ECOC 2025 Tech Focus: Marvell on Coherent Lite Technology and Market Outlook

Introduction

As data centers enter a new AI-driven era, traditional IMDD (Intensity Modulation Direct Detection) is increasingly hitting speed and reach bottlenecks, while full coherent technology, though powerful, is too costly and power-hungry.

Sitting between the two, Coherent Lite has emerged to fill the application gap between 2-20 km with a low-cost, low-power, low-latency design. At ECOC 2025, Marvell presented its view of Coherent Lite, covering the technology fundamentals, use cases, market forecast and standardization progress.

Content

1. Optical Connectivity Needs Within and Between Data Centers

• Intra-DC

  • Dominated by IMDD, with more than 60 million optical modules shipped in 2024.

  • Key requirements: low cost, low power, low latency; typical reach <10 km.

  • Representative technology: 800G PAM4.

• Inter-DC

  • Dominated by full coherent, with only about 1 million ports shipped in 2024.

  • Key requirements: high spectral efficiency, long reach (100 km+).

  • Representative technologies: 400ZR → 800ZR → 1.6ZR.

👉 The problem: as speeds push to 1.6T / 3.2T, IMDD's effective reach shrinks sharply (800G can reach 10 km, but 1.6T only ~2.5 km), while coherent solutions are too costly and power-hungry - leaving a technology gap in between.

2. Where Coherent Lite Fits

• Core value: in short-reach (<20 km) scenarios, replace IMDD with simplified coherent technology, combining cost efficiency and scalability.

• Four technology pillars:

  1. Low-power DSP

     • Requires only minimal chromatic dispersion compensation and a simplified equalizer.

     • Relaxed requirements for polarization tracking (SOP tracking).

  2. Distortion and noise compensation

     • Includes Tx/Rx IQ impairments mitigation.

  3. Low-latency FEC

     • Uses a simplified outer code (e.g., KP4) plus a lightweight inner code (BCH/Hamming).

  4. Low-cost module design

     • Uses O-band lasers, relaxing frequency accuracy requirements.

     • Enables a more flexible optoelectronic supply chain.

3. Real-World Example: Marvell 1.6T Coherent Lite DSP

• Architecture: supports 800G-1.6T and can pair with O-band silicon photonics PICs.

• Features:

  • Integrated drivers, lowering BOM cost.

  • Low power (< 5-10% of GPU power).

  • Supports 2-40 km transmission.

• Demo: Marvell showed a live demo at ECOC 2025, highlighting its latency and power advantages.

4. Key Use Cases

1. AI campus networks (10-20 km)

   • Connecting multiple AI data halls.

   • Supports a scalable 6.4T → 25T architecture.

   • Expected to be a high-volume use case.

2. OCS (optical circuit switching) architectures

   • Requires 2-10 km reach with a 6-8 dB link budget.

   • Strong potential when paired with emerging AI fabric technologies.

3. FR (2 km class)

   • The most power- and cost-sensitive scenario.

   • Beyond 1.6T, Coherent Lite is expected to replace IMDD.

5. Market Forecast

• 800G Coherent Lite: currently focused on LR (10-20 km) applications.

• 1.6T Coherent Lite: gradual rollout in 2027-2028, with shipments potentially surpassing IMDD LR.

• 3.2T and beyond: becomes one of the main technologies for high-capacity AI data centers, with the market expected to scale rapidly after 2028.

6. Standardization Progress

• OIF: driving standardization of 1.6T Coherent Lite (1600CL), in parallel with ZR/ZR+.

• Technical issues:

  • Symbol rate premium: is a higher symbol rate than IMDD needed?

  • ETCC (Extended Transmitter Constellation Closure): a new TDECQ-like metric under discussion at IEEE/ITU-T/OIF.

• Goal: establish an IMDD-compatible "reference receiver" and test methods to lower the industry's adoption barrier.

Summary

Marvell's view at ECOC 2025 can be summarized as:

1. IMDD is running out of road: higher speeds shrink effective reach, making it hard to support 1.6T / 3.2T.

2. Coherent Lite fills the gap: low-power DSP and simplified coherent technology support 2-20 km short-reach links.

3. Broad applications: campus networks, OCS architectures and FR applications all show potential.

4. Huge market potential: after 2028 it may replace IMDD in short-reach markets and become the mainstream for next-generation optical modules.

5. Standardization underway: OIF has launched the 1600CL project to build an industry ecosystem for Coherent Lite.

Overall, Coherent Lite is not a replacement for IMDD or full coherent, but fills the "gap" between them. As AI data centers expand, it will be one of the most noteworthy emerging technologies of the next 3-5 years.