ARM & GlobalFoundries Demo 2.5GHz+ 28nm Cortex A9 & 20nm Test Vehicleby Anand Lal Shimpi on December 14, 2011 7:34 PM EST
- Posted in
GlobalFoundries sent over a PR showcasing two significant milestones in its march towards being a major foundry player in the mobile space. The first is the announcement of a dual-core Cortex A9 test chip built on GF's 28nm HPP (High Performance Plus) process. The test chip operates at 2.5GHz although it is apparently capable of higher frequencies according to the ambiguous statement from GF. The chip's operating voltage is a low 0.85V. Both the frequency and voltage targets are good for a Cortex A9 implementation, although again this is only a test chip.
Many companies are expecting to break the 2GHz barrier on high performance 28nm processes starting late next year and moving on into 2013. These designs aren't likely to be used in smartphones, but instead we'll see them in tablets and netbook replacements (e.g. Windows 8 on ARM).
The second milestone is the the tapeout of a 20nm Technology Qualification Vehicle (TQV). This is a test chip designed to, well, test GF's 20nm process as it would be used by a vendor for producing Cortex A9 based SoCs. The move to 20nm is much further out for GF, but there's lots of development work prior to release that necessitates the production of test silicon.
Both announcements are designed to showcase GF's close partnership with ARM, as well as its continued momentum in the manufacturing space. Process technology in general is going to become a very important piece of the puzzle in the mobile space, particularly as companies like Intel enter the market next year.
Post Your CommentPlease log in or sign up to comment.
View All Comments
Hector2 - Wednesday, December 14, 2011 - linkThis is certainly an interesting development. At face value, it implies that Global Foundaries has working ARM silicon at 28nm. As is everything, the devil is in the details.
Is that fully functional ? Is leakage at 28nm sufficiently low to be sustainable in the long run ? or are short channel effects bad enough to require major additional development to correct ? How does GF's definition of 28nm channel length compare to Intel's definition at 22nm -- which apparently is how thin the FinFETs are ?
I read a few days ago that Intel has 14nm FinFETs working in the lab now. Does "working at 14nm" at Intel an equivalent milestone to GF's "28nm operating at 2.8GHz" ? Only time will tell.
retrospooty - Thursday, December 15, 2011 - linkI am wondering how these newer 2+ghz chips do in real world performance. How much faster are things vs' todays 1-1.2ghz chips, is it noticeable, or are we at the point of diminishing returns.
Hector2 - Thursday, December 15, 2011 - linkIt's about power & die cost, too. Intel's new 22nm FinFET transistors are not only faster but smaller as well -- so cheaper to make. Much more impressive, though, is that the new transistors are ultra effiicient with leakage power and offers a 20X improvement over their 32nm process. That translates directly to much longer battery life for ultramobile devices.
retrospooty - Thursday, December 15, 2011 - linkYup... thats kind of why I want to see if there is any performance gains... I could see these clocked lower and being extremely miserly on the juice.
shriganesh - Thursday, December 15, 2011 - linkI bet Rory Read and AMD must be banging their heads to move away from GloFo!!!
silverblue - Thursday, December 15, 2011 - linkNot really. This isn't quite the same as actual mass production at 28nm like TSMC are doing right now.
haukionkannel - Thursday, December 15, 2011 - linkYep. GloFo 32 nm and 28 nm production is not the same.
silverblue - Thursday, December 15, 2011 - linkI was referring to the fact that this is a 28nm demo, whereas TSMC are mass producing 28nm chips. GF's inability to keep up with TSMC is more than just yields, it's process as well.
Mugur - Thursday, December 15, 2011 - linkNo, just bang the heads of GF people for not being able to make a good 32nm process for their FX...
fteoath64 - Friday, December 16, 2011 - link"not being able to make a good 32nm process for their FX... " What ?. Releasing a 3.6Ghz stock chip @32nm is very good progress. It is just unfortunate that AMD did not optimize the micro-architecture of BullDozer for it. Hence, the overall performance duffers. I think using shared FPU is a bad idea. They had a lot of work to get the memory bandwidth up and the cache bandwidth to be comparable or better than i7.
And considering the lower price of their CPUs, I would expect more dual socket boards to be out there. These seemed rare these days.