ECTC 2026 | Sumitomo Electric | Photonic-Electronic Integration on Glass Substrate with Temperature-Stabilized Vertical Optical Coupling by Resin-Encapsulated Collimation Mirror

2 days ago

5 min read

What Sumitomo Electric and FICT tackle at ECTC 2026 is the thing optical coupling fears most: alignment sensitivity. A typical PIC output has a mode field diameter (MFD) of only 3 µm, so a 3 µm misalignment causes catastrophic loss. They use collimation mirrors (off-axis parabolic mirrors) to expand the beam into a 32 µm collimated beam, cutting the loss from a 3 µm misalignment to just 0.2 dB. The mirrors are 3D-printed into trenches on both the PIC and the glass waveguide substrate; the PIC is flip-chip (FC) bonded onto the glass, then protected with index-matched (n=1.46) transparent resin. The measured coupling efficiency of the FC module is -2.30 dB. More importantly, the beam-expanding structure keeps it stable at high temperature: under structural deformation from 25 to 80°C, the maximum displacement at any point is only 0.34 µm. In one sentence: expanding the beam tames both coupling killers - "alignment" and "temperature" - at once.

1. Background: Coupling PICs on Glass Waveguide Substrates

This work is a collaboration between Sumitomo Electric (Kanagawa, Japan) and FICT, presented at the 2026 IEEE 76th ECTC. To raise optical signal integration density, optical waveguide substrates capable of fan-in/out are highly effective, and the technology for connecting PICs to them is key. Sumitomo chose glass as the waveguide substrate - low CTE, flat, compatible with both electrical and optical functions, and with little warpage under temperature changes - making it suitable for large-scale multi-chip integration.

For the value of glass as a substrate, see Glass substrates are no longer slideware: the TGV race; and for why keeping fiber connectors detachable at the edge is a CPO volume-production hurdle, see The detachable fiber battle.

Schematic and cross-section of photonic-electronic integrated packaging of a PIC on a glass waveguide substrate, using vertical optical coupling via collimation mirrors. | Source: Sumitomo Electric, ECTC 2026 - Fig. 1

2. The Core Problem: The Whole Paper in One Sentence

The problem this paper solves: how to make optical coupling tolerant of misalignment, stable at high temperature, and protectable with resin encapsulation when a PIC is flip-chip bonded onto a glass waveguide substrate.

The answer: expand the beam with collimation mirrors (beam expanding), suppressing both alignment and temperature sensitivity together.

3. Walking Through the Key Figures

3.1 This figure shows that "expanding the beam tolerates misalignment"

These figures (Fig. 2, Fig. 5) cover the collimation mirror design.

A typical PIC has an MFD of only 3 µm, so a 3 µm misalignment causes catastrophic loss. Using off-axis parabolic collimation mirrors, Sumitomo expands the output beam to a 32 µm MFD, targeting just 0.2 dB of added loss for a 3 µm misalignment. The mirrors are formed in trenches on both the PIC and glass substrate with a metal reflective coating, and each mirror's curvature can compensate for the MFD mismatch between the PIC and glass waveguides. Wavelength dependence is low, supporting about 100 nm of bandwidth (suitable for CWDM). Simulated coupling non-uniformity across the O-band (1260-1360 nm) is <0.1 dB, and including gold reflection absorption (about 2.5% per reflection), overall coupling efficiency is estimated at -1.1 dB.

3.2 This figure shows "real coupling efficiency after flip-chip assembly"

These figures (Fig. 6) show the FC module assembly. A 10 mm square glass waveguide substrate has 3D-printed collimation mirrors in its trenches, and the PIC is thermocompression bonded to the glass with Au stud bumps. Measured mirror misalignment was 1.6 µm in the waveguide direction and 2.3 µm in the metal trace direction, within the FC bonder's ±3 µm placement accuracy. After filling with index-matched (n=1.46) transparent resin and vacuum degassing, coupling efficiency at 1310 nm is -2.30 dB. Achieving this coupling despite the unavoidable misalignment of FC bonding proves that the tolerance advantage of beam expanding really works.

SEM of a PIC flip-chip bonded onto the glass waveguide substrate, with collimation mirrors facing each other on the PIC and glass sides; measured coupling at 1310 nm is -2.30 dB. | Source: Sumitomo Electric, ECTC 2026 - Fig. 6

3.3 This figure shows "the structure stays stable at high temperature"

These figures (Fig. 7, Table I) present the high-temperature structural analysis. Transparent resin usually has a larger CTE than underfill (UF) (since filler is hard to add given transparency requirements), making it temperature sensitive. But the beam-expanding structure makes coupling insensitive to tiny displacements: under deformation from 25 to 80°C, the maximum displacement at any mirror/waveguide point is only 0.34 µm, with a maximum angular displacement of 0.05° - far smaller than what the 32 µm expanded beam can tolerate. This is evidence that beam expanding also solves "temperature stability."

4. Technical Highlights

The first highlight is expanding the beam from 3 µm to 32 µm with collimation mirrors, solving two coupling killers in one move: alignment (a 3 µm error costs only 0.2 dB) and temperature (displacement <0.34 µm from 25 to 80°C has no impact). Beam expansion greatly relaxes the coupling's sensitivity to both "position" and "deformation."

The second highlight is that the whole scheme is compatible with existing flip-chip processes: mirrors are 3D-printed in trenches, the PIC uses standard FC thermocompression bonding, and transparent resin encapsulation follows the regular FC flow. The paper is also candid about what remains: the electrical region needs UF while the optical region needs transparent resin, and partitioning the two resins (or developing a transparent UF) is a hurdle for practical deployment.

5. Industry Relevance: How Far from Volume Production? Who Benefits?

Distance: this is at the "FC module proof-of-concept" stage - -2.30 dB coupling and high-temperature structural stability are both verified, but UF/transparent resin partitioning and volume yield still have a way to go. Sumitomo's prior work cited here (3D-printed beam-expanding lenses, ECTC 2024) shows this is a route under continuous development.

Beneficiaries: most directly, the chain of glass waveguide substrates + 3D-printed micro-optics + flip-chip packaging (Sumitomo Electric is a major fiber and optical component maker); next, CPO/NPO packaging that wants to use "PIC flip-chipped onto glass + beam-expanded vertical coupling." This approach, KYOCERA's curved-mirror coupling and Intel's glass coupler represent different schools of "reducing coupling alignment sensitivity" - beam expansion vs. curvature compensation vs. process lock-in, each with its own strengths.

6. Conclusion

The one line to remember from this paper: expanding the beam tames both optical-coupling killers - "alignment" and "temperature" - at once. Sumitomo expands the beam to 32 µm with collimation mirrors, so a 3 µm misalignment costs only 0.2 dB and displacement from 25 to 80°C stays below 0.34 µm. For those tracking CPO packaging, the point to watch: the competitiveness of vertical optical coupling schemes lies in "tolerance after beam expansion × high-temperature structural stability × compatibility with existing FC processes," not coupling loss alone.

References

Shingo Nakamura, Yasutaka Mizuno, Kunio Kobayashi, et al. (Sumitomo Electric), Mio Emura, et al. (FICT), Yoichiro Kurita (Institute of Science Tokyo), "Photonic-Electronic Integration on Glass Substrate with Temperature-Stabilized Vertical Optical Coupling by Resin-Encapsulated Collimation Mirror," 2026 IEEE 76th ECTC.
Sumitomo Electric Industries, Ltd., Kanagawa, Japan; FICT Limited, Nagano, Japan