ECTC 2026 | Intel | V-groove Based Edge Coupling Enabled by Optical Glass Coupler Attach for Co-packaged Optics
Updated: Oct 7
Intel's ECTC 2026 paper does something that sounds trivial: it replaces the "flexible fiber array" used for CPO edge coupling with a "rigid glass coupler." But this swap tackles the hardest hidden hurdle in CPO volume production: fiber alignment. A flexible fiber array relies on self-alignment and its positions drift. Intel instead uses its own low-loss glass, etching inverse V-groove posts 127 µm in diameter at a 250 µm pitch to mimic a fixed fiber bundle, so the relative channel positions are locked by the process and self-alignment is no longer needed. Combined with vision-only passive alignment, force-feedback attach, and fast UV curing, the whole flow becomes modelable, traceable, and manufacturable. The reliability data is there too: IL increase <0.7 dB after 450 thermal cycles and <0.6 dB after 500 hours of storage at 150°C. In one sentence: Intel has turned the detachable CPO connector from a craft into a production line.
1. Background: Intel brings the glass coupler into its own advanced packaging line
The paper comes from Intel's Advanced Packaging Technology Manufacturing team in Chandler, USA, and the authors include Nicholas Psaila, a leading figure in glass micro-optics (who joined Intel through its acquisition of Optoscribe). It was presented at the 76th IEEE ECTC in 2026. The combination itself is a signal: Intel is not publishing an academic paper; it is pulling the glass coupler into the "manufacturing" flow of its own advanced packaging.
Why does it matter now? Because electrical interconnect is hitting a bottleneck. As data rates climb, copper becomes short-reach, high-loss, and power-hungry, turning into the bottleneck of high-performance AI clusters. Co-Packaged Optics (CPO) breaks through by heterogeneously integrating the optical engine with silicon in the same package. But the real volume-production hurdle for CPO is how to make the edge-coupling interface between the Photonic Integrated Circuit (PIC) and the fiber low-loss, repeatable, and detachable.
The traditional approach etches V-grooves into bulk silicon and bonds a Fiber Array Unit (FAU) directly. Intel argues for using a rigid glass coupler as an intermediary instead: not only better mode matching and lower Insertion Loss (IL), but also a stable alignment process, a detachable connector, and, crucially, an attach process that can be dimensionally modeled.
Detachable connectors are a hidden gating item for CPO volume production. In Optical Packaging in Transition (5): The Detachable Fiber Battle we surveyed the approaches from Teramount, Senko, Intel, and ICON; this paper is the engineering deep dive into Intel's approach.

2. The core problem: the whole paper in one sentence
The problem this paper solves is: how to move CPO edge-coupling fiber alignment away from flexible fiber arrays that depend on self-alignment, toward a rigid, position-locked, manufacturable, detachable glass coupler process that passes reliability testing.
The key move is shifting the "alignment degrees of freedom" from the assembly floor into the "glass process": channel positions are defined when the glass is etched, so assembly only needs to align the whole glass piece to the PIC, with no per-channel self-alignment.
3. Key figures, one by one
3.1 This figure shows how the glass coupler replaces the flexible fiber array

This figure (Fig. 1) is the core structure. Intel builds the coupler from an in-house glass with low optical loss, high mechanical stability, and a coefficient of thermal expansion (CTE) matched to silicon; single-pass waveguide propagation loss is about 0.2 dB/cm, comparable to polymer waveguides. Cylindrical posts 127 µm in diameter, forming inverse V-grooves at a 250 µm pitch, are etched into the glass to mimic a bundled fiber array.
The key point: unlike a flexible fiber array, the relative positions of these channels are fixed and defined by the glass process. With that degree of freedom removed, per-channel self-alignment is no longer needed, which is exactly what makes the process stable and manufacturable. The PIC side has 24 channels at a 250 µm pitch, paired with a metamaterial Spot Size Converter (SSC).
3.2 This figure shows how precise the alignment gets: IL repeatability <0.1 dB
This figure (Fig. 7) is a half-day repeatability test of optical IL measurement across 15 multi-channel units. Intel uses a non-contact, non-invasive optical probing system with a 1310 nm source and active alignment at nanometer resolution, selecting channels to measure IL without reconnecting fibers. The result: even with environmental variation, the IL repeatability standard deviation is <0.1 dB.
What this number means is that the measurement itself is stable enough to serve as feedback on process health. Without stable measurement, all subsequent process optimization is just blind tuning.

3.3 The attach-force dilemma: too much force warps the glass and edge channels lose 2-3 dB

This figure (Fig. 9) plots loopback IL against attach force. Attach force is critical to yield, but it cuts both ways: too little force cannot overcome friction, the inverse V-grooves fail to seat in the V-grooves, and self-alignment fails; too much force warps the thin glass, causing misalignment and optical degradation, and even the bond head itself deforms under high force. In measurements under high force (more than 20x the low-force condition), loopback channels near the edge showed 2-3 dB higher IL.
3.4 The choice of mechanical epoxy makes or breaks it


These figures (Fig. 10, Fig. 11) cover the Mechanical Epoxy (ME). The ME acts as a dam that keeps the optical epoxy (OE) from flowing under the glass. But ME cure shrinkage induces warpage: switching to a low-shrinkage material cut glass warpage in half. Worse, the ME must also survive subsequent thermal processes. Intel co-optimized the ME material with the downstream flip-chip bonding process, reducing ME delamination and in-line IL degradation by more than 0.5 dB.
4. Technical highlights: two points worth remembering
The first highlight is moving alignment degrees of freedom from the assembly floor into the glass process. Channels in a flexible fiber array each drift and must converge through self-alignment; Intel's fixed post array etched into glass locks relative channel positions at the process level, leaving assembly with a single action: align the whole piece to the PIC. This makes attach a vision-only passive alignment step that can be controlled by force feedback, modeled dimensionally, and traced end to end.
The second highlight is the complete reliability data. After 450 cycles of Thermal Cycling G (-40°C to 125°C), IL increased <0.7 dB/dir on all channels; after 500 hours of high-temperature storage at 150°C, <0.6 dB; after 275 hours of uHAST (110°C, 85% RH), most units stayed <1 dB per channel. For a connector that must go into data centers and be qualified by customers, these numbers are the ticket to entry.
5. Industry link: how far from volume production, and who benefits?
Distance. This paper's maturity leans toward "manufacturing validation": there are both unit-level and wafer-level processes, a production-oriented vision passive alignment tool, repeatability across 15 units, and three types of reliability stress tests. Intel itself says the learnings will extend to its next-generation "expanded-beam vertical coupling" solution. The V-groove glass coupler is Intel's present tense; expanded-beam vertical coupling is the next step.
Beneficiaries. The most direct is Intel's own CPO/advanced packaging platform: bringing the glass coupler into the line means owning in-house capability for the detachable connector. Second is the glass materials and micro-optics ecosystem: once rigid glass couplers are proven manufacturable and reliable, glass moves from a supporting role to a leading role at the CPO interface, in the same broad direction as Corning using glass for optical waveguide substrates (for glass's electrical value in advanced packaging, see Why the TGV Race Comes Together in 2026).
A note of caution: rigid glass buys stability at the cost of warpage sensitivity, a narrow attach-force window, and ME material choices that ripple through everything. And Intel already positions it as a "transition": the next generation moves to expanded-beam vertical coupling. So the value of this solution is more about making the current generation of detachable connectors manufacturable than about being the end-state architecture.
6. Summary
The one sentence to remember from this paper: Intel has moved alignment for detachable CPO connectors from assembly-floor craft into the dimensional control of the glass process. The rigidity of the glass coupler makes alignment modelable, traceable, and manufacturable, and with the reliability data now in place, this is a pragmatic engineering step in pushing a lab coupling solution toward the production line.
For those tracking the CPO supply chain, the watch point is clear: to judge whether a coupling solution can reach volume, don't just look at the IL number; check whether alignment is "process-locked" rather than "self-aligned on site," and whether it has passed thermal cycling and uHAST. Detachable connectors are one of the last few bottlenecks before CPO ramps.
References
Zhou Yang et al., "V-groove Based Edge Coupling Enabled by Optical Glass Coupler Attach for Co-packaged Optics," 2026 IEEE 76th ECTC, pp. 1079–1082. Intel Corporation, Chandler, USA.
Further reading: N. Psaila et al., JLT 41(19) 2023; JLT 42(15) 2024.
Related reading
Optical Packaging in Transition (5): The Detachable Fiber Battle: the broader context for Intel's approach
Glass Substrates Are No Longer Slideware: The TGV Race: another value line for glass in advanced packaging
Polymer Waveguide Holds +20 dBm for Six Hours: the polymer approach to the same coupling puzzle



Comments