GlobalFoundries’ SCALE optical module solution for co-packaged optics (CPO) is a platform designed to meet OCI MSA specifications. It delivers high-bandwidth, energy-efficient AI interconnects. Built on GF’s silicon photonics foundation, SCALE brings together advanced photonics and modern packaging. This helps speed up design-to-volume production for next-generation AI systems.
Overview: SCALE and the CPO revolution
GlobalFoundries’ SCALE platform—the Silicon Photonics Co-Packaged Advanced Light Engine—stands out as the industry’s first CPO solution that’s compatible with the Optical Compute Interconnect Multi-Source Agreement (OCI MSA). By merging optics and electronics into a single module, SCALE aims to meet the rising demand for scalable AI compute. It offers better bandwidth density, lower latency, and better energy efficiency.
The architecture is ready to support future AI workloads that need rapidly expandable interconnect bandwidth. It’s also built for reliability in data-center environments. SCALE marks a shift away from copper toward optical interconnects, using co-packaged optics to shrink footprint, cut power, and lower signaling latency.
SCALE’s compatibility with OCI MSA opens up broader ecosystem interoperability. Customers can move more easily from design to full-scale production.
Optical performance and WDM capabilities
SCALE supports both CWDM and DWDM for bidirectional transmission over a single fiber. This setup boosts bandwidth density and keeps things flexible for different AI workloads. The architecture can handle future expansions in spectral channels and still stay serviceable.
GF has shown native 8λ– and 16λ-bidirectional DWDM on the platform. That’s a big step toward high-channel-count interconnects in CPO-enabled systems. With a broadband approach across the CWDM band, SCALE keeps performance flat as channels scale up.
It’s easy to move from four wavelengths per direction to eight and beyond, and insertion losses stay manageable for system designers.
Integrated photonics devices and capabilities
The SCALE solution packs in a suite of qualified photonic devices essential for high-speed interconnects. You’ll find 50 Gbps and 100 Gbps microring modulators, coupled ring resonators, and integrated photodiodes. These components allow for efficient modulation, precise wavelength handling, and high-sensitivity detection.
They’re all critical for keeping signal integrity strong in dense multi-wavelength networks used by AI accelerators.
Packaging architecture and silicon interposers
Advanced packaging sits at the heart of SCALE. The design uses through-silicon vias (TSVs) to route high-speed signals and power, keeping parasitics low.
Copper pad pitches range from 110 μm down to under 45 μm, which enables 2.5D/3D stacking on silicon interposers. This creates compact, high-performance modules that fit right into data centers.
By including electrical ICs on advanced single-digit node processes, compute and optics work together without sacrificing performance or efficiency.
Removable broadband fiber approach and maintainability
SCALE features a removable broadband fiber approach, offering flat insertion loss across the CWDM band. This move improves serviceability and makes it easier to test known-good dies (KGDs).
It helps with troubleshooting and speeds up fault isolation. The architecture is set up for scalable channel counts, so reliability stays high as the interconnect grows from 4λ per direction to 8λ and beyond.
Impact on AI infrastructure and industry adoption
The SCALE platform looks set to speed up the transition from design to volume production for new AI interconnects. By sticking to the OCI MSA, SCALE gives a path to interoperable, scalable optics that can keep up with the throughput and energy-efficiency needs of modern AI workloads.
The approach aims to cut total cost of ownership while delivering the performance that data-center operators and AI model developers want.
Executive perspective and market readiness
Mike Hogan, GF’s Chief Business Officer, points to the company’s decade-plus photonics expertise and their readiness to scale high-bandwidth, energy-efficient connectivity for AI infrastructure.
What this means for researchers and engineers
- Researchers and engineers can now tap into better interoperability with OCI MSA compatibility. This speeds up design cycles and makes ecosystem collaboration less of a headache.
- DWDM/CWDM technology unlocks high-bandwidth, low-latency AI interconnects. The 8λ and 16λ bidirectional channels are already up and running.
- Advanced packaging—think TSVs and fine copper pitches—lets you pack in dense 2.5D/3D integration right on silicon interposers.
- Flat CWDM insertion loss supports robust testability and known good dies (KGDs). This definitely makes validation in development and manufacturing less painful.
- AI workloads can hit volume production faster than before. That means less waiting around for new accelerators and systems to hit the market.
AI models just keep getting bigger and hungrier for data. SCALE feels like a smart move toward real, scalable optical interconnects that might finally keep up with the crazy demands of today’s AI infrastructure.
Here is the source article for this story: GlobalFoundries: Accelerating introduction of co-packaged optics for modern AI data centers with SCALE optical module solution