• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

AMD going 20nm this year?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

mjw21a

Member
Joined
Sep 29, 2004
Copied from the Inquirer. I know it specifies GPU's, but how much longer to move the CPU's to this also.... I find it a bit odd considering TSMC can't even meet demand on their 28nm process node. :-/

Globalfoundries looks to Mentor Graphics for 20nm fill techniques
20nm silicon expected in 2H 2012
By Lawrence Latif
Tue May 29 2012, 15:39

WAFER BAKER Globalfoundries will collaborate with Mentor Graphics on its upcoming 20nm process node.

globalfoundriesMentor Graphics works on the hardware and software that ends up in fabs and the Smartfill technology in its Calibre Yieldenhancer will be used by Globalfoundries for fill techniques in its 20nm process node. According to Mentor Graphics, its Calibre Yieldenhancer will help Globalfoundries hit integrated circuit fill constraints in a single pass, and that should reduce tape-out time.

Maq Mannan, director of PDK development at Globalfoundries said, "At 20nm the fill strategy becomes much more sophisticated because it's no longer just about planarity. Besides CMP-ECD [chemical mechanical planarisation or polishing - electron capture dissociation] issues, you have to consider a whole range of interrelated effects, such as etch, lithography, stress, rapid thermal annealing (RTA) and other issues essential to successful manufacturing. Calibre's integrated fill analysis and cell-based approach makes fill more precise, while achieving target runtime and output file size."

Mentor Graphics claims its Calibre Yieldenchancer can be integrated into existing design processes. The firm's Smartfill product allows chip designers to define a set of critical nets that require attention during the fill procedure.

Globalfoundries and Mentor Graphics have been working together for some time and Globalfoundries is busy preparing its 20nm process node, which increases requirements for circuit fill techniques. According to Mentor Graphics, Globalfoundries will start delivering 20nm silicon in the second half of this year. µ
 
Very nice. I would guestimate that what were seeing is AMD leveraging their new found flexibility from the non exclusivity agreement with GF. Since this gen of APU / CPU parts will be on 28nm and TSMC is already manufacturing at that level they can pull down orders with GF while they retool for the 20nm process to follow it up with. Either way I wouldnt expect to see 20nm AMD APU/CPU parts to hit the market until next summer at best. Just because GF will be setup to run 20nm by the end of this year doesnt mean that the process will be full scale production ready AND that AMD will have there 20nm designs nailed down for production any sooner than next spring.
 
Well, it still seems to be a much faster change between process nodes than I'm accustomed to seeing from AMD..... I'm not sure if I believe it. I'd like to..... AMD's major problem is how far they are behind in process tech. Makes them uncompetitive.
 
Oh how i was there was competition... reading the article on tomshardware last week about how amd is "dropping out" of the speed race against intel made me frown... if amd isn't pushing intel, intel gets away with putting TIM in the heatspreader instead of fluxless solder...and ivy bridge runs hotter and doesn't have as much overclocking headroom...and i cry a little...
 
Pft, they've already stopped competing directly. Thats what APU's are about. Once more code is written for OpenCL they're going to look a whole lot stronger than they presently do.

I don't believe we'll ever see AMD beat Intel on pure x86 performance ever again, where the GPU is leveraged though they'll hold their own. They just need to stay close to Intel with x86 performance, not beat them. The GPU can do the rest.
 
even intel is getting into this compute area, the Larrabee project brought a lot for intel. it may have never made it into the video card market, it does however live in the specialized compute department like nv's telsa cards. after a bit of looking the prototype card was call "Knights Ferry" and a production one called "Knights Corner" is due out sometime 2012 or later. amd is challanging them just in a different area we may not see. as the gpus will be able to do the same HPC type work as knights corner. i just cant recall if intel was pushing for OpenCL or using something else to allow programing for knights corner.
 
Hopeully its better than Larrabee. A humongous number of simple x86 cores just means more power in those areas where x86 already excels. Those things it does poorly however it would also do poorly at. I think nvidia and AMD are going down the right path here. I'll have to wait and see what Intel brings to the table.
 
I kind of doubt it though. Historically nobody has ever beat Intel on process shrinks.
 
I'd say they're just setting some aggressive goals to work towards. They won't meet them, but it amy allow them to get to that process sooner than they otherwise might have done.
 
To be honest i doubt AMD will move to TSMC for there CPU's any time soon. they don't have the critical (SOI) fab, as far as i know.

GlowFlow do, and that's what AMD want as there pursuing ever higher clock frequencies, I can't find the article now but the last thing i read on PD is that they were going to be hitting 4.5Ghz (stock)

It looks to me it will remain CPU / APU @ GlowFlow, GPU @ TSMC.

In the same article it also stated that GlowFlow are aiming for 10nm by 2014.

After all the troubles and hate between AMD and GlowFlow in recent times, whats the bet they will fall in love again? perhaps renew there vows?
 
AMD have already stated that they intend to move to bulk process sometime in the future for all their products. SOI is only going to be used for this current generation as far as I'm aware.
 
AMD have already stated that they intend to move to bulk process sometime in the future for all their products. SOI is only going to be used for this current generation as far as I'm aware.

Current must include PD as there SOI.

I don't know what to make of that, its cheaper for them but there probably not going to keep those new found clock speeds up, they had better hope they can dramatically improve IPC instead.
 
Yes, but PD is 32nm. A die shrink will mean a different code name so it woun't be PD anymore, even if it is the same architecture at that time.
 
Yes, but PD is 32nm. A die shrink will mean a different code name so it woun't be PD anymore, even if it is the same architecture at that time.

Perhaps Steamroller, but the SR was expected to be 28nm yet that article was written 3 months ago, which was probably before GF's 20nm announcement.
In it AMD refused to say if it was going to be made at GF or TSMC, at that time of tension between GF and AMD, and keeping in mind TSMC had the 28nm node,- that's probably where they intended it to end up.

AMD paid GF out of their fixed contract, GF are probably using that money to upgrade their fab.

I do wish AMD weren't so secretive, it would not hurt them to simply say this is our up coming chip, it will be ??nm and #### will make it and this is what we hope it will do, they must know we are not execting it to kill Intel, simply for it to move in the right direction and be competitive at AMD's traditional value for money.

Instead we are left speculating on leaked road maps and hungry for some real news. and articles like that making bold claims like this,-

Another important factor that should be considered is that the Steamroller-based central processing units for high-performance servers and desktops [due in very late 2013 or in 2014] will compete not against Intel's current-generation or next-generation processors, but against server offerings based on Haswell and desktop offerings powered by Rockwell/Broadwell, which will be made using 22nm and 14nm process technologies, respectively.
 
Last edited:
Yeah, with aly luck there will be enough capacity to move their entire line across. I imagine 20nm would be sufficient to complete with Intels 22nm trigate process.

I'm not so sure about that. If AMD don't release details, then it can't be said that their products are released late as they never gave a time frame in the first place. I've never ever seen AMD manage to release something on time when they've said they'd do so.

I really do hope we see AMD moving to 20nm for everything soonish. I believe it could bring them up to performance parity with Intel (pack more stuff into their chips).
 
I really do hope we see AMD moving to 20nm for everything soonish. I believe it could bring them up to performance parity with Intel (pack more stuff into their chips).

Also 20nm run cooler and use less power.:cool::popcorn:
 
Back