The Quest for Ultra-Low-Power Implementation of Transformer-Based LLMs: Opportunities for Optoelectronics

The rapid evolution of Transformer-based Large Language Models (LLMs) is transforming AI applications, but their growing computational complexity is creating an equally pressing challenge: energy efficiency. As model size, context length, and inference demand continue to increase, the energy required to move and process vast amounts of data is becoming a fundamental constraint on the scalability of AI systems.

This presentation explores the quest for ultra-low-power implementation of Transformer-based LLMs and examines how advances in optoelectronic devices, integration, and system architectures could offer fundamentally new approaches to addressing these challenges, complementing electronics where conventional approaches face increasing efficiency constraints. More broadly, it considers the potential of optoelectronics as an enabling technology for energy-efficient AI computing —from optical interconnect to emerging approaches for accelerating the core operations of Transformer architectures.