Avicena unveils ultra-low power microLED optical interconnects for next-generation AI data centers

Avicena has announced its LightBundle microLED-based optical links are now operating at 4 Gbps per lane with transmitter (Tx) currents as low as 100 μA per LED. According to Avicena, this breakthrough delivers a raw bit error rate (BER) of 1 x 10-12 and achieves an energy efficiency of 80 fJ/bit per LED—without relying on Forward Error Correction (FEC). The company claims this performance addresses the increasing bandwidth and reliability demands of future AI infrastructure in data centers, especially for hyperscale environments.

The LightBundle platform uses microLEDs as direct light sources instead of lasers, enabling lower transmit power since operation is not limited by a lasing threshold. Avicena highlights that while silicon photonics may require splitting external laser sources and utilize complex control mechanisms, microLEDs generate light directly, minimizing packaging complexity. The transmitters are only a few microns in size, do not require temperature stabilization, and can be densely arrayed to support terabit-level aggregate bandwidth.

This latest milestone was made possible by Avicena’s newly developed high-sensitivity receiver technology, built with manufacturing partners and optimized photodetectors derived from high-volume image-sensor production processes. The company reports that LightBundle technology transmits raw parallel data directly, bypassing the need to serialize lower-speed on-chip data to higher-speed optical lanes. This simplifies integrated circuit (IC) input/output architectures and supports low-latency, low-power, wide-bandwidth links across various silicon nodes.

LightBundle chiplet transceivers are designed to integrate into common packaging architectures for data center deployments, including co-packaged optics (CPO), on-board optics (OBO), pluggable modules, and wide memory interconnects. Avicena states that these transceivers can scale up to support large AI clusters spanning multiple racks and thousands of GPUs, directly targeting the hyperscale data center and AI accelerator space.

“We already demonstrated an efficient microLED link in a live demo at ECOC 2025 in September. By further optimizing our highly sensitive receivers, we have managed to further reduce the operating currents of the microLEDs and obtained Tx energy consumption down further to tens of femtojoules for this part of the link. Combined with the unique properties of microLEDs, we can achieve unmatched energy efficiency in our LightBundle interconnects,” said Rob Kalman, Chief Scientist of Avicena. “This benchmark shows the scalability of our roadmap, how microLED technology can replace legacy laser-based links with a simpler, more reliable and far lower power solution.”

Source: Avicena

Get Data Center Engineering News In Your Inbox:

Popular Posts:

1765906506220
Tritium launches 800 VDC bidirectional inverter for data centers and renewable energy sites
Water_From_Air_Data_Center_Rendering_Nov_2025
iMasons selects Water From Air for data center waste heat-to-water innovation
Screenshot
Five AI data centers to reach 1 GW power capacity in 2026, new analysis shows
68e79d30a17eea847251fae6_img-home-product-liquidjet-main
Frore Systems updates LiquidJet direct-to-chip coldplate for 1,950 W NVIDIA Rubin data center GPUs
River Bend Rendering
Hut 8 announces $7 billion, 245 MW hyperscale data center IT lease with Fluidstack backed by Google

Share Your Data Center Engineering News

Do you have a new product announcement, webinar, whitepaper, or article topic? 

Get Data Center Engineering News In Your Inbox: