Avatar

Service providers, web scale companies, and equipment manufacturers have long struggled with networks comprised of multiple unique architectures driving up huge operational complexity and cost. Power-hungry routers, feature-limited web scale switches, and an ever-expanding footprint have them seeking new alternatives. Unfortunately, it seems the only solution is to wait for another chip family release once every 24 months.

Well, we have some good news. Those days are over. In December of 2019, we made a bold announcement about how we’ll forever change the economics of the Internet. Two of the key elements in that announcement were the introduction of Cisco Silicon One and a new business model. For the first time in the industry, customers can buy our state-of-the-art silicon and discrete optics parts directly from us.

The 10Tbps Cisco Silicon One Q100 is the first device in a family of chips that mark a fundamental shift in the routing silicon market. Whereas it once took at least five pieces of routing silicon to build a complete system, this can now be accomplished with just one. More than ever before, customers have the freedom to innovate at their own pace. We designed the new Cisco Silicon One architecture to be endlessly scalable, enabling faster iterations of future generations. Naturally there were skeptics – after all, this defied the norm.

Today, less than 10 months later, we’re proud to announce six new 7nm additions to the Cisco Silicon One portfolio: three highly efficient web scale switching devices, and three high-scale and deep-buffered routing devices covering an expanded performance range between 3.2Tbps and 12.8Tbps. The Cisco Silicon One portfolio now covers the entire space from service provider and web scale routers all the way to the Top of Rack (TOR) switches and everything in between. With one architecture, one Software Development Kit (SDK), and one P4 forwarding code, Cisco Silicon One unifies the network across routing and web scale switching.

Cisco Silicon One

The Cisco Silicon One Q200 extends our market leadership in routing silicon with a 7nm high performance, large scale, run-to-completion P4 engine and deep buffers. At 12.8Tbps, it enables customers to build a 32x400GE 1RU router with a single piece of silicon while others on the market require between two and 10 pieces of silicon. This dramatic difference alters the fundamental economics of building fixed-box routers.

Multiple Q200 devices can then be interconnected to create massive multi-petabit systems with a fully scheduled Ingress Virtual Output Queueing (VOQ) fabric. The Q201 and Q202 extend the portfolio down to enable lower bandwidth and less expensive 64x100GE and 32x100GE 1RU systems. Cisco Silicon One now covers the entire routing market, from 3.2Tbps to multiple petabits.

Today we also expand the market addressability of Cisco Silicon One in a big way by adding purpose-built devices for the web scale switching market. The Q200L is a 7nm 12.8Tbps web scale switching device featuring high performance, run-to-completion P4 engines and a fully unified on-die buffer. The Q200L is also foot-print compatible with the Q200, which enables one hardware design to support a 32x400GE router with the Q200 or a 32x400GE web scale switch with the Q200L, simplifying the development costs for our customers.

Although the leading switching devices are at rates of 12.8Tbps (and soon 25.6Tbps), we found that many customers are searching for highly efficient 3.2Tbps and 6.4Tbps to play different roles in their network. Because Cisco Silicon One is built with a scalable, slice-based architecture, we can easily create fully optimized, lower-speed devices from the same architecture. The Q201L at 6.4Tbps enables a highly efficient 64x100GE web scale switch and Q202L at 3.2Tbps can be used to build a 32x100GE web scale switch.

The efficiency of the Cisco Silicon One architecture combined with 7nm creates a truly unique offering in the web scale switching market.

Cisco Silicon One erases the hard-architectural boundaries that have existed for decades in the industry between routing and web scale switching devices by providing a single architecture that optimally spans both market segments. Customers can port one SDK and address multiple markets, while network operators need only train their support staff once to troubleshoot one architecture across their entire network. But in the end, this unique level of convergence wouldn’t be so interesting if each of the devices weren’t best of breed themselves. What sets Cisco Silicon One apart from the competition is how it brings convergence without compromise.

Cisco Silicon One was created by the team that built the silicon carrying more than 90 percent of the world’s Internet traffic and we believe this clean sheet architecture will serve the industry for the next several decades. To do this, we had to solve many of the problems that other architectures stumble over, but just as important, we’ve created an architecture that is highly efficient and scalable.

The impacts of a unifying architecture are clearly visible today. Most silicon vendors announce new silicon once every few years, whereas today we’re announcing six new devices in less than a year, just as we promised back in December 2019.

But as exciting as these new devices are, this is just the beginning. We’ve started a journey and we have more devices on the way. Stay tuned because there will be exciting news ahead!

Learn more about Cisco Silicon One architecture, devices, and benefits.



Authors

Eyal Dagan

Senior Vice President

Common Hardware Group