I think that any of you don't have slightest clue of the nature of AMD/Intel cross-licensing agreement!
Printable View
I think that any of you don't have slightest clue of the nature of AMD/Intel cross-licensing agreement!
I think you should read it. Its even been posted several times.Quote:
Originally Posted by Nedjo
Quote:
3.2. Intel License to AMD. Subject to the terms and conditions of this
--------------------
Agreement, Intel hereby grants to AMD a non-exclusive,
non-transferable ***** worldwide license, without the right to
sublicense, under Intel's Patents to:
(a) make, use, sell (directly or indirectly), offer to sell,
import and otherwise dispose of all AMD Licensed Products;
(b) make, have made, use and/or import any equipment and practice
any method or process for the manufacture, use and/or sale of
AMD Licensed Products; and
(c) have made ***** AMD Licensed Products by another manufacturer
for supply solely to AMD for use, import, sale, offer for sale
or disposition by AMD pursuant to the license granted above in
Section 3.2(a).
1. Depending on which process. TSMC typically release low power process before high performance due to the nature of their high volume customers (cell phones, wireless, etc).Quote:
Originally Posted by JumpingJack
2. True, but margins could speak for themselves.
I really don't. Please educate me.Quote:
Originally Posted by eleeter
In semiconductor manufacturing, size is an absolute requirement for technological advance. And so far the only 3 players that are poised to compete are Intel, Samsung and TSMC. All of them are a majority market share leader of their own.Quote:
Originally Posted by Nedjo
While the 45nm productions in TSMC is still in small quantities, they do have actual production samples for Qualcomm and Texas Instruments. AMD? As you suggested its still a good 3 months away.
Please cut the bullcrap about 'highly complex chips' as the design/manufacturing requirement for all chips are different, i.e., Intel's process would probably go kaboom if its used for flash manufacturing. AMD could probably get a dedicated fab at TSMC with tweaked process too. AFAIK, Spansion has a dedicated TSMC fab in Tainan, Taiwan, for flash and Omnivision has a dedicated TSMC line for its specialty CMOS image sensor process.
While my days in AMD is several years past, I hope my understanding and information are still relevant. :rofl:
This is one thing AMD (or their potential acquirers and partners) could wrestle against via all the on-going anti-competitive lawsuits against Intel.Quote:
Originally Posted by Nedjo
Just because a process is smaller does not make it more advanced. Theres a reason why ATI's and nVidias 600MHz chips used 120+ watts of power, while AMDs and Intel's 2.4+ GHz chips only use <75 watts, and as of lately, the transistor counts are similar as well.Quote:
Originally Posted by vitaminc
What are you talking about? I don't see any AMD chips randomly dieing. I've seen plenty of nVidia and ATI GPUs die all of a sudden, however.Quote:
2. AMD could make more as TSMC's quality control is top notch vs. big variances at AMD's own fab.
The only variances in AMD's CPUs is their weekly steppings, which is done to refine the process for better power usage at the designed frequencies. Boohoo if one stepping can't overclock better then another, that has nothing to do with quality control.
AMD's chip process is among some of the most advanced in the world, not many fabs have the capability to make changes to the process on the fly like AMD can. AMD actually had to send over some of their software and hardware in order for chartered to be able to produce their CPUs and be able to have weekly steppings.
Quote:
Size does matter in the manufacturing business, and the only guys that could afford to charge forward and lead in manufacturing technologies are Intel, Samsung and TSMC. IBM/Hitachi could have patents but that's only paperwork. AMD? Nowhere on the map.
AMD and IBM work together to develop technologies. Copper interconnects and SOI come to mind.
TSMC's processes are not advanced in anyway, like it was said, they are a very generic process. If they had the refinement like Intel and AMD did, you can bet that GPUs wouldn't be using near the amount of power they are right now, especially at the clocks they are running
Well , it has to be generic since they serve so many different customers.Some want density , some want performance , some want low power , some want as cheap as possible.Quote:
Originally Posted by [XC] Lead Head
Intel and AMD can tweak the process to their design and vice versa , a liberty that a foundry does not have.
As a matter of fact Intel does flash on its older FABs ( n-1 and n-2 ).As I've said it is a matter of marring the process to the design.Quote:
Originally Posted by vitaminc
Intel typically goes for the highest performance possible for its transistors , as a result others generally manage to outpace them in cell density at a given process node.
That's highly unlikely IMO.At best , AMD could get some lines inside a FAB , they need the newest process anyway.And it's a problem of capacity.Quote:
AMD could probably get a dedicated fab at TSMC with tweaked process too. AFAIK, Spansion has a dedicated TSMC fab in Tainan, Taiwan, for flash and Omnivision has a dedicated TSMC line for its specialty CMOS image sensor process.
I must say I'm extremely surprised at TSMC going SOI for AMD; I thought if anything it will mean AMD transitioning some of its CPU products to bulk.
Smaller geometry is more advanced, doesn't matter how you cut it. Otherwise everyone would just be pumping out chips in 32nm/22nm right now. The investments in equipments, the tuning on OPC recipes, etc are much harder than you asssumed. If making transistors smaller is such a no-brainer as you suggested, we wouldn't have a 18-24 months cycle for die-shrinks.Quote:
Originally Posted by [XC] Lead Head
GPUs just have poor power consumption designs compare to CPUs; it was never the priority compare to chips used for cell-phones or gadgets. The memory access methodology is so far behind (and very high power consumption) that Nvidia has a hard time getting traction in cell phones compare to Imagination Tech. AFAIK, one of the biggest downfalls/drawbacks of Nvidia's design has been memory bus usage where it has to regularly going out of local cache to fetch data.
TSMC process is *NOT* generic as you believed. As I stated above, they have a process dedicated for low power and a process for higher performance logic. They have not dedicated a lot of efforts at high performance logic due to the lack of volumes (Nvidia and ATI volumes are small); they have dedicated a lot of efforts on low power for high volume customers such as Qualcomm, TI, Broadcom, etc.
Their high performance logic process do need work; it is usually 6-months later than their low power process and yields are usually sub par (partly due to larger die size from Nvidia, ATI and Altera).
Intel is transitioning out of the n-2 fab filler strategy (n-1 is mostly used for chipsets) because the transistor designs do not fit well with smaller geometries. As witnessed with their JV with Micron for NAND and ST Micro for NOR.Quote:
Originally Posted by savantu
If Spansion could get a dedicated fab, AMD is very likely to get a dedicated fab if it can committee enough volume.
TSMC going SOI has been speculated for well more than 5 years now (and has been 'evaluating' SOI for the said amount of time).
you do realize that you are comparing a cpu arch with a gpu arch? you also realise that a gpu has much more logic then any cpu and that there is a different in quantity of transistors?Quote:
Originally Posted by [XC] Lead Head
90nm G80:681m transistors core freq: 575 shader freq: 1,35Ghz
TDP: 185W
65nm G92: 754m transistors core freq: 600 shader freq: 1,5Ghz
TDP: 145W
45nm QX9770: 820m transistors freq:3,2GHz TDP: 135W
TSMC process looks quite compared to intels if you ask me
Heh, the shader part is basicly like P4 ALUs. And it would quickly be 600Mhz vs 3.2ghz. Or if you lower it abit, a 2billion trasistor IA64 at 95W at 1.6-2Ghz vs 750million G92 at 500Mhz at about 110W. All at 65nm.Quote:
Originally Posted by GoThr3k
TSMCs process tech is compared to AMD/Intel is garbage. But its cheap and extremely high volume. TSMC is basicly making cheap Lada/Skodas while AMD/Intel is making expensive lamboginis and ferraries. Why do you think GPUs are so cheap aswell? To compare, TSMC even still uses low-K in their 45nm because high K and metal gates are too advanced. Why do you think GPUs easily die (even by themselves) but CPUs hardly if even do?
Hell even TSMC writes a 40% speed increase compared to 65nm for generel purpose process. While Intels is 51% and AMDs will also be somewhere near that. Not even to talk about leakage that is TSMC major issue. TSMC also got a 30nm gatelength and Intel 35nm. Simply to help cut cost, not to boost performance. The oxide thinkness for TSMC is 1.25nm and 1nm for Intel. The 2 processes are worlds apart. And AMDs will look alot more like Intels than TSMC.
This also helps to illustrate the differences:
Note: TSMC is NOT making 32nm low power chips. But they released information about what they will do.
http://www.realworldtech.com/include...edm-2007-2.gif
As long good product cost as normal product, im fine. :)
I dont care if intel goin to make ther chips Nec fabs.
I dont care if amd goin to make ther chips Atmel fabs.
I whant ultra fast pc stuffs for less $ as posible.
In this time i awoid certain products, wich having bad and big= slow drivers!, no supports, bad quality hw.
PPl did make way much better(smarter cpus and gpus)
And we talk abaut 30 year old tech x86 wich came from z80 by Zilog (some guys left Zilog and make x86 tech).
Motorola 1979 year cpu 68000-16 bit at 8 mhz cpu was way much better then: year 1985 286 in 8 mhz in that time.
Old tech x86-tech is for jokers, and whole world is using that, incl me.
according to your picture TSMC has a tremendous advantage in producing SRAM cells?Quote:
Originally Posted by Shintai
i dont argue it's not that good as Intel's or AMD's processes, but it is far from garbage...
Advantage? In density yes. In performance no. TSMC optimises for cheap, not performance.Quote:
Originally Posted by GoThr3k
I would bet if you made say a Core 2 on their 65 or 45nm process it would barely break 2-2½Ghz and use alot of power. But it would be cheaper.
But if they would optimize their process for performance we might see a good manufacturer. because an SRAM cell at that size is quite insane if you ask me. Maybe thats why all other specs are so bad in comparison to intel/AMD. But i am pretty sure if they would make a cpu or something else fusion like they would do that on a special developped process. Or a slightly more optimized process.Quote:
Originally Posted by Shintai
So actually TSMC is far from crap, it just has other objectives then AMD/intel
Optimize for performance=they lose they cell density. You cant blow air and have flour in your mouth the same time. You choose one path. And for TSMC cheap is the path. Compete in terms of performance and they will lose badly.Quote:
Originally Posted by GoThr3k
Would you want to risk the entire business to make CPUs that wouldnt even take up 5% of your production?
TSMC is crap when we talk about performance. Not discount production with lower quality and less lifetime. Dead/semidead GPUs, NICs etc that most if not all know about on a closer relation.
Offc they will lose the cel density if they would optimize for performance, no doubt about that. But if they would optimize a line especially dedicated for performance they would get great results i thinkQuote:
Originally Posted by Shintai
And use what, 1-2billion to do it and put the company in red? And then get slammed hard by other companies that are much better at it?Quote:
Originally Posted by GoThr3k
Its easier and cheaper to make a Ford sedan than it is to make a Ferrari formula 1 racer.
You know the production line needs to be customized to each design or you end up on the short end of the stick. They could at best make the very most lowend AMD CPUs. But for what profit? They might even have worse cell density then than what AMD got itself. Do what you are good at, not try to do something others are much better at.
You have a fairly misleading analogy and is spreading a good amount of misinformation. You have not understand or have avoided the single key issue of the manufacturing model. That is, IDM will always have better process and design integration than fabless. The key here is integration, not the ability to produce better processes or better designs. A better analogy would be comparing bespoke/custom-made clothing to off-the-rack clothing; the former will always fit better than the latter.Quote:
Originally Posted by Shintai
Other than that, here are some counter to your points:
1) Please name one single company that is better than TSMC in contract manufacturing semiconductor chips. I just want to know by whom will TSMC "slammed hard by".
2) Too bad the only 'Ferrari' maker in the semiconductor industry got its ass kicked so hard that it has to exit the 'Ford sedan' DRAM business. Guess it's just easier and cheaper to make a 'Ferrari' than a 'Ford sedan' using your moronic analogy without any basis.
3) There are always trade-offs in manufacturing process and transistor design and you can't optimize for all parameters and have your yields too. High density leads to low yield, high performance leads to low density, low leakage leads to low yield, etc.
4) A company of TSMC's size could easily afford $1-2B investments over a 12-18 month horizon. And contrary to what you believe, it is not production line that has to customized but the process recipe. You can spend thousands of dollars buying William Sonoma gears and still won't make any French recipe spicy.
You fail to understand companies with their own fabs can DESIGN aswell as optimize for a process THEY control. Same reason Intel says Nehalem will be the first chip to really exploit their 45nm process. Its designed for it. 45nm Core 2 aint.
1. TSMC is the biggest one. But look on their revenue. They have the same revenue as Intel got in 1 single quarter. And having a profit that cant even build 1 45nm factory per year. Yet TSMC produce over 10 times the amount of wafers. TSMC dont make many 45nm chips yet you know. TSMC makes cheap discount chips on a cheap discount process. Thats how their business works and what their consumers want.
2. DRAM is a multitude different business. It makes absolutely no sense what you write. Intel/AMD could make some very expensive very high performance memory in very low quantities. The businessplan would be what? DRAM production lines are simple and extremely cheap in a ultra low profit market.
3. High density leads to low yields? No. Low leakage to low yields? No. I think you mix things up greatly in lack of understanding who they make chips for.
4. So TSMC should use half their profits so AMD can use 5% or less of all their manufactoring lines to make what and for what huge profit? You got even less business knowledge than Hector. :rofl:
Oh, and remember it would be BULK and not SOI. And AMD would have to respin and redesign large parts of their chips.
Altough you have some point there shintai, but let's put some thing into consideration
3. The high density part is a typo i think, but high leakage may result to low yields. If you want your semiconductor device to have a maximum TDP of X, and because of the leakage you get a TDP of X+15 on some part, you will have to throw that part away.
4. Maybe optimizing a few nodes in that process line isnt that difficult as you think? it's not because they would have to build CPU's the would have to built new clean rooms or an entire new production line....
Bulk vs SOI => you dont need anything to retool in your fab, only wafers change
Quote:
Originally Posted by GoThr3k
High leakage= low HIGH performance yields with high power requirements. But your yields for what they make most of TSMCs customer base its very good. So the only bad parts would be highend GPUs and such. Same reason TSMC launches its low power process first. Generel Purpose process got low priority.
Optimizing a few nodes=redo your entire research, change of tools etc. Its not tweaking later on. Its the basics its wrong with. Its like halfway reinventing your nodetype.
Wafers aint the only thing that change.
Don't need to rephrase what I said. In case you didn't catch the bold line in my previous post, here it is again: IDM will always have better process and design integration than fabless.Quote:
Originally Posted by Shintai
I guess that's the reason why AMD can't compete because it has lower revenue than TSMC and much less profitable at the same time. :ROTF:Quote:
Originally Posted by Shintai
So what was your point? AMD should stay in IDM model compete against Intel in both process technology and design? Makes so much sense when it can't even rival TSMC for both revenue/profit.
You meant I made absolutely no sense because Intel started its business making DRAMs, got almost wiped out by other 'cheap' and 'low cost' manufacturers, and found nirvana in their single IBM design win?Quote:
Originally Posted by Shintai
Last time I checked, Samsung's 2008 semiconductor CapEx is $8.3Bn but Intel's CapEx is $5.2Bn. So much for extremely cheap and simiple production lines!!! Unless my math or data goes wrong somewhere but $8.3Bn is more than $5.2Bn.
High density => higher errors from OPC (and use of immersion), which leads to lower initial yields, assuming the same die size. Low leakage = wider gates = lower yield, assuming the same transistor count. Low leakage using SOI = higher inherent defects from SOI poly = lower yields.Quote:
Originally Posted by Shintai
However, those are initial yields and will be improved as the process matures.
Or AMD could loan out some of its IP to TSMC for a dedicated production line. There are a lot of different opportunities if you learn to permutate. :)Quote:
Originally Posted by Shintai
Oh and btw, stop that BULK vs. SOI marketing bullcrap. Intel's 'BULK' is as much different from TSMC's BULK as from AMD's SOI. Charter's BULK/SOI and UMC's BULK are all different too. OPC models for all fabs are all different. But I guess you believe that you can take a GDS2 layout to both TSMC and UMC and expect the same chip spinning out of fabs!!!
uh, have I not already stated that there's not enough demand from TSMC's customers for high performance node for it to develop the process? Qualcomm, TI and Broadcom/Marvell are much higher volume customers than, say, Altera and Nvidia...Quote:
Originally Posted by Shintai
Quote:
Originally Posted by [XC] Lead Head
copper interconnects came from motorola not IBM
I take it you have never worked in a semiconductor fab or studied the progress of microelectronics in any depth beyond benchmark data. I think you are confusing the fabrication process with fabricated processor!Quote:
Originally Posted by [XC] Lead Head
Here are a few of the added steps or advancements at some recent CMOS process node transitions in or layman's terms a die shrink:
250nm -> 120nm
Chemical-Mechanical-Polishing, Copper Interconnect
120nm -> 90nm
Strained silicon
90nm -> 65nm
Deep UV Lithography
65nm -> 45nm
Hi-K dielectric, Metal Gate
45nm -> 32nm
Stacked CMOS, Copper vias
And of course at each node the number of interconnect layers increases
AMD needs its own factories to compete against Intel. And TSMCs process is cheaper and alot different. What part of that didnt you understand? Do you even know how many fabs TSMC got and what kind? The issue for AMD is 180nm CPUs dont sell well!Quote:
Originally Posted by vitaminc
Samsung semiconductor makes what and how much? Intel makes what and how much? Samsung Semiconductor makes alot and alot of cheap simple devices. Along with something you forgot, LCDs!Quote:
Originally Posted by vitaminc
http://www.samsung.com/us/business/s...ochure0607.pdf
So yes, 5.2B focused is more than 8.3B very wide and thin spread. Samsung dont even use 1B on the Logic side.
Samsung Semiconductor is about 20B revenue, Intel 30B.
Asume the same diesize with different density?Quote:
Originally Posted by vitaminc
Read again, I said the exact opposite.Quote:
Originally Posted by vitaminc
And you basicly repeat what I already said. Why would TSMC change its entire business to handle a tiny tiny bit from AMD? Besides in some fanboys dreams to get a quick way to 45nm.Quote:
Originally Posted by vitaminc
You seem to skip on the reading.
Dream on, but you are probably aware that AMD's cash is drying up quickly.Quote:
Originally Posted by Shintai
Even though TSMC's most lucrative business is in their 110/130/150/180nm fabs, but they sure know how to pump out 55/65nm chips for real profitable chip makers, such as Qualcomm or Broadcom. *Correction* TSMC has receive 40nm orders from Qualcomm, Nvidia, ATI/AMD and Altera.
So if you could do me a favor, explain to me why would TI stopped its own logic fab and sole source to TSMC starting at 45nm and beyond? And please enlighten me on the 'cheap'ness of TSMC's process, in real engineering terms instead of marketing gimmicks.
ASP has no direct correlation to complexity of the design and manufacturing; it takes as much skill to make things fast than to make things cheap.Quote:
Originally Posted by Shintai
$6.3B on memory. $0.81B on logic. That's a combined $7.11B CapEx vs. Intel's $5.2B. Both uses the same front end tools but different recipe. And btw, quite a bit $0.81B is spending on converting the n-1/n-2 nodes to logic for fab fillers (with less success, but that's another story on Korean ethics).
So would you now [bend over and] agree that memory fabrication is a complex and expensive process or you still hold your previous argument that fabricating memory is a 'simple' and 'extremely cheap' process? :welcome:
And do not avoid/ignore the fact that Intel got almost wiped out of the 'simple' and 'extremely cheap' DRAM business back in the early 80s.
In case you forgot, die size doesn't go down with time because gate counts are higher. In some cases die size even increases as geometry shrinks.Quote:
Originally Posted by Shintai
Why would you exaggerate? TSMC fab for AMD does not require a change to its 'entire' business; TSMC would still remain in the contract manufacturing side of things (aside from its other supporting businesses such as IP licensing or packaging). There are a variety of permutations of business combinations to have TSMC fab for AMD, but you would rather hide under a rock than think.Quote:
Originally Posted by Shintai