28nm @ TSMC: Very expensive or just wafer-limited?

[AnarchX]

Latest MSRPs of 28nm products are indicating, that 28nm could be 4-times expensive as 40nm comparing costs per mm^2, if the margins are equal to 40nm products.

Could be this real or is 28nm just wafer-limited and the prices are only so high to secure supply and to shift the margin in direction of manufactures instead of retail.

What could be a good estimate how expensive 28nm could be?

 

[CarstenS]

Linky-link pls?

 

[ninelven]

Don't know about 28nm, but this article was interesting... (would seem to indicate wafer-limited)

http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

EDIT: Based on some of the images there, one could estimate ~1.5x cost/mm^2 over 40nm (~2x when 7970 launched).

 

[silent_guy]

Don't know about 28nm, but this article was interesting... (would seem to indicate wafer-limited)

http://www.extremetech.com/computing/123529-nvidia-deeply-unhappy-with-tsmc-claims-22nm-essentially-worthless

EDIT: Based on some of the images there, one could estimate ~1.5x cost/mm^2 over 40nm (~2x when 7970 launched).

Wow. That's an astonishing presentation.

 

[Babel-17]

"40 percent"?

http://semimd.com/blog/tag/umc/

Over time, however, 28nm will represent a “longer-lasting node in the industry,” Sun said. On average, a 28nm process with high-k is 40 percent more expensive than 40nm. And 20nm is a staggering 50 percent more expensive than 28nm, he said. “If you look at 20nm, it will require double patterning. That’s very expensive,” he said.

 

[AnarchX]

http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

EDIT: Based on some of the images there, one could estimate ~1.5x cost/mm^2 over 40nm (~2x when 7970 launched).

Great piece of information.

This also explains why 40nm GPUs at similar size were never so cheap like 55nm ones.

 

[Man from Atlantis]

so nv's expecting 28nm costs settle on Q412, october release for biggieK should be ready till then..

 

[Megadrive1988]

Does this put the Maxwell architecture in danger?

 

[MfA]

Intel is the only one really succeeding at shrinks nowadays (for logic, flash and DRAM have less problems). The rest of the semiconductor industry is flailing wildly and praying for EUV power source problems to be finally fixed.

 

[Alexko]

Does this put the Maxwell architecture in danger?

In danger of being late, perhaps, but what else should we worry about?

Intel is the only one really succeeding at shrinks nowadays (for logic, flash and DRAM have less problems). The rest of the semiconductor industry is flailing wildly and praying for EUV power source problems to be finally fixed.

Intel is ahead, almost by a full node, and although their transitions are a bit quicker, I'm not sure they're fundamentally different. But they have complete control over both their processes and physical design, which helps, and since everything is internal, you never hear Intel complaining about Intel screwing Intel with poor yields or slow ramp-ups.

Actually, when it comes to complaining about foundries, NVIDIA seems to be, by far, the most vocal company out there.

 

[MfA]

I'm sure they have problems, but their lead has been expanding ... and given the volume they are always able to ship I don't think slow ramp ups are a big issue for them.

I wouldn't be surprised if no one but Intel manages to make 22 nm commercially viable until EUV's power source problems are fixed, and I think it's highly unlikely that anyone will get below that with immersion lithography except for Intel.

 

[UniversalTruth]

Actually, when it comes to complaining about foundries, NVIDIA seems to be, by far, the most vocal company out there.

Yes, they are kind of vocal, customers (like us) are vocal too, because it's obvious we need fair (and low) prices.
My question- is it really TSMC responsible for this pricing explosion? I mean- they do get the tools and machines from somewhere else, right? Those ones are guilty.

 

[Alexko]

I'm sure they have problems, but their lead has been expanding ... and given the volume they are always able to ship I don't think slow ramp ups are a big issue for them.

I wouldn't be surprised if no one but Intel manages to make 22 nm commercially viable until EUV's power source problems are fixed, and I think it's highly unlikely that anyone will get below that with immersion lithography except for Intel.

I don't know, people at the Common Platform conference 2012 seemed to be pretty optimistic, but then again this is what such conferences are for.

Yes, they are kind of vocal, customers (like us) are vocal too, because it's obvious we need fair (and low) prices.
My question- is it really TSMC responsible for this pricing explosion? I mean- they do get the tools and machines from somewhere else, right? Those ones are guilty.

I'm afraid the main culprit is just physics.

 

[jimbo75]

I dunno why Nvidia is crying about "fair partnership" and that stuff - if they can't make profitable parts then they need to change what they are doing, ie make smaller parts. That's what AMD has had to do at TSMC since long.

If Apple or Qualcomm or somebody else is able to pay more for the same wafer then TSMC is gonna take it. Nvidia would take it and not blink an eyelid about "fairness" for their partners so long as they were making money out of it, neither do they think it's unfair to get more wafers than AMD at lower prices either. Do they really think anyone cares?

 

[silent_guy]

I dunno why Nvidia is crying about "fair partnership" and that stuff - if they can't make profitable parts then they need to change what they are doing, ie make smaller parts.

Because Kepler parts are much larger than GCN parts?

If Apple or Qualcomm or somebody else is able to pay more for the same wafer then TSMC is gonna take it. Nvidia would take it and not blink an eyelid about "fairness" for their partners so long as they were making money out of it, neither do they think it's unfair to get more wafers than AMD at lower prices either. Do they really think anyone cares?

I think you must have missed the key point of the presentation: the one about the cost cross-over point between different technologies. That doesn't materially change by making your dies a bit smaller.

 

[jimbo75]

Because Kepler parts are much larger than GCN parts?

Yes?

I think you must have missed the key point of the presentation: the one about the cost cross-over point between different technologies. That doesn't materially change by making your dies a bit smaller.

The point of the presentation was to cry like a baby over what they deem "unfair". I don't see anyone else crying over it. If they can't afford it any longer the onus is on them to find ways to make it affordable.

You know their real problem is constantly burning bridges with everyone. After the mauling they gave TSMC over 40nm why would they be surprised that TSMC is raising prices that possibly only Apple and Qualcomm can really afford? If you were TSMC would you be feeling particularly charitable towards Nvidia?

AMD have stated that they are going to slow down process transitions. You know, adapting? Why does Nvidia just cry and cry over their persecution complex? It's tough for everyone but Nvidia is the ONLY company whining over it.

They had a chance at GF and blew it - GF would have taken their business in a heartbeat but they flippantly disregarded them as "AMD's fab". Nvidia's problem is they actually believe that TSMC is *their* fab, but its not. They are no more important than anyone else who is after TSMC's extremely highly sought-after manufacturing capability.

 

[ninelven]

Yes?

So you are saying 300 > 365 ???

 

[Man from Atlantis]

Yeah if the projected trend continues you'll have to pay same price for 100mm2@20nm and 200mm2@28nm.. And it'll keep doubling for next nodes

 

[ninelven]

Exactly, if transistor density doubles, but cost/transistor remains constant... you have a real problem in the not too distant future (at least if you want to actually increase performance).

 

[jimbo75]

So you are saying 300 > 365 ???

No I'm saying that Nvidia's planned performance chip was 296mm2 compared to AMD's planned performance chip at 212mm2.

Their planned enthusiast chip will have a similar size increase over AMD's. They are extremely lucky that Tahiti is such a mediocre performer else they'd be in a horrible position.