Next-Gen 25.6T Platforms Usher in 112 Gbps SerDes and 800G Connectivity

by

2020 saw many key technology announcements, including 25.6T ASICs based on 112 Gbps SerDes. The big question is, how will these next-generation products get deployed by Cloud, Telco SP, and Enterprise customers, and how will these next-generation ASICs differ from the 400 Gbps upgrade cycle.

In our forecast projections, we model 25.6T ASICs to have two form factors: 2RU 64-port 400 Gbps systems and 1RU 32-port 800 Gbps systems. In our forecasts, we believe the 1RU 32-port system will ship the most volume.

We believe 25.6T will take a slightly different path to the same 1RU 32-port systems. 400 Gbps and the 8-lane architecture per port were not only a step up in speed, but also an increase in lanes. The two previous technologies, 40 Gbps and 100 Gbps were both 4-lane architectures (4X10 and 4X25, respectively). At the same time, process geometry improvements in networking have outpaced general compute. Networking ASICs were historically at least two generations behind the most cutting edge process node when 40 Gbps and 100 Gbps systems first launched during the previous decade. When we look at 400 Gbps and upcoming 800 Gbps products, the networking ASIC is right at the bleeding edge using the same tools as the highest-end GPU or smartphone processor. By doing much of the heavy lifting in 400 Gbps as well as a rapid improvement in process geometry, it is likely that 25.6T ASICs will have the highest volume in 1RU 32-port systems.

Optics availability for 400 Gbps disappointed and delayed adoption for the industry. 800 Gbps optics seem to be in better shape in the development process at this stage; again, the heavy lifting of moving for 4-lane to 8-lane happened in the previous speed transition. The industry, with multiple optics manufacturers, is actively working on DR/FR flavor modules and came together with the 800G QSFP-DD800 MSA to agree on form factor. Multiple 400G optics modules are already at 100G/lambda. The need to move the electrical from 56 Gbps to 112 Gbps was a straightforward engineering problem that has been solved. Samples of 800 Gbps optics are here now and were showcased at ECOC in Dec 2020. Not only 800G optics, but copper DAC cable samples are also available for 800G connectivity.

System vendors, serving Cloud, Telco SP, and Enterprise customers, can design 1RU systems today and move directly to DR/FR flavors of 800G OSFP and QSFP-DD800 optics and copper cables for shorter distances. We believe many customers will opt for this option to keep with the uniformity of past architectures. We believe the market is well poised to begin shipments in 2021 and widespread adoption into 2022. The 10 Gbps->40 Gbps->100 Gbps upgrade drove massive adoption of 100 Gbps network infrastructure. Putting 800 Gbps in the same reference 100 Gbps ->400 Gbps -> 800Gbps will provide a historical and contextual reference point for the upcoming transition.

It is crucial to frame the transition by customer size. Hyperscalers continue to expand their infrastructures rapidly. Therefore, the volumes of the higher speeds continue to exceed the previous port speeds. 800 Gbps and systems based on X-lanes of 112 Gbps SerDes will set volume records in the middle portion of this decade as shown in the projections below. Hyperscalers are several times larger today than during 100 Gbps upgrade cycle and will be even larger in the future.