nVidia increases 40nm for 2009

[thilanliyan]

http://www.digitimes.com/news/...ith_tsmc_for_2009.html

Hopefully this relates to G300 as well. I'd like to see both G300 and RV870 out by the end of the year...I'm getting bored with what we have right now.

 

[Idontcare]

Nvidia has also expressed interest in becoming an initial client of TSMC's 28nm process, which will start initial production in first-quarter 2010, the sources noted.

We've now had multiple accounts of TSMC starting 28nm production in Q1 2010...including TSMC stating it...I find this to be breathtakingly aggressive.

Can they do it? 9 months after customers start shipping 40nm product they intend to have 28nm in production? If true that definitely deserves a "simply...wow" comment.

28nm will debut HK/MG for TSMC as well...although they stated they will make available a traditional SiON/doped-Si xtor for customers who don't want to pay for the HK/MG wafer tech.

This time next year will we be seeing HK/MG GPU's shipping out as a shrink/refresh of GT300? Seems just so ridiculously aggressive of a timetable to be true...

 

[Elfear]

Originally posted by: Idontcare

Nvidia has also expressed interest in becoming an initial client of TSMC's 28nm process, which will start initial production in first-quarter 2010, the sources noted.

We've now had multiple accounts of TSMC starting 28nm production in Q1 2010...including TSMC stating it...I find this to be breathtakingly aggressive.

Can they do it? 9 months after customers start shipping 40nm product they intend to have 28nm in production? If true that definitely deserves a "simply...wow" comment.

28nm will debut HK/MG for TSMC as well...although they stated they will make available a traditional SiON/doped-Si xtor for customers who don't want to pay for the HK/MG wafer tech.

This time next year will we be seeing HK/MG GPU's shipping out as a shrink/refresh of GT300? Seems just so ridiculously aggressive of a timetable to be true...

Wow. That would be impressive.

 

[Lonyo]

Are TSMC going out of step with the CPU players and doing the traditionally called half node processes as their normal processes? (AMD/Intel go 45 -> 32 while TSMC goes 40 -> 28)

 

[OCGuy]

I thought they had major leakage with 40nm? They are confident moving ahead with 28?


Interesting times indeed.

 

[ViRGE]

Originally posted by: OCguy
I thought they had major leakage with 40nm? They are confident moving ahead with 28?


Interesting times indeed.

Well we know that Intel has been having great success with their 32nm efforts, so perhaps TSMC has discovered Intel's big secret.

Although I find it much more likely that TSMC is just full of it.

 

[Idontcare]

Originally posted by: Lonyo
Are TSMC going out of step with the CPU players and doing the traditionally called half node processes as their normal processes? (AMD/Intel go 45 -> 32 while TSMC goes 40 -> 28)

Doesn't matter what they call it when you think about it, the performance and cost of the chips it generates is all that really matters.

There are professional teardown/reverse-engineer outfits out there that basically dissect every chip that gets put out on the market - ChipWorks and Semiconductor Insights are the two more well known in the business.

You can bet they will cross-section TSMC's 28nm and determine whether or not it is superior to Intel's and GF's 32nm.

But yes, at the superficial node label level TSMC renamed their 45nm process node as the 40nm node, relabeled what was to be their 40nm half-node as now their 32nm node, and relabeled what was going to be their 32nm node (w/HKMG, etc) as now being the 28nm node.

All the node labels just took a half-node shuffle down the queue, if they didn't do this then it would have seemed like they (TSMC) had a sizable 18+ month lag behind Intel on 45nm which in turn would have been to their customer's disservice in product markets where their customers parts compete with Intel and GlobalFoundries customers.

It's all marketing. Totally harmless. I could care less whether TSMC releases their upcoming Ketchup node slightly after Intel releases their upcoming Mustard node, what is relevant to you and me is how well does GT300 shrink perform on TSMC's Ketchup node versus Larrabee shrink on Intel's Mustard node. (and when do we get to find out?)

edit: Yeah it's Fudzilla, so you know apply usual caveats:

Globalfoundries to fight TSMC on 28nm

Who can get it faster, yield better

Sources close to Globalfoundries have informed us that TSMC won?t be the only company having the 28nm half node in early 2010.

http://www.fudzilla.com/content/view/13833/1/

 

[nemesismk2]

This is good news because it gives me more options for future upgrades.

 

[Grinja]

Originally posted by: Idontcare

We've now had multiple accounts of TSMC starting 28nm production in Q1 2010...including TSMC stating it...I find this to be breathtakingly aggressive.

Can they do it? 9 months after customers start shipping 40nm product they intend to have 28nm in production? If true that definitely deserves a "simply...wow" comment.

28nm will debut HK/MG for TSMC as well...although they stated they will make available a traditional SiON/doped-Si xtor for customers who don't want to pay for the HK/MG wafer tech.

This time next year will we be seeing HK/MG GPU's shipping out as a shrink/refresh of GT300? Seems just so ridiculously aggressive of a timetable to be true...

My guess is that they need to be aggressive in order to ensure that Global Foundries doesn't steal their business. If all things are equal, ATI would go with Global Foundries, if Nvidia followed suite then TSMC would be in a tight spot no?

This is of course assuming Global Foundries will be ready with a descent offering at 28nm ...

 

[Idontcare]

Originally posted by: Grinja
My guess is that they need to be aggressive in order to ensure that Global Foundries doesn't steal their business. If all things are equal, ATI would go with Global Foundries, if Nvidia followed suite then TSMC would be in a tight spot no?

This is of course assuming Global Foundries will be ready with a descent offering at 28nm ...

I don't know how much of a tight spot TSMC would be in if both ATI and NV business went away, around 75% of TSMC's sales are still from 90nm and larger tech nodes.

To be sure it isn't good to loose that publicity halo of having highly visible and high profile clients like ATI and NV move their business to another foundry, but it wouldn't be the end of the world for TSMC.

But you are right, they have got to aggressively go after keeping GlobalFoundries away from their leading-edge customers...the best defense is a good offense type of cliche.

 

[1ManArmY]

How long does it take for the retail product once tape-out has been completed?

 

[Keysplayr]

Originally posted by: 1ManArmY
How long does it take for the retail product once tape-out has been completed?

Normally 120 days. Or so I have heard.

 

[1ManArmY]

Originally posted by: Keysplayr

Originally posted by: 1ManArmY
How long does it take for the retail product once tape-out has been completed?

Normally 120 days. Or so I have heard.

I was close, I was guessing 90 days give or take

 

[cbn]

Originally posted by: Idontcare

Nvidia has also expressed interest in becoming an initial client of TSMC's 28nm process, which will start initial production in first-quarter 2010, the sources noted.

We've now had multiple accounts of TSMC starting 28nm production in Q1 2010...including TSMC stating it...I find this to be breathtakingly aggressive.

Can they do it? 9 months after customers start shipping 40nm product they intend to have 28nm in production? If true that definitely deserves a "simply...wow" comment.

28nm will debut HK/MG for TSMC as well...although they stated they will make available a traditional SiON/doped-Si xtor for customers who don't want to pay for the HK/MG wafer tech.

This time next year will we be seeing HK/MG GPU's shipping out as a shrink/refresh of GT300? Seems just so ridiculously aggressive of a timetable to be true...

With this economy I would imagine die shrinks are more important than ever before with this product segement in particular (re: cost , power efficiency, performance to $ dollar ratio all improve)

 

[Keysplayr]

Originally posted by: Just learning

Originally posted by: Idontcare

Nvidia has also expressed interest in becoming an initial client of TSMC's 28nm process, which will start initial production in first-quarter 2010, the sources noted.

We've now had multiple accounts of TSMC starting 28nm production in Q1 2010...including TSMC stating it...I find this to be breathtakingly aggressive.

Can they do it? 9 months after customers start shipping 40nm product they intend to have 28nm in production? If true that definitely deserves a "simply...wow" comment.

28nm will debut HK/MG for TSMC as well...although they stated they will make available a traditional SiON/doped-Si xtor for customers who don't want to pay for the HK/MG wafer tech.

This time next year will we be seeing HK/MG GPU's shipping out as a shrink/refresh of GT300? Seems just so ridiculously aggressive of a timetable to be true...

With this economy I would imagine die shrinks are more important than ever before with this product segement in particular (re: cost , power efficiency, performance to $ dollar ratio all improve)

In this economy, that's for sure. But there is a snag. Retooling these fabs costs billions.

 

[Jacen]

let us hope they don't suffer from the bump issues a lot of their current stuff does. =/

 

[Idontcare]

Originally posted by: Jacen
let us hope they don't suffer from the bump issues a lot of their current stuff does. =/

Bump issues were all 65nm stuff I thought, doesn't afflict the 55nm stuff does it?

 

[Jacen]

Originally posted by: Idontcare

Originally posted by: Jacen
let us hope they don't suffer from the bump issues a lot of their current stuff does. =/

Bump issues were all 65nm stuff I thought, doesn't afflict the 55nm stuff does it?

They haven't been confirmed by Nvidia but there are theories that they may very well be facing them also.

 

[soccerballtux]

Originally posted by: Idontcare

Nvidia has also expressed interest in becoming an initial client of TSMC's 28nm process, which will start initial production in first-quarter 2010, the sources noted.

We've now had multiple accounts of TSMC starting 28nm production in Q1 2010...including TSMC stating it...I find this to be breathtakingly aggressive.

Can they do it? 9 months after customers start shipping 40nm product they intend to have 28nm in production? If true that definitely deserves a "simply...wow" comment.

28nm will debut HK/MG for TSMC as well...although they stated they will make available a traditional SiON/doped-Si xtor for customers who don't want to pay for the HK/MG wafer tech.

This time next year will we be seeing HK/MG GPU's shipping out as a shrink/refresh of GT300? Seems just so ridiculously aggressive of a timetable to be true...

I don't know the finances behind this, but doesn't each node shrink require new litho gear? How can they justify doing this after only having 40 going for 9 months?

Wonder if they're thinking of making the investment now to capitalize on the incoming inflation (like I think Intel is planning on).

 

[Idontcare]

Originally posted by: soccerballtux
I don't know the finances behind this, but doesn't each node shrink require new litho gear? How can they justify doing this after only having 40 going for 9 months?

Wonder if they're thinking of making the investment now to capitalize on the incoming inflation (like I think Intel is planning on).

No that's not true, you get a good 2nodes (or more) out of leading edge litho while it is still leading edge. 193nm dry litho scanners worked just fine for printing ever-smaller critical features for three nodes for example.

Then the litho equipment continues to be used for successive nodes for those masksets for which the minimum features are large enough to still be printable with the litho equipment in question.

For instance at 32nm Intel will still use their existing/installed 193nm dry litho for many many mask sets, immersion litho will only be used for maybe 5 or 6 masksets. And for the higher-up metal levels they will actually use even older litho tech, 248nm and 365nm.

The right tool for the job, not every job needs immersion litho even at 28nm.

It's a common misperception that new nodes mean new equipment across the board. 75-80% of the equipment installed and used for 45nm will be re-used for 32nm, just as a good 75-80% of the equipment used in 45nm came from existing 65nm processes.