Purely decorative image, appearing on Electronics Industry Monthly in an article entitled IBM Pioneers Co-Packaged Optics with PWG for AI Data Centers.

IBM Pioneers Co-Packaged Optics with PWG for AI Data Centers

Artificial Intelligence Products of the Month R&D Breakthrough Semiconductor Sensors and Transducers

IBM has unveiled a breakthrough in co-packaged optics (CPO) technology, leveraging polymer optical waveguides (PWG) to significantly enhance AI data center performance. This advancement represents the first successful implementation of PWG to support CPO, as detailed in IBM Research’s recently published technical paper. The innovation enables high-bandwidth optical connectivity, offering faster AI model training, improved energy efficiency, and reduced data transmission delays.

“As generative AI demands more energy and processing power, the data center must evolve – and co-packaged optics can make these data centers future-proof,” said Dario Gil, Senior Vice President and Director of Research at IBM. “With this breakthrough, tomorrow’s chips will communicate much like how fiber optics cables carry data in and out of data centers, ushering in a new era of faster, more sustainable communications that can handle the AI workloads of the future.”


Closing the Gap Between AI Demands and Data Center Limitations

Purely decorative image, appearing on Electronics Industry Monthly in an article entitled IBM Pioneers Co-Packaged Optics with PWG for AI Data Centers.
IBM’s co-packaged optics module with PWG technology enables high-bandwidth, energy-efficient data center communication. (Photo Courtesy of IBM)

Fiber optics currently transmit data across long distances globally but are rarely used within data centers, where copper-based electrical wires predominate. This reliance on electrical interconnects often results in slowdowns, particularly for GPU accelerators that spend significant time waiting for data during distributed AI training processes.

IBM’s CPO technology replaces these electrical interconnects with optics, providing faster, more efficient data transfers. This shift allows data centers to scale AI operations while addressing growing energy consumption concerns.

Key benefits of IBM’s CPO innovation include:

  • Energy Efficiency: IBM’s optical solution consumes five times less energy than mid-range electrical interconnects, potentially saving energy equivalent to 5,000 U.S. homes’ annual consumption for each AI model trained.
  • Accelerated Training: Large language models (LLMs) can be trained up to five times faster, reducing training time from months to weeks.
  • Extended Reach: Optical connections can extend from centimeters to hundreds of meters, enabling greater scalability.

Polymer Optical Waveguides Enable High Bandwidth

Central to this innovation is IBM’s development of high-density polymer optical waveguides (PWG). These waveguides, integrated with silicon photonics at a 50-micrometer pitch, allow for 80 times the bandwidth density of conventional electrical connections. IBM’s research demonstrates the ability to add six times more optical fibers to the edges of a silicon photonics chip, increasing the number of optical pathways while maintaining reliability.

These components have undergone rigorous stress testing, including exposure to high humidity, extreme temperatures (-40°C to 125°C), and mechanical durability tests. The results confirm their suitability for real-world manufacturing and deployment. Researchers also demonstrated the ability to achieve an 18-micrometer pitch, paving the way for up to 128 connectivity channels when stacking multiple waveguides.


Scaling AI with Sustainable Optics

IBM’s innovation in CPO technology aligns with its long-standing leadership in semiconductor research, which includes milestones such as the first 2 nm node chip and nanosheet transistors. By integrating optics into data center interconnections, IBM offers a sustainable solution to meet the performance demands of AI applications.

“Fiber optics have long transformed external communications, and now they are poised to revolutionize internal data center operations,” said Gil. “This advancement reflects IBM’s commitment to creating scalable, energy-efficient solutions for the AI era.”


Research and Development in the Northeast Semiconductor Corridor

IBM’s groundbreaking co-packaged optics (CPO) prototype was developed within the Northeast Semiconductor Corridor, a key hub for semiconductor innovation spanning the United States and Canada. The design, modeling, and simulation work for the CPO technology took place at IBM’s Albany, New York research center, home to the National Semiconductor Technology Center (NSTC) EUV Accelerator, which supports advanced semiconductor research and development.

Purely decorative photo appearing in an article on Electronics Industry Monthly entitled
Researchers stress-test an optics module at IBM Research’s optics lab in Yorktown Heights, NY. Credit: Ryan Lavine for IBM

Prototyping and testing were carried out at IBM’s Bromont, Quebec facility, one of North America’s largest and most advanced chip assembly and testing sites. Known for its global leadership in chip packaging, the Bromont lab validated the durability and performance of the optical components under real-world conditions. Together, these facilities exemplify IBM’s ongoing commitment to pushing the boundaries of semiconductor technology.

This effort builds on IBM’s long-standing history of innovation in semiconductor research, including breakthroughs such as the first 2 nm node chip, the development of nanosheet and vertical transistors (VTFET), and advancements in 7 nm and 5 nm process technologies. With its pioneering CPO technology, IBM continues to deliver solutions that address the growing performance demands of AI and next-generation computing.

IBM’s introduction of CPO with PWG technology provides a roadmap for data centers to scale AI operations while reducing energy use and operational delays. By replacing traditional electrical interconnects with optics, IBM is addressing the evolving demands of generative AI while ensuring sustainable, high-performance solutions for the future of computing.


About IBM

IBM provides hybrid cloud, artificial intelligence (AI), and consulting solutions to clients in over 175 countries. The company supports businesses and organizations in sectors such as financial services, telecommunications, and healthcare by helping them utilize data insights, optimize business processes, and achieve digital transformation securely and efficiently. IBM’s research and innovation extend across AI, quantum computing, and industry-specific cloud solutions, offering open and flexible options to address diverse business needs. Its hybrid cloud platform, including Red Hat OpenShift, is used by more than 4,000 government and corporate entities globally. IBM is committed to transparency, responsibility, inclusivity, and service, delivering solutions designed to meet the challenges of a rapidly evolving technological landscape.

For more information, please click here.

Source/Image: IBM


(Editor’s Note: All trademarks mentioned in this article, including company names, product names, and logos, are the property of their respective owners. Use of these trademarks is for informational purposes only and does not imply any endorsement.)

Tagged