One of the unspoken rules of customer-foundry relations is that you virtually never see the former speak poorly of the latter. Only when things have seriously hit the fan do partners like AMD or Nvidia admit to manufacturing problems, and typically only after postponed launches and poor availability have made protestations that everything is fine unsustainable.
That’s why we were surprised — and our source testified to being stunned — that Nvidia gave the following presentation at the International Trade Partner Conference (ITPC) forum last November. Many of the company’s complaints regarding its current partnership with TSMC are exactly what you’d expect given the manufacturing problems the entire industry is facing. What’s surprising are Nvidia’s remarks concerning TSMC’s current cost curves and manufacturing ramps. This is normally the sort of information discussed quietly between a foundry and its customers or by the press with help from various anonymous sources. Discussing the problems publicly is a sign of just how frustrated the company has become.
TSMC builds hardware for a huge number of companies, but those customers have very different needs and use a wide range of process technologies. Historically, Nvidia (and ATI/AMD) have been regular early adopters. The nature of graphics is that it can easily soak up new processes and the higher transistor counts they enable.
The flip side of that situation is that companies like AMD and Nvidia have also been responsible for assuming the risks associated with “risk production” and footing a hefty bill for the privilege. As those risks mount and costs skyrocket, Nvidia is increasingly unhappy with being asked to shoulder the burden. Nvidia’s slides talk about the need for “real” understanding, compromises on “rough justice,” and a closer relationship that looks more like that of an IDM. For those of you who don’t know the term, Intel is an IDM — it handles both manufacturing and design. AMD used to be.
When AMD spun GlobalFoundries off, one of the things GF promised to provide that would distinguish it from TSMC was high levels of IDM-style integration. At TSMC, the customization work that is available is highly monetized; specialized work is expensive and time-consuming. In reality, GF’s ability to provide the amount of IDM-like flexibility that it wanted to offer has been sharply constrained by the problems associated with Llano and Bulldozer; our sources tell us that the foundry devoted enormous resources to bringing AMD’s 32nm APU back on track.
According to Nvidia, the current model is unsustainable.
Lot more at the link. You're going to want to read it all:
> Few days back there was an article titled "A Giant Threat to Intel" in which the author
> suggested falling CPU prices as a giant threat to Intel even if it succeeds in
> smartphones/tablets market.
> Though the point is valid, I was wondering how long Intel competitors will be able
> to sell processors at low prices? How will they recover the cost of increasingly
> powerful processors required for Windows 8 devices and future tablets/smartphones?
> How will they fund move to next-gen processes, and who will? How will they fund
> R&D to keep up with Intel?
Very true - Instruction Set IP, R&D and manufacturing costs are real and staggeringly high. Any semiconductor vendors has to be able to pony up these costs to do business. In fact, Intel's costs are lower than any ARM-SoC because it has integrated "IP+Mfg" and has a huge process lead (similar to what Apple has done in devices "HW+OS+some bundled SW").
Further, Intel has repeatedly made the point that they have low-cost CPUs but people tend to migrate upward. Just think - would you buy a Celeron rather than an i3 and i5 for your laptop? There will always be a market for the lowest-end, but the bulk of the market will tend to be in the mid-range.
Intel, being Intel, will be able to put a CPU or SoC for the whole range....at a better price point than any of its competitors.
"If TSMC cannot deliver 28nm at a price that its customers consider viable, then the whole model is broken."
Few days back there was an article titled "A Giant Threat to Intel" in which the author suggested falling CPU prices as a giant threat to Intel even if it succeeds in smartphones/tablets market.
Though the point is valid, I was wondering how long Intel competitors will be able to sell processors at low prices? How will they recover the cost of increasingly powerful processors required for Windows 8 devices and future tablets/smartphones? How will they fund move to next-gen processes, and who will? How will they fund R&D to keep up with Intel?
And you can back to your argument that pricing is the only issue, TSMC's 28nm process is working well.
I would argue that these 2 issues are integral to the success of the customer-foundry model and business relationships. If TSMC cannot deliver 28nm at a price that its customers consider viable, then the whole model is broken. And Nvidia will not squeal this loudly unless they are being squeezed pretty badly. Further, TSMC can pick and choose its customers leaving the lower-end ones at the kerb. Not a great position for a fabless semiconductor to be in. (Nvidia now, Qualcomm next?)
And what do these node-wise price increases bode for 22/20nm and beyond? It can only get worse, far worse.
They can't do it before that, because planning has to be done a couple of years in advance.
I think that's more realistic - Samsung is out don't you think?
Would NVDA touch Samsung NVDA is paranoid - NVDA neither
In the meantime relax...
I know that's the common thinking. But Apple has always had secret backroom projects. I wouldn't put it past them to break out something earlier.
Then again seeing how they don't like to change much of anything lately I can see them holding off on switching to Intel. I can see the sub $200 Intel based smartphone without contract would get them angry as they watch the premium handset market dry up.
This article definitely has merit but there is much more to this story. I will post a more detailed response tomorrow on www.semiwiki.com
To me this a simple case of wafer pricing negotiations gone wild. Since TSMC is in such a dominant position at 28nm and 20nm the industry is telling them, in the Year of the Dragon, to be more like a teddy bear and collaborate on pricing in the same way they collaborate on technology.
But, for any fabless semiconductor company to say that they will not aggressively transition to the coming process nodes, 20nm, 14nm, etc… is ridiculous! I remember back in the day buying an AMD 40MHZ based PC versus an Intel 36MHZ. Seriously, did I really need that extra 4MHZ? Did I pay extra for it? Of course!
"At AMD’s Financial Analyst Day, CEO Rory Read made a point of saying that the company no longer intends to aggressively transition to new process nodes given the diminishing marginal returns from doing so........"
Okay, let me clarify, any leading edge fabless semiconductor company that wants to STAY in business will aggressively transition to the coming semiconductor nodes, absolutely. Just my opinion of course.
"Okay, let me clarify, any leading edge fabless semiconductor company that wants to STAY in business will aggressively transition to the coming semiconductor nodes, absolutely. Just my opinion of course."
Yes, this is ABSOLUTELY true. And this is why Nvidia's statements carry such weight and import.
The problem is that TSMC is no longer on the leading edge. Intel's move into trigate has left them in the dust. Now they need trigate technology and likely EUV and 450mm wafers all at the same time.
The cost of this is incredible and the time to do it is slight. They don't have the technology or the money. The fabless model is broken. It simply doesn't provide the huge sums necessary to keep up.
Huge volumes are necessary for the move to the next level to be economic. TSMC can't move fast enough to generate those volumes. The struggle with 28nm clearly illustrates this.
Intel simply has a superior model. As an IDM they both design and manufacture. They don't have the fabless problem and they have the volumes necessary for every move up to be economic.
AMD and now Nvidia have both thrown in the towel. Well, Nvidia is really doing it for TSMC since TSMC is unlikely to admit any of this.
To say that this is a simple case of wafer pricing negotiations gone wild is just silly. It's a case of the underlying dynamics falling apart at a very, very fundamental level.
Nvidia had the guts to admit it. What's your problem?
The impact of what Nvidia is saying cannot be overstated.
They are saying that all of the ARM foundries have reached the point where the pursuit of further reductions in die sizes is no longer economically feasible.
This means that, just like AMD, they will no longer be able to pursue any type of regular fabrication cycle improvements.
This leaves one and only company that can still economically pursue regular fabrication cycle improvements -> Intel.
This means that the ARM world can look forward to using those great big heat producing chips for a long time into the future.
Or switch to Intel. Which is what the Nvidia CEO wishes he could do. He said exactly that...
You certainly forgot Samsung;
NAND is @19nm - this does not mean NAND is more advanced than logic - each has its own challenge but Samsung can learn from NAND and DRAM and apply it to logic and vice versa.
NAND is also going towards 3D cell structure different though - so good for INTC to continue working with Micron.
I think that's how it ultimately will shake out to and there would be plenty of business for both Intel and Samsung.
NVDA has a very aggressive design which looks good on paper but it's too difficult to manufacture.
There is a long history of tension between NVDA and TSMC.
NVDA can't go to GF nor Samsung because they settled with TSMC on gate last at 28nm, if NVDA would go to Samsung or GF they need to make changes to gate first but we know already that Samsung and GF will change to gate last at 20/22nm.
It's a mess.
It's cheap on NVDA's side to blame TSMC (they are no dummies) blame NVDA not coming up with a more production worthy design.
The manta of design is more important than manufacturing touted by the ARM people hit a brick wall
On the other hand I don't hear ALTR or XLNX crying
<Nvidia is increasingly unhappy with being asked to shoulder the burden. Nvidia’s slides talk about the need for “real” understanding, compromises on “rough justice,” and a closer relationship that looks more like that of an IDM>
So, now you see why NVidia wants Intel to open up it's fabs to them. They want to grind down Intel on price and use their new relationship to grind down TSMC. Nvidia wants to ramp up production at 28nm and they really have no other alternatives. I can only imagine how expensive those Kepler 680 GPU's must be!
If this CEO keeps talking smack about TSMC they may refuse to do business with the company until he is replaced. You can only be a loud mouth in this industry for so long.
I would love to see GloFo's P&L - when Siemens was doing semiconductors the word was that semiconductors were Siemens most expensive hobby - I don't think they ever made money.
Actually the ex Siemens/Infineon guy is working for small foundry in Malaysia I believe or is it China.
SMIC also working on 28nm HKMG - not sure if this is good or bad - all I know more arms to be sold.
Makes you wonder why AMD paid GF money
I am also an ARMS dealer which means my picks provide the tools - so I have to try to stay objective.
Regarding FPGA I found a link INTC/ALTR - they already have an offering.
I think I understand now why Intel picked those two players.
The foundry landscape is currently a mess - the key foundry does Gate last - Samsung, GF do Gate First and my understanding is that is NOT transparent to customers like NVDA - so I think they are in a real pickle.
IBM fan club spent all that money only to switch to gate last messy, messy