Recent developments in single-mode polymer waveguides are reinforcing their role as a vital component in co-packaged optics (CPO)—a critical technology for ultra-high-bandwidth, low-power data center interconnects.
Why polymer waveguides are making waves in CPO
Co-packaged optics integrate photonic components directly with electronic ASICs (CPUs/GPUs), dramatically cutting power consumption and latency compared to traditional copper or pluggable optics.
Polymer waveguides, fabricated on glass-epoxy (FR-4) substrates, serve as flexible, low-cost, and efficient optical bridges between external laser sources and silicon photonic ICs.
They offer strong compatibility with existing electrical board tech, making them ideal for external laser source (ELS)-based CPO setups.
June 2025 Data Confirms Performance Targets
-
Latest IEEE Journal of Lightwave Technology reports polymer waveguides trimmed to ~11mm length (core: ~9µm × 7µm) match commercial single-mode fiber dimensions.
-
Polarization-dependent loss is consistently below 0.5dB; differential group delay is under 0.2% ps; polarization extinction ratio stays above 20dB across CWDM4 wavelengths (1271–1331nm).
-
Under continuous 6-hour exposure at +20dBm optical power, waveguides showed negligible signal degradation and minimal thermal impact (<5°C rise), confirming high robustness.
Industry moves toward scalable integration
A technical collaboration between imec and Ghent University (March 2025) validated high-density polymer–silicon coupling methods, achieving sub-2dB loss from chip-to-chip and chip-to-fiber using CMOS-compatible processes.
This “heterogeneous integration” hints at future CPO packages with dense optical I/O, streamlined assembly, and mass-production feasibility.
From bench to system: expert viewpoints
-
Dr. Satoshi Suda (AIST): “These polymer waveguides demonstrate the low-loss, high-power stability foundations needed for real-world CPO deployment.”
-
Mark Gardner (Intel) highlights how co-packaging reduces SerDes channel penalties—from ~20dB in board-level links to ~1–2dB—citing CPO’s path to <1?pJ/bit efficiency.
-
Vikas Gupta (GlobalFoundries) explains that polymer waveguides help shrink system energy costs from ~15pJ/bit (pluggable) to around 5pJ/bit—and even lower in optimized designs.
What’s next: coherent formats and board designs
A recent preprint titled “C2PO” outlines a coherent modulation strategy using offset QAM16 and microring modulators in silicon PICs, coexisting with polymer delivery lines to hit 400?Gb/s data rates at moderate power.
Polymer waveguides don’t modulate light but remain essential as sturdy optical conduits in such hybrid systems.
Market implications and use-case drivers
Trade events like OFC 2025 and GTC 2025 highlighted CPO as the inevitable path for high-speed, high-density AI and HPC systems. Industry leaders such as Jensen Huang (NVIDIA) and Broadcom demonstrated >100?Tb/s optical engine prototypes, driven by the limitations of traditional electrical interconnects.
Hurdles to clear
- Volume fabrication: Scaling direct-laser-write on FR4 to production volumes
- Reliability testing: Extended aging studies (beyond six hours)
- Standard formats: Defining interfaces for connectors, coupling, and substrates
- Field repairability: Learning from LPO advocates on CPO service models
Roadmap highlights
- March 2025: imec/Ghent Uni paper solidifies dense-waveguide under-2?dB coupling
- June 2025: IEEE reports endorse performance even at +20?dBm and high temperatures
- Mid-late 2025: Expect system trials combining coherent CPO (e.g., C2PO) and polymer–silicon photonic packages
- 2026: Initial CPO pilot deployments by large scale hyperscalers (Microsoft, Meta, Google)
Big picture
Polymer waveguides are emerging from labs into system-level relevance. Meeting key industry requirements—low optical loss, high polarization fidelity, power stability, and scalability—the technology is poised to underpin next-gen CPO platforms delivering:
- 5–10× energy reductions per bit
- Multi-terabit optical I/O at chip edge
- Integration into AI/HPC systems with footprints under thermal and latency budgets
Continued progress on manufacturing, long-duration tests, and connector standards will determine whether polymer waveguides become the backbone of mass-market optical compute packages by 2026.
