Key Technology Trends at ECOC 2025
I had the good fortune of attending the 51st European Conference on Optical Communications, ECOC 2025, which was held in Copenhagen from September 28th through October 2nd. The conference included 500+ presentations and over 2,000 conference delegates from 70+ countries. Nokia contributed with 2 workshops, over 30 papers/presentations, 4 posters and a 100-years of Bell Labs event.
Running alongside the conference from September 30th through October 1st was the ECOC exhibition, with around 6,000 visitors and 350 exhibitors. The Nokia booth hosted demonstrations of our latest innovations including the Nokia 1830 GX hyperscale OLS, thin transponders with ICE-X 800ZR/ZR+ pluggables, coherent routing with C+L ICE-X 800G pluggables in Nokia and third-party routers, ICE-X 400G coherent breakout (4x100G), ICE-D intra-datacenter technology, optical fiber sensing and anomaly detection, optical spectrum as a service, and closed-loop energy optimization.
Having attended all five days of the technical conference, several Market Focus and Product Focus sessions in the exhibition hall, and having also spent time observing what was new in the exhibitor booths, these are my key takeaways on the major optical technology trends:
Higher Baud Rates
Several conference sessions looked at how far we could go with baud rates and more importantly, when, including looking at the key challenges and potential stumbling blocks. With 250+ GBaud pluggables (1600ZR/ZR+) under development and expected to come to market around 2027/2028, the question becomes can we double the baud rate again to 500 GBaud for single wavelength 3.2T pluggables? And if so, by when? A key challenge here is the ADC/DAC inside the digital ASIC/DSP, which is limited by the cutoff frequency (fT) that can be supported by the CMOS process node. To hit 500 GBaud the fT would need to be around 1,000GHz. CMOS technology that can support this was forecast to be available around 2035.
One option to address this limit before then is ADC/DAC (de-)multiplexing, which was a topic in the presentation of Di Che from Nokia Bell Labs. This ADC/DAC (de-)multiplexing typically leverages alternative higher frequency semiconductors, such as indium phosphide or silicon germanium, on a separate chip outside the silicon digital ASIC/DSP. These approaches have an impact in terms of cost and power consumption making them unlikely to be used in pluggables. 3.2T coherent pluggables are therefore likely to be dual wavelength at least at first, before later transitioning to single wavelength when the CMOS frequency is able to support it. And what about 1,000GBaud? Sometime after 2040, probably leveraging novel modulators (such as plasmonic ones) and ADC/DAC (de-)multiplexing.
Modulator Technology
Given the modulator bandwidth limitations of silicon, indium phosphide is the only modulator material with proven volume production and reliability that can support next generation coherent at 250+ GBaud. That said, while challenges remain, thin film lithium niobate (TFLN) is maturing as both a technology and as an ecosystem with new TFLN foundries appearing at the show, providing another potential option for high baud rate modulators. Haïk Mardoyan from Nokia Bell Labs talked about several of their experiments and tests on TFLN. I also heard about thin film lithium tantalate (TFLT), whose proponents claim has several advantages relative to TFLN, including the ability to support higher power. Another modulator trend was silicon photonics micro ring modulators. These provide an alternative to the Mach-Zehnder modulators used in coherent optical engines, with advantages including much smaller size and lower power consumption, though with lower bandwidth. The big news at the show was that Nvidia are using micro-ring modulators in their recently announced CPO products. Plasmonic modulators also came up but are still primarily in the realm of universities and start-ups.
Coherent Pluggables
Coherent pluggables were another hot topic. Nokia’s Rob Shore, Fady Masoud and Julia Larikova gave Market Focus and Product Focus presentations, with Rob and Fady focusing on 800ZR/ZR+ applications including thin transponders, while Julia focused on the components and packaging used to build the best-in class ICE-X 800G pluggables. Nokia and others also showed 800ZR/ZR+ multi-vendor interoperability in the OIF demonstration area.
A killer application for 800ZR/ZR+ would appear to be scaling AI training across multiple metro datacenters, an application some are now referring to as “scale across”. Jeff Rahn from Meta gave a very interesting presentation on how and why Meta is deploying 800ZR/ZR+ pluggables in their network, with the pluggables deployed directly in the routers, the need for interoperability and backwards compatibility, and a 600Gb/s mode running on a 1,200km route from Portland to Sacramento.
1600ZR, 1600ZR+ and 1600CL were also covered in multiple presentations with the industry making good progress towards availability in the 2027/2028 timeframe. As discussed previously the availability of 3.2T pluggables and whether these would be one wavelength or two was also discussed, alongside the need for liquid cooling to cope with their high total power consumption. And finally, whilst not coherent, it was interesting to hear about a quantum key distribution (QKD) QSFP28 pluggable.
Multi-rail
With coherent optical engine technology now close to the Shannon limit, one approach to scaling capacity that is gaining traction is multi-rail. That is running multiple parallel fiber pairs between nodes. This creates significant challenges at the ILA huts that are used to amplify the wavelengths between terminal/add-drop/ROADM nodes. These huts are typically limited to 3KW per rack and maybe 4 racks. Providing the ILAs for large numbers of fiber pairs is driving innovation at both systems vendors such as Nokia and the vendors that make components such as amplifiers, pump lasers, optical channel monitors (OCM), dynamic gain equalizers (DGE) and optical time-domain reflectometer (OTDR). Lasers are getting smaller and more power-efficient with multi-emitter packages and uncooled lasers. The OCM, DGE and OTDR are reducing the cost, space and power by sharing these components across a larger number of rails. This is driving the number of ILAs per rack unit from around 1 today to 2, 4, 8 or even 16.
Hollow Core Fiber (HCF)
The loss of HCF keeps coming down, with around 0.05dB/km presented by YOFC and Linfiber. Microsoft also presented a post deadline paper showing 11,439.2km (25.6Tb/s) and 2878.4km (20.6Tb/s) transmission through a 3x120km HCF loop. However, my key observation was the maturing of the HCF ecosystem. In addition to Microsoft, YOFC, with whom Nokia collaborated on two HCF papers at the conference, and OFS, with whom Nokia participated in a 2024 HCF trial, Coherent surprised a lot of people with a spool of HCF on their stand with a sign stating sampling in 2026. I learnt that Microsoft now has partnerships with Corning and Heraeus to scale HCF production. Amazon Web Services (AWS) also presented that they are planning to deploy HCF.
Multi-Core Fiber (MCF)
While HCF had been getting most of the attention, with MCF traction previously limited to submarine applications with dual core fiber, demand is heating up for four-core MCF inside the datacenter in order to reduce the size, footprint and weight of cables. A major fiber manufacturer is rumored to be entering this market, and several intra-datacenter pluggable vendors are making pluggables with integrated MCF connectors.
Summary
Overall, ECOC 2025 was a great event that also covered a wide range of additional interesting topics including co-packaged optics, 400G per lane, coherent lite, sensing including longitudinal power monitoring and OFDR-based subsea earthquake detection, automation, quantum and, of course, AI.
