At the end of the AMD to Acquire Xilinx Continuing Consolidation article this week, I noted that Marvell needed a dance partner in the discussion. A passage in that article read “The big question is if I drive down 237 to Marvell’s campus today whether they have a big “Open house” or “For sale” or “Buying semis” sign off the freeway.” The next day we heard a leak about a deal that was confirmed shortly thereafter that Marvell reached a deal to purchase Inphi. This deal fits in the broader narrative of chip deals, and I wanted to take a moment to discuss why.
Talking AMD-Xilinx and Marvell-Inphi
We also have a video discussing both the AMD-Xilinx and Marvell-Inphi deals that were announced this week, and why they have a key theme.
Our suggestion, since that is a lot of discussions rather than showing products, is to open the video on YouTube and listen along.
Marvell and Inphi Background
Marvell for its part makes a number of connectivity options. These range from PHYs, to NICs that go anywhere from desktops to servers and specialized communication equipment. Marvell also has a line of switches for the data center and edge.
Marvell also makes processors that sit in 5G infrastructure that you can learn more about in our New Marvell OCTEON TX2 and Fusion CNF95xx 5G SoCs article.
There are other parts of the business at Marvell, but a key theme in the company’s direction is connectivity. While companies like Intel are focused on building the best processors, Marvell is working on next-gen disaggregated data center connectivity even in seemingly unrelated parts of their business. As a good example, Marvell and Kioxia Ethernet SSD solutions are now sampling.
While Marvell has some traditional controller businesses and other areas, we are increasingly seeing the company focus on connectivity. That brings us to the Marvell-Inphi deal.
Inphi generally makes products designed for the optical networking space with products such as DSPs, transceivers, and drivers.
While the company may not be a household name, it is better known in the infrastructure markets, specifically on the networking side.
In short, Inphi makes a lot of the products you need a stable of PhDs to design and yet are often taken for granted outside of the networking space.
In the network switching space, as well as some of the 5G edge deployment areas, there is an enormous demand for next-gen and higher-speed data fabrics. As speeds double, power consumption also rises which creates practical packaging challenges. Earlier this year, we did a piece just before lockdown (2 days prior in the Silicon Valley) on Hands-on with the Intel Co-Packaged Optics and Silicon Photonics Switch. In that video we go into the challenges of next-gen networks, and what Intel is doing with their Barefoot acquisition as well as their silicon photonics business.
Marvell has switch chips and many parts of the infrastructure market. As the company is looking ahead to the future demand, it needed a story around optical transport. Inphi effectively gives Marvell this story through a well-regarded company that has many of the components Marvell will need to integrate into next-generation assemblies.
The key markets are telecom and infrastructure as well as cloud providers. Inside a cloud data center, optical runs can easily reach 1-2km before even getting outside the walls of the facility. As speeds rise, even the shorter in-rack runs will become challenging if not practically impossible to service with copper DACs. The industry knows we are moving to more optics.
Going beyond the obvious network connectivity, there is another layer to the discussion. We know that current PCB signaling hit a major hurdle with the PCIe Gen4 era. We showed off the future of servers as requiring many cables to do what PCB runs previously did simply due to better signal integrity.
Like the network side, companies are in research phases now of what future chip-to-chip connectivity looks like. This has been promised for years, but the next speed jumps are going to turn optical connectivity from a “nice to have” to a requirement in future generations. I asked Intel’s chief architect about this at Intel Architecture Day 2020 and he told me, not commenting on specific products, we would need to see this start to be implemented by 2025.
While long haul networking is always exciting, optics are moving into even shorter runs within racks, but also even to within servers. Moving to optical engines within chips means we can get connectivity for disaggregated computing in the CXL 2.0 era or just beyond.
While at first the AMD-Xilinx and Marvell-Inphi deals may seem completely different, they share at least three key themes.
- They are focused on scaling semiconductor companies
- Target markets are data center and 5G deployments
- They have strong connectivity stories
While some may question why Inphi versus other options, at some point there are only so many companies out there that Marvell could purchase and Inphi adds what is mostly a net-new product portfolio where the Inphi solutions are often designed to operate next to existing and future Marvell products.
As Marvell transforms its business to making customized solutions for key customers in the data center and 5G edge markets, there is one fact it can be sure of: all of these customers will use optical networking, and all of them can potentially use Inphi products as well. Looking further ahead, Marvell needed an optics story for their chip business so this move makes sense.