Corning Glass Bridge for CPO: How Passive Fiber-to-PIC Coupling Could Change Optical Packaging

Co-packaged optics places optical engines close to switch ASICs, GPUs, or other high-bandwidth processors, shortening the electrical path between processing silicon and the optical interface. This tighter integration shifts more of the packaging burden to fiber attachment, optical alignment, mechanical tolerance, thermal control, and manufacturing repeatability.

Corning GlassBridge addresses one part of this challenge: connecting external optical fibers to a photonic integrated circuit. It does not replace the complete optical engine or the other optical, electronic, thermal, and packaging functions of the module. Its significance lies in using wafer-fabricated glass waveguides, passive alignment, and a detachable physical-contact interface to perform the Fiber-to-PIC connection differently from a conventional Fiber Array Unit.

Corning GlassBridge is a detachable, wafer-based Fiber-to-PIC connector platform that uses ion-exchange glass waveguides and passive mechanical alignment to connect external fibers to a photonic integrated circuit. It is intended for high-density NPO, CPO, and photonic-module architectures rather than functioning as a complete optical engine or data-center solution.

A photonic integrated circuit can generate, modulate, route, receive, or process optical signals, but it still needs a physical interface to the fibers carrying those signals outside the package. Each fiber channel must be positioned relative to the corresponding optical structure on the PIC while maintaining acceptable coupling loss.

This role is traditionally performed by a Fiber Array Unit, or FAU. A conventional FAU arranges fibers at controlled positions, commonly through precision V-groove structures. Depending on the coupling architecture, it may also work with lenses, polished fiber faces, or other micro-optical elements.

GlassBridge and a traditional FAU therefore overlap at the functional level. The major differences concern how the optical paths are formed, how final alignment is achieved, how the interface is fixed or remated, and how the design scales as channel count increases.

                                                Fiber-to-PIC Connection Architecture

GlassBridge should not be treated as another name for GlassWorks AI.

Corning launched GlassWorks AI in March 2025 as a broader portfolio for dense AI data-center infrastructure. It includes fiber, cable, connectivity hardware, network planning, design, and deployment support.

GlassBridge occupies a narrower technical position. It provides a compact interface between external fiber and the PIC edge, while the wider CPO system still requires photonic and electronic chips, optical engines, substrates, thermal management, power delivery, fiber harnesses, and system-level connectivity.

In a CPO architecture, optical engines operate close to the main processing device rather than at a distant pluggable interface. This increases integration density but places the fiber connection inside a compact package where optical, mechanical, and thermal tolerances must be managed together.

The challenge is not simply bringing a fiber close to a chip. The optical mode leaving the fiber must overlap sufficiently with the coupler or waveguide on the PIC. Small positional or angular changes can alter coupling performance.

A conventional FAU controls fiber pitch, fiber-core position, and end-face geometry. During final attachment, it must be positioned relative to the PIC or optical engine.

The FAU itself is passive, but installation may use active alignment. Light is launched or monitored while the fiber assembly is moved across several axes. When an acceptable optical position is found, the assembly is fixed, often by adhesive bonding and curing.

This method is technically mature, but the final result depends on several separately manufactured parts. Fiber position, V-groove dimensions, chip placement, adhesive thickness, package flatness, and alignment-equipment accuracy can all affect coupling.

Active alignment requires optical feedback, precision motion control, and a defined acceptance threshold. In multi-channel assemblies, the position that optimizes one channel may not produce identical results across all channels.

Traditional alignment is sometimes described as a minutes-scale operation, while passive connection is presented as a seconds-scale step. These figures are not universal benchmarks. Actual cycle time depends on channel count, coupling geometry, automation, curing, inspection, and rework.

The more reliable distinction is:

• Active alignment adjusts the completed interface through live optical feedback.

• Passive alignment relies on manufactured optical paths and mechanical references.

Moving precision into a wafer-fabricated glass element may reduce repeated adjustment at final assembly, but it does not remove the need for precision from the wider manufacturing process.

                                            Active Alignment vs. Passive Alignment Workflow

TSMC’s COUPE platform, or Compact Universal Photonic Engine, integrates an electronic IC and photonic IC within a compact photonic-engine structure. It supports both grating-coupler and edge-coupler configurations and can be integrated with a host ASIC.

A commonly shown COUPE diagram labels the EIC as a 6 nm device and the PIC as a 65 nm SOI device. These process nodes illustrate the heterogeneous integration level of the package, but they do not directly define Fiber-to-PIC alignment tolerance.

Optical tolerance is determined by the fiber mode, PIC coupler design, waveguide geometry, package stack, thermal behavior, and acceptable loss variation—not by the semiconductor process node alone.

Traditional FAUs and GlassBridge address the same Fiber-to-PIC interface through different alignment, fixation, and manufacturing approaches.

                                                       Traditional FAU vs. GlassBridge

GlassBridge uses ion-exchange waveguides formed within a glass element. The relative optical routes are established during wafer processing rather than being created only through final fiber positioning.

Mechanical references then locate the connector relative to the PIC interface. This allows final attachment to rely more heavily on repeatable geometry and less heavily on live optical optimization.

Passive alignment does not mean alignment accuracy is no longer important. Precision remains necessary in waveguide fabrication, ferrule manufacturing, PIC coupler placement, connector geometry, package reference surfaces, and final assembly.

A traditional FAU is commonly bonded after alignment. Once the adhesive is cured, removal may be difficult.

GlassBridge uses a rematable physical-contact structure based on a standard TMT ferrule format. Corning’s current design specifies a TMT ferrule with a 125 μm hole and presents the interface as detachable.

This can support more flexible assembly, testing, rework, and replacement. It does not automatically prove a specific service life or maintenance-cost reduction. Remating repeatability, contamination, retention, vibration, and thermal stability still require validation.

A traditional FAU can achieve precise fiber positioning, but the complete interface still includes several tolerance contributors, including fiber-core location, V-groove accuracy, chip placement, adhesive thickness, mounting surfaces, and final alignment.

GlassBridge moves part of this problem into wafer-based glass processing. Multiple waveguide channels can be formed relative to one another within the same manufacturing sequence.

Wafer processing does not eliminate tolerance. It changes where tolerance is generated and controlled. Waveguide uniformity, glass dimensions, ferrule fit, PIC placement, surface quality, and package references remain important.

In its March 2026 GlassBridge brochure, Corning reports demonstrated 1.5 dB O-band Fiber-to-PIC coupling.

The result is technically relevant, but it should not be treated as a universal guarantee. The published material does not define a complete production distribution, sample count, channel variation, aging result, or maximum acceptance limit.

It also does not prove lower loss than every FAU. FAU performance varies with fiber type, PIC coupler, mode-field conversion, wavelength, polishing, and alignment quality.

An optical waveguide confines light inside a region with a controlled refractive-index profile. In an ion-exchange process, mobile ions in selected areas of the glass are replaced by other ions, changing the local refractive index and forming a light-guiding path.

A 2021 review published in the peer-reviewed journal Applied Sciences traces ion-exchanged glass waveguides to the early 1970s and documents their long use in planar photonic circuits, telecommunications, and optical sensing.

This distinction matters:

• Ion-exchange glass-waveguide physics is established.

• A detachable, high-density Fiber-to-PIC connector using that technology is a newer packaging application.

                                                  IOX Waveguide and Pitch Conversion

The glass waveguide can route light between different channel pitches. This is useful because the preferred external connector pitch may differ from the optical shoreline pitch on the PIC.

Corning lists example PIC pitches of:

• 40 μm;

• 80 μm;

• 127 μm;

• 165 μm.

The current platform also publishes the following characteristics:

These are published product characteristics rather than universal specifications for every future implementation.

A Through-Glass Via is a precision opening through a glass substrate that can be metallized to route an electrical connection from one side to the other.

Corning’s semiconductor-glass platform presents TGVs as a method of routing electrical connections through glass.

IOX waveguides and TGVs perform different functions:

                                               IOX, GlassBridge and TGV Functional Roles

Corning has documented capabilities in ion-exchange waveguides, glass wafers, fiber arrays, optical connectivity, and TGV structures. These capabilities are complementary because advanced photonic packages require both optical and electrical interconnection.

However, this does not prove that every GlassBridge configuration already combines IOX waveguides and TGVs on the same commercial substrate.

The broader opportunity is that Corning can approach photonic packaging through several related capabilities rather than through one connector alone. The exact combination will depend on the PIC, package, foundry platform, and customer architecture.

GlassBridge could replace an FAU-based interface where it satisfies the required channel count, pitch, coupling geometry, loss budget, package process, reliability, and cost.

That does not mean every FAU application will migrate to GlassBridge.

In May 2025, Corning announced that it had become a qualified supplier for the optical infrastructure used with Broadcom’s Bailly CPO system. The Broadcom Bailly announcement describes fiber harnesses containing FAUs that connect fibers to silicon-photonics optical engines.

This shows that advanced FAUs remain relevant in current CPO systems. GlassBridge and FAUs are therefore more likely to coexist across different architectures than to follow an immediate industry-wide replacement cycle.

Adoption also depends on:

• passive-alignment repeatability;

• channel uniformity;

• wafer-process yield;

• remating stability;

• contamination control;

• PIC compatibility;

• inspection and rework;

• production scalability;

• total cost;

• customer qualification.

No single coupling-loss value can determine commercial adoption.

GlassBridge has moved beyond a laboratory-only concept.

Corning has published product specifications, defined connector dimensions and pitch options, reported an O-band coupling result, and developed solutions for the GF Fotonix silicon-photonics platform.

The Corning–GlobalFoundries collaboration confirms development of detachable edge- and vertical-coupling solutions and public demonstrations in 2025.

These milestones establish a defined product and demonstration stage. They do not establish universal compatibility or broad high-volume deployment.

                                GlassBridge Technology Readiness and Evaluation Framework

Application-specific validation is still needed for:

• coupling-loss distribution;

• channel uniformity;

• remating repeatability;

• contamination sensitivity;

• thermal and mechanical reliability;

• reflow stability;

• production consistency;

• PIC compatibility;

• rework procedures;

• customer qualification;

• total manufacturing cost.

GlassBridge has published specifications and foundry-platform milestones, but broad customer qualification, sustained production volume, and long-term field reliability have not yet been publicly confirmed.

Corning GlassBridge addresses a real optical-packaging problem: connecting more fibers to a PIC without allowing active alignment, accumulated tolerance, permanent bonding, and channel-count scaling to become increasingly difficult.

Its technical proposition combines:

• wafer-based IOX glass waveguides;

• passive alignment;

• pitch conversion;

• a TMT physical-contact interface;

• detachable assembly;

• multi-element scaling.

These features create a credible alternative to conventional FAU coupling in selected high-density architectures. They do not establish that FAUs will disappear.

The wider strategic opportunity lies in glass as an integration platform. Whether GlassBridge becomes a major CPO interface will depend on production yield, channel uniformity, remating stability, package compatibility, customer qualification, total cost, and the development of a broader manufacturing ecosystem.

It connects external optical fibers to a photonic integrated circuit in high-density NPO, CPO, and photonic-module designs.

A traditional FAU commonly uses precision fiber positioning and active alignment. GlassBridge uses wafer-fabricated glass waveguides, passive alignment, pitch conversion, and a detachable interface.

It can reduce or eliminate active adjustment at the final connector interface, but precision is still required throughout manufacturing and package assembly.

Corning reports demonstrated 1.5 dB O-band Fiber-to-PIC coupling. This is a published result, not a universal maximum for every configuration.

It can replace FAU-based interfaces in some designs, but FAUs remain widely relevant. The two approaches are likely to coexist.

It has published specifications and demonstration milestones, but broad customer qualification and sustained high-volume deployment have not yet been publicly confirmed.