Memory Wall
The growing disparity between processor speed and memory bandwidth that limits system performance in computing.
Cache-Coherent Interconnect
Cache-coherent interconnect — A system that helps data consistency between processors, accelerators, and memory in a SoC or MCM.
Wavelength Division Multiplexing (WDM)
Wavelength division multiplexing (WDM) — a way to efficiently transmit multiple optical signals in a single fiber by using different wavelengths (i.e., colors) of light.
UCIe
An open specification for die-to-die interconnect enabling high bandwidth density, low-latency, low-power communication between chiplets within a package.
Transformer Models
Neural network architectures that transform input sequences by learning contextual relationships.
Training (AI/ML)
Training (AI/ML) — The process of teaching a machine learning model to make accurate predictions or decisions based on data.
Tokens (AI)
Tokens (AI) — Tokens are fundamental units of text or data that AI models process to understand and generate language.
TeraPHY™ Optical I/O Chiplet
TeraPHY™ optical I/O chiplet — Ayar Labs’ proprietary optical I/O chiplet with both electronic and photonic elements.
Tensor Parallelism
Tensor parallelism — A technique that splits model computations across multiple devices for efficient processing.
SuperNova™ Light Source
SuperNova™ light source — Ayar Labs’ multi-wavelength, multi-port light source.