TSMC Shows Path to 16nm, Beyond

[ShintaiDK]

TSMC 16nm is essentially just 20nm with finfets.

 

[khon]

Here's a newsflash in case people haven't caught on yet, the XX nm numbers that companies quote have no meaning anymore. It's not an actual physical measurement of anything.

You cannot compare two companies by saying company A produces XX nm while company B produces YY nm, it's not an objective parameter.

I'm in the industry, and I know for a fact that one company decided to simply rebrand XX nm as YY nm, even though they did not do a process shrink, they just thought it would be better for marketing reasons.

 

[Homeles]

I think all of us know that, at this point. Idontcare has hammered that in for years now

 

[Nothingness]

An easy to read article about that: http://spectrum.ieee.org/semiconductors/devices/the-status-of-moores-law-its-complicated

 

[Fjodor2001]

Update from Oct 30, 2013:

http://semiaccurate.com/2013/10/30/tsmc-shows-production-20nm-16nm-dev-wafers/

So by how much does TSMC's 16 nm trail Intel's 14 nm? About 1 year or so?

No, from what I've read - TSMC's 16nm will be no where near Intel's 14nm in terms of density and electrostatics. Of course, TSMC will do just fine - their real competition is Samsung and GFL, who are pretty much following the same path but still up river compared to TSMC.

Just to clarify, I was not comparing 16 nm vs 14 nm process technology. Instead I was comparing the time difference between when the production using those two process technologies are expected to begin according to what has been communicated.

And the article says that TSMC will begin (risk) production of 16 nm in late 2014. Intel has previously said it will begin production of 14 nm in Q1 2014. So that's less than a 1 year difference, and it could be as little as around 6 months depending on if it's the beginning or end of the quarters they are talking about.

Intel did not state if they meant "risk" production or not though when they communicated the Q1 2014 timeframe. But since they are still having yield problems (they have said they "think" the problems have been solved), or have just solved them, I guess their 14 nm production should be quite "risky" too in Q1 2014 which is just a few months from now.

 

[Meghan54]

Intel did not state if they meant "risk" production or not though when they communicated the Q1 2014 timeframe. But since they are still having yield problems (they have said they "think" the problems have been solved), or have just solved them, I guess their 14 nm production should be quite "risky" too in Q1 2014 which is just a few months from now.

Not this profanity removedagain. Are you really being this obtuse or are you just genuinely just stupid? I'd let you plead ignorance, but you seem to have at least a tenuous grasp of the subject matter, so ignorance isn't an avenue of escape.

So, maybe you just are so obtuse and willfully stupid. Only answer I can come up with, considering the last thread you used the "think" profanity removed.




If you have an issue with a poster, use the report button. You do not get to attack him in the thread.

Also, you cannot use profanity in the tech side of these forums


esquared
Anandtech Forum Director

 

[Homeles]

Intel did not state if they meant "risk" production or not though when they communicated the Q1 2014 timeframe. But since they are still having yield problems (they have said they "think" the problems have been solved), or have just solved them, I guess their 14 nm production should be quite "risky" too in Q1 2014 which is just a few months from now.

Risk production is a term that is essentially inapplicable to Intel. Also, I believe that they specifically stated that Broadwell would be what is going into production in Q1, which is absolutely a product that would be ramped during volume production, and no earlier.

 

[Ajay]

Can we read it too?

I'm not sure what sources he's read, but it's essentially common knowledge. TSMC and the other foundries can't keep up with Intel's metal pitches. GloFo and TSMC will be using their 20nm BEOL on their 14nm and 16nm processes, respectively (likely to keep costs down).

I read from a variety of sources on the web, and hence can't recall specific articles from trade mags or threads from places like RWT or even here on AT. Sorry, I haven't figured out what to do with information overload yet

 

[Ajay]

Risk production is a term that is essentially inapplicable to Intel. Also, I believe that they specifically stated that Broadwell would be what is going into production in Q1, which is absolutely a product that would be ramped during volume production, and no earlier.

Hmm, that brings up and interesting point. Pure fabs can sell some of their wafers during ramp up as risk production. And IDM like Intel has to 'eat' all those wafers (can they be written off?).

 

[Homeles]

Hmm, that brings up and interesting point. Pure fabs can sell some of their wafers during ramp up as risk production. And IDM like Intel has to 'eat' all those wafers (can they be written off?).

They're not going to be running those things 24/7. The raw materials cost and electrical costs would be huge, for no purpose. The only thing that Intel would be using those fabs for during that phase, other than fine tuning the process, would be producing engineering samples and the like.

At least that's how I'd imagine it is.

 

[Khato]

Intel did not state if they meant "risk" production or not though when they communicated the Q1 2014 timeframe. But since they are still having yield problems (they have said they "think" the problems have been solved), or have just solved them, I guess their 14 nm production should be quite "risky" too in Q1 2014 which is just a few months from now.

Remember the articles from about a year ago about Intel taping out Broadwell? That's what should be compared against TSMC's risk production schedule. (For comparison, TSMC started risk production of 20nm at the start of this year roughly.)

 

[CHADBOGA]

Just to clarify, I was not comparing 16 nm vs 14 nm process technology. Instead I was comparing the time difference between when the production using those two process technologies are expected to begin according to what has been communicated.

And the article says that TSMC will begin (risk) production of 16 nm in late 2014. Intel has previously said it will begin production of 14 nm in Q1 2014. So that's less than a 1 year difference, and it could be as little as around 6 months depending on if it's the beginning or end of the quarters they are talking about.

Do you really think we are going to see widespread 16nm products fabbed out of TSMC within 18 months of when Intel's 14nm products are widely available?

It appears you do, which just shows how much difficulty you have in separating what you wish for and what surely you must know(if you have been paying attention for the last decade) will be the case.

 

[Exophase]

Do you really think we are going to see widespread 16nm products fabbed out of TSMC within 18 months of when Intel's 14nm products are widely available?

It appears you do, which just shows how much difficulty you have in separating what you wish for and what surely you must know(if you have been paying attention for the last decade) will be the case.

Why is it so unfathomable that TSMC's 16nm could take less time to come out than a full node transition usually does, seeing as how it's not a full node transition? And not comparable to Intel's 14nm...

 

[Homeles]

Why is it so unfathomable that TSMC's 16nm could take less time to come out than a full node transition usually does, seeing as how it's not a full node transition? And not comparable to Intel's 14nm...

I agree. I'd imagine reduced time to market was a significant deciding factor in "choosing" to go with the 20nm BEOL.

 

[Phynaz]

Hmm, that brings up and interesting point. Pure fabs can sell some of their wafers during ramp up as risk production. And IDM like Intel has to 'eat' all those wafers (can they be written off?).

They don't have to eat them if it's worth while to harvest the chips.

But Intel doesn't do risk production. Look up their "copy everywhere" manufacturing strategy. Intel doesn't release a manufacturing process to production until it's fully baked.

 

[Phynaz]

Why is it so unfathomable that TSMC's 16nm could take less time to come out than a full node transition usually does, seeing as how it's not a full node transition? And not comparable to Intel's 14nm...

Because FinFet's are very, very difficult to get right. You're talking about atomic level manufacturing in three dimensions.

 

[Homeles]

Because FinFet's are very, very difficult to get right. You're talking about atomic level manufacturing in three dimensions.

18 months after Intel releases Broadwell is a big stretch, though.

 

[ShintaiDK]

We all know we got 28nm GPUs and CPUs in 2010, because thats when TSMC had "risk production"

 

[Exophase]

Because FinFet's are very, very difficult to get right. You're talking about atomic level manufacturing in three dimensions.

And I'm sure that, like Intel, they've been researching FinFET development in parallel with the development of several nodes. That sort of explanation isn't useful in trying to estimate how much time will elapse between 20nm and 16nm.

 

[Exophase]

Re: the whole risk production thing and estimating the time from it to release, frankly I think S|A's story doesn't add up at all with what TSMC actually said:

“We will begin volume production of 20nms in the first quarter 2014. That's 90 days from now. 16nm will follow 20nm in one year. We view both 20nm and 16nm as virtually one node,” said Morris Chang, chief executive officer and chairman of TSMC, during a conference call with financial analysts.

"On 16 FinFET, technological development is progressing well. Risk production is on schedule by the end of this year."

http://www.xbitlabs.com/news/other/..._Regarding_16nm_FinFET_and_20nm_Progress.html

Now normally I give Anton Shilov even less credence than Charlie, but I'm willing to give him the benefit of the doubt that he isn't being completely fraudulent with actual quotes, especially these quite lengthy ones which would have to be have been pretty thoroughly imagined if not true. And frankly, I don't even think Shilov could manage to convincingly make up something like this.

So yeah, sounds like terrible reporting on Charlie's part, and that in fact:

a) You can't actually buy 20nm products now, months before the most optimistic estimations, and instead they'll end volume production Q1 2014, meaning they'll probably be out Q2 2014 like TSMC has been saying forever
b) FinFET will follow volume, not RISK production a year later, ie Q1 2015. TSMC has for a while said FinFET risk production will begin 2013 and has just now reasserted it, making a late 2014 start anything but "on schedule." That's also in line with the fact that TSMC taped out a Cortex-A57 test chip on 16nm earlier this year.

 

[CHADBOGA]

Why is it so unfathomable that TSMC's 16nm could take less time to come out than a full node transition usually does, seeing as how it's not a full node transition? And not comparable to Intel's 14nm...

If it was a full node transition, I suspect it would be 24 to 30 months later.

It seems to me that the foundries are struggling to introduce new processes that ramp properly, within the same time frames as Intel has been doing.

TSMC might show off a product at some point to claim they have their new process out, but just as it took forever for 28nm to be widely available, I expect they will have similar problems with 20nm and certainly with what they call 16nm.

 

[Soulkeeper]

http://spectrum.ieee.org/semiconductors/devices/the-status-of-moores-law-its-complicated
http://spectrum.ieee.org/tech-talk/...life-extension-tech-goes-into-full-production

I just read those two articles, some interesting stuff there.

 

[Exophase]

If it was a full node transition, I suspect it would be 24 to 30 months later.

It seems to me that the foundries are struggling to introduce new processes that ramp properly, within the same time frames as Intel has been doing.

TSMC might show off a product at some point to claim they have their new process out, but just as it took forever for 28nm to be widely available, I expect they will have similar problems with 20nm and certainly with what they call 16nm.

In my book Krait SoCs qualified as "widely available" early 2012. Just because you had a hard time buying a GCN or Kepler GPU or whatever doesn't mean that 20nm is only going to be out on some nothing volume product for a long time. TSMC themselves said that 20nm will have a more aggressive ramp than 28nm, I'm sure that claim isn't coming from nowhere but you must think it is.

You need to put some context behind their claims instead of just saying everything's a lie and everything will be at least 2 years later than when they say it'll be. If they've been continually hitting their milestones and continually keeping to the same schedule in public reports you should expect their schedule to be more and more accurate (or the likely slip to be less and less). There's a difference between being 2 years off on a roadmap made 5 years in advance and made 1 year in advance. Do you think that their 90 days schedule estimate for volume production of 20nm is also off by 1-2 years still?

Right now TSMC is still better qualified to estimate their schedule than lay people saying that they must take longer because FinFETs are hard.

 

[bullzz]

http://www.theverge.com/2013/11/6/5072242/samsun-2014-phones-will-have-78-percent-more-pixels

samsung roadmap mentions their AP will be on 3d finfet in 2014. either they are using intel as foundry or tsmc has cracked this. interesting

 

[Soulkeeper]

http://www.theverge.com/2013/11/6/5072242/samsun-2014-phones-will-have-78-percent-more-pixels

samsung roadmap mentions their AP will be on 3d finfet in 2014. either they are using intel as foundry or tsmc has cracked this. interesting

IBM, samsung, and Global foundries are collaborating (each having their own cutting edge manufacturing capacity)
http://www.commonplatform.com/about/

dunno, maybe the exynos6 will see finfets ?
some talk of it here http://www.dailytech.com/Report+Sam...+Exynos+6+Chip+for+Galaxy+S5/article33645.htm