ARM, GlobalFoundries Announce Extended Parternship On 20nm, FinFET Technologies

rated by 0 users
This post has 6 Replies | 0 Followers

Top 10 Contributor
Posts 26,504
Points 1,197,000
Joined: Sep 2007
ForumsAdministrator
News Posted: Tue, Aug 14 2012 3:42 PM
ARM and GlobalFoundries have been working together ever since AMD spun GlobalFoundries off as an independent business, but the two companies are taking steps to further expand their joint development efforts. As part of the deal, ARM has committed to creating a "full platform of ARM Artisan® Physical IP, including standard cell libraries, memory compilers and POP™ IP solutions."

We typically discuss ARM as selling licenses to various companies like Samsung, Texas Instruments, and Nvidia, but licenses aren't the company's only product. When ARM talks about physical IP, it's referring to all the other components that go into the SoC and make it tick. Need an L2 cache implementation, a bus, and a memory controller? ARM can sell you those. Need a GPU? ARM can sell you one of those, too. Qualcomm, for example, has an ARM license but designs its own SOCs. Other companies that want to save on R&D costs and bring a product to market more quickly, might prefer to opt for an end-to-end predesigned solution. The SoC won't be as customized, but you know exactly what you're getting and how it works.



This new agreement is important because GlobalFoundries' path from 28nm to 22/20nm isn't as straightforward as TSMC's. Back when 28nm was under development, GlobalFoundries, Samsung, and IBM committed to a 'gate-first' approach to manufacturing on the nascent process. TSMC and Intel, on the other hand, went for gate-last. The two approaches are not compatible, and GF announced quite some time ago that it will adopt gate-last at the 22/20nm node.

The other major part of the announcement is that GF and ARM are teaming up to ensure that the post-20nm FinFET transition goes as smoothly as possible. Intel's Ivy Bridge uses a specific type of FinFET the company refers to as Tri-Gate, and neither TSMC nor GlobalFoundries will catch up on this front any time soon. TSMC has stated publicly that it won't adopt FinFET until the 14nm node, with 20nm production expected to begin in 2014. GlobalFoundries seems likely to follow a similar trajectory, with FinFET technology shipping in volume 3-5 years from now.

“ARM technologies are at the heart of many of the world’s highest volume product categories, and we believe will only grow in importance for our customers in the years ahead,” said Mike Noonen, executive vice president, worldwide marketing and sales at GLOBALFOUNDRIES. “By leveraging our implementation knowledge and applying it to a next-generation, energy-efficient ARM processor and graphics processing unit, we believe we can jointly offer a compelling differentiation to our mutual customers that will power innovation into the next two generations.”
  • | Post Points: 95
Top 500 Contributor
Posts 257
Points 3,190
Joined: Aug 2012
Jaybk26 replied on Tue, Aug 14 2012 5:02 PM

20nm? That's incredible! It's almost difficult to think how tiny that is.

  • | Post Points: 5
Top 100 Contributor
Posts 1,077
Points 11,665
Joined: Jul 2009
Joel H replied on Tue, Aug 14 2012 6:46 PM

Here's something that'll really bake your noodle: A sodium atom is about 0.16nm wide. That's not unusually large as such things are reckoned.

Now, obviously we'll never push CMOS to anything like that small -- but at 20nm, we can talk about atomic diameter without it being foolishly off-scale. The distance between the traces of a 20nm processor, expressed in sodium atoms, is 125 sodium atoms wide.

What that *does* mean is that when we talk about doping, the amount of dopant deposited matters down to 10s of atoms. There's a difference between 125 atoms of X and 100 atoms of X. That's how tiny the frame of reference has gotten, and how exact we need tools to be to keep pace with it.

  • | Post Points: 5
Top 10 Contributor
Posts 4,839
Points 45,835
Joined: Feb 2008
Location: Kennesaw
rapid1 replied on Tue, Aug 14 2012 9:26 PM

20nm come on you can do 18 be a real chip maker buncha wusses

ASUS Z87C
i7-4770K
Xonar DGX
Intel Gigabit CT
Geforce GTX 770 4GB
G.Skill X1600/1754 2x8GB 7/8/8/24
  • | Post Points: 5
Top 500 Contributor
Posts 290
Points 2,255
Joined: Aug 2012

Wow thats remarkble. 2 heads are better than 1

  • | Post Points: 5
Top 150 Contributor
Posts 501
Points 4,625
Joined: Dec 2011
Location: centennial park az
AKnudson replied on Wed, Aug 15 2012 12:36 AM

Joel your right that is incredible, Bringing chips to the atomic scale is amazing but doesn't it make developing chips like a finite geometry?

If we keep making them smaller wont we eventually hit atomic level? we might get really close and make chips a tiny bit smaller (tiny a scientific term i made up to mean equaling 1/100 of a nm) but that wont be able to make computers much faster, they will eventually stop being able to make chips smaller. The result of which everyone will scramble to do more with the speed they've got, i am talking revamping hardware, circuits, cooling systems and basic code to create an easier more efficient computer interface.

  • | Post Points: 5
Top 100 Contributor
Posts 1,077
Points 11,665
Joined: Jul 2009
Joel H replied on Wed, Aug 15 2012 10:08 AM

AKnud,

We started hitting those problems in 2005. Everyone talks about Moore's Law (the tendency for transistor density to double every 18-24 months) but Moore's Law wasn't actually the factor that kept performance moving. It was Dennard scaling -- the tendency for smaller transistors to draw less power and dissipate less heat, that truly kept performance moving upwards.

Long term, CMOS is finished. NAND flash hits hard limits around 14nm, IIRC. Intel has talked about sub-10nm scaling for certain cases, TSMC and GF haven't said much about scaling below 14nm. But when I say long-term, I mean it. We've got another decade of scaling.

  • | Post Points: 5
Page 1 of 1 (7 items) | RSS