3D photonic chip module. Credit: Keren Bergman.

A Columbia Engineering team addresses critical challenges in data communications that persistently block AI hardware systems performance.

Artificial intelligence (AI) systems promise transformative advancements, yet their growth has been limited by energy inefficiencies and bottlenecks in data transfer.

Researchers at Columbia Engineering have unveiled a groundbreaking solution: a 3D photonic-electronic platform that achieves unprecedented energy efficiency and bandwidth density, paving the way for next-generation AI hardware.

The research, led by Keren Bergman, Charles Batchelor Professor of Electrical Engineering at Columbia, details a pioneering method that integrates photonics with advanced complementary-metal- oxide-semiconductor (CMOS) electronics to redefine energy-efficient, high-bandwidth data communication.

This innovation addresses critical challenges in data movement, a persistent obstacle to realising faster and more efficient AI technologies.

Bergman said: “In this work, we present a technology capable of transferring vast volumes of data with unprecedentedly low energy consumption.

“This innovation breaks through the long-standing energy barrier that has limited data movement in traditional computer and AI systems.”

Michael Cullen, electrical engineering graduate student and co-author of the paper, works with Keren Bergman (foreground) in the Lightwave Research Laboratory. Credit: Timothy Lee/ Columbia Engineering.

Breakthrough in data communication

The Columbia Engineering team collaborated with Alyosha Christopher Molnar, Ilda and Charles Lee Professor of Engineering at Cornell University, to develop a 3D-integrated photonic-electronic chip that boasts a high density of 80 photonic transmitters and receivers within a compact chip footprint.

This platform delivers high bandwidth (800 Gb/s) with exceptional energy efficiency, consuming just 120 femtojoules per bit.

With a bandwidth density of 5.3 Tb/s/mm², this innovation far exceeds existing benchmarks.

Designed for low cost, the chip integrates photonic devices with CMOS electronic circuits and leverages components manufactured in commercial foundries, setting the stage for widespread industry adoption.

Revolutionising AI hardware

The team’s research redefines how data is transmitted between compute nodes, addressing the long-standing bottlenecks in energy efficiency and scalability.

By 3D integrating photonic and electronic chips, this technology achieves unmatched energy savings and high bandwidth density, breaking free from traditional data locality constraints.

This innovative platform enables AI systems to efficiently transfer vast volumes of data, supporting distributed architectures that were previously impractical due to energy and latency limitations. 

The resulting advancements are poised to unlock unprecedented levels of performance, making this technology a cornerstone of future computing systems across applications, from large-scale AI models to real-time data processing in autonomous systems.

Beyond AI, this approach holds transformative potential for high-performance computing, telecommunications, and disaggregated memory systems, signalling a new era of energy-efficient, high-speed computing infrastructure.

The study is published in Nature Photonics.