Yield challenges in FinFET foundry land

[Arachnotronic]

From chip equipment vendor KLA-Tencor:

In the near term however, demand remains fluid, particularly in foundry and logic, where we have recently seen orders from select customers slated for sub 20-nanometer, originally expected in the March quarter, pushed to later in the calendar year. We believe these delayed orders reflect yield and process stability issues associated with bringing these advanced device architectures to market.

http://seekingalpha.com/article/284...e-on-q2-2015-results-earnings-call-transcript

 

[Enigmoid]

Exactly as I said.

Take anyone in the chip fab industry (intel, GF, TSMC, Samsung) and multiply their expected timeline by 125% minimum (1.25x) to get the actual dates.

Everyone is giving the absolute best case. Nobody is giving their realistic timeline.

Delay, delay, delay.

 

[III-V]

Well TSMC delayed their 16nm a bit, right? That'd certainly explain a good chunk of it.

Exactly how plausible is it that Samsung has 14nm totally up and running right now, with devices showing up next quarter?

Also, the delay in orders may have been as slight as from Q1 to Q2 (standard, not fiscal). The wording is purposefully vague, it seems, to protect foundries and their customers.

 

[Arachnotronic]

Well TSMC delayed their 16nm a bit, right? That'd certainly explain a good chunk of it.

Exactly how plausible is it that Samsung has 14nm totally up and running right now, with devices showing up next quarter?

Also, the delay in orders may have been as slight as from Q1 to Q2 (standard, not fiscal). The wording is purposefully vague, it seems, to protect foundries and their customers.

Samsung claims to have "fed in" production wafers by the end of 2014 and an executive is on record saying that the first 14nm production wafers will ship in "early 2015."

My guess is that we see a 14nm Exynos in some variants of the Galaxy S6.

 

[Idontcare]

Samsung claims to have "fed in" production wafers by the end of 2014 and an executive is on record saying that the first 14nm production wafers will ship in "early 2015."

My guess is that we see a 14nm Exynos in some variants of the Galaxy S6.

Will be a real coup for Samsung over Apple and Qualcomm if true. Not to mention the situation with Intel and what will be left looking like a "bungled" 14nm rollout.

 

[mrmt]

Exactly how plausible is it that Samsung has 14nm totally up and running right now, with devices showing up next quarter?

Just for the sake of curiosity, did any of the equipment makers reported firm orders from Samsung related to its 14nm Finfet process? It would be an Apple process and Samsung will take 70% of its orders, it ought to have make some equipment maker very, very happy.

 

[aigomorla]

Exactly as I said.

Take anyone in the chip fab industry (intel, GF, TSMC, Samsung) and multiply their expected timeline by 125% minimum (1.25x) to get the actual dates.

Everyone is giving the absolute best case. Nobody is giving their realistic timeline.

Delay, delay, delay.

not necessarily true for intel.

The other competitors DO NOT BUILD NEW FAB's for new nodes, where each new node intel builds a NEW FAB.

Usually intel is always on time, they only started to show some delays because of the lack of market, and hence no need to rush, along with milking every last bit of efficiency at an already highly efficient level.

If intel had any form of pressure, they would even be early...

 

[jpiniero]

Will be a real coup for Samsung over Apple and Qualcomm if true. Not to mention the situation with Intel and what will be left looking like a "bungled" 14nm rollout.

In fairness, Samsung's "14 nm" is much worse than Intel's 14... it really should be called 20FF.

 

[aigomorla]

In fairness, Samsung's "14 nm" is much worse than Intel's 14... it really should be called 20FF.

i just find it really amusing that people actually think the other fab's can be considered on the same level as intel's.

creating an ARM SoC, is not on the same level as a x86 processor.

Also if you really knew what went behind closed doors at intel, you would truely understand why they call intel the giant blue ogre.

 

[Arachnotronic]

In fairness, Samsung's "14 nm" is much worse than Intel's 14... it really should be called 20FF.

I don't know about much worse, but there's definitely an appreciable technology gap. Intel's 14nm process offers tighter metal/gate pitches, air gaps, and probably higher Idsats. I'd bet that Intel's yields are better, too.

Too bad Intel isn't bringing this process greatness to competitive mobile products to help the likes of HTC/LG/Xiaomi fight the 14nm Exynos chips.

 

[aigomorla]

Too bad Intel isn't bringing this process greatness to competitive mobile products to help the likes of HTC/LG/Xiaomi fight the 14nm Exynos chips.

This is mostly due to them needing to change a big branch internally.

Its too much of an investment at this point with too much market saturation from other big companies.

Can intel do it? Yes they can, but they would not see great returns in short term.

Also the introduction of big sharks like samsung as competitors does not sit well in any board room meeting.
This is why intel even paired up with micron in the SSD department, because they did not want to pick up a large competitor where the future results were not guaranteed.

 

[Idontcare]

In fairness, Samsung's "14 nm" is much worse than Intel's 14... it really should be called 20FF.

They could call it the banana node for all I care, one is over hyped and under delivering, the other is apparently under hyped and over delivering (if the analysts are right).

In terms of business execution, one company's management is executing whereas the other company's management has failed to inspire confidence.

Combine that with the sudden "uh, we are going dark on all discussions related to 10nm, but don't worry it isn't because of the yields or anything, no no, uhm, quite the opposite in fact, we are ahead of schedule! couldn't be going any better! but seriously, don't ask us because we aren't going to talk about it, wouldn't want to tip our hand now would we?"

:| :hmm:

 

[videogames101]

They could call it the banana node for all I care, one is over hyped and under delivering, the other is apparently under hyped and over delivering (if the analysts are right).

In terms of business execution, one company's management is executing whereas the other company's management has failed to inspire confidence.

Combine that with the sudden "uh, we are going dark on all discussions related to 10nm, but don't worry it isn't because of the yields or anything, no no, uhm, quite the opposite in fact, we are ahead of schedule! couldn't be going any better! but seriously, don't ask us because we aren't going to talk about it, wouldn't want to tip our hand now would we?"

:| :hmm:

I think it's important to distinguish between products produced on a node and the technical merits of a node itself. Although I don't think the technical merits are of much importance when the cards are played. Because as I am quite sure you are aware the design of a chip is enabled by a process node, but it's market success is really determined by a massive number of other factors.

 

[witeken]

They could call it the banana node for all I care, one is over hyped and under delivering, the other is apparently under hyped and over delivering (if the analysts are right).

In terms of business execution, one company's management is executing whereas the other company's management has failed to inspire confidence.

Combine that with the sudden "uh, we are going dark on all discussions related to 10nm, but don't worry it isn't because of the yields or anything, no no, uhm, quite the opposite in fact, we are ahead of schedule! couldn't be going any better! but seriously, don't ask us because we aren't going to talk about it, wouldn't want to tip our hand now would we?"

:| :hmm:

But it's true. Intel has been very quiet about 10nm and also Skylake. And not because those will under deliver.

Q2:

We have done no changes or shift to our 10-nanometer schedule but we won’t really talk about 10-nanometer schedules until next year.

Q3 earnings call:

But you are seeing in the fourth quarter, you’re seeing the front edge of the startup cost associated with the 10-nanometer and that’s kind of right in line with the historical timing of what you’d expect. We’ll go through more excellent where we talk about how that might look over the next couple of years but we’re seeing at least the front edge at the time I think you’d expect and that’s about a point of gross margin next quarter.

Investor Meeting:
http://intelstudios.edgesuite.net/im/2014/archive/qa1/archive.html (8:30, should really watch)

“We felt like we went on a little early with 14nm as far as timing and performance and features and we saw actually competitors adjust to that. So we're gonna be a little bit more prudent, a little smarter about signalling to the industry exactly when, what and where. And you'll have to trust a little bit the 50 year history we have with Moore's Law and that we should be able to keep it going for 51 or 52 years. So we're gonna be a little careful there about that signalling exactly when, what and where.” --Brian Krzanich, CEO Intel, IM’14

“So I think you talk about the cost chart where at the very end I put a few little dotted points. That was intentially ambiguous. What we want to make clear to you is that we believe that the cost reduction historical trend can be continued. So we're extending the dotted line on the historical trend. What we didn't say is exactly how far below that line we think we'll be and how that trend will be [compared to] the previous generation exactly because that stuff we don't know. It will be below the historical trend, that much we know. [...] Yes [we can get below that line without EUV], but I don't want to: I want to be further below the line.” William Holt, Intel, IM’14

“We do think that we've been giving too much insight too far in advance and so we'll talk about 10nm some time in the next 12 or 18 months [from Nov '14] and when it's appropriate.” Stacy Smith, CFO Intel, IM’14

“But you are seeing in the fourth quarter, you’re seeing the front edge of the startup cost associated with the 10-nanometer and that’s kind of right in line with the historical timing of what you’d expect.” Stacy Smith, CFO Intel, IM’14

IEDM:

In development and research, we see scaling continues at least another 10 years, which is the same answer we gave 10 and 30 years ago. It’s hard to see beyond 10 years.

We don’t expect we’ll have similar problems at 10 nm, because we’ve learned and we’re trying harder.

Q4:

We are timing on 10-nanometers. We are not going to come out with when we’ll be introducing a 10-nanometer to the marketplace in general, probably until the end of this year. So again as we go through the year probably by the investor meeting in November, we’ll give you an outlook on how and what timing is for 10-nanometers.

:sneaky:

 

[Fjodor2001]

Intel has been very quiet about 10nm and also Skylake. And not because those will under deliver.

Just wondering, how do you know that? Do you have some secret insider info on Intel's 10 nm process and Skylake?

 

[witeken]

Just wondering, how do you know that? Do you have some secret insider info on Intel's 10 nm process and Skylake?

Read carefully. They might under deliver, but that's not the reason Intel's not talking about them. But for 10nm we know that traditional scaling and cost per transistor will continue at a slightly accelerated pace, and some new transistor innovation should continue the transistor efficiency trend.

 

[NTMBK]

But for 10nm we know that traditional scaling and cost per transistor will continue at a slightly accelerated pace, and some new transistor innovation should continue the transistor efficiency trend.

We don't know that, because even Intel don't know that! They think that it will based on their current predictions and estimates, but they don't know. Always be cautious in your expectations

 

[TreVader]

The excuses... Now the "merits of a process" are totally unrelated to the products produced on it? You guys are bending over backwards mentally to explain the last five years of intel's foundry progress.

You know when I was thinking about the massive leaps in performance ARM is giving (I.e. 100% or often more improvement without even a change in process) I thought about intel's history and for the life of me I can't think of a single generation of Intel chips where the next one (I.e. Haswell to Broadwell i7) gives 100% improvement even once!

Intel has always been a terrible processor company with a decent foundry. The problem is that so much of their business is built on bullying competitors, lying, and outright removing any competition that they never learned to make a decent product.

What they did learn, is how to convince a generation of AT members, supposedly educated individuals, that 2+2=5. Do the math! Find me one generation of processor where Intel can give even just %100 improvement!

I won't ask you to give the 200-500% increases apple often gives.

Cue the rabble of Intel apologists to come explain why all the benchmarks are unfair, Intel is still the fastest and finally they really do have a new faster architecture, they are just unloading this crappy one on those silly customers for margin and will soon release skylake which will blow us all away.

 

[witeken]

We don't know that, because even Intel don't know that! They think that it will based on their current predictions and estimates, but they don't know. Always be cautious in your expectations

It's normalized, so they do know.

 

[Arachnotronic]

They could call it the banana node for all I care, one is over hyped and under delivering, the other is apparently under hyped and over delivering (if the analysts are right).

In terms of business execution, one company's management is executing whereas the other company's management has failed to inspire confidence.

Combine that with the sudden "uh, we are going dark on all discussions related to 10nm, but don't worry it isn't because of the yields or anything, no no, uhm, quite the opposite in fact, we are ahead of schedule! couldn't be going any better! but seriously, don't ask us because we aren't going to talk about it, wouldn't want to tip our hand now would we?"

:| :hmm:

IDC,

I would be very curious as to any insights on why Intel might be quiet about 10nm in light of the quotes witeken posted. Their R&D head is claiming that 10nm won't have the same 14nm yield challenges, their CFO is saying that the startup costs will be on a "normal" cadence, and at the 2013 investor meeting they already said they're aiming for Q4'15 production.

And yet, they don't want to directly answer questions on 10nm right now.

What's your take on this?

 

[witeken]

You know when I was thinking about the massive leaps in performance ARM is giving (I.e. 100% or often more improvement without even a change in process) I thought about intel's history and for the life of me I can't think of a single generation of Intel chips where the next one (I.e. Haswell to Broadwell i7) gives 100% improvement even once!

Because SoCs learned from the PC space. It's that easy. When Intel is designing a new process node, they do it for the first time. When Intel designs a new architecture, it's the fastest in the world, so something that no one has done before.

But when Apple hires people from AMD, and other companies, those people already know how to deal with OoO and superscalar architectures. And when Apple decides to go from 45nm to 32nm, the 32nm process has already been used for a long period of them. Now you see Apple entering uncharted territory, and immediately they give up and add a third core because that's the easy way. Low hanging fruit's been picked.

And lastly, they also benefit from not giving power consumption numbers. When Qualcomm boosts clock speeds to 1.7GHz (LP) or 2.3GHz (HPM), they also push voltage to insane levels.

So instead of taking a slow core and doubling its performance every year, you can take a high-end one, Core, and scale it down to tablets, and you see that, after all the doublings, they still can't compete.

 

[videogames101]

The excuses... Now the "merits of a process" are totally unrelated to the products produced on it? You guys are bending over backwards mentally to explain the last five years of intel's foundry progress.

You know when I was thinking about the massive leaps in performance ARM is giving (I.e. 100% or often more improvement without even a change in process) I thought about intel's history and for the life of me I can't think of a single generation of Intel chips where the next one (I.e. Haswell to Broadwell i7) gives 100% improvement even once!

Intel has always been a terrible processor company with a decent foundry. The problem is that so much of their business is built on bullying competitors, lying, and outright removing any competition that they never learned to make a decent product.

What they did learn, is how to convince a generation of AT members, supposedly educated individuals, that 2+2=5. Do the math! Find me one generation of processor where Intel can give even just %100 improvement!

I won't ask you to give the 200-500% increases apple often gives.

Cue the rabble of Intel apologists to come explain why all the benchmarks are unfair, Intel is still the fastest and finally they really do have a new faster architecture, they are just unloading this crappy one on those silly customers for margin and will soon release skylake which will blow us all away.

1. My point about the technical merits of a process was only to point out that you can in fact have inferior products on a superior node, because the node is only an enabler for your chip design. It is not the sole and probably not even the most important factor.

2. I think you don't have a good idea about how similar CPUs really are. It's not like AMD or Intel or ARM or Apple has some secret sauce or significantly better engineers that makes their CPU significantly faster than everybody else's. What sets them apart is the tradeoffs they choose to make during the design process and how well those tradeoffs fit their product into market segments. Do you think Apple/ARM or anybody else will keep giving you 100% increases per generation? The answer is clearly no if you have any understanding of how the industry as a whole is progressing. So when you say "Intel is a terrible processor company", I have a hard time understanding your position. They have released the best desktop/server class CPUs for years now. However I would not disagree that they don't have their mobile products positioned low enough on the power scale to compete with ARM and Apple designs. That doesn't mean they're a "terrible processor company", it means they're a processor company who has traditionally focused on frequency targets and has had trouble shifting their focus onto power targets. Although it boggles my mind why Intel is even trying, you don't make 60% margins in mobile.

 

[witeken]

IDC,

I would be very curious as to any insights on why Intel might be quiet about 10nm in light of the quotes witeken posted. Their R&D head is claiming that 10nm won't have the same 14nm yield challenges, their CFO is saying that the startup costs will be on a "normal" cadence, and at the 2013 investor meeting they already said they're aiming for Q4'15 production.

And yet, they don't want to directly answer questions on 10nm right now.

What's your take on this?

It's still possible to run into yield issues. I once read some comment from Bohr I think, he said that he's often surprised by how much the yield improves at the last moments.

As you can see, there were severe issues around June and July of '13. So this could also still happen with 10nm later this year. But as you can also see, 14nm was already lower on the curve than 22nm, which had and has stellar yields, but if we believe Bohr, that's not the case now so yield learning should be on track.

 

[Arachnotronic]

It's still possible to run into yield issues. I once read some comment from Bohr I think, he said that he's often surprised by how much the yield improves at the last moments.

As you can see, there were severe issues around June and July of '13. So this could also still happen with 10nm later this year. But as you can also see, 14nm was already lower on the curve than 22nm, which had and has stellar yields, but if we believe Bohr, that's not the case now so yield learning should be on track.

Bohr was interviewed at IEDM 2014 (Dec. 2014). Intel's last public statement on 10nm was that the plan was to go into production in Q4 2015. Here in Q1 2015, Intel should be close to the finish line if they are indeed going to keep to their original schedule.

That said, Intel expected 14nm to be production viable by Q1 2014, but the process did not qualify for production until Q2 2014. I think this will cause at least a 1 quarter push-out of 10nm with production start in Q1 or Q2 of 2016.

Given that TSMC will go into production on its 10nm in Q4 2016, I would expect Intel to have an ace up their sleeve for this process (crazy tight gate/metal pitches, III-V materials, something...) in order to maintain the "process lead" that it seems so confident that it has.

 

[AtenRa]

I think this will cause at least a 1 quarter push-out of 10nm with production start in Q1 or Q2 of 2016. Given that TSMC will go into production on its 10nm in Q4 2016, I would expect Intel to have an ace up their sleeve for this process (crazy tight gate/metal pitches, III-V materials, something...) in order to maintain the "process lead" that it seems so confident that it has.

Is this for high volume production or testing 10nm production ??