Globalfoundries May Deliver First 28nm Chips in the First Half of 2011.

[PraxxtorCruel]

28nm Graphics Chips from ATI Can Emerge Early Next Year

Globalfoundries said that it would tape out the first chips using 28nm production technology as early as late this year and would initiate risk production early in 2011. It is highly likely that among the very first customers to use Globalfoundries' 28nm process technology will be ATI, graphics business unit of Advanced Micro Devices.

"For the 28nm process - aimed at both high performance and lower power devices - we will tape out products by the end of the year and start production in early 2011. The timing of when you will see products is customer-dependent but it is possible you could see them in [the first half of 2011]," an official for Globalfoundries said in an interview with The Register web-site.

Globalfoundries’ 28nm fabrication process will be available in two variants:

•The 28nm-HP (High Performance) variant will be optimized for leading-edge applications in such areas as graphics, game consoles, storage, networking and media encoding.
•The 28nm-SLP (Super Low Power) variant is optimized for wireless mobile applications such as baseband, application processors, and other handheld functions that require long battery lifetime.
It is known that the 28nm-SLP process will be used to manufacture certain ARM-based system-on-chip (SoCs) devices. Meanwhile, the most likely candidate for 28-HP technology is ATI graphics chips designer, which is known to have selected Globalfoundries' 28nm process to make certain of its next-generation chips.

Unfortunately, it is unknown which graphics processors will be made in Dresden, Germany, but manufacturing of rather complex graphics chips using thin 28nm-HP process will be a benefit for ATI, provided that it is indeed among the first clients for the node. What is unlikely is that AMD starts to use a brand-new HKMG fabrication technology for a wide range of products and especially very complex GPUs for high-end market.

 

[Klo]

What is unlikely is that AMD starts to use a brand-new HKMG fabrication technology for a wide range of products and especially very complex GPUs for high-end market.

But otherwise good news!

 

[drunkenmaster]

My guess would be a , 6670/6770, basically a midrange test card on 28nm, though its debatable.

I wouldn't be suprised to see a 6xxx series midrange part tested and then the 7xxx parts with two designs, one for TSMC, one for GloFo, well without question they'll have two designs, the question is, will they produce them at both, or just one, and if one, which one.

The GloFo process will almost certainly be significantly better than TSMC's, but even then that doesn't necessarily let you know what they might produce where.

A super high quality process with much higher yields, you might want to make the highest end cards there, get higher yields and more cores. However if TSMC sucks(it will) with low yields, the high end has a higher margin of profit as long as you can get to a certain yield, which AMD have shown for 3 years now they seem capable of getting, however if the low end with tiny tiny margins have reduced yields, it can turn something from $5-10 profit with 5million in volume, into a no profit part, and thats a worse problem.

SO we might see the low/mid end high volume being made on the better process and the high end chips made at TSMC, its completely up in the air.

Companies always have a plan B so they won't be just assuming TSMC are a no go for the next gen.

If though GloFo transition works great and the 7series, produced at both places comes off without a hitch I can see AMD moving almost exclusively to GloFo for the 8xxx series cards.

But its ask expected, the process is just about ready, we've seen test wafers at trade shows so they can make them, but the production line is probably not up and running. The rumour/guess for the last year has been GloFo 28nm testing in the last quarter, and some risk production in Q1 next year, with products maybe available as soon as Q2 next year.

 

[bru]

I do hope your right and GLoFo does manage to get their 28nm process sorted out so its better than the bound to be dismal TSMC, it might make TSMC pull their finger out and sort its self out.

 

[twist3d0n3]

any chance NV will switch to glofo?

 

[Orcvader]

My guess would be a , 6670/6770, basically a midrange test card on 28nm, though its debatable.

I think that's a good point there. If we can say the 7000 series are coming out in late Q4 11, having a midrange 6000 card to help test and improve the process could help AMD.
I doubt AMD will stick with TSMC for 28nm. having two manufacters might result in two different grades of cards, plus as we've seen from before they're failing to meet with deadlines.

 

[jigger]

I think what will probably happen is ATI will move the most problematic, but 28nm viable cards over to GF and then use the extra capacity at TSMC to build the 40nm 6000 range.

 

[Freddie1980]

Does the OP want to provide and link and give credit to those who wrote the article?

 

[drunkenmaster]

I think what will probably happen is ATI will move the most problematic, but 28nm viable cards over to GF and then use the extra capacity at TSMC to build the 40nm 6000 range.

Well thats the thing, with any new cards the older cards are still being made, largely because Dell and co still sell old cards by the millions, and they don't update all their systems at once, likewise new process's usually have significantly limited capacity so they will still be making plenty of 40nm cards.

But the 40nm won't be replaced by 28nm, 28nm will be put into another fab likely replacing older equipment as they run several process's at once.

Its very likely some of the range will be split and they'll happily make more eslewhere.

AS you said theres a possibility of having 2 different versions of a 6870, ones from GloFo that go like they're on crack, and crap ones from TSMC, or the other way around.

AMD have proven one thing in the past 3 years, adaptability and not shooting themselves in the foot, so they'll have plans for TSMC's and GloFo's process, the design won't change much but the layout and potential size of the cores changes.

IE one process might promise transistor size within say 0.1microns and another 0.3, which means you have to build more leeway into the design. Thats one of the fundamental flaws with TSMC at the moment, Nvidia had awful yields as they didn't account for TSMC's 40nm process being highly variable in transistor size, while AMD did, which accounted for apparently a 10-15% die size increase.

Whats likely to happen though is the riskier chips will be aiming for GloFo as AMD have FAR closer ties with GloFo and are probably heavily involved in all the test wafer production as GloFo's biggest customer and first and largest 28nm customer. AMD would probably have a lot more confidence, and know almost exactly where GloFo are.

Remember also that Glofo's 28nm could well have less capacity than TSMC's, even if its earlier, and if AMD pull out of TSMC completely its likely TSMC will have a lot of spare capacity, which means AMD with new options can probably get a decent deal on TSMC production as they essentially move into a much more powerful negotiating position. Then theres the simple fact that AMD might be able to sell everything both TSMC and GloFo can make for them, and if they can, why wouldn't AMD use that capacity.

Its more likely they'll try and limit one line to one factory, so all 6870's and 6770 type cards at glofo and do all the low end at TSMC, or the other way around, or whatever. Mixing up different cores for the same product could get confusing, though with the same specs 99% of people buying them wouldn't care less.

 

[Freddie1980]

Once 28nm is in full swing we might see video cards being sold at lower prices due to competition between Globalfoundries and TSMC fighting it out with each other to attract customers.

 

[tbh]

Once 28nm is in full swing we might see video cards being sold at lower prices due to competition between Globalfoundries and TSMC fighting it out with each other to attract customers.

Or we may see lower prices because Global are capable of making a decent process that doesn't have dismal yields.

 

[bru]

everybody seems to think that GloFo will make a good 28nm process, and I really do hope your right. of course TSMC track record speaks for itself, they will be crap as they always are, but theirs always a chance that TSMC will be the better process.

oh my god...... what am i thinking....of course GloFo will be better....it has to be doesn't it.....?

 

[drunkenmaster]

Its not as simple as there being one 28nm process and one company can get it better than the other. At any process you can use any number of different methods to combat leakage and ensure uniform transistor lengths.

I'll be honest and not claim to know a huge amount about it, or what some of the terms mean, but TSMC have gone for a "gate first" method to fight leakage, and its basically the utterly cheapo method of doing it and getting it ready quickly, and its a large part of the reason their 40nm is so crap, so its almost certainly going to be a "cheap" 40nm process, cheap as in R&D, not cheap as in high yields so cheap to make cores.

GloFo have really done a "proper" process, gate last, hugely higher quality all throughout, and frankly AMD/GloFo have amongst the best process's in the world, the only reason they've been behind is limited fab space/money to expand.

Intel have 5-6 fabs, running at 70-80% capacity, and to move to 32nm they just outfit one fab which is shut down for a while and it doesn't matter.

AMD had/have 2 fabs, they run at basically 100% capacity and constantly make updates to increase capacity, they can't easily shut down a fab to update it to the latest process. THe time taken to move to 65 then 45 nm isn't to do with quality, just logistics. Because they have limited capacity they really are forced to be just about the best in the world to make sure yields are right up there.

Frankly its very likely that GloFo's 28 nm will have both higher yields, less leakage, lower power usage AND be literally smaller. They have estimated that a similar chip on GLoFo's 28nm vs TSMC's 28nm could be up to 30% smaller, simply because with less leakage, less problems and more uniform transistor size you can pack everything in much much tighter. Which could mean higher yields, but the a wafer of the same chips could already have 30% more chips on it anyway.

I think pretty much everyone in the entire world will be surprised if TSMC somehow make the better process.

Soon as TSMC saw GLoFo were going to be a real threat, and that Samsung have pledged to get into the game and want to have 20-30% of the fabbing market in 5 year or something, they massively increase R&D spending from I believe early last year. But that R&D investment really won't be seen/paying off for a while as most of that spending will be going into 22 and 18nm at this point.

TSMC have a couple fabs in the process of being built, and have more spare capacity than GloFo will for now. It will likely be late 2011 when GloFo is supposed to start risk production at the New York State fab that GloFo will suddenly have what will literally be the best fab on earth, with huge capacity, that cost $8billion + that they can really start pumping out other chips.