ChatGPT at the Speed of Light: Optical Comb-Based Monolithic Photonic-Electronic Linear-Algebra Accelerators

TL;DR

This paper introduces a novel monolithic silicon-photonics chip that accelerates linear algebra computations using optical comb technology, significantly improving density and energy efficiency for deep learning applications.

Contribution

It presents a new integrated photonic-electronic system-on-chip design leveraging optical combs for high-dimensional matrix operations, advancing the state-of-the-art in AI hardware acceleration.

Findings

Achieves high-density matrix-matrix multiplication with optical comb technology.

Demonstrates energy-efficient computation suitable for large neural networks.

Addresses practical integration challenges through holistic co-design.

Abstract

This paper proposes to adopt advanced monolithic silicon-photonics integrated-circuits manufacturing capabilities to achieve a system-on-chip photonic-electronic linear-algebra accelerator with the features of optical comb-based broadband incoherent photo-detections and high-dimensional operations of consecutive matrix-matrix multiplications to enable substantial leaps in computation density and energy efficiency, with practical considerations of power/area overhead due to photonic-electronic on-chip conversions, integrations, and calibrations through holistic co-design approaches to support attention-head mechanism based deep-learning neural networks used in Large Language Models and other emergent applications.