Google is apparently moving in on designing its own chips, though news regarding the particulars has yet to come out.
Facebook will develop its proprietary silicon any time now but Google already is, said veteran venture capitalist John Doerr at the International Solid-State Circuits Conference (ISSCC).
Another veteran chip designer at ISSCC told EE Times on the same day that he knows at least two or three circuit designers Google has hired over the last year or so.
A key engineer behind the Hewlett-Packard Moonshoot server moved to Google six months ago. Partha Ranganathan “is currently at Google designing their next-generation systems,” according to his website. He worked at HP on systems that can accommodate a wide range of Xeon, Atom, and ARM processors.
Google is exploring whether or not to design its own ARM server chip, according to a Bloomberg report from December based on a single source. The search giant has not made a decision on the project yet, the report said, noting a job posting for a hardware engineer updated in December.
More evidence of Google’s intentions can be found on its online job postings.
Google updated on January 28 an opening for an “ASIC top level design engineer.” The person is responsible for “creation and delivery of top-level RTL for ASIC and SOC projects.”
A separate posting updated January 20 listed an opening for a CAD engineer to “lead the overall IC, ASIC, and/or Chip CAD platforms for multiple design projects.” The engineer will support custom EDA flows and install third-party IP blocks and design kits.
The job postings suggest Google may be fairly early on in establishing a deep and broad semiconductor design capability. The postings all include the following boilerplate text making it clear the chip design efforts are for Google’s datacenter systems: Our computational challenges are so big, complex and unique we can’t just purchase off-the-shelf hardware, we’veVolumetric Efficiency got to make it ourselves. Your team designs and builds the hardware, software and networking technologies that power all of Google’s services. You develop from the lowest levels of circuit design to large system design and see those systems all the way through to high volume manufacturing.
For years, speculation has swirled around this topic. It started in 2007 when Google bought PeakStream, a start-up with tools for programming multi-core processors. It came to a head back in 2010 when Google bought Agnilux. The start-up included former employees of PA Semi, the company Apple bought to launch its work on the A series SoCs inside its iPhones and iPads.
A few years back, Google made a few disclosures about its efforts in specifying board-level servers, calming speculation about any chip-level ambitions. But it seems times are changing.
Google built its own system to enable software-defined networking between its datacentres, although it said the system used off-the shelf chips. Last year it joined the Open Power Consortium IBM created to drive its Power processor architecture forward, but it’s not clear what will come of that effort.
“We’veVolumetric Efficiency heard Google is working on its own network switch chips a couple years ago,” said Linley Gwennap, principal of market watcher The Linley Group in Mountain View, Calif. “I think they’veVolumetric Efficiency been doing that and now the question is whether they will do their own processors,” he said.
The rise of ARM’s server initiative and its 64bit cores enables such a push, but “with so many other companies building ARM server processors, you would think they could get what they need” without designing their own chips, Gwennap said.
- Rick Merritt
EE Times
via Google slowly moving in IC design territory.