[DigiTimes] TSMC 10 nm trial production in 2015, mass production in 2016

[Homeles]

What are you saying exactly? There are literally 20nm chips in products right now..

And to all of you saying no 20nm SoCs in 2014, I guess you expect iPhone 6 to be using a 28nm SoC. We'll see.

I think he's talking about SOI.

And I think any concerns of the A8 not being on 20nm were dispelled with Qualcomm's 20nm modem finally appearing in devices.

 

[Exophase]

I think he's talking about SOI.

I was wondering that too, it just didn't seem to fit with the mention of 16nm (AFAIK only TSMC is offering a node with that name and they're not interested in FDSOI)

And I think any concerns of the A8 not being on 20nm were dispelled with Qualcomm's 20nm modem finally appearing in devices.

And yet some are still saying it

 

[Homeles]

And yet some are still saying it

I can't think of any other reason why TSMC would be ramping 20nm so aggressively. I know you're not one of the doubters, but I think the signs are very obvious.

 

[jdubs03]

It is pretty apparent the patterns that are occurring, the timing is just trickier in terms of product releases for each node progression on a medium to longer term.

 

[krumme]

Its just great news anyway, that the ramping is so fast. It probably means QQ post a57 arch on 16nm h2 2015, and the same for some freaking fast 16nm Apple whatever the fab used.

Its interesting to note the S5 mini uses 4* a7. They dont use the mm2 for cpu thats for sure lol. The new modem and dsp tech is going to pull the consumer adoption.

 

[Arachnotronic]

Its just great news anyway, that the ramping is so fast. It probably means QQ post a57 arch on 16nm h2 2015, and the same for some freaking fast 16nm Apple whatever the fab used.

Its interesting to note the S5 mini uses 4* a7. They dont use the mm2 for cpu thats for sure lol. The new modem and dsp tech is going to pull the consumer adoption.

I personally wouldn't expect a 16nm Snapdragon from Qualcomm until 1H 2016, though a Gobi modem in 2H 2015 would make sense. Qualcomm needs to get its return on those 20nm designs after all

 

[Exophase]

I personally wouldn't expect a 16nm Snapdragon from Qualcomm until 1H 2016, though a Gobi modem in 2H 2015 would make sense. Qualcomm needs to get its return on those 20nm designs after all

I was going to say the exact same thing.

I think some SoC makers and other fabless companies are however skipping 20nm entirely, some may go for 16nm as aggressively as possible.

 

[mavere]

Given the earlier speculations on how 20nm could reach 20+% of TSMC's 4th quarter revenue, I thought these expectations were interesting:

We believe TSMC should see a very aggressive ramp up of 20nm revenues in 3Q14, with wafer out for Apple rising from 30-40k wfpq to 120-130k wfpq, with quite high yields of a process at such early stage of ramp up.
[...]
We are expecting the wafer consumption from Apple to rise to 150-160k wfpq in 4Q14. In addition, 20nm wafer orders for other leading edge customers (Xilinx, AMD, Qualcomm) should also start from late 3Q14 onwards, driving 20nm revenue contribution to 20%+ in 4Q14.

Apple is expected to contribute to 13% and 15% of TSMC’s third-quarter and fourth-quarter revenue, versus only 4% in the second quarter.

 

[sm625]

So if Apple buys 150,000 wafers, how many chips is that?

 

[witeken]

50M 100mm² chips at 50% yield.

 

[mavere]

July 31st report from an Asian financial firm on TSMC's outlook. Besides the usual money blahblahblahs, there were some interesting, if pessimistic, bits:

with QCOM diversifying its 14nm production to Samsung (and potentially Apple) in FY15, TSMC stands to lose USD2b in revenue, which could result in sales growth slowing down to mid-single digit YoY. Under the worst case, should both QCOM/Apple outsource half of their production to Samsung, TSMC could lose up to USD5b of business and its sales may decline by 2-3% YoY in FY15

TSMC will not bring up the 16nm+ pilot line until 3Q15, which is 6-9 months behind customers’ roadmap. Such delay was caused by TSMC’s decision to introduce an enhanced version of the 16nm node (namely 16nm +) in late 2013 to improve the device performance and cost reduction in order to match Intel’s and Samsung’s 14nm baselines.

TSMC will still commence its 16nm pilot production in 4Q14 but we believe demand from customers for this node is rather muted as most are planning to adopt 16nm+/14nm in late-2015

We believe from the design rule standpoint, TSMC’s 16nm+ is essentially identical to Intel’s and Samsung’s 14nm, although some industry experts think the latter two still offer 10% smaller die size than TSMC’s 16nm+.

And lastly, they mentioned that Apple's A8 yields at TSMC are currently 60-70%.

 

[sm625]

TSMC wants to claim that they've been in mass production of 20nm since the beginning of this year... and yet it's only going to account for a measly 20% of their revenue in Q4. By comparison 22nm accounted for a quarter of Intel's revenue in Q2 2012

I think that is probably the best comparison metric. When does a company achieve 20% of its revenue from a particular process? That is the quarter that should be used for all comparisons. Intel is pretty damn far ahead. 2.5 years.

 

[Ajay]

July 31st report from an Asian financial firm on TSMC's outlook. Besides the usual money blahblahblahs, there were some interesting, if pessimistic, bits:

And lastly, they mentioned that Apple's A8 yields at TSMC are currently 60-70%.

Interesting info - thanks!

 

[Fjodor2001]

I think that is probably the best comparison metric. When does a company achieve 20% of its revenue from a particular process? That is the quarter that should be used for all comparisons.

By that metric Intel won't be releasing 14 nm until sometime in 2015...

Intel is pretty damn far ahead. 2.5 years.

Intel is 9-12 months ahead of Samsung/GF on 14 nm.

 

[Homeles]

By that metric Intel won't be releasing 14 nm until sometime in 2015...

Right.

Intel is 9-12 months ahead of Samsung/GF on 14 nm.

They're also ahead in density, and undoubtedly performance as well.

 

[Fjodor2001]

They're also ahead in density, and undoubtedly performance as well.

Are you sure about that? Then please show us numbers for transistor density and performance on Samsung 14 nm vs Intel 14 nm. Also, perhaps you have data on transistor cost and power consumption for each alternative too?

 

[Khato]

Intel is 9-12 months ahead of Samsung/GF on 14 nm.

If and only if all parties deliver on their current roadmaps. Well, and if what Samsung is calling 14nm has any relation to Intel's 14nm.

 

[Ajay]

If and only if all parties deliver on their current roadmaps. Well, and if what Samsung is calling 14nm has any relation to Intel's 14nm.

Yeah, I wonder what Samsung's 14FF is, since it wasn't derived from their 20nm node AFAIK.

 

[lefty2]

Yeah that report is interesting, but two things they got wrong.
Firstly, Samsung said in their earnings call that 14nm would start with Exynos production. Only in 2nd half of 2015 would they start production for external customers.
Secondly, when TSMC said a certain competitor would have more marketshare of 14nm FinFET in 2015 they were talking about Intel, not Samsung.

 

[mavere]

TSMC said a certain competitor would have more marketshare of 14nm FinFET in 2015 they were talking about Intel, not Samsung.

I originally thought the same, but I've already reread that part of the call repeatedly, and Chang's contradictory statements pointed to an imaginary competitor that is not Intel, GloFo, or Samsung. I think the only obvious conclusion is that he misspoke.

Post-call comments from the various call participants only discussed TSMC vs Samsung, so at least they were on the same page. Also, from even before the call, Samsung's 14nm output were expected to reach 40k+ wpm by Q4'15, and I would be floored if Intel's list of foundry customers would even order anything near that amount.

 

[lefty2]

I originally thought the same, but I've already reread that part of the call repeatedly, and Chang's contradictory statements pointed to an imaginary competitor that is not Intel, GloFo, or Samsung. I think the only obvious conclusion is that he misspoke.

Post-call comments from the various call participants only discussed TSMC vs Samsung, so at least they were on the same page. Also, from even before the call, Samsung's 14nm output were expected to reach 40k+ wpm by Q4'15, and I would be floored if Intel's list of foundry customers would even order anything near that amount.

There is a very important statement in the earnings call that you missed:

Elizabeth Sun
Randy's question is with respect to Chairman's comment on 2015's market share is lower than a major competitor in 2015. So Randy's asking
why will it be lower and what is the impact to TSMC if we have a lower market share. And what gives us the confidence that we will regain the
market share in following year?

Morris Chang - TSMC - Chairman
Oh, okay. Well, we need to go back to history a little bit. 32 -- 28-nanometer followed 32 and that particular major competitor that I referred to, chose 32 and skipped 28. And then of course we came to 20 and 16, 16 for us, 14 for him. And we chose to do both. Actually we chose to do 20 first and 16 about a year or so later, but it was a pretty quick succession. And this major competitor skipped 20 and went on to 16.

Now, we all know that Samsung has a 28nm node and Intel is the only foundry that skipped 28nm, so the competitor Chang is referring to is clearly Intel.

 

[Khato]

Now, we all know that Samsung has a 28nm node and Intel is the only foundry that skipped 28nm, so the competitor Chang is referring to is clearly Intel.

It's hard to say that when you take the entire quote, because the portion that you put in bold which implies Intel contradicts the last sentence, "And this major competitor skipped 20 and went on to 16." Which actually describes no one since TSMC is the only one calling their node 16 anything, but if you expand the context to +/- 2, well, in that respect Intel did do '20' and '16' while Samsung is the only one to skip the '20' and go on to '16'.

Regardless, the fab competition is already getting somewhat tiresome for the simple fact that it's been a war of powerpoint presentations for awhile now. Maybe we'll actually get a taste of some of these upcoming processes next month...

 

[mavere]

There is a very important statement in the earnings call that you missed: [...]

That's the part I was referring to when I mentioned rereading something repeatedly.

Khato already replied with the necessary points, and it still defies reason that Intel's 14nm as a foundry service would be noteworthy in terms of volume until it signs up other companies (and I doubt Morris Chang is willing to admit intimate knowledge of Intel's nascent deals in a conference call of all things). Meanwhile, the report above claims that Samsung's output will be 50+% of the available 14/16nm foundry market in FY15. Intel's will be ~20%.

 

[jdubs03]

If Samsung can get 14LPE Exynos chips out in 1H2015, I would be impressed, they have a shot at making Exynos a lot more relevant in chip design+process. Just the switch in process alone will help them leapfrog in performance, add A57 into that and we have an impressive CPU (even though the A57-based Exynos 5433 figures remain to be seen, Exynos 6/Infinity is their real shot). But count me as somewhat skeptical; the Galaxy Note series would seem to make the most sense; the galaxy tab pro series makes sense too, but they really should just merge the two (to make larger notes), and offer the stylus.

Yeah, I wonder what Samsung's 14FF is, since it wasn't derived from their 20nm node AFAIK.

http://www.eetimes.com/document.asp?doc_id=1321974

Based off that and the TSMC information in their most recent transcripts (+both foundries stated 28nm -> 20nm performance increased by 30%), Samsung's performance increase or power consumption reduction is lower compared to 16FF:http://forwardthinking.pcmag.com/none/322862-get-ready-for-14nm-and-16nm-chips, and the 14LPP designation has the same increase as 16FF+, so still lower performance. The density reduction for the 14 LPE process is as here:

http://electronics360.globalspec.com/article/4171/samsung-globalfoundries-form-14nm-alliance-to-fight-tsmc

I expect LPP to be the same density. As we've seen the density improvement from Intel's 14nm is much higher than TSMC's 16FF/+, so I anticipate Intel's 14nm to be of higher performance compared to both TSMC and Samsung.

I originally thought the same, but I've already reread that part of the call repeatedly, and Chang's contradictory statements pointed to an imaginary competitor that is not Intel, GloFo, or Samsung. I think the only obvious conclusion is that he misspoke.

Post-call comments from the various call participants only discussed TSMC vs Samsung, so at least they were on the same page. Also, from even before the call, Samsung's 14nm output were expected to reach 40k+ wpm by Q4'15, and I would be floored if Intel's list of foundry customers would even order anything near that amount.

TSMC's conference calls are so damn confusing it is hard to figure out exactly what their process perf/w improvements are, for example they state 16FF+ is 15% faster than 16FF with 16FF being 30% faster than 20nm planar, but they state that 16FF+ is 40% faster than 20nm. That doesn't make sense: 1.3 * 1.15 = 1.495, so it decreases clarity, maybe that is what they want to do in the PR wars.

 

[ancientarcher]

By that metric Intel won't be releasing 14 nm until sometime in 2015...



Intel is 9-12 months ahead of Samsung/GF on 14 nm.

Just talking about products on shelves, Intel will have products on shelves by 4Q14. Whereas Samsung said they will have 14nm in mass production by 4Q14 and on the shelves by 2Q15 (probably the Galaxy S6)

That gives Intel a 6 month lead over Samsung/GF.

About performance, we will cross that bridge when we come to it. For the moment, its all about powerpoint battles