Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/more-finfet-delays.4792/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

More FinFET delays

L

lefty

Active member
The CEO of Ultratech made some interesting comments in their latest earning call. Looks like either Samsung, or TSMC are having problems with 14/16nm:
Once again a major logic manufacturer delayed their FinFET ramp.
Click to expand...

We see that occurring later or in 2015 and it’s really hard to project when. The current anticipation is they’re all being overly optimistic as to when they’re going to solve their problems. But the yields on the major companies right now is in the 10% to 20%.
Click to expand...

Ultratech's (UTEK) CEO Art Zafiropoulo on Q3 2014 Results - Earnings Call Transcript | Seeking Alpha


<script src="//platform.linkedin.com/in.js" type="text/javascript">
lang: en_US
</script>
<script type="IN/Share" data-counter="right"></script>
 
Last edited by a moderator:
It is interesting to see the equipment guys out their customers. NDA's mean nothing in this business. My guess is that it is Samsung. I had heard that 14nm silicon is not correlating with the process models. Since TSMC is in full production at 20nm and 16nm uses the same fabs/process equipment I don't foresee yield problems. As I mentioned in my blog:

https://www.semiwiki.com/forum/conte...-next-soc.html

If FinFETs are delayed TSMC is going to make a big bucket of money on 20nm. My guess is that Samsung and TSMC will both be shipping 14/16nm wafers just in time for Apple's next iProducts in 2015. That would be Q2 2015. There is just too much money at stake here and Samsung can eat bad yield if they have to without disclosing it. TSMC not so much.
 
Daniel Nenni said:
My guess is that Samsung and TSMC will both be shipping 14/16nm wafers just in time for Apple's next iProducts in 2015. That would be Q2 2015.
Click to expand...
20nm started January this year and it had just ramped enough for the iPhone6 to be ready in September. Given that TSMC say FinFET plus production starts June-ish, Apple would have to delay the launch of the iPhone 7 til December if they want it on FinFET plus.
 
Daniel Nenni said:
It is interesting to see the equipment guys out their customers. NDA's mean nothing in this business. My guess is that it is Samsung. I had heard that 14nm silicon is not correlating with the process models. Since TSMC is in full production at 20nm and 16nm uses the same fabs/process equipment I don't foresee yield problems. As I mentioned in my blog:

https://www.semiwiki.com/forum/conte...-next-soc.html

If FinFETs are delayed TSMC is going to make a big bucket of money on 20nm. My guess is that Samsung and TSMC will both be shipping 14/16nm wafers just in time for Apple's next iProducts in 2015. That would be Q2 2015. There is just too much money at stake here and Samsung can eat bad yield if they have to without disclosing it. TSMC not so much.
Click to expand...

Daniel, your comments don't stack up. TSMC is certainly not in full mass production yet with 16nm and neither is Samsung, it's their first FinFET process in both cases, either could be having a yield problem if it's in the FEOL, and neither would admit it if they were. If the yield problem is BEOL then maybe it's less likely to be TSMC because they have more 20nm experience than Samsung, but there's no proof of this. In fact reading the report, the delay could equally well be Intel (which is well-known) not TSMC or Samsung...

Intel had initial yield problems with both 22nm and 14nm FinFET processes but could hide it because of the huge margins on the CPUs they fab in them. As an IDM Samsung can hide it where they're their own customer because they can lose money on the wafers/chips but more than make it back on the phone. As a pure-play foundry it's hard for TSMC to hide such problems, except that neither they or their bleeding-edge customers are likely to admit a problem in public even if there is one.

What I've heard is that from 20nm down double-patterning (and especially FinFETs) are proving problematic for *everyone*, regardless of what's said in public -- presumably some more than others, but nobody's having a smooth ride here...
 
Last edited:
Makes sense given how much trouble Intel had (still has?) at 14nm. Considering the differences between IDM and fabless processes, I think we'll find when 14nm is said and done that Intel's lead in process technology is very much alive and intact.
 
IanD said:
Daniel, your comments don't stack up. TSMC is certainly not in full mass production yet with 16nm and neither is Samsung, it's their first FinFET process in both cases, either could be having a yield problem if it's in the FEOL, and neither would admit it if they were. If the yield problem is BEOL then maybe it's less likely to be TSMC because they have more 20nm experience than Samsung, but there's no proof of this. In fact reading the report, the delay could equally well be Intel (which is well-known) not TSMC or Samsung...

Intel had initial yield problems with both 22nm and 14nm FinFET processes but could hide it because of the huge margins on the CPUs they fab in them. As an IDM Samsung can hide it where they're their own customer because they can lose money on the wafers/chips but more than make it back on the phone. As a pure-play foundry it's hard for TSMC to hide such problems, except that neither they or their bleeding-edge customers are likely to admit a problem in public even if there is one.

What I've heard is that from 20nm down double-patterning (and especially FinFETs) are proving problematic for *everyone*, regardless of what's said in public -- presumably some more than others, but nobody's having a smooth ride here...
Click to expand...

Having just returned from Taiwan I feel my information is much more up to date than what I have read around the internet thus far. This is a very fluid situation so cutting and pasting comments from conference calls or rumor websites is not going to get you anywhere except lost. I will do a more detailed write up as soon as I can. Unfortunately I'm still digging out from my trip so it could be a day or two. In the mean time I'm very interested in what others have to say here and if you have specific questions I will do my best to get the answers.
 
If yield rate is a common problem for every foundry at 14/16nm, will Apple decide to stay with 20nm for their A9 processor? Assume Apple insists to keep the same summer/fall schedule for their annual new product announcement.
 
hist78 said:
If yield rate is a common problem for every foundry at 14/16nm, will Apple decide to stay with 20nm for their A9 processor? Assume Apple insists to keep the same summer/fall schedule for their annual new product announcement.
Click to expand...

Comparing TSMC FinFET capabilities to Samsung’s is unwise. TSMC pioneered FinFET technology and developed the 16nm process from the ground up. Samsung on the other hand licensed their 14nm FinFET process from IBM. Samsung also got the 28nm process from IBM which did not yield properly and Samung 20nm silicon is still missing in action. So don’t lump these two companies together in regards to FinFETs.
 
Daniel Nenni said:
Comparing TSMC FinFET capabilities to Samsung’s is unwise. TSMC pioneered FinFET technology and developed the 16nm process from the ground up. Samsung on the other hand licensed their 14nm FinFET process from IBM. Samsung also got the 28nm process from IBM which did not yield properly and Samung 20nm silicon is still missing in action. So don’t lump these two companies together in regards to FinFETs.
Click to expand...
Daniel, you're starting to sound like you work for TSMC...
 
Daniel Nenni said:
Having just returned from Taiwan I feel my information is much more up to date than what I have read around the internet thus far. This is a very fluid situation so cutting and pasting comments from conference calls or rumor websites is not going to get you anywhere except lost. I will do a more detailed write up as soon as I can. Unfortunately I'm still digging out from my trip so it could be a day or two. In the mean time I'm very interested in what others have to say here and if you have specific questions I will do my best to get the answers.
Click to expand...

You should be careful about casting doubt on other people's sources while making out that your own are impeccable. My comment came not from rumour websites but from work done evaluating technologies and talking openly (but under NDA) to technology people inside the foundries -- not PR guys -- about the difficulties going below 20nm with double patterning and FinFETs, including discussions on how many double-patterned metal layers are really needed to get the best density/yield tradeoff. So let me repeat again that *nobody* is having an easy time here, this is really challenging stuff.

Any information you have found and are allowed to disclose here is unlikely to throw any light on this, there's always a positive spin on released information, and everybody does this, TSMC and Intel included.
 
Last edited:
Everyone, please cool down a little bit. This discussion can become a religious war and we know how destructive it is. I can sense many of you are real experts and know some facts from various sources but you can't reveal them freely. It makes our discussion more interesting yet more difficult.

I think we should require everyone to contribute a funny or cheerful story too. Bad idea? Don't kill me!!
 
Here are my 16N FinFET notes from Dr. Mark Liu, president and co-CEO at the TSMC OIP Forum held this month:


  • Today 20nm production has a monthly volume of 60,000 wafers with good defect density
  • The yield learning on 20nm production will directly benefit 16nm production
  • 20nm capacity can quickly support the coming 16nm ramp up
  • More than 90 percent of TSMC's equipment for the established 20nm node is being reused at 16nm.
  • TSMC's 16nm defect learning has reached a similar level as 20nm (they are less than six months apart)
  • 10 customer 16nm tape-outs in 2014 so far, more than 45 are expected in 2015
  • TSMC is already in production with a 16nm FinFET network processor for HiSilicon Technologies Co. Ltd.
The TSMC OIP Forum is invitation only for TSMC customers and partners. Most of these people already have 16nm silicon so there is no fooling them with ambiguous slides and double speak. Just my opinion of course.
 
Daniel Nenni said:
Here are my 16N FinFET notes from Dr. Mark Liu, president and co-CEO at the TSMC OIP Forum held this month:


  • Today 20nm production has a monthly volume of 60,000 wafers with good defect density
  • The yield learning on 20nm production will directly benefit 16nm production
  • 20nm capacity can quickly support the coming 16nm ramp up
  • More than 90 percent of TSMC's equipment for the established 20nm node is being reused at 16nm.
  • TSMC's 16nm defect learning has reached a similar level as 20nm (they are less than six months apart)
  • 10 customer 16nm tape-outs in 2014 so far, more than 45 are expected in 2015
  • TSMC is already in production with a 16nm FinFET network processor for HiSilicon Technologies Co. Ltd.
The TSMC OIP Forum is invitation only for TSMC customers and partners. Most of these people already have 16nm silicon so there is no fooling them with ambiguous slides and double speak. Just my opinion of course.
Click to expand...

Not a religious war, I'm not biased towards or against any foundry (and have been involved in process/foundry selection) -- but just a couple of points need making...

"10 customer 16nm tapeouts in 2014 so far" -- so maybe 5 customers have actually got 16FF silicon in their hands right now (see below)...
"TSMC is already in production with a 16nm FinFET network processor for HiSilicon" -- they've just delivered first samples, this is not "in production"

"Most of these people already have 16nm silicon" -- if 5=most (see above) it must be a pretty small event then, presumably only 7 or 8 companies attending?
"no fooling them with ambiguous slides and double speak" -- I know at least one of said companies is unhappy with 16FF not delivering claimed power savings -- so were they fooled?

From the TSMC OIP website: "With this forum, TSMC offers you an effective marketing venue..."

In other words, it's a marketing event for TSMC to promote their technology, and ecosystem (IP/software) partners to market their products to current and potential customers. Absolutely nothing wrong with that -- TSMC is a fantastic foundry with massive support -- but don't expect to see any shadow of doubt about how wonderful the technology and the ecosystem are and that all the problems are solved. I'm equally sure that if you went here:

GLOBALFOUNDRIES to Host Inaugural Global Technology Conference

you'd see exactly the same kind of presentations but with TSMC replaced by Globalfoundries, and everything being similarly wonderful. Nobody should be under any illusion that marketing events like these are in any way impartial, just put on the free rose-tinted glasses when you walk through the door.

As you say, your *opinion*... ;-)
 
Last edited:
"Samsung on the other hand licensed their 14nm FinFET process from IBM. Samsung also got the 28nm process from IBM which did not yield properly"

No, Samsung did not license FinFET from IBM, and BTW their 28nm is what powers iphone 5s.

All companies have yiled problems once in a while and that's the nature of this industry. Suppose TSMC has problem ramping 16FF in time for next Apple prodcut. So what? They might as well ship the next product on 20nm (no one waits for an uncertain yield ramp to tape-out a product anyway).

Marketing always tend to show things rosy. But looks like we are entring a new era with the introduction of FinFET. Honestly, I'd feel much better if foundries would call their first FinFET technology a 20nm FF. Same thing that they did at 28nm. Each company has several versions and yet they are all called 28nm, because they share the same base GR. That was the first place things went wrong. Then came the performance claims. Virtually everyone in the industry made a claim of 30-50% power saving. If that's the case why there is a need to 16FF+? I was looking at some of the charts a top tier fabless shared with me two years ago when they were evaluating 16FF and to my surprise their post layout was faster than pre-layout! The third thing that went wrong, IMO, was several fabless companies insisting in skipping 20nm and moving directly from 28nm to FinFET. I know they spent time and resources just to realize it's better to ship 20nm products. And now is the 4th thing; tying the industries success to the success of FinFET. If TSMC says they share 90% of the 20nm equipment with FinFET (which should be the case for most processes and for all foundries) why Ultratech is claiming their shipment is delayed beacuse of FinFET ramp delay?

TSMC's strength is the ability to serve a wide range of customers across many nodes and I admire them for that. I don't care if they are first to ramp 16nm, 10nm, or whatever. That's a game a certain company likes to play; As I commented somewhere else, to see where the industry stands you should not go to IEDM, ISSCC, or companies sponsored conferences. Go to Apple store, Best buy, or whatever your favorit store is.
 
You must log in or register to reply here.
Back
Top