Optical interconnect platform delivers over 1 Tbps/mm I/O density
Avicena has announced its new scalable LightBundle™ interconnect platform at OFC 2025, which supports over 1 Tbps/mm shoreline density and extends ultra-high-density die-to-die (D2D) connections to more than 10 meters with class-leading sub-pJ/bit energy efficiency.
The optical interconnect platform will enable AI Scale-Up networks to support large clusters of GPUs across multiple racks, eliminating the reach limitations of current copper interconnects while drastically reducing power consumption.
The LightBundle interconnect features a modular transceiver chiplet-based architecture. Transceiver chiplets contain integrated microLED and PD arrays connected to other LightBundle transceiver chiplets via multi-core fibre bundles. They support dense, low-power D2D interfaces such as UCIe or BOW. Chiplet transceivers are suitable for various packaging architectures, including co-packaged optics (CPO), on-board optics (OBO), pluggable optical modules, and active optical cables (AOC). In contrast to silicon photonics, LightBundle interfaces can be integrated into nearly any IC process from a diverse range of foundries.
The increasing demand for compute and memory performance driven by advanced AI models requires efficient interconnects with higher density and lower power consumption. In AI data centres, traditional passive copper interconnects limit GPU clusters to a single rack due to their short reach of about 1 meter. The LightBundle interconnect technology overcomes this limitation by extending the reach to over 10 meters, allowing clusters to scale to hundreds or thousands of GPUs across multiple racks while significantly reducing power consumption and easing cooling requirements.
“At Avicena, we are excited to announce our modular, ultra-low power, scalable chiplet interconnect based on our LightBundle technology,” says Bardia Pezeshki, Co-Founder and CEO of Avicena. “Working with our hyperscale data centre partners, we have optimised our platform to address the specific needs of large clusters of GPUs in future Scale Up networks. We support shoreline densities of greater than 1Tbps/mm at power efficiencies at least 5x better than competing solutions.”
“As generative AI continues to evolve, the role of high bandwidth-density, low-power and low latency interconnects in scale-up networks cannot be overstated”, says Michael Fox, Founder & Managing Partner of InflectionPoint Research. “Avicena’s microLED-based interconnect technology has the potential to deliver a paradigm shift in this domain of hyperscale AI Scale-Up networks.”
https://avicena.tech
www.ofcconference.org
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
