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

[mrmt]

Agreed. And we'll only have more of the story once we finally see the entire 14nm product portfolio launched. The fact they did a Haswell Refresh was a harbinger.

I think the major shift happened in 14nm in terms of scope complexity for the node development project. Intel is no longer aiming for the most efficient transistor on the market, but the most efficient transistor on the market with costs and density comparable to TSMC and Samsung. 10nm should pose an even greater challenge for them, and the complexity should be mind-boggling.

I wonder if they somehow plan to do something different like split their node development pathfinding for a high performance node and another one heavily optimized for costs.

 

[jpiniero]

I wonder if they somehow plan to do something different like split their node development pathfinding for a high performance node and another one heavily optimized for costs.

Does Intel really need a 'high performance' node? Their cash cows are corporate laptops and multicore servers.

 

[mrmt]

Does Intel really need a 'high performance' node? Their cash cows are corporate laptops and multicore servers.

And do you think you can get these without the highly efficient transistors they do now?

 

[Arachnotronic]

I think the major shift happened in 14nm in terms of scope complexity for the node development project. Intel is no longer aiming for the most efficient transistor on the market, but the most efficient transistor on the market with costs and density comparable to TSMC and Samsung. 10nm should pose an even greater challenge for them, and the complexity should be mind-boggling.

I wonder if they somehow plan to do something different like split their node development pathfinding for a high performance node and another one heavily optimized for costs.

I think Intel wants to have higher performance and far greater density than comparable foundry nodes, all at a competitive cost structure.

You're definitely right that the scope of Intel's node development has changed, and perhaps this explains the 14nm yield issues.

 

[witeken]

I think Intel wants to have higher performance and far greater density than comparable foundry nodes, all at a competitive cost structure.

You're definitely right that the scope of Intel's node development has changed, and perhaps this explains the 14nm yield issues.

I feel like, although there isn't a single showstopper according to them, the main issue, like they also said, it the (multiple) patterning. Intel now uses triple patterning.

 

[Fjodor2001]

Apparently Intel intends to launch 10 nm in 2017:

http://gulfnews.com/business/technology/intel-to-launch-10nm-chips-in-early-2017-1.1443856

"The 10nm chips are expected to be launched early 2017,” said Taha Khalifa, general manager for Intel in the Middle East and North Africa region.

So does that mean TSMC 10 nm and Intel 10 nm will go head to head sometime in 2017?

 

[witeken]

Apparently Intel intends to launch 10 nm in 2017:

http://gulfnews.com/business/technology/intel-to-launch-10nm-chips-in-early-2017-1.1443856

So does that mean TSMC 10 nm and Intel 10 nm will go head to head sometime in 2017?

For a second I almost believed the date he said, but then I noticed how *extremely* conservative he is:

He said that Cherry Trial is the 14nm for smartphone and tablets and will be available in the second half of the year.

Intel is already promoting Skylake — successor to Broadwell — that can wirelessly charge laptops and connect monitors, printers and external storages.
Khalifa said that Skylake architecture will be launched by end of this year.

He's 2 times off by almost 6 months. Skylake seems to be planned for early Q3, and CT is shipping right now. Of course he's not pinning down any specific or realistic date since Intel hasn't announced that, so you might want to remove 6 months from this early '17 date he gives. At least that will give a realistic time span, although within what most people here say (except someone who said Q4 for HVM).

In any case, pretty good news that within 2 years from now, in line with Moore's Law, there will be plenty of 10nm SKUs on the market, which is something e.g. AMD could only have dreamed of when they launched 28nm exactly 3 years ago... 3 years ago (and soon enough it will be 4 or 5 if there won't be any 20nm ones).

 

[NTMBK]

Regarding Cherry Trail- that quote may mean "devices containing Cherry Trail", while Intel is "shipping" Cherry Trail right now in low volume to OEMs.

 

[Fjodor2001]

For a second I almost believed the date he said, but then I noticed how *extremely* conservative he is:

He's 2 times off by almost 6 months. Skylake seems to be planned for early Q3,

Latest news from Intel is that Skylake will launch in 2015H2. That was presented at CES 2015. This is what the Intel representative in article I linked to earlier also reconfirms.

Of course he's not pinning down any specific or realistic date since Intel hasn't announced that, so you might want to remove 6 months from this early '17 date he gives.

That's a pretty bold statement. What inside info do you have that makes you better suited at providing launch date estimates than Intel representatives?

 

[Idontcare]

Apparently Intel intends to launch 10 nm in 2017:

http://gulfnews.com/business/technology/intel-to-launch-10nm-chips-in-early-2017-1.1443856

So does that mean TSMC 10 nm and Intel 10 nm will go head to head sometime in 2017?

Probably, but TSMC's "10nm" will be more like Intel's 14nm and not have much at all in common with the electrical or physical characteristics of Intel's 10nm.

Why are you seemingly hung-up on pitting these businesses together on the basis of node labels which you've already been educated to know are not an apples-to-apples comparison? Its like you are seeking conflict and drama, even when there is none to be had.

 

[Fjodor2001]

Probably, but TSMC's "10nm" will be more like Intel's 14nm and not have much at all in common with the electrical or physical characteristics of Intel's 10nm.

Why are you seemingly hung-up on pitting these businesses together on the basis of node labels which you've already been educated to know are not an apples-to-apples comparison? Its like you are seeking conflict and drama, even when there is none to be had.

I'm just confirming that these are the two process techs that will be competing with each other. It's important to know when analyzing what TSMC's 10 nm competition will look like. What characteristics and pros/cons each of the process techs have is another story, and that is something which is up for discussion.

Is there any other way to put this that you'd prefer?

 

[witeken]

Latest news from Intel is that Skylake will launch in 2015H2. That was presented at CES 2015. This is what the Intel representative in article I linked to earlier also reconfirms.

That's a pretty bold statement. What inside info do you have that makes you better suited at providing launch date estimates than Intel representatives?

I'm not going to repeat what I already said. Do you think Krzanich, Holt, Bohr or Smith were just joking when they said they wouldn't talk about 10nm until IM'15? Of course not, you can't expect the most detailed from a company that has decided not to talk about the product, unless you didn't believe yet that there would be 10nm in 2017. He could as well have said that the sucessor of Cannonlake will be available in early 2018. No duh Sherlock.

 

[Fjodor2001]

I'm not going to repeat what I already said. Do you think Krzanich, Holt, Bohr or Smith were just joking when they said they wouldn't talk about 10nm until IM'15? Of course not, you can't expect the most detailed from a company that has decided not to talk about the product, unless you didn't believe yet that there would be 10nm in 2017.

Here is what Krzanich and Smith said at Intel Investor's Meeting '14:

http://forwardthinking.pcmag.com/none/329835-intel-sees-path-to-extend-moore-s-law-to-7nm

"Krzanich said the company thinks it signaled too much of its intentions to the industry about its 14nm plans, so "we'll be a bit more prudent in releasing information" about new manufacturing nodes. He wouldn't commit to the company's familiar Tick/Tock cadence of releasing a new process node one year and a new architecture the following year, though Smith said the company expects to be on a "fairly normal cadence" and "will talk about 10 nm in the next 12 or 18 months when appropriate.""

As I interpret this, Intel says they will talk more about 10 nm during the period from the 2014 Investors Meeting to 12-18 months ahead of that. I.e. we may see info popping up along the way during this period. So far we've not got much info on 10 nm, except a rough estimate of an early 2017 launch that this Intel representative now indicated in the article I linked to earlier.

He could as well have said that the sucessor of Cannonlake will be available in early 2018. No duh Sherlock.

So you are saying the Intel representative just made everything up, lied, or intentionally provided a launch date estimate far ahead of what Intel actually expects? Why would he do that? Wouldn't it be better to just not comment on the 10 nm launch date at all in that case?

Also, how come we should believe your estimates to be better than the ones provided by the Intel representative anyway? :hmm:

 

[Idontcare]

I'm just confirming that these are the two process techs that will be competing with each other. It's important to know when analyzing what TSMC's 10 nm competition will look like. What characteristics and pros/cons each of the process techs have is another story, and that is something which is up for discussion.

Is there any other way to put this that you'd prefer?

Ah, I see what you mean now. Yeah you put it just fine, I misunderstood the discussion.

10nm will be even more exciting than all the current 16/14 industry drama. But first we have to get past all the 16/14 drama

 

[Arachnotronic]

10nm will be even more exciting than all the current 16/14 industry drama. But first we have to get past all the 16/14 drama

Can you offer us a sneak preview?

 

[Idontcare]

Here is what Krzanich and Smith said at Intel Investor's Meeting '14:

http://forwardthinking.pcmag.com/none/329835-intel-sees-path-to-extend-moore-s-law-to-7nm

"Krzanich said the company thinks it signaled too much of its intentions to the industry about its 14nm plans, so "we'll be a bit more prudent in releasing information" about new manufacturing nodes. He wouldn't commit to the company's familiar Tick/Tock cadence of releasing a new process node one year and a new architecture the following year, though Smith said the company expects to be on a "fairly normal cadence" and "will talk about 10 nm in the next 12 or 18 months when appropriate.""

As I interpret this, Intel says they will talk more about 10 nm during the period from the 2014 Investors Meeting to 12-18 months ahead of that. I.e. we may see info popping up along the way during this period. So far we've not got much info on 10 nm, except a rough estimate of an early 2017 launch that this Intel representative now indicated in the article I linked to earlier.

Intel was super-quiet about their xtor technology plans twice before, on the eve of announcing they were putting HK/MG into production at 45nm, and then again on the eve of announcing they were putting tri-gate (finfet) into production at 22nm.

It would not surprise me at all if Intel were being hush-hush about 10nm because they don't want to give the competition any more of a chance to change their development roadmaps to re-align to whatever Intel is doing (as that is what keeps happening).

Doesn't mean 10nm will be III-V or Ge, but it would mean it won't be standard finfet shrink. But that doesn't jive with what everyone is doing in the supply chain, it is all finfet right now for the foreseeable future. Time will tell, of course.

 

[Arachnotronic]

Intel was super-quiet about their xtor technology plans twice before, on the eve of announcing they were putting HK/MG into production at 45nm, and then again on the eve of announcing they were putting tri-gate (finfet) into production at 22nm.

It would not surprise me at all if Intel were being hush-hush about 10nm because they don't want to give the competition any more of a chance to change their development roadmaps to re-align to whatever Intel is doing (as that is what keeps happening).

Doesn't mean 10nm will be III-V or Ge, but it would mean it won't be standard finfet shrink. But that doesn't jive with what everyone is doing in the supply chain, it is all finfet right now for the foreseeable future. Time will tell, of course.

Idontcare, if you poke around on LinkedIn, you can find profiles of process development engineers at Intel refer to 10nm as "10nm TriGate".

Intel also stated when it launched 22nm that the FinFET technology could be "extended out for a couple of generations" (paraphrasing).

10nm is definitely FinFET as you say.

 

[TechFan1]

My interpretation of their comments is that they will keep their mouths shut until 10nm is starting to ramp or about to ramp, so they don't have news articles saying they had to delay production.

 

[witeken]

10nm will be even more exciting than all the current 16/14 industry drama. But first we have to get past all the 16/14 drama

Why?

 

[witeken]

Idontcare, if you poke around on LinkedIn, you can find profiles of process development engineers at Intel refer to 10nm as "10nm TriGate".

Intel also stated when it launched 22nm that the FinFET technology could be "extended out for a couple of generations" (paraphrasing).

10nm is definitely FinFET as you say.

It's too soon for GAA.

SE: Some believe that gate-all-around is the frontrunner in the industry at 7nm. Any comments?

Bohr: That wouldn’t surprise me greatly. I won’t comment beyond that in terms of what Intel sees as the favorite.

Probably at 5nm.

10nm is definitely a new channel material, 'cause silicon has run out of juice.

 

[Idontcare]

Why?

Because of the Heaviside step function that has occurred in all three major players in terms of their R&D funding and efforts for 10nm over that which was allocated for 16/14.

And they'll all be on the same node label, 10nm, which will make for fantastic marketing and PR blitz's on all sides as the attempts to make apples-to-apples comparisons will be taken to an all time high in intensity.

 

[witeken]

Because of the Heaviside step function that has occurred in all three major players in terms of their R&D funding and efforts for 10nm over that which was allocated for 16/14.

Not completely sure what you mean with that.

And they'll all be on the same node label, 10nm, which will make for fantastic marketing and PR blitz's on all sides as the attempts to make apples-to-apples comparisons will be taken to an all time high in intensity.

I suspect TSMC and Samsung's 10nm will compete against Intel's 7nano-units for most of their bleeding edge lifetime. I also suspect many people will uncritically think Intel's technology lead has been eroded away.

I wonder if EUV will be first deployed at 10,000pm and by which company first.

 

[Fjodor2001]

I suspect TSMC and Samsung's 10nm will compete against Intel's 7nano-units for most of their bleeding edge lifetime.

Hmm... how come? Latest info is that both Intel and TSMC will release 10 nm to the public in 2017 (although those 10 nm process techs will have different characteristics). Can Intel really release 7 nm almost at the same time as 10 nm?

 

[Idontcare]

Not completely sure what you mean with that.

Heaviside step function

What most people invoke as a "quantum leap" is really more appropriately described as a Heaviside step function whenever the individual invoking the phrase "quantum leap" is doing so to imply a gap or step-up in whatever is being discussed versus the less relevant equivocation of the physics entailed in a quantum transition.

If you are looking for a specific example, TSMC doubled their entire R&D team (headcount and resources) for 10nm development.

I suspect TSMC and Samsung's 10nm will compete against Intel's 7nano-units for most of their bleeding edge lifetime. I also suspect many people will uncritically think Intel's technology lead has been eroded away.

2 years ago I would have taken that bet. But I'm not seeing Intel running the same way under BK that it did under Otellini. I think we are going to see 10nm run head-to-head, and then 7nm, etc., even if by virtue of marketing manipulation of the node label. If Intel gets too far ahead, again, then the foundries will just pull another 16/14nm out (claim a new node exists where none was before) and they'll be back to "node parity".

As for those who will rush to internalizing that as an erosion in Intel's lead...true but that is Intel's problem. Intel has no issue hyping their technology prowess for the benefit of analyst and shareholder alike. I'm confident they won't sit idly by and let their shareholders and analysts down. They'll find a marketing way to ensure the market is fully and properly educated on why their 10nm (or 7, or 5) is superior to the competition's.

I wonder if EUV will be first deployed at 10,000pm and by which company first.

Ha ha, 10,000pm We stuck fast to the micron node label well near the 0.1um node before finally switching vernacular and referring to them in nanometer units. Even the early days of 90nm development it was still referred to as 0.09um. I doubt we'll ever see nm node labeling give way to pm node labels, the numbers are just to big to make for good marketing.

It will be nm all the way down to 0.1nm (equivalence, of course, the same as we use for gate oxide scaling which is given an EOT rating in Angtroms).

EUV, if it happens in production at 10nm, will likely be limited to just a single litho step at the very most critical mask in the entire flow, with all the remaining less-critical (but still paramount) masks being done in 2, 3 or 4 multi-patterning. But it is a necessary "first" for EUV to gain the production environment time it needs to work out all the gatekeeping issues before full production rollout at 7nm can happen.

Hmm... how come? Latest info is that both Intel and TSMC will release 10 nm to the public in 2017 (although those 10 nm process techs will have different characteristics). Can Intel really release 7 nm almost at the same time as 10 nm?

Intel could, but nothing is keeping TSMC and Samsung from deciding they are skipping 10nm and are going to go straight to 7nm...the key being their 7nm would look surprising similar to what they were previously calling their 10nm node :hmm: :sneaky: (kinda like how 16FF/14nm look an awful lot like what they should have delivered with their 20nm nodes in the first place...)

 

[podspi]

Because of the Heaviside step function that has occurred in all three major players in terms of their R&D funding and efforts for 10nm over that which was allocated for 16/14.

And they'll all be on the same node label, 10nm, which will make for fantastic marketing and PR blitz's on all sides as the attempts to make apples-to-apples comparisons will be taken to an all time high in intensity.

I haven't been able to wrap my head around this. Yes, more cores sells, because many non-technical people want computers, they know cores are good, and more = better, right?

The people who buy process tech (rather, the people who hire fabs) are not (I assume) non-technical. They make their choices based on the price and technical characteristics. Why would they care whether the label is 10nm, 7nm, 42nm, etc?

Is it just a jockeying game for investors?