell’s light engine portfolio, is designed for low-power, high-bandwidth applications using linear-drive pluggable optics (LPO) and on-board optics. It supports eight channels of 200 Gbps PAM4 optical signaling, enabling a total throughput of 1.6 terabits per second.

The light engine integrates key components including linear drivers, transimpedance amplifiers, a silicon photonics chip, and an embedded microcontroller, all within a single compact package. It consumes less than 5 picojoules per bit under typical conditions and is optimized for use in LPO modules, retimed optics, and as a foundation for co-packaged optics (CPO) systems.

The technology offers a power-efficient alternative to passive copper for rack-scale and row-scale connectivity in AI systems. Its modular design helps speed up system integration and time to market for hyperscalers and optical module vendors.

Currently sampling to select customers, the 1.6T light engine positions Marvell to lead the transition from copper to optical connectivity in future data center infrastructure.