The current surge in artificial intelligence (AI) represents a unique and unparalleled hype cycle. While parallels can be drawn to the dotcom explosion in the late 1990s and the mobile revolution of the early 2000s, the level of investment and ambitious forecasts regarding transformative societal impacts associated with AI overshadow previous tech booms.
However, this rapid advancement comes with significant challenges. Developers in the AI sector are facing obstacles such as securing sufficient energy for the extensive data centers required, navigating a shortage of memory chips, and improving the efficiency of data transfer among AI chips and systems.
An innovative technology, known as photonics, presents a potential solution for addressing data transfer issues.
Photonics technology leverages light to facilitate data movement between components like graphics processing units (GPUs), memory units, networking chips, servers, and data centers, rather than depending on electrical signals transmitted through copper wires. Some applications of photonics are already in place, most notably in fiber optic connections.
Nevertheless, a considerable amount of data connectivity within AI servers and racks still relies on copper wiring, which can limit speed and escalate energy expenses.
"One of the primary bottlenecks affecting AI model performance is the speed of communication among chips and between chip servers," explained Gil Luria, head of technology research at D.A. Davidson.
He further elaborated, "The quicker the communication, the faster users receive their answers or have tasks completed. Transitioning to optical connections between chips and servers could lead to a substantial enhancement in model performance."
