This article was co-written by Chris Lyon, President of the Ethernet Alliance and Cloud Business Development Manager at Amphenol, and Michael Klempa, Ethernet Alliance member and Product Marketing Specialist at Alphawave Semi.
"You won't lag behind, because you'll have the speed.
You'll pass the whole gang and you'll soon take the lead.
Wherever you fly, you'll be best of the best.
Wherever you go, you'll top all the rest."
To paraphrase the witty brilliance of the late Dr. Seuss, Ethernet is off to Great Places (again). During its last five decades, Ethernet has become a fixture in all aspects of our daily lives. When viewed through the lens of its 50 years of unmitigated success, Ethernet’s future is seemingly boundless.
With the conclusion of 2023, so too came the end of Ethernet’s 50th Anniversary celebrations. During the past fifty years, Ethernet grew from an enigmatic idea sketched on the back of a napkin to a ubiquitous technology staple so pervasive that you can’t swing an Ethernet cable around without hitting something that relies on its infallible reliability, interoperability, and speed.
With an initial blazing speed of 2.94 Mbps in 1982, to the forthcoming 200Gb per lane specification, Ethernet has, is, and always will be racing towards the future. The specification for 200Gb per lane is progressing in the IEEE P802.3dj Task Force, and there are already pre-spec products under development, all in anticipation of current and emerging hyperscale compute demands. Artificial intelligence (AI) and machine learning (ML) deployments have exploded and there is seemingly no limit to the requirements for high-speed interconnections.
Like the nimble architects that they are, IEEE and a myriad of Multi-Source Agreement (MSA) groups are doing their best to keep up with this rising demand. They’re rapidly advancing many new Ethernet signaling and communication specifications over multiple data rates and application spaces that will ensure burgeoning deployments can take advantage of Ethernet’s hallmark “you plug it in and it works” ethos.
Looking down the road to the next fifty years and beyond, here are a few trends that will carry Ethernet to those Great Places and new heights in performance.
#1: Ethernet shifts towards I/O chiplets enabled by UCIe for copper and optical interconnectivity
2023 was an inflection point for chiplets, with multiple big-name chip announcements addressing AI compute solutions based on heterogeneous chip packages. In 2024, we will likely see more ground-shaking announcements, complemented by other players who were historically boxed out of these advanced interconnect strategies, with specific solutions facilitated by the democratization of chiplets via Universal Chiplet Interconnect Express (UCIe).
Like Ethernet, UCIe focuses on interoperability and holds the key to enabling greater flexibility and efficiency in connecting chiplets within a system. This standardization facilitates the integration of different functionalities, including Ethernet connectivity, onto separate chiplets that can seamlessly communicate. Copper and optical interconnectivity advancements within chiplets can yield higher bandwidth, lower latency, and improved power efficiency for Ethernet-enabled devices. We can expect to see modular, optical, or electrical Ethernet PHYs placed on top of the UCIe Die-to-Die link, permitting highly efficient Ethernet links within the data center to power AI and HPC interconnectivity in 2024.
#2: Emergence of tailored low latency Ethernet solutions
As cloud infrastructure changes to adapt to generative AI workloads, the Ethernet ecosystem is quickly following suit by blazing a trail that will accelerate Ethernet’s role in this growing interconnectivity space. Entwining the performance of HPC with the support and accessibility of Ethernet through faster data rates and robust performance will mean bringing low latency solutions within Ethernet’s purview, a niche previously left to other technologies.
#3: 200GbE gains momentum as standardization progresses
IEEE 802.3dj is slated to make significant progress in 2024, and we can expect the first draft of the 200GbE specification to begin circulating in early 2024. This tracks with the condensed timeline – which is 20 percent quicker than the 100GbE project – proposed at the outset in order to keep pace with the cadence of the three-year cycle of doubling new switch silicon speeds needed to meet the demands of the cloud and AI/HPC. Expect to see more 200G-capable products rolling out as the specification shakes itself out (we say “more” because 200G is already here), as companies have demonstrated. We further anticipate seeing even more solutions pushing the envelope for the next generation of interconnectivity.
#4: Industry turns its eyes beyond 200Gb per lane links within the datacenter
As more and more work goes into defining 200Gb per lane solutions – transmitters, receivers, equalization schemes, error correction schemes, connectors, channels and the management software to stitch these things together – it is increasingly apparent that whatever comes next will require a paradigm shift in how we think about Ethernet. As losses pile up inside the server just to get to the front panel connector, the above mentioned switch to chiplets and optical connectivity are both looking increasingly appealing. And just as the industry shifted from NRZ signaling to PAM4 to squeeze more bits out of a unit interval, which is now under ten picoseconds, perhaps Ethernet might once again shift to a higher-order modulation scheme to keep increasing bandwidth density inside the datacenter.
Higher bandwidths, and lower latency with faster speeds now and in the near future are just two of the industry's requirements that must be met to help it get to where it wants to go. There will be many more, but how can we know what they will be? Look to the Ethernet Alliance.
Through industry events such as tradeshows where it exhibits the latest in Ethernet technology, and Technology Exploration Forums that play host to industry leaders sharing what is and what can be, the Ethernet Alliance remains Ethernet’s best advocate and the leading industry voice of Ethernet. By listening to what the industry needs in the near future and hopes for in the long-term, and then sharing that information with those helping to shape its future, the Alliance is ensuring Ethernet’s longevity. And to once again paraphrase Dr. Seuss:
"And will Ethernet succeed?
Yes! It will, indeed!
(98 and ¾ percent guaranteed!)"
Michael Klempa is a Product Marketing Specialist for high-speed applications such as 112G and 224G CEI and Ethernet, and Gen6 and Gen7 PCIe. He is currently serving as the Optical Internetworking Forum's (OIF) Physical & Link Layer Interoperability Working Group Chair. He contributes to the 802.3 Working Group, including editing the 802.3bz specification.
Chris Lyon is the Cloud Business Development Manager for the Americas at Amphenol CS. With over 30 years’ experience in the industry, Chris has expertise in storage, networking, high speed interconnect, go-to-market strategy, team building, and professional services. He currently serves on the Ethernet Alliance board of directors as President and is also Chairman for FCIA.
