Insights
Key terms in co-packaged optics and AI infrastructure.
Integration of optical components directly into the semiconductor package, eliminating the need for pluggable optical transceivers and enabling higher bandwidth, lower latency, and improved power efficiency.
Ayar Labs' co-packaged optical I/O chiplet that delivers over 8 Tbps of bandwidth per chiplet with sub-25ns end-to-end latency.
Ayar Labs' remote light source in the industry-standard ELSFP form factor, providing efficient laser power to TeraPHY optical engines across the rack.
Technology that uses silicon as an optical medium to create photonic devices and circuits on a chip using standard CMOS manufacturing processes.
General term for accelerator chips used in AI workloads, including GPUs, TPUs, and other AI-specific processors.
The amount of data bandwidth available per unit area or unit power, critical for AI scale-up architectures.
Universal Chiplet Interconnect Express — an industry standard for die-to-die connectivity that Ayar Labs supports for its optical chiplet integration.
Enhanced Large SFP — an industry-standard form factor for optical transceivers used by the SuperNova light source.
A photonic device pioneered by Ayar Labs that enables high-density, energy-efficient optical modulation for co-packaged optics.
The process of connecting AI accelerators within a single cluster to operate as one unified system, requiring high-bandwidth, low-latency optical interconnects.