Your take on Global Foundries

JAYDEEJOHN · Apr 17, 2009

Weve all read about their recent announcement of the half node 28nm process. We all have our opinions as to what this means overall to AMD. The way I see it, it has tradeoffs. Lets try to keep this one civil. My main question here is 2 fold.
Going half node, whats your opinions on the capability,availability for GF to 1: get nVidia and others on board (such as ARM) using the 28nm HKMG second gen process, and 2: Do you think AMD can/could use this on their cpus?
My reasoning for this is, itd put AMD in a good position, and very competitive, say, even if Intel is at 22nm in that timeframe, 28 will have advantages for AMD
I understand about production limits etc, but at these process sizes, itll be 2.5 times that of the current 45nm process, plus the added fab in NY, so I think it can be done, and maybe there wont be anything exclusive other than AMD products of course, meaning that ARM or nVidia would have to go elsewhere for this process on occasion

fazers_on_stun · Apr 17, 2009

JAYDEEJOHN :

Hmm, maybe I missed the news - whatever happened to Intel's pulling the x86 license over the GF spinoff again?

Anyway, I thought the half-nodes were for GPU's only, not CPU's since the half-nodes are bulk Si and not SOI. 32nm with 1st-gen HKMG needs to be done first. TBH, I think IBM is just blowing smoke since Intel released ES Clarksdales already at 32nm, and hence beat IBM and the consortium once again.

Also, I would think GF's process would have to be significantly better/cheaper than TSMC in order to get nVidia's business, plus NV would probably be worried about IP issues. Although I'm no process engineer, my understanding is that when you outsource your designs there's a learning curve to optimize it with the DFM process rules. That's what happened with AMD initially at 65nm, from what I've read - they rushed Barcie out the door as it was 9 months late already, when it really needed another 7-8 months to be optimized for the process.

Finally, some recent artilces I've read concerning the UAE indicate all is not well there, and the global recession is having some effects there. So it seems the 'unlimited petrodollars' myth is just that..

DXRick · Apr 17, 2009

I read somewhere that Intel spent 7 billion to retool their plants for their next CPUs. So, we can assume that GF will also need to spend billions on the tools needed to develop CPUs that AMD designs. Where is this $$ going to come from?

Will GF and/or AMD invest that much $$ to develop FABs for a product that will likely not sell enough to cover that expense?

EDIT: I just saw this on DailyTech: IBM Technology Alliance Unveils 28nm CMOS Technolgoy

sub mesa · Apr 17, 2009

Maybe AMD is trying to revolt against the limitations the x86-agreement imposes on AMD; which has proven to be very imposing on its ability to gain market share: it cannot outsource production on a large scale so it has few flexibility. Having two companies (one for chip design the other for chip production) might make seem sense, as it reduces the risk of having the production facilities bankrupt because the chip division cannot compete effectively with intel. Not saying that is the case though, but they sure have a hard time.

sub mesa · Apr 17, 2009

DXrick: AMD has a new financial partner from the Arabic Emirates. As far as i understand, they insisted on AMD seperating their production. So it might not even be that AMD had a choice in their struggle for survival. Its a shame though, if Intel didn't massively neglect the law in the days AMD was having a much better product and actually got some good cash to build on a stable foundation, their market share only went up after Intel already retook the performance crown with the Conroe CPU - and Intel has leaked lots of juicy details about this CPU, perhaps to prevent people from having bought an AMD system which was then the best choice, but instead to wait for the Conroe to come out. Marketing can be smart, and for the Conroe there was lots to talk about long before the product actually came out.

Their court struggle against Intel might be the place to re-negotiate their agreements, but i have no idea what AMD has its mind on to be honest. As far as i see, they are in troubling water regarding the legal agreements with Intel.

DXRick · Apr 17, 2009

sub mesa :

I disagree with the claim that Intel is to blame for AMD's lack of product recognition. AMD did nothing to advertise their CPUs, relying on people to go to THG to learn about them. Meanwhile Intel did an advertising blitz with the "Intel Inside" thing and used the Blue Men in ads to generate product recognition. Dell and others were slow to offer AMD computers, because of the lack of demand for them. By the time AMD computers became available, Intel released the Conroe architecture.

It also didn't help that AMD relied on third parties to develop chipsets for the CPUs. Initial mobos were lousy.

I chose to build an Intel P4 3.0 computer instead of AMD Athlon, because the differences in gaming weren't that big, Intel was faster in apps (like Photoshop) where clock speed mattered, AMD CPUs where more expensive, and I read about too many problems with the available mobos. I am still using this computer, tho I am now thinking its time for a i7 build.

Clearly, AMD's CPU and mobo offerings are much better today, but they still suffer from product awareness for those that aren't gamers and/or enthusiasts. And Intel has billions to invest in new tech.

sub mesa · Apr 17, 2009

Both SiS and ULi made excellent chipsets during the Socket A age, the very best i would actually dare to say. But those names were of no marketing value, so they don't get used.

And that's where Intel's better at: marketing, product management, overall business strategy. They are smart, but AMD has always tried to compete by delivering a better product: same or higher performance for less money. While not all of the periods in last decade's rollercoaster ride were in favor of AMD, they did often give you the best value for your money.

And its a secret that high-performance "enthusiast" grade hardware is often not profitable at all because few people buy them. But its not the sales of these products which matters; its the marketing value of having THE FASTEST ... <insert here>. And that works, see users will read about performance tests showing Intel faster than AMD, or nVidia faster than ATi on the high-end side of their product range. So the user wants an Intel with nVidia, but ofcourse not those expensive ones in the tests so something less expensive. Even though they do not realise it, they stepped into another market range (there are: budget, value, mainstream, performance, high-end, enthusiast). The problem is, Intel might not be the best product in that price range, and ATi might also be better in that price range. So marketing obviously has done its job here. And the buyer is happy because he associates his product with those high benchmarks he has seen "intel is good".

And no this is not about Intel or AMD or ATi, its all about customer psychology. And Intel knows about that alot more than AMD does; that was my point.

Those online test

sonoran · Apr 17, 2009

JAYDEEJOHN :

Short answer, yes.

Long answer - having a process developed does NOT equate to having factories retooled to produce product using that new process. Even if AMD had a chip design for that process in hand today, you're still talking possibly billions of dollars in new equipment purchases and equipment upgrade costs, along with the disruption caused by all that tool install and conversion activity. Let me assure you, even planning for all that activity is not simple, much less the process of doing all those upgrades. If they started the conversion process today, it'd be at least a year, maybe two, before they'd be producing product in quantity on the new process.

JAYDEEJOHN · Apr 18, 2009

OK, but the NY fab will be the best one for the new processes ever made, as itll incorporate 32 from the get go. No retooling needed there at all, as itll be that way when they open production. Thats something thats being overlooked. Its not as if itll need any more than already planned, its Dresden and the other plant thatll need it. It could come in with HKMG already to run, also.
Im just wondering, if Intel is courting 22nm, and AMD is just getting onto 32, I was wondering if a 28nm would not only be possible, but would also offer AMD a catchup mode, til 22 was ready.
I understand people wanting to know if cpus were selling as much as gas, and that gas also has slowed in sales in this economy. But no one needs a cpu, however they do need gas, so sales havnt suffered like tech has. And as the economy gets better, itll show immediately selling petrolium products, as everyone uses them for everything, and those sales are met first, then new jobs, higher production for the rest of us. Just like we need to eat. So, no need to go there, as its not a reality in the equation of whether UAE is going to hold to its end or not. Interesting comments tho, as if this is somehow connected in any way.
I believe AMD, which has already taken on the expense of its imeersion costs for its new fab process may have this ability built in using that new immersion tech, however, a good chunk of Intels upgrading will have to go into this cost, which AMD is already using.
So, once they all get on the same node, having HKMG, using the new immersion tech, AMD may have an option Intel doesnt

jimmysmitty · Apr 18, 2009

fazers_on_stun :

Intel beating IBM and the others is no suprise. Intel has had wroking 32nm for a while and are now running 22nm SRAM chips that can clock pretty damn high on near 1v.

The only way any consortium would manage to beat Intel to a node is IF they skipped one which has its ups and downs. The ups are having it first and maturity with it. The down is not having a previous node experience that you can usually learn a lot from.

Intel learned a lot from their 90nm and applied some useful things to 65nm. Thats why their 65nm was so good. Intel has learned a lot with 45nm and its HK/MG and will apply not only what they have learned from that but also improved designs.

DXRick :

Their money will come from the UAE investors. Thats the only place it can come from. AMD doesn't have enough money to spend on that or they would look into doing their own FABs.

sub mesa :

Even if it was the case AMD signed the agreement and there is no way to fight it. Its like when you agree to something in software. Thats your comitment to your end of it.

And TBH, its not as restrictive as you would think. I completely understand why Intel would not want something like the GF to happen. Giving someone else other than the company who has the license access to your patents and the ability for them to copy them and mass produce them for some other area is scary since it would cut your profits.

How would you feel if you owned the rights to something, let another company use that design and they just gave it away to another company to produce? Would you enjoy watching your product be replicated and you not get any of the income to your idea? Most people wouldn't Any company would do the same thing. If it was IBM they would too.

And AMD has plenty of room to compete. If not then they never would have had their glorious K8 days. If AMD had kept pace instead of buying ATI and milking K8 they probably would have been a very viable competator. If they had released Phenom II before Intel released Core 2 (I skip Phenom because Phenom II is what Phenom should have been) then they would have gained even more market share. But they decided to milk K* just as Intel milked P4.

But I am glad to see Intel is at least pushing out new stuff every year be it a die shrink or new tech. Maybe AMD will as well that will accelerate our technology boosts faster and give better prices faster.

As for the GF, all I know is that its a good idea but I still don't like it. I think they will get to the point where AMD is not their main focus which could be bad for AMD as the quality may go down causing more bad chips per wafer. While AMD can still sell those chips like they do with the X3s, X2s or Semprons I doubt they would be able to discount them as much seeing as the GF would like to profit.

It may also raise the price of AMDs chips even if their product is not the best one because the GF has to profit. Their investors will not take too much in loss and if they do they will hit a point where they will withdraw leaving AMD in a bad place.

JAYDEEJOHN · Apr 18, 2009

My understanding is, using half nodes, its usually limited to gpus, phones etc, but having their new immersion tech, Im wondering if a half node may be seen in a cpu. It appears ARM may end up at GF, and is why I mentioned awhile ago in a different thread that this is only another reason Intel doesnt want GF around.
As far as ability and making AMD equal customer, it wont happen, as all the things Ive read shows AMD, then we will see what else we can run. Besides, as Ive been saying all along, any profits made are also profits shared with AMD, as their portion of GF.
I know IBM hasnt exactly led the way in process, and they like to steal Intels thunder for bits and giggles, but at least now with GF, theres actually going to be something there to push the better processes ahead, as it seems IBM isnt in a big hurry, as theres really no one tick tocking away as top their competition, but that changes of course with AMD, and changes yet again by GF having higher capabilities than AMD had.
If they dont spend the time with AMD, that also means they wont be doing so with other companies, which is just bad business. Itd be like saying, theres so many of these damn ATOMs per wafer, why bother doing it well at Intel.

sub mesa · Apr 18, 2009

Even if it was the case AMD signed the agreement and there is no way to fight it.

I would not say that. There are competition laws in each country, though USA was always less strict than say Europe. Anyway, AMD may state that x86 processors have become the standard in modern processors and needing a restricitive license in order to produce these means the company can't compete on a fair level with Intel. Why does Intel not have any limitations on their production etc, while they can use all the technology from AMD for free. Heck their EM64T-document doesn't even mention AMD; though it was like a 99% copy. The point is: is x86-processors have become so common they cannot be considered a special implementation anymore, and both Intel and AMD have extended this architecture and will have to do serve the future demand of x86-cpu's. If AMD couldn't build x86-cpu's anymore, there would be no competitor to Intel; and that is not good for the consumer because free market societies with only one market player being very dominant, is going to be a disaster.

Also note, if AMD threatens to break the license, Intel might not be selling their products as well; they would have to workaround many patents, perhaps their IMC and most probably their amd64 implementation Intel had got from its cross-license agreement; if that license is disbound that means Intel looses access to AMD technology and both companies can sue eachother and demand each to stop its production. Oh this could get ugly...

fazers_on_stun · Apr 18, 2009

JAYDEEJOHN :

We won't know how well Luther Forest will work until it's up and running and the bugs worked out, obviously. And you never know - could be a NY taxpayer revolt and they throw all the bums outta office

. Or if the downward trend in world-wide chip shipments continues, there may be no need for it. But assuming things turn out OK, then yes it'll be quite the 'state of the art' I'm sure.

Im just wondering, if Intel is courting 22nm, and AMD is just getting onto 32, I was wondering if a 28nm would not only be possible, but would also offer AMD a catchup mode, til 22 was ready.

Mask sets are expensive to develop and I would think 28nm's design rules would require a new mask set, testing & validation, etc. If AMD wanted to go to 28nm for CPUs, then I would think they would bypass 32nm entirely. If you're asking if AMD is working on both nodes for CPUs, the returns would have to justify the expense.

I understand people wanting to know if cpus were selling as much as gas, and that gas also has slowed in sales in this economy. But no one needs a cpu, however they do need gas, so sales havnt suffered like tech has. And as the economy gets better, itll show immediately selling petrolium products, as everyone uses them for everything, and those sales are met first, then new jobs, higher production for the rest of us. Just like we need to eat. So, no need to go there, as its not a reality in the equation of whether UAE is going to hold to its end or not.

It's not just the crude oil business, which I believe accounts for about a third of the UAE's economy, but also the downturn in banking, investments, etc. I'll see if I can find the link I mentioned above, mentioning the problems in the UAE. Besides, there was another article on MSN to the effect that oil may not recover its former astronomic pricing for years, due to cheating by OPEC members and by the non-OPEC producers like Russia.

I believe AMD, which has already taken on the expense of its imeersion costs for its new fab process may have this ability built in using that new immersion tech, however, a good chunk of Intels upgrading will have to go into this cost, which AMD is already using.
So, once they all get on the same node, having HKMG, using the new immersion tech, AMD may have an option Intel doesnt

I don't think immersion is that expensive, nor difficult to achieve. From what I've read, going to HKMG is a bigger deal and Intel is ahead there in the experience factor - they're already on their 2nd-gen HKMG process. And Intel didn't need immersion at 45nm, so they saved some $$ and increased their profit margin. Rumor has it that some significant price reductions on the Penryn quad-cores are coming soon, so a better & cheaper 45nm fab process gives Intel more pricing leeway.

fazers_on_stun · Apr 18, 2009

jimmysmitty :

Yep - according to the press release by Intel last month, 32nm is going great which is why they are pushing it full-speed, with the ES samples out to the mobo manufacturers already. Of course, AMD/IBM/GF etc counter with more announcements and roadmaps, instead of actual hardware - we'll see. They did do well and timely with 45nm - hopefully that is the trend and not a fluke

.

And, it's funny that AMD made so much noise about "fusion" with the ATI purchase, and now Intel is beating them to market with an actual product

. And Larrabee is not too far off either, hopefully. I guess we could say now AMD is copying Intel??

JAYDEEJOHN · Apr 19, 2009

Intel said it needed HKMG at 45nm, and many people took that and ran with it, and said AMD was doomed without it. That turned out to be false . It is expensive to go immersion, and possibly cheaper than than HKMG, tho, since currently only Intel needs it for bulk, we again, just have to take their word on it.
Either way, its good to have both, and since both companies are currently only using 1 of each. its hard to draw conclusions, of course we could go back to the "AMD needs HKMG too" days, but time will tell.
If petro doesnt raise to those prices for years, this is a good thing, as it was one of the contributors of the current economic problems weve had. But UAE was a banking mecca before 100$ a barrel oil, thats just that much more monies. Its not as though having an oil filed is bad business, or is that what youre saying? Poor profits on oil today? Or are you only talking markets? One which is needed, unlike cpus? Im sure theyre raking in the bucks, even with "cheap" oil

JAYDEEJOHN · Apr 19, 2009

I posted in another thread, that Intels IGP on die may be 2x as large, if theres alot of redundancy in certain things going 2x, it may well have 2x or more the capabilities of thier old and well known crap IGP. I welcome this
But I also see this as an opportunity for Intel to market their on die solution as superior due to just the IGP improvements alone, and no one would know the difference, as its on die, and not singly tested without the ondie connects.
As for LRB, Ive found a few scenarios where it will have some slow downs inherited in it. Its FP will be halved from full potential due to latencies, but even so, itll still, or should still do well

fazers_on_stun · Apr 19, 2009

JAYDEEJOHN :

JDJ, I think you may be confusing immersion photolithography and SOI, used by AMD/GF, vs. double-patterning and strained silicon as used by Intel. Intel said it could have gone to immersion at 45nm, but found that they could just use double-patterning instead on a few critical layers and delay the expense of the immersion equipment until 32nm.

As for HKMG, Intel found that it greatly improved speed and power due to reduced leakage on their strained silicon substrate, whereas AMD found that they didn't need it with their SOI substrate. My understanding is that SOI is about 20% more expensive to buy in wafer form, and has diminishing utility at the smaller nodes anyway where due to the thinner dielectric, significant leakage starts occuring through the gate, compared to the channel and substrate.

AMD itself has planned to use HKMG at 32nm, except that the IBM-developed process is the opposite of Intels - either gate first or gate last, I forget!

Intel is already on its 2nd generation HKMG implementation, on 32nm, and has reported significant transistor improvement over the 1st-gen process at 45nm, as reported in the EETimes and elsewhere.

You may have seen this already, but it's a good article on how AMD got to 45nm without HKMG: http://www.eetimes.com/showArticle.jhtml?articleID=212002243&pgno=3

For AMD, the goal was not only to shrink the physical dimensions but also to increase performance. Scaling transistors means better performance, but most of the dimensional scaling at 45 nm is related to gate and metal pitches rather than to length of the transistor gate or channel. The minimum gate length on Shanghai is 38 nm—a reduction of only seven percent from the 65-nm node. But the transistor performance is 19 percent better for the NFET and 23 percent better for the PFET compared with 65-nm generation transistors.

How'd they do that? The answer, in a word, is optimization. Although there are no new materials or techniques like adding an additional stressor for strained silicon engineering, AMD improved transistor performance by squeezing every last drop out of the performance-enhancing structures already in use at 65 nm. As usual, the starting point was a silicon-on-insulator (SOI) wafer as opposed to the bulk wafer technology used elsewhere. The rest of the transistor performance story relates to strain engineering.

As for the UAE, check out this 2009 prognosis from no less than Dubai's minister of the economy: http://www.middle-east-online.com/english/?id=31068

Oil (including shipping) only accounts for about a third of the UAE economy, and finance & investing is actually a larger chunk from what I've read. The worldwide recession has affected both.

JAYDEEJOHN · Apr 19, 2009

I think youre confusing that AMD said it doesnt need HKMG at 45nm, and itll simply be using it at 32nm, and wont incur the expenses having to do it in 2 gens, as its not needed.
As to the UAE, compare it to all the other countries in that region. Youll find the GDP of those countries with a much higher % makeup of their oil revenues, which only goes to the credit of UAE, and not the other way around.

poul · May 4, 2009

I hope they get a lot of customers fast. TSMC and UMC have delayed the introduction of new technology for a while, played there own little game, now it backfires. Now it comes out, that they could have introduced HKMG at the 45nm node, but didn't and so on. So I wish GF all the best. Now TSMC is bying equipment for billions and hireing engineers in hundreds. Now all of a sudden they are moving faster than even Intel is. I can't help but thinking about Doug Grose and others from Global Foundries world tour, as the Thechnology Bus to the future.

jimmysmitty · May 5, 2009

sub mesa :

It still doesn't matter. Think of it this way. A company creates something new and unique. They allow other companies to use them of course at a charge for the patent use rights. That company breaks the agreement allowing some other company to use the patent. thats breaking the contract. its common law.

Now of course AMD and Intel will work around it. I never said AMD would not be around. All I said was that they have to abide by it since they signed it. And as I said I am sure a new agreement will come out of it.

And probably yes. Intel would have to stop using AMDs 64. But the IMC, not so much. There is no real copyright on that since its a technology that was on the NB but now on the CPU. In fact there have been plenty of IMCs that predate AMDs. Intel even had one developing on the 486 and other companies had some before that.

Thats like saying QPI is the same as HTT when its really not. They both opperate differently and in their own ways while doing the same end job.

Your take on Global Foundries

Champion

Splendid

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Champion

Champion

Champion

Distinguished

Splendid

Splendid

Champion

Champion

Splendid

Champion

Distinguished

Champion

Share this page