counter for iweb
Website
Silicon Photonics

Published book, click here

« OFC 2024 industry reflections: Final part | Main | OFC 2024 industry reflections: Part 3 »
Monday
May062024

OFC 2024 industry reflections: Part 4

Gazettabyte is asking industry figures for their thoughts after attending the recent OFC show in San Diego. This penultimate part includes the thoughts of Cisco’s Ron Horan, Coherent’s Dr. Sanjai Parthasarathi, and Adtran’s Jörg-Peter Elbers.


Ron Horan, Vice President Product Management, Client Optics Group, Cisco   

Several years ago, no one could have predicted how extensive the network infrastructure required to support artificial intelligence (AI) and machine learning (ML) back-end networks in data centres would be. This year's OFC answered that question. In a word, immense.

By 2025, the optics total addressable market for AI/ML back-end networks is expected to equal the already substantial front-end network optics market. By 2027, the back-end network optics total addressable market is projected to significantly exceed that of the front-end network. Additionally, the adoption of higher speeds and interface densities in the AI/ML back-end network will likely surpass that of the front-end.

Last year, linear pluggable optics (LPO) advocates heralded the power and cost savings associated with removing the digital signal processor (DSP) from an optics module and driving it directly from the host ASIC. Cisco and others have shown, using data and demos, that the overall power and cost savings are significant. However, in the last year, enthusiasm for this disruptive technology has been checked as concerns about link robustness and accountability have surfaced.

Enter linear receive optics (LRO), where the transmit path gets retimed while the high-power module receiver path moves to a linear receiver, which drives directly to the host ASIC. While not as power or cost friendly as linear pluggable optics, it does reduce power and some cost from the module compared to a fully retimed module while providing some diagnostic support for the link.

Only time and significant interoperability testing will determine whether linear pluggable optics or linear receive links will be robust enough to make them deployable at scale. Additionally, today's linear pluggable and linear receive solutions have only been shown at 100 gigabits per lane. It is unclear whether 200 gigabits per lane for both approaches can work. Many think not. If not, then 100 gigabit per lane linear pluggable and linear receive optics may be a one-generation technology that is never optimal. The LPO-MSA, an industry effort that included many of the industry's key companies, was announced before OFC to specify and resolve interoperability and link accountability concerns.  

The overall concern about reducing power in the data centre was a strong theme at the show. The linear pluggable optics/ linear receive optics theme was born from this concern. As optics, switches, routers, and GPU servers become faster and denser, data centres cannot support the insatiable need for more power.

However, end users and equipment manufacturers seek alternative ways to lower power, such as liquid cooling and immersion. Liquid cooling uses liquid-filled pipes to remove the heat, which can help cool the optics.  Liquid immersion further amplifies the cooling approach by immersing the optics, along with the host switch or GPU server, directly into an inert cooling fluid or placing them just above the fluid in the vapour layer. The ultimate result is to operate the optics at a lower case temperature and save power. It seems each customer is approaching this problem differently.  

Last year's OFC produced the first optics with 200 gigabit per optical lane technology. These solutions assumed a gearbox to a host interface that used 100-gigabit electrical channels. While some early adopters will use systems and optics with this configuration, a more optimal solution using 200 gigabits per lane electrical channels between the host and optics will likely be where we see 200 gigabits per lane optics hit their stride. This year's show revealed a broader set of optics at 200 gigabit per lane rates. The technology maturity was markedly improved from last year's early feasibility demos.

This is an exciting time in the optics industry. I look forward to learning what technologies will be introduced at OFC 2025.

 

Dr. Sanjai Parthasarathi, Chief Marketing Officer, Coherent

The progress in making 200-gigabit VCSELs ready for 200-gigabit PAM-4 optical transmission was a pleasant surprise of the event.

We at Coherent presented a paper on our lithographic aperture VCSEL, while Broadcom's presentation outlined the technical feasibility of 200-gigabit PAM4 links. While both mentioned that more work is needed, the historic success of VCSEL-based links in short-reach interconnects suggests that the arrival of 200G-capable VCSELs will significantly impact the datacom market.

The feasibility of linear pluggable optics has likely delayed the market acceptance of co-packaged optics. There seems to be widespread consensus that LPO can reduce cost and power while retaining all the advantages of pluggable transceivers – a vibrant ecosystem, deployment flexibility, and a clear distinction of link accountability.

 

Jörg-Peter Elbers, senior vice president, advanced technology, standards and IPR, Adtran.

At this year's OFC, discussions were much hotter than the weather. Who would have anticipated rain, winds and chilly temperatures in an always-sunny San Diego?

AI infrastructure created the most buzz at OFC. Accelerated compute clusters for generative AI are expected to drive massive demands for high-speed interconnects inside cloud-scale data centres. Consequently, 800-gigabit, 1.6-terabit, and future 3.2-terabit pluggable optical transceivers for front-end and back-end data centre fabrics stirred a lot of interest. Progress on co-packaged optics was also exciting, yet the technology will only go into deployments where and when pluggable transceivers hit unsurmountable challenges.

Silicon Photonics, indium phosphide, thin-film lithium niobate and VCSEL-based optics compete for design slots in a very competitive intra-data centre market, leading to new partnerships across the pluggable transceiver value chain. Linear receive optics and linear transmit & receive pluggable optics offer opportunities to reduce or eliminate DSP functions where electrical signal integrity permits.

While green ICT (information and communications technology) received a lot of attention at the conference, comments at the OFC Rump Session on this topic were somewhat disenchanting: time-to-market and total-cost-of-ownership drive deployment decisions at hyperscale data centres; lower energy consumption of optics is welcome but not a sufficient driver for architectural change.  

On the inter-data centre side, a range of companies announced or demonstrated 800G-ZR/ZR+ transceivers at the show. More surprising was the number of transceiver vendors - including those not traditionally active in this market domain - who have added 400G-ZR QSFP-DD transceivers to their product portfolio. This indicates that the prices of these transceivers may decline faster than anticipated.

As for the next generation, industry consensus is building up behind a single-wavelength 1.6T ZR/ZR+ ecosystem using a symbol rate of  some 240 gigabaud. There was a period in which indium phosphide and silicon photonics seemed to have taken over, and LiNbO3 appeared old-fashioned. With the move to higher symbol rates, LiNbO3 - in the form of thin-film Lithium Niobate - is celebrating a comeback: "Lithium Niobate is dead - long live Lithium Niobate!" 

The OIF's largest ever interop demo impressively showed how 400G-ZR+ modules can seamlessly interoperate over long-haul distances using an open-line system optimized for best performance and user-friendly operation. Monitoring and controlling such pluggable modules in IPoWDM scenarios can create operational and organizational challenges and is the subject of ongoing debates in IETF, TIP and OIF. A lean demarcation unit device  can be a pragmatic solution to overcome these challenges in the near term. In the access/aggregation domain, the interest in energy-efficient 100G-ZR solutions keeps growing. 

As the related OFC workshop showed, growing is also the support for a coherent single-carrier PON solution as the next step in the PON roadmap after 50Gbps very high-speed PON (VHSP).

Overall, there was excitement and momentum at OFC, with the conference and show floor returning to pre-Covid levels.

This is a good basis for the 50th anniversary edition of ECOC, taking place in Frankfurt, Germany, on September 22-26, 2024.

Reader Comments

There are no comments for this journal entry. To create a new comment, use the form below.

PostPost a New Comment

Enter your information below to add a new comment.
Author Email (optional):
Author URL (optional):
Post:
 
Some HTML allowed: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <code> <em> <i> <strike> <strong>