3D photonic chip module. Credit: Keren Bergman
Artificial Intelligence (AI) systems promise transformative advocates, yet their growth in data transfer has been limited by energy disabilities and hurdles. Colombia engineering researchers have unveiled a groundbreaking solution: a 3D photonic-electronic platform that attains unprecedented energy efficiency and bandwidth density, paves the way for the next generation AI hardware.
the study“3D Photonics for Ultra-Lo Energy, high bandwidth-density chip data link,” led by Keren Bergman, Charles bachelor professors of Electrical Engineering, have been published. Nature photonics,
Research gives a description of a leading method that integrates photonics with advanced supplement-info-oxide-semiconductor (CMOS) electronics to re-define energy-skilled, high-bandwidth data communication. This innovation addresses important challenges in data movement, rapidly and a frequent obstruction to realize more efficient AI technologies.
“In this work, we present a technique that is capable of transferring vast versions of data with unprecedentedly low energy consumption,” Bergman said. “This innovation breaks through a long -standing energy barrier with a limited data movement in traditional computer and AI systems.”
Michael Kulene, Electrical Engineering Graduate student and co-writer, work with Kerin Bergman (foreground) at Lightwave Research Laboratory. Credit: Timothy Lee/Columbia Engineering
Success in data communication
The Columbia Engineering Team collaborated with the engineering ElioSha Christopher Molnar, ILLS and Charles Lee Professor at Cornell University to claim 80 photonic transmitters within a compact chip footprint and high density of the receiver within a compact chip footprint.
This platform offers high bandwidth (800 GB/s) with extraordinary energy efficiency, which consumes only 120 fematosals per bit. 5.3 With a bandwidth density of TB/S/MM2This innovation is more than the existing benchmark.
Designed for low cost, chip integrates photonic devices with CMOS electronic circuits and takes advantage of components manufactured in commercial foundry, which establishes the phase to adopt a comprehensive industry.
Revolution in AI Hardware
Team research again reveals how data is broadcast between compute nodes, addressing long -bound hurdles in energy efficiency and scalability.
By integrating 3D photonic and electronic chips, this technique achieves unmatched energy saving and high bandwidth density, free from traditional data area obstacles. It enables the innovative platform AI system to move vast versions of data efficiently, supporting distributed architecture that were previously impractical due to energy and delay boundaries.
The resulting progress is ready to unlock the unprecedented levels of performance, from which this technique becomes the cornerstone of future computing systems in applications, from large -scale AI models to real -time data processing.
Beyond AI, this approach has a transforming capacity for high-demonstration computing, telecommunications, and individual memory systems, indicating a new era of energy-efficient, high-speed computing infrastructure.
The associate research included the contribution of Molnar Lab, Air Force Research Laboratory and Dartmouth College of Cornell University.
More information:
Stuart Dudlin et al, 3D photonics for ultra-lo energy, high bandwidth-gradient chip data link, Nature photonics (2025). Doi: 10.1038/s41566-025-01633-0 But arxiv, Doi: 10.48550/arxiv.2310.01615
Citation: Next generation AI Hardware: 3D increases photonic-electronic platform efficiency and bandwidth (2025, March 23)
This document is subject to copyright. In addition to any impartial behavior for the purpose of private studies or research, no part can be re -introduced without written permission. The content is provided for information purposes only.
