[S|A] TSMC rumored to have killed its 32nm node

Aerwidh · 11-26-2009, 10:00 AM

Quote:

WORD JUST IN from the far east says that TSMC may have canceled its 32nm process node. On the surface this might seem very bad, but it is much more nuanced than that.

Update: The knifing of 32 has been confirmed by multiple sources.

SemiAccurate has not heard it directly from TSMC yet, but given the circumstances, we believe this report to be accurate. As we pointed out earlier in the week, some of the 32nm processes at TSMC are delayed and somewhat underperforming. Given how well 40nm has shaped up, or not, the earlier news is likely only the tip of the iceberg. As far as we can tell, this will be the first time TSMC has canceled a node entirely.

AliG · 11-26-2009, 10:12 AM

Hmm. Big win for ati? I mean if ati can have global foundries fab their gpus at 32nm, while NV is stuck at 40nm, this could potentially drastically change control of the market.

Helloworld_98 · 11-26-2009, 10:35 AM

major win for GF and ATi

Mescalamba · 11-26-2009, 10:46 AM

Major loose for everyone who likes "healthy market" (eg. everyone whos not stupid). But I won´t believe in that until its really confirmed... just rumor.

Piotrsama · 11-26-2009, 10:47 AM

Read the article guys, TSMC is killing the 32nm process in favor of the 28nm.

Eson · 11-26-2009, 10:57 AM

Quote:

Originally Posted by Piotrsama

Read the article guys, TSMC is killing the 32nm process in favor of the 28nm.

If its not in the quote - it never happend.

AliG · 11-26-2009, 11:06 AM

Quote:

Originally Posted by Piotrsama

Read the article guys, TSMC is killing the 32nm process in favor of the 28nm.

Yes but the time frame won't change much, not to mention skipping nodes is generally a bad idea. Why do you think with as much money Intel has, they don't skip nodes?

eric66 · 11-26-2009, 11:06 AM

and why this news don't suprise me. given tmscs 40 nm late success one can also expect another triumph from them :P

AliG · 11-26-2009, 11:23 AM

I disagree eric. 40nm had its problems, but those were minor issues (relative, of course) with calibration as I understand. If TSMC has to scrap 32nm entirely, that means their issues are such that the overall cost of fixing it or releasing a very late product outweighs the revenue.

In other words, these must be some serious problems. That's a lot of potential money lost by waiting until 28nm. Because 28nm is a half node of 32nm, there's a good chance they'll be using the same tools. If that's the case, then you can compare this case to AMD's 65nm. It was garbage, had lots of leakage problems and no one could overclock the phenoms worth a damn. That's because they were using a lot of their 90nm tools, which simply were not fit to do the job. Either TSMC must feel they'll have enough time to fix the issues by 28nm or make entirely new tools.

Piotrsama · 11-26-2009, 11:45 AM

Quote:

Originally Posted by AliG

Yes but the time frame won't change much, not to mention skipping nodes is generally a bad idea. Why do you think with as much money Intel has, they don't skip nodes?

Again, going by what the article says, 32nm was skipping and approaching 28nm dates.
Also, TSMC's consumers weren't very interested on 32, but were on 28.

Intel does not use half nodes (for CPUs at least), your comparison is not valid because missing a node for Intel would be way more significant.

AliG · 11-26-2009, 11:48 AM

Regardless of whatever comparison you like, you use the same tools for 28nm as 32nm (this true for all half nodes), meaning that if there's a problem with the 32nm tools, it will affect the 28nm product.

safan80 · 11-26-2009, 03:10 PM

How can they do 32nm when they can barely even handle 40nm?

Sadasius · 11-26-2009, 03:33 PM

Quote:

Originally Posted by safan80

How can they do 32nm when the can barely even handle 40nm?

Exactly what I was thinking.

Machinus · 11-26-2009, 03:44 PM

Quote:

Originally Posted by AliG

Yes but the time frame won't change much, not to mention skipping nodes is generally a bad idea. Why do you think with as much money Intel has, they don't skip nodes?

I think smaller companies benefit more from taking the risk of skipping nodes. A large company like intel has a huge opportunity cost to not saturate every market all the time. They need incremental advantages to feed their branding and OEM contracts. If you wanted to challenge intel, you would not fight them at their game - you would take the opportunity to cancel a node with low profit projections and get a head start on the next one, with the hope of beating them to market by a year.

Sadasius · 11-26-2009, 03:53 PM

Quote:

Originally Posted by Machinus

I think smaller companies benefit more from taking the risk of skipping nodes. A large company like intel has a huge opportunity cost to not saturate every market all the time. They need incremental advantages to feed their branding and OEM contracts. If you wanted to challenge intel, you would not fight them at their game - you would take the opportunity to cancel a node with low profit projections and get a head start on the next one, with the hope of beating them to market by a year.

That is grabbing a really big bull by the horns while wearing a red gown with no underwear. It's grabbing a lot of trouble and getting ready to get raped.

highoctane · 11-26-2009, 04:12 PM

I'm guessing on that the release of 32nm and 28nm are close enough together release wise along with customer demand for the smaller node that they're going where the money's is.

Quote:

In the end, the truth is probably closer to the happy version. 32nm has weak uptake everywhere other than the CPU world, and TSMC is not the only one with that problem. 28nm is very close behind, and that is where the customers and volume are, again, industry wide.

highoctane · 11-26-2009, 04:15 PM

Quote:

Originally Posted by AliG

Regardless of whatever comparison you like, you use the same tools for 28nm as 32nm (this true for all half nodes), meaning that if there's a problem with the 32nm tools, it will affect the 28nm product.

On the other side of the coin 32nm might have shown good preliminary results which makes the move to 28nm possible in a timely manner, meaning that the 32nm tools are rockin and 28nm shouldn't be a problem now.

Manicdan · 11-26-2009, 04:51 PM

how many nodes did intel skip for their IGPs? they are still producing some at 130nm right?

RADCOM · 11-26-2009, 05:14 PM

Quote:

Originally Posted by highoctane

On the other side of the coin 32nm might have shown good preliminary results which makes the move to 28nm possible in a timely manner, meaning that the 32nm tools are rockin and 28nm shouldn't be a problem now.

I like this ....the glass is 1/2 full attitude

zalbard · 11-27-2009, 05:54 AM

Well, gfx cards often use different node sizes compared to CPUs, no big news here.