Gazettabyte is asking industry figures for their thoughts after attending the recent 50th-anniversary ECOC show in Frankfurt. Here are contributions from Nubis Communications' Dan Harding, imec's Peter Ossieur, and Chris Cole.
Dan Harding, CEO, Nubis Communications
Our biggest takeaway from ECOC is the increased confidence not just in 200-gigabit electrical and optical interfaces but also in 400 gigabit. It is becoming clear that in 2025 and 2026, the industry will broadly launch platforms using a 200 gigabit per lane serdes [serialiser/deserialiser interfaces] that will connect to 200 gigabit per lane optics.
At ECOC, we were shown demonstrations of 400-gigabit serdes. We had several discussions with industry leaders who expressed confidence that serdes can scale to 400 gigabit per lane and that the industry will need optics to support this in the next few years. Different optics approaches were shown at ECOC, but the main takeaway was that serdes speeds continue to advance, and optics needs to figure out a way to keep up.
Our second takeaway is that advancements in materials have significantly reduced loss across printed circuit boards (PCBs), so linear pluggable optics (LPO) at 200 gigabit looks increasingly feasible with vertical line cards and even with traditional ones.
Generally, the 200 gigabit per lane generation will be more similar to the 100-gigabit generation than we thought a year ago. That said, the transition to each new service speed is becoming fuzzy, such that 100 gigabits per lane will have years of overlap with 200 gigabits per lane. The data centre operators and system vendors remain committed to copper for short-reach links, even at 200 gigabit per lane. However, there will be more "active" copper links, so the mix between passive and active copper will shift to more active at 200 gigabit per lane.
As a supplier of optics for AI/ machine-learning networks, the third big takeaway for us is that the speed at which new architectures are being deployed puts an extreme focus on delivering a solution that can quickly move from first samples into volume production. That means we must constantly consider our entire development flow to support this.
Lastly, we were encouraged to see progress on a new form factor for pluggable optics to eliminate the "gold fingers." This will help the optics industry take advantage of silicon photonics and the density it can deliver. Let's see how quickly this form factor work progresses, but this is the right direction for the optics industry.
Peter Ossieur, Program Manager, High-Speed Transceivers at imec-IDLab
I noted the speed at which the industry has embraced the concept of linear pluggable optics (LPO). But I'm still unsure how linear pluggable optics will play out.
At ECOC, it was evident that linear pluggable optics are now driving rapid adoption of new materials, notably thin-film lithium niobate. This is excellent news for imec, which is putting significant effort into integrating lithium niobate on its 200mm silicon photonics platform.
As for surprises, one is that co-packaged optics continues to struggle. Another is that the industry's focus is already turning to 400 gigabit per lane.
Chris Cole, Optical Communications Advisor
The overall ECOC impression was like this year's OFC show; optics are back with a vengeance. The excitement, buzz, and optimism were infectious. Also uplifting was the focus of the technical conference and the exhibition on solving tough engineering challenges and going after new markets rather than eking out a living. This period may yet be a bubble, but it is a fun ride while it lasts.
There were two important technical trends of note. First, parallelism will become increasingly important, including more fibre and wavelengths. Second, reliability must be approached holistically, and today's data centre paradigm of swapping failed modules is inadequate to support AI/ machine-learning growth, especially for training.
Also necessary are significantly lower Failures in Time (FIT) for all parts of an optical link, along with system-level redundancy schemes. Achieving this will likely require a shift to fab process-based integration, replacing current discrete assembly methods. This transition is not just a suggestion but a crucial step towards ensuring the reliability and efficiency of optical systems.