EETimes: TSMC estimates they hold 90% of the world's pending 28nm tape-outs

[jimbo75]

Do you realize that 32nm/28nm are basically the same? They are just marketing it differently this time.

And a year and a half later, lets not forget how 28nm was coming in H1 2010.

http://www.xbitlabs.com/news/other/...s_32nm_28nm_Process_Technologies_Roadmap.html

It's "only" a year late I guess. That's assuming it's actually in production now and it's not just more lies from them.

 

[jimbo75]

Yes, it's all part of a huge conspiracy that nobody mentioned it in this thread...

Is it not relevant? Surely if we are to believe what TSMC is stating, their previous history regarding new nodes should be taken into consideration especially before claims of GF's utter defeat are mentioned?

I know who the smart money is on.

 

[Terzo]

It's pretty cool reading about these kinds of things, even if I understand half of it at best. Keep the stories coming. I'd really like to hear more about Bulldozer. Even though Computex/E3 is only a month away, it seems like an eternity.

 

[Martimus]

Both GF and TSMC canceled their bulk 32nm node. They both gave the exact same reason: lack of interest (customers) in the node to make it worth pursuing versus going directly to the 28nm node.

http://news.softpedia.com/news/Globalfoundries-32nm-Bulk-Process-Canceled-139063.shtml
http://www.xbitlabs.com/news/other/...ies_Scraps_32nm_Bulk_Fabrication_Process.html
http://www.cdrinfo.com/sections/news/Details.aspx?NewsId=26392

It is entirely possible that there were other more prevalent reasons for canceling the 32nm node, but the fact remains that both foundries canceled their 32nm bulk production plans.

 

[piesquared]

Do you realize that 32nm/28nm are basically the same? They are just marketing it differently this time.

"Basically the same" means exactly zero if it causes products to be canceled and roadmaps to be altered.

 

[Tuna-Fish]

Can you by any chance show me a link to where AMD says they are moving the 28nm Bobcat variant to GF? I don't think that is out of the realm of possibility, but I thought I read a quote from AMD that they were going to produce the 28nm variant at TSMC. I don't have that handy, but I am hoping that you have your quote handy so you can set my thoughts on the matter straight.

Wasn't bobcat supposed to be a completely synthesized design, making switching fabs (relatively) very cheap? I don't understand that much about the costs of moving chips between fabs, but if they are low enough, I'd expect at least a test run on both to hedge the risk.

Also, what exactly is known about the relative qualities of the GloFo and TSMC processes? As far as I understood it, gate first/last should make TSMC chips faster but GloFo chips denser and leak less? Or did I get it all wrong?

Have anyone published for a SRAM macro on GloFo 28nm high performance? According to Kanter, TSMC is at .130/.127 um^2 for high and low power, respectively, while GloFo 28nm low power is at .120, with no data for high performance.

 

[Martimus]

Do you realize that 32nm/28nm are basically the same? They are just marketing it differently this time.

That isn't exactly true. TSMC’s 32nm bulk fabrication process did not feature HKMG and was largely considered a shrink of the company’s 40nm fabrication process. TSMC's 28nm bulk fabrication process was always planning on using Gate Last HKMG (well originally they planned to use "gate-first" but changed to "gate-last" probably about 2 years ago).

(Source: http://www.xbitlabs.com/news/other/...ies_Scraps_32nm_Bulk_Fabrication_Process.html)

 

[busydude]

"Basically the same" means exactly zero if it causes products to be canceled and roadmaps to be altered.

Look at post #29. TSMC did not go for 32nm cos it did not make any economic sense to them.

 

[busydude]

That isn't exactly true. TSMC’s 32nm bulk fabrication process did not feature HKMG and was largely considered as shrink of the company’s 40nm fabrication process. TSMC's 28nm bulk fabrication process was always planning on using Gate Last HKMG (well originally they planned to use "gate-first" but changed to "gate-last" probably about 2 years ago).

(Source: http://www.xbitlabs.com/news/other/...ies_Scraps_32nm_Bulk_Fabrication_Process.html)

Thanks for the correction. Will stay out of this thread.... cos it is beyond my capability.

 

[Martimus]

Wasn't bobcat supposed to be a completely synthesized design, making switching fabs (relatively) very cheap? I don't understand that much about the costs of moving chips between fabs, but if they are low enough, I'd expect at least a test run on both to hedge the risk.

Also, what exactly is known about the relative qualities of the GloFo and TSMC processes? As far as I understood it, gate first/last should make TSMC chips faster but GloFo chips denser and leak less? Or did I get it all wrong?

Have anyone published for a SRAM macro on GloFo 28nm high performance? According to Kanter, TSMC is at .130/.127 um^2 for high and low power, respectively, while GloFo 28nm low power is at .120, with no data for high performance.

Gate-last should provide performance benefits as it allows you to customize the characteristics on each gate, and allows more exotic materials to be used since the materials don't need to survive the entire manufacturing process (they are added later in the process).

Gate-first should be more dense than Gate-last though (since the gates don't need to be removed, then filled with the more exotic HKMG material), and far cheaper to implement per wafer (for the same reason).

IDC is the resident expert on the matter, so he would likely give you a better overview on the differences. I will go back into my notes and try to write out a better explanation as well.

As for the other question, yes, Bobcat was supposedly entirely synthesized on the core level, so you are likely right that it would be easier to convert to a different process. Although having to buy two sets of masks (one for TSMC and one for GF) at ~$10M per mask might be an issue with them unless they have a relatively large volume of parts.

 

[Phynaz]

Why is it that the people with the low post counts are the ones derailing CPU threads...

It's not the post counts, it's the invasion from the Zone.

 

[piesquared]

Look at post #29. TSMC did not go for 32nm cos it did not make any economic sense to them.

I'm looking at your post that states 28nm and 32nm are basically the same. Again, that means absolutely zero if it causes entire product lines and roadmaps to be altered. And sticking with the theme of the topic, it certainly doesn't add to their credibility.

 

[Idontcare]

Because somebody has to counter the intel fanboy crap that is so prevalent in this forum, obviously.

Ah, the old "the ends justifies the means" approach that the rest of us just call trolling and thread-crapping.

Why would ANYBODY have confidence in TSMC? Take a look at older roadmaps when 28nm was already out 6 months ago according to them. TSMC is full of it.

Because outside of 40nm they have a stellar record in the industry, their marketshare proves this. Do you think they have customers for no good reason?

To understand what happened with 40nm at TSMC you have to understand some of the history of TSMC.

First the founder of TSMC, Dr. Morris Chang, stepped down and ceded control of the company to Rick Tsai in 2005. This is key because Rick had a pivotal role in the early of development stages for 45/40nm.

Second was that Dr. S.Y. Chiang left TSMC in 2006, he was the head of R&D process development for many years (a decade) and was also responsible for the sustained node cadence that got TSMC the customers they have. His absence and subsequent change in who directed R&D no doubt had its impact on 45/40nm as well as 32nm.

So 40nm was the first node at TSMC to be managed under new corporate management and developed under new R&D management.

Why is 28nm expected to be different?

Firstly, Morris Chang came back and replaced Rick Tsai in 2009 after seeing how badly things were going since he left in 2005. In time to impact the development of 28nm.

Secondly, Chang brought back Chiang almost immediately in late 2009, but still in time to impact 28nm.

So what happened at TSMC which accounts for the 40nm debacle is a changing of the guard at both the corporate as well as the R&D level, and this was reversed on both accounts in hopes of 28nm not replicating 40nm's woes.

That breeds confidence, which translates into customers, which TSMC is observing.

Nobody in the entire industry has any confidence in TSMC.

Utter bollocks. Your hyperbole and rhetoric may play well in forums but it has no basis in reality in the industry.

However, GF needs to PROVE they can do it first and IDC knows this, which is why this entire thread is just more intel fanboy crap.

I've got more friends, great people, working at GloFo than working at TSMC. My heart goes out to my friends, I've been pulling for them and their families, many of whom I've been in their house and dined with, and if you read my posts from the past years you will see that I've been staunchly hopeful that they would come out of the gate on 28nm roaring like a tiger.

But it is a sad harsh reality that it just isn't happening, according to TSMC, and the ball is in GloFo's court to prove them wrong.

There's business and there's pleasure, I've got horses in both races. I was hoping everyone would get to eat from 28nm, a 50/50 split of the TAM. That is not looking to be a reasonable happy ending at this time.

 

[busydude]

I'm looking at your post that states 28nm and 32nm are basically the same. Again, that means absolutely zero if it causes entire product lines and roadmaps to be altered. And sticking with the theme of the topic, it certainly doesn't add to their credibility.

I was mistaken and Martimus corrected it. But you have to keep in mind that Nvidia and AMD are not the only customers of TSMC .. and if customers did not show any interest in their 32nm process.. why would they go ahead with the plans? I don't think their credibility is at stake.

 

[Phynaz]

Thanks for the background story IDC.

 

[Idontcare]

I was mistaken and Martimus corrected it. But you have to keep in mind that Nvidia and AMD are not the only customers of TSMC .. and if customers did not show any interest in their 32nm process.. why would they go ahead with the plans? I don't think their credibility is at stake.

TSMC approached their 32nm customers and told them that the costs to develop 32nm would be $XYZ and that the customers should expect this development cost to translate into a $/wafer cost of $ABC. The 32nm customers at the time collectively said "then don't bother, just double down and focus on 28nm, we'll do the same".

It was purely a business decision made by all parties involved based on simple accounting. There were no jilted lovers left standing at the alter on this one, contrary to the drama that gets threaded into the stories that are bandied about on popular rumor-mongering sites.

 

[jimbo75]

Ah, the old "the ends justifies the means" approach that the rest of us just call trolling and thread-crapping.

Oh come on you started the thread with comments which you must have known would be construed as trolling AMD fanboys.

Because outside of 40nm they have a stellar record in the industry, their marketshare proves this. Do you think they have customers for no good reason?

They have never had any real competition. UMC and Chartered? That's not competition.

So what happened at TSMC which accounts for the 40nm debacle is a changing of the guard at both the corporate as well as the R&D level, and this was reversed on both accounts in hopes of 28nm not replicating 40nm's woes.

That breeds confidence, which translates into customers, which TSMC is observing.

Let's be clear here. TSMC's current node is the laughable chamber-mismatch 40nm bulk. GF's is 32nm SOI HKMG.

Are you seriously telling me that customers have more confidence in TSMC, considering how awful their current node is?

But it is a sad harsh reality that it just isn't happening, according to TSMC, and the ball is in GloFo's court to prove them wrong.

As much as you'd love to believe it's about lack of confidence or some disaster on GF's part, it's a lot less dramatic. On top of the obvious one which I mentioned earlier, there is simply no way GF could currently manage the capacity for TSMC's customers.

That 90% figure is bollocks btw. They have 89 tapeouts meaning only 8 tapeouts for the rest of the industry? While AMD has 4 at GF, meaning only 4 for the entire rest of the industry?

Actually according to your link she did say "pending tapeouts". Maybe they aren't including the ones GF already have taped out.

 

[Phynaz]

While AMD has 4 at GF

Could you provide a link to AMD stating they have 28nm devices that taped out (that are being manufactured at GloFo)?
Thanks.

 

[Bearach]

Oh come on you started the thread with comments which you must have known would be construed as trolling AMD fanboys.

They have never had any real competition. UMC and Chartered? That's not competition.

Let's be clear here. TSMC's current node is the laughable chamber-mismatch 40nm bulk. GF's is 32nm SOI HKMG.

Are you seriously telling me that customers have more confidence in TSMC, considering how awful their current node is?

As much as you'd love to believe it's about lack of confidence or some disaster on GF's part, it's a lot less dramatic. On top of the obvious one which I mentioned earlier, there is simply no way GF could currently manage the capacity for TSMC's customers.

That 90% figure is bollocks btw. They have 89 tapeouts meaning only 8 tapeouts for the rest of the industry? While AMD has 4 at GF, meaning only 4 for the entire rest of the industry?

Actually according to your link she did say "pending tapeouts". Maybe they aren't including the ones GF already have taped out.

These claims are TSMC not IDC, so I don't know why you're attacking the OP. Sure, their claims could be misleading, but that is now up to GloFo to rectify. It's good for TSMC to announce this, it fortifies their position, and increases confidence.

I do find it hard to believe they have 90% of the market, but they could possibly have that. Till we have figures, lists of contracts and wins for both foundries we won't know for certain. Think of this as TSMC sending a arrow across GloFo's bow, a taunt as it were.

Also, it's not about AMD or Intel, this is about GloFo and TSMC.

 

[jimbo75]

These claims are TSMC not IDC, so I don't know why you're attacking the OP. Sure, their claims could be misleading, but that is now up to GloFo to rectify. It's good for TSMC to announce this, it fortifies their position, and increases confidence.

Well I think "attacking" is a bit strong. I would seriously question what his motives are however, considering what he knows about the industry he doesn't seem to have much common sense when it comes to logically figuring out why TSMC has more customers than GF.

Do you think it's because GF is failing at 28nm or because GF is actually FAR smaller than TSMC is at present?

I do find it hard to believe they have 90% of the market, but they could possibly have that. Till we have figures, lists of contracts and wins for both foundries we won't know for certain. Think of this as TSMC sending a arrow across GloFo's bow, a taunt as it were.

Also, it's not about AMD or Intel, this is about GloFo and TSMC.

GF announced $5.4 billion of spending a few months ago, based on customer requests.

http://www.bloomberg.com/news/2011-...ble-spending-on-plants-equipment-in-2011.html

Does that sound to you like a company that is struggling for orders? Again, I would have to seriously question the OP's motives because he should know this.

 

[Chiropteran]

90% of 28nm tapeouts!?

If this is true then this is nothing short of a total landslide defeat for GloFo at 28nm

As someone who rarely ever pays attention to fabs, I read this completely differently from the way the thread has progressed. It sounded to me like TSMC has 90% of the PENDING 28nm work because the process was backed up so long that the orders are piling up and can't be filled. Which doesn't sound like a positive, it sounds like a negative... they have work they can't fill and those companies with work waiting around are going to go elsewhere if they want to get it done in a timely manner.

 

[Bearach]

Well I think "attacking" is a bit strong. I would seriously question what his motives are however, considering what he knows about the industry he doesn't seem to have much common sense when it comes to logically figuring out why TSMC has more customers than GF.

Do you think it's because GF is failing at 28nm or because GF is actually FAR smaller than TSMC is at present?



GF announced $5 billion of spending a few months ago, based on customer requests.

http://www.bloomberg.com/news/2011-...ble-spending-on-plants-equipment-in-2011.html

Does that sound to you like a company that is struggling for orders? Again, I would have to seriously question the OP's motives because he should know this.

Probably was a bit strong, sorry. I agree in terms of a foundry they are FAR smaller. I would think that GloFo, still has some growing to do. We are talking about a new foundry. Nevertheless this announcement by TSMC will hurt GloFo, and it does give them a problem.

I believe all IDC's intentions were, was to notify people of the announcement, it is interesting that TSMC can claim such a high percentage of the market. I don't think his intentions were to stir up any body in regards to brand. He also states "if true", meaning he hasn't accepted it as fact yet.

 

[Martimus]

Oh come on you started the thread with comments which you must have known would be construed as trolling AMD fanboys.

As much as you'd love to believe it's about lack of confidence or some disaster on GF's part, it's a lot less dramatic. On top of the obvious one which I mentioned earlier, there is simply no way GF could currently manage the capacity for TSMC's customers.

That 90% figure is bollocks btw. They have 89 tapeouts meaning only 8 tapeouts for the rest of the industry? While AMD has 4 at GF, meaning only 4 for the entire rest of the industry?

Actually according to your link she did say "pending tapeouts". Maybe they aren't including the ones GF already have taped out.

I would appreciate it if you stopped attacking Idontcare. As much as it is against the rules fo this forum to state it here, and not just report this post, I feel the need to stand up for my friend in public. He doesn't deserve to be treated this way.

He in no way wishes that Global Foundries will fail, and your insistance that he does is befuddling. He believes that based on the information that is available, they are doing a poor job of selling their 28nm manufacturing to customers. That doesn't mean he wants them to do poorly, it just means that he thinks they are doing poorly. I can tell you that I know he wants them to succeed, so I know that your insistance that he hates them is completely baseless.

As shown by an earlier chart, TSMC had 4 times the revenue of GF last year. GF was third by a slight margin behind UMC. For all intents and purposes UMC and GF had about the same revenue last year. Although UMC lost perhaps their biggest costumer for 28nnm in Xilinx to USMC (http://www.eetimes.com/electronics-news/4087890/Xilinx-confirms-Samsung-TSMC-in-UMC-out-at-28-nm) so they seem to be hurting for custers on their 28nm node as well.

I do really like their hybrid approach to HKMG though. They actually run a gate-first process, with HKMG material tailored to n-type gates, then remove the material at the p-type gates and replace it with p-type tailored HKMG material. (IIRC, I can't seem to open any of the articles about this from behind my firewall at the moment).

 

[Janooo]

Could you provide a link to AMD stating they have 28nm devices that taped out (that are being manufactured at GloFo)?
Thanks.

http://www.xbitlabs.com/news/other/..._Expects_Two_28nm_Tape_Outs_This_Quarter.html

 

[Phynaz]

http://www.xbitlabs.com/news/other/..._Expects_Two_28nm_Tape_Outs_This_Quarter.html

Key words: "Expects", not "have".

Also they don't state what they are and where they will be manufactured - although we can guess it will be GloFo.