Why is our technology vital to the world?
Lack of energy efficient interconnects is limiting Large Language Models in AI
Ever since OpenAI released Chat-GPT 4 to the public in late 2022 the world has been fascinated with exploring new applications where these Large Language Models (LLMs) could be put to use. What most people don’t realize is that these latest LLMs use hundreds of billions of parameters, and that low power low-latency interconnects in the underlying High-Performance Compute (HPC) architectures are key to further advances in performance.
While electrical interconnects work well over short distances of a few millimeters, only optical connections can provide suitable links over longer distances. However, conventional optical interconnects using lasers are bulky and consume lots of energy. This is where Avicena’s LightBundle™ microLED based interconnects will offer a paradigm shift in ultra-low power and low latency which will transform the High-Performance Computer (HPC) industry.
Learn more about how Avicena LightBundle™ technology can help solve the power challenge in AI/ML HPC.
How Does Our Technology Work?
LightBundle™ — Using microLEDs to “Move Data”
Avicena’s LightBundle™ interconnects use arrays of microLEDs connected via multi-core fiber bundles to Si Detectors on CMOS ICs. This enables ultra-low power links of < 1pJ/bit with up to 10m reach.
Want to learn even more about how Avicena LightBundle™ technology can revolutionize interconnects?
Avicena Technology Highlights
High Energy Efficiency < 1 pJ/bit < 0.5 pJ/bit
High Bandwidth Density > 1 Tbps/mm > 6 Tbps/mm
Low Latency < 2 ns + ToF < 2 ns + ToF
High Temperature Tolerance > 125ºC > 150ºC
How does it relate to interconnect technology?
Avicena LightBundle™ Power and Density are >10x Better
The performance of an interconnect can be expressed in terms of a figure of merit (FoM) combining bandwidth density on the edge of a chip (Gbps/mm) with energy efficiency (pJ/bit). This FoM is dependent on reach. As is the shown in the chart below our microLED based interconnects extend the performance of electrical links from a few mm up to about 10m thus allowing much more efficient disaggregation of high-performance compute (HPC) architectures.