GlobalFoundries and Corning’s collaboration advances co-packaged optics technology to meet exponential bandwidth demands of artificial intelligence computing
In the race to satisfy artificial intelligence’s insatiable appetite for data, a fundamental bottleneck has emerged not in processing power, but in the microscopic realm where silicon chips must interface with optical fibers. The exponential growth of AI workloads has pushed datacenter connectivity requirements beyond the capabilities of traditional electronic interconnects, driving a technological revolution toward photonic integration. Now, a collaboration between semiconductor manufacturer GlobalFoundries and glass innovator Corning promises to address this challenge through breakthrough detachable fiber connector solutions that seamlessly merge silicon photonics with advanced glass waveguide technology.
The partnership, announced this week, centers on Corning’s GlassBridge solution—a glass-waveguide-based edge coupler designed to integrate with GlobalFoundries’ silicon photonics platform. This seemingly incremental advance addresses one of the most complex engineering challenges in modern computing: creating reliable, high-bandwidth optical connections between photonic integrated circuits and the fiber optic networks that carry data across datacenters and between computing facilities.
Silicon photonics represents a convergence of two mature technologies that promises to revolutionize data transmission. By fabricating optical components using the same silicon-on-insulator processes employed for conventional semiconductors, engineers can create photonic integrated circuits (PICs) that manipulate light with the precision and scale of modern microelectronics. These circuits can encode data onto optical signals, route those signals through microscopic waveguides, and decode optical information back into electronic form—all within chip-scale dimensions compatible with existing semiconductor manufacturing infrastructure.
The physics underlying silicon photonics exploits silicon’s transparency to infrared light at telecommunications wavelengths around 1.55 micrometers. At these wavelengths, silicon exhibits a high refractive index contrast with silicon dioxide, enabling tight optical confinement within waveguides just hundreds of nanometers wide. This confinement allows complex optical circuits to be fabricated within the footprint of conventional electronic chips, dramatically reducing the size, power consumption, and cost of optical communication systems.
However, transitioning optical signals between on-chip silicon waveguides and external optical fibers has remained a persistent engineering challenge. The fundamental issue lies in mode field mismatch—silicon waveguides confine light to cross-sectional areas roughly 100 times smaller than the cores of standard optical fibers. This dimensional discrepancy creates significant optical losses when coupling light between silicon chips and fiber networks, limiting system performance and reliability.
Traditional approaches to fiber-to-chip coupling have relied on either grating couplers that redirect light vertically from chip surface to fiber, or edge couplers that transition light horizontally from waveguide to fiber end. Both methods require precise alignment tolerances measured in fractions of micrometers, making assembly complex and expensive while limiting the practicality of detachable connections essential for system maintenance and upgrades.
Corning’s GlassBridge technology represents a sophisticated solution to these coupling challenges. By fabricating glass waveguides with precisely controlled refractive index profiles, the system creates intermediate optical structures that gradually transition modal characteristics between silicon waveguides and optical fibers. This adiabatic transformation minimizes reflection losses that occur when light encounters abrupt changes in waveguide geometry, enabling efficient power transfer across the silicon-to-fiber interface.
The glass substrate approach offers several advantages over alternative coupling technologies. Glass materials can be engineered with refractive indices and thermal expansion coefficients optimized for specific applications, while advanced processing techniques including ion exchange (IOX) and laser processing enable precise three-dimensional waveguide formation. These capabilities allow Corning to create coupling structures with optical properties tailored to minimize losses for specific wavelengths and polarizations critical to datacenter applications.
The collaboration extends beyond edge coupling to include vertically-coupled detachable fiber-to-PIC solutions, demonstrating the versatility of glass-based optical interconnect approaches. Vertical coupling offers advantages for high-density packaging scenarios where horizontal space is constrained, while maintaining the detachable connectivity essential for practical datacenter deployment. The ability to support multiple coupling geometries within a single platform provides system designers with flexibility to optimize interconnect architectures for specific applications.
Kevin Soukup, senior vice president of GlobalFoundries’ silicon photonics product line, emphasized the performance implications: “Corning’s cutting-edge fiber technology, integrated with our silicon-proven platform, delivers the performance and flexibility required for enabling scalable, high-density optical packaging for AI datacenters.” This scalability becomes crucial as AI workloads continue expanding, with some estimates suggesting that advanced machine learning models may require interconnect bandwidths exceeding terabits per second within individual computing systems.
The timing of this collaboration reflects the urgency of addressing AI connectivity bottlenecks. Modern artificial intelligence training and inference workloads involve massive parallel computations distributed across thousands of processing units, requiring constant data exchange between processors, memory systems, and storage networks. Electronic interconnects that served previous generations of computing systems cannot provide the bandwidth density and energy efficiency required for these AI applications, driving rapid adoption of optical communication technologies.
Co-packaged optics (CPO) represents the architectural approach enabled by technologies like GlassBridge. Rather than connecting optical transceivers to processing chips through printed circuit board traces and connectors, CPO integrates photonic devices directly within processor packaging, eliminating intermediate electrical connections that limit bandwidth and consume power. This integration requires reliable, high-performance fiber connections that can be assembled and maintained within the constrained environments of advanced semiconductor packages.
The energy efficiency advantages of optical interconnects become particularly important for AI datacenters, where power consumption represents a major operational cost and environmental concern. Electronic signaling over copper interconnects consumes power proportional to data rate and transmission distance, making high-bandwidth, long-distance electrical connections prohibitively energy-intensive. Optical signaling, once converted from electrical form, can traverse arbitrary distances with minimal additional power consumption, enabling more efficient datacenter architectures.
All-fiber sensor technologies have demonstrated the potential for sophisticated optical signal processing without electronic conversion, suggesting pathways for even more advanced optical computing architectures. Similarly, developments in photonic lattice structures indicate the potential for complex optical signal processing directly within photonic integrated circuits.
Dr. Claudio Mazzali, vice president of global research at Corning, highlighted the broader implications: “There’s something truly powerful in combining GlobalFoundries’ and Corning’s expertise in silicon process and optical connectivity—together, we’re enabling new possibilities for the AI-powered industries of tomorrow.” This combination leverages Corning’s 170-year history in glass science and optical innovation with GlobalFoundries’ advanced semiconductor manufacturing capabilities.
The collaboration also addresses supply chain considerations critical for large-scale AI infrastructure deployment. Corning’s established optical fiber and connectivity supply chains provide the manufacturing scale necessary to support the massive fiber count requirements of next-generation datacenters, while GlobalFoundries’ high-volume semiconductor fabrication capabilities enable cost-effective production of photonic integrated circuits.
Manufacturing considerations become particularly important for detachable fiber connector solutions, which must maintain optical performance through repeated connection and disconnection cycles while withstanding the environmental stresses of datacenter operation. The precision required for optical alignment, combined with the mechanical robustness necessary for practical deployment, demands sophisticated materials engineering and manufacturing process control.
The international scope of the collaboration reflects the global nature of AI infrastructure development. With demonstrations planned for the European Conference on Optical Communication (ECOC) in Copenhagen and GlobalFoundries’ Technology Summit in Munich, the partnership aims to engage the worldwide community of optical communication engineers and system architects developing next-generation AI computing platforms.
Looking forward, the success of glass-waveguide-based coupling technologies could accelerate broader adoption of co-packaged optics across computing applications beyond AI. High-performance computing, 5G wireless infrastructure, and autonomous vehicle systems all face similar connectivity challenges that could benefit from advanced optical interconnect solutions. The modular, detachable nature of the GlassBridge approach provides deployment flexibility essential for diverse application requirements.
The emergence of practical, high-performance fiber-to-chip coupling solutions also enables more ambitious optical computing architectures. As quantum photonics research continues advancing and optical signal processing capabilities expand, the boundary between optical communication and optical computation may continue blurring, potentially leading to hybrid electronic-photonic systems that leverage the advantages of both domains.
The GlobalFoundries-Corning collaboration thus represents more than an incremental advance in optical connectivity—it demonstrates the maturation of silicon photonics toward practical, large-scale deployment in the AI infrastructure that will define the next generation of computing capability. By solving the fundamental challenge of efficient, reliable fiber-to-chip optical coupling, this partnership removes a critical bottleneck constraining the evolution of AI datacenters toward the optical architectures necessary for continued performance scaling.
As artificial intelligence continues its rapid evolution from research curiosity to transformative technology, the optical connectivity solutions emerging from collaborations like this will prove essential for realizing AI’s full potential across scientific discovery, technological innovation, and societal advancement.
Demonstrations of the GlassBridge edge-coupled glass-waveguide based detachable fiber connector solution will be showcased at ECOC 2025 in Copenhagen (Corning booth #2118) and GlobalFoundries’ Technology Summit in Munich, Germany.
