Seamless optical cloud computing across edge-metro network for generative AI

Area of Science:

• Computer Science
• Optical Engineering
• Artificial Intelligence

Background:

• Generative artificial intelligence (AI) advancements necessitate powerful computational architectures.
• Current cloud computing models face challenges in power consumption and security.
• Edge-metro networks require efficient solutions for distributed AI tasks.

Purpose of the Study:

• To propose and demonstrate an optical cloud computing system for generative AI.
• To reduce power consumption and enhance computational scale in cloud computing.
• To enable seamless deployment across edge-metro networks.

Main Methods:

• Modulating AI inputs and models into light signals.
• Developing an optical computing center accessible via edge-metro networks.
• Experimental validation of the optical cloud computing architecture.

Main Results:

• Achieved an energy efficiency of 118.6 mW/TOPs.
• Reduced energy consumption by two orders of magnitude compared to electronic solutions.
• Successfully performed complex generative AI models for image generation tasks.

Conclusions:

• The proposed optical cloud computing system offers a revolutionary, energy-efficient solution for generative AI.
• This architecture can be deployed across edge-metro networks, overcoming limitations of traditional cloud computing.
• Optical cloud computing paves the way for scalable and sustainable AI development.