Global Foundries delays 32nm SOI

[Phynaz]

45nm bulk also slips a quarter

 

[ilkhan]

whats the "bulk" process used for?
And: so looks like 1H 2011 at the earliest for AMD 32nm? Or even later?

 

[drizek]

video cards, usually. I don't think GF is producing any of those though.

 

[Fox5]

Originally posted by: drizek
video cards, usually. I don't think GF is producing any of those though.

A lot more than just video cards. Video cards could be a target for AMD's SOI process anyway.

And I'd bet on ATI (and possibly nvidia) transitioning to GF as it should be superior to TSMC, at least on performance. The time frame is right for ati to have designed a gpu from the ground up for GF's process by the time 45nm launches.

 

[ilkhan]

I think you mean 32nm? They're already launching 40nm from TSMC.

 

[Idontcare]

Originally posted by: ilkhan
whats the "bulk" process used for?

Basically all those chips that TSMC/UMC/Chartered/SMIC sell are being sold to bulk process customers.

Globalfoundries wants to be able to court that business, and not all of it (in fact a rather small percentage of it) is leading edge business.

Having a 45/40nm bulk Si process available next year still gets you (the foundry, not you you the person) early enough to market that you get to court after probably 90-95% of the 40nm volume that will eventually develop in the foundry space.

Checkout slide 8 (revenue by technology) of TSMC's 2Q09 earnings presentation to see what I'm talking about.

All that existing revenue volume you see there at 55nm and larger is potential customer revenue for 40nm (for TSMC, for UMC, for GF, for any foundry at the right price) as those customers slowly migrate their product lines to the smaller process technology.

I use TSMC as the example because they report their revenue by technology in such a nice way, and they represent about 60-70% of the foundry market by revenue so they are a good proxy of what is out there for GF to "steal" away as they grow their customer base.

 

[Phynaz]

Check out this PR doublespeak...

''Our roadmap for 32-nm SOI has not slipped,'' according to a spokesman for the company. ''Yes, the timeframe for introduction has been altered slightly compared to the roadmap we showed you in July, but that is not because of any issues with the technology.

I wonder if I can use this...Boss, the project isn't late, the timeframe is slightly altered.

 

[DrBombcrater]

What GF are saying makes perfect sense, bearing in mind that 32nm SOI is only going to be used by AMD. If AMD have had to delay Bulldozer then the timeframe for initial production of 32nm chips slips even if the process is 100% ready on time, purely because there's nothing to build. 'not because of any issues with the technology' sounds like a very broad hit that this is what's happened.

 

[Idontcare]

Originally posted by: Phynaz
Check out this PR doublespeak...

''Our roadmap for 32-nm SOI has not slipped,'' according to a spokesman for the company. ''Yes, the timeframe for introduction has been altered slightly compared to the roadmap we showed you in July, but that is not because of any issues with the technology.

I wonder if I can use this...Boss, the project isn't late, the timeframe is slightly altered.

From the article:

That process, devised especially for Advanced Micro Devices Inc.'s microprocessors, was supposed to move into ''tape out'' or ''risk production'' in the first quarter of 2010. Now, the technology is targeted for ''risk production'' in the third quarter of 2010.

Am I reading that right in that this says this particular milestone (risk production of 32nm process tech for AMD) has been delayed/slipped by 2 quarters (6months give or take where in the quarters things actually were to fall)?

How that impacts AMD is unclear, although the processor maker could fall further behind Intel Corp. GlobalFoundries denied its roadmap has slipped, however.

I really don't understand...either the roadmap slipped/changed or it didn't. It can't be both. That means either GF is wrong or Mark LaPedus of EETimes is wrong in regards to either the previous roadmap or the current roadmap.

Which is it?

GlobalFoundries denied its roadmap has slipped. ''Our roadmap for 32-nm SOI has not slipped,'' according to a spokesman for the company. ''Yes, the timeframe for introduction has been altered slightly compared to the roadmap we showed you in July, but that is not because of any issues with the technology. The roadmap has simply been adjusted to align with AMD's product needs. We have solid natural yields ramping up every week, and we have high confidence in our ability to demonstrate the same robust yields and manufacturing capability on 32-nm that we have historically had.''

Tisk tisk phynaz you got me with your selective quoting ...not that you meant to do it, but yes I think I can now rationalize what GF is saying now that I've seen the other two sentences in the quote.

There is a difference between having a node ready and having an IC design ready to be produced on that node. The two happen in parallel, and is part of the reason why roadmaps (both process tech and IC) are so critical in their timeline definitions as they must be aligned or else you risk wasting resources on one or the other in terms of each reaching their milestones.

If the IC design team gets to tapeout before the process node is ready to support a risk production run then that means the IC design team was over-resourced (too much money spent) all those preceding years leading up to tape-out. That's not bad from the employee standpoint but that is bad from the project management and shareholder standpoint, so there are active checks and balances in place to minimize such over-resourcing from occurring.

(there is another equally held view on this type of outcome, and that is that the project scale and scope must not have been aggressive enough, not enough risk was taken at the outset of the project's definition and as such the company's decision makers fell short in not taking advantage of the opportunity to deliver a more aggressive product)

Vice versa if the process node development team gets the node in question to production-ready status prior to any actual customer designs being ready for production then that means the node development team was over-resourced or under-challenged.

Xtor parameters should have been targeted more aggressively for the node, metal pitch should have been scaled more aggressively (I have personal experience here, a story if you like to hear) or less R&D resources should have been allocated to that particular project as those excess resources undoubtedly came at the expense of under-resourcing other parallel development projects at the time (node N+2 development for example, also personal experience here).

What I read here is that GF's is on track to have ready/available their 32nm SOI + HKMG technology in Q1 2010 but that they are not expecting AMD to have ready a design taped out for production runs until Q3 2010. That could be viewed as meaning GF's over-resourced their 32nm SOI + HKMG development team or that they under-challenged them in their scaling metrics for 32nm's FOMs. Not ideal but really this is better than the opposite being the situation (process tech is late, ala TSMC's 45nm -> 40nm switcheroo).

 

[Phynaz]

Tisk tisk phynaz you got me with your selective quoting ...not that you meant to do it

Of course I meant to do it. :laugh:

I was just pointing out what the PR mouth was saying, "It didn't slip, it's just changed". I found that really humorous is all.

 

[Idontcare]

Yeah it definitely twisted my mind in a wonky way as I attempted to rationalize what manner of context the PR person was thinking they were addressing when making that statement.

I concluded drugs had to have been involved, and still aren't too sure I should rule them out entirely either.

 

[Idontcare]

Originally posted by: Idontcare

Originally posted by: ilkhan
whats the "bulk" process used for?

Basically all those chips that TSMC/UMC/Chartered/SMIC sell are being sold to bulk process customers.

Globalfoundries wants to be able to court that business, and not all of it (in fact a rather small percentage of it) is leading edge business.

Having a 45/40nm bulk Si process available next year still gets you (the foundry, not you you the person) early enough to market that you get to court after probably 90-95% of the 40nm volume that will eventually develop in the foundry space.

Checkout slide 8 (revenue by technology) of TSMC's 2Q09 earnings presentation to see what I'm talking about.

All that existing revenue volume you see there at 55nm and larger is potential customer revenue for 40nm (for TSMC, for UMC, for GF, for any foundry at the right price) as those customers slowly migrate their product lines to the smaller process technology.

I use TSMC as the example because they report their revenue by technology in such a nice way, and they represent about 60-70% of the foundry market by revenue so they are a good proxy of what is out there for GF to "steal" away as they grow their customer base.

ilkhan I went ahead and pulled the data and graphed it in a way that I hope better communicates why I say Globalfoundries pursuit of establishing a bulk 45/40nm process flow has opportunity to grab a healthy chunk of the global annual foundry revenues even if it isn't ready for another year yet.

Graph of TSMC's Revenue by Node over the past 12 yrs

First thing to notice is just how much existing revenue is derived from what we enthusiasts consider to be decade old technology...look at the 0.15um numbers, heck even look at how much revenue still ships on 90nm.

Now notice just how tiny of a blip 45nm has made to date in that upper right hand corner. Compare that to the tiny blip that 65nm was just 2 yrs prior. Now look at how little the revenue by node landscape changed in the 2yrs since 65nm (and 55nm) was introduced. In 2yrs that 65nm process tech at best snagged about 25-26% of the revenue for TSMC, the rest of their business was quite happy to stick with the older more mature technologies.

So how much of this revenue by node landscape can we really expect to change over the next year or two? At best it will change by the same amount as it did in the past two. Meaning whenever Globalfoundries brings their bulk 45/40nm process online there will be PLENTY of customers out there still on 65nm and older technology to go courting after as they slowly transition some of their products to 45/40nm node tech.

As for legitimizing my previous claim that analyzing TSMC revenue data is a good proxy for estimating attributes of the global foundry market as whole, here is the most recent iSuppli foundry rankings by revenue: http://www.isuppli.com/Publish...2009-06-08_Foundry.jpg

And the chart at the bottom of this edn article contains the same iSuppli data only they fleshed the table out to include annual stats as well: http://www.edn.com/article/CA6663913.html

Taking the 2008 annual data and ranking by revenue we get:

Rank.......Company.......Total 2008 revenue
..1.............TSMC.............$10,253
..2..............UMC..............$2,939
..3..........Chartered..........$1,778
..4.............SMIC..............$1,354
..5..........Vanguard...........$515
..6........DongbuHi Tek.......$433
..7..............X Fab.............$368
..8............HHNEC.............$367
..9............Tower..............$253
..10...........SSMC.............$176
................Others............$1,405
.................Total.............$19,841

TSMC just dominates the foundry business by revenue, representing more than half the entire market. Adding Globalfoundries to this list will add another $3-4B and put them in 2nd place, but almost 100% of that revenue will be AMD on 45nm SOI so it wouldn't really be much value to add them to the comparison here.