Avicena demonstrated a fully operational microLED-based optical link at ECOC 2025 in Copenhagen, achieving an industry milestone of 200 femtojoules per bit (fJ/bit) at <1E-12 raw bit error rate without forward error correction. The live demonstration underscores the efficiency of Avicena’s LightBundle chiplet platform for scale-up GPU clusters and next-generation memory interconnects in AI and HPC data centers.
The prototype system used a microLED running at 0.25 mA drive current and 4 Gb/s, paired with a hybrid-bonded camera sensor and discrete TIA. By leveraging the absence of threshold current in LEDs, Avicena demonstrated energy-efficient operation at dramatically reduced drive currents. The LightBundle approach bypasses high-speed serialization by transmitting raw parallel data directly, simplifying architectures and enabling massive microLED arrays with low latency, low power, and high aggregate bandwidth.
Avicena positions LightBundle as a platform spanning co-packaged optics (CPO), on-board optics (OBO), pluggable modules, and high-bandwidth memory (HBM) interfaces. The company highlighted the technology’s ability to address GPU-to-GPU and HBM-to-processor connectivity, targeting scale-up clusters of thousands of GPUs.
• Live demo: 200 fJ/bit microLED optical link at <1E-12 BER
• LightBundle chiplet platform for CPO, OBO, pluggables, and memory interconnects
• Direct parallel transmission avoids high-speed serialization
• Hybrid-bonded receiver based on high-volume camera process
• Energy-efficient solution for GPU clusters and HBM interfaces
“By leveraging a highly sensitive receiver, a minor modification to a high-volume camera process, and the unique properties of microLEDs, we can achieve unmatched energy efficiency in our LightBundle interconnects,” said Bardia Pezeshki, CTO of Avicena.
🌐 Analysis: Avicena’s demonstration highlights microLEDs as a disruptive alternative to laser-based optical interconnects, offering ultra-low power and low-latency parallel transmission. While still at early stages compared to established laser-driven VCSEL and silicon photonics solutions, microLED approaches could be particularly compelling for HBM memory interfaces, where wide, parallel buses are critical. As co-packaged optics and memory-to-processor interconnects evolve, Avicena’s work signals growing interest in diversifying beyond traditional laser-based platforms.
🌐 We’re tracking the latest developments at ECOC25 in Copenhagen. Follow our ongoing coverage at https://convergedigest.com/tag/ecoc25/
🌐 We’re launching the “Data Center Networking for AI” series on NextGenInfra.io and inviting companies building real solutions—silicon, optics, fabrics, switches, software, orchestration—to share their views on video and in our expert report. To get involved, send a note to jcarroll@convergedigest.com or info@nextgeninfra.io.
