The word on the street in Bratislava, Slovakia, is that Intel's foundry operation has started or will start to make chips for Cisco. What is more the deal could be worth as much as $1 billion.
That's what we heard at the International Electronics Forum held in Bratislava last week and organized by market analysis firm Future Horizons. It did come couched as a "just a rumor" from one of the speakers, but in addition analysts at investment bank Piper Jaffray have reportedly told clients in a note that Cisco could cut a foundry partnership deal with Intel worth $1 billion.
Known users of Intel foundry services include FPGA companies Achronix and Tabula and network processor company Netronome.
[Get a 10% discount on ARM TechCon 2012 conference passes by using promo code EDIT. Click here to learn about the show and register.]
Most of Cisco's silicon right now is at the 40-nm level with a number of 28-nm chip designs in the pipeline. But Intel is known to have been offering its 22-nm FinFET process as a foundry option for some time.
So it would be the right time for Cisco, which employs about 750 chip designers, to be working on IC designs intended for implementation at the next node. And it would seem that the step up to high-performance Intel silicon could make sense for a networking equipment and chip company that is likely to bring all its own intellectual property and IP cores to the design process.
And who was it that mentioned the possibility of a billion-dollar Intel foundry deal to the assembled executives at the IEF? It was non-other than John Lofton Holt, chairman and founder of Intel foundry user Achronix.
In the dim and distant past (2004) Cisco used IBM as a foundry supplier of its chips. But a billion dollars does make you stop and think. Even though it may be spread over several years a billion dollars is a significant amount of other foundries' lunch that Intel would be eating.
Intel's 22 nm transistor density won't be better than foundry 28 nm, because the fin increases minimum width to at least 90 nm. The fin pitch has to be at least halved, where double patterning is not enough. So I think cisco would still need to get on the 28 nm lines at all the foundries.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.