While the speed of high-speed interfaces (SerDes, PCIe, UCIe, DDR, HBM,...) doubles every two years, the size of AI models increases at a much faster rate, doubling every 4 months. There is a need for parallelism to close this performance gap and continue to scale the computing and memory resources while addressing latency, density, energy efficiency and cost.

​

Our ability to integrate laser arrays with silicon photonics is ideal for channel multiplexing in pluggable or co-package optics for even denser and lower power implementation.