Intel Investor Meeting 2015: November 19

[Arachnotronic]

So interesting information that does lower my confidence in III-V at 10nm, even though the vast majority of the data points point to it.

Don't expect III-V materials at 10nm. Like I said, the good stuff happens at 7nm.

 

[witeken]

Don't expect III-V materials at 10nm. Like I said, the good stuff happens at 7nm.

C'mon, you're basing that on one guy who used "futuristic" as buzzword. 7nm is a long way out, so I'll dig into that when time comes. I believe Intel might move to spintronics around 5 or 3nm.

Also, look at the edit.

But if you’re at 22nm, 14nm isn’t very far away [Bohr just said finFET is scalable to 14nm], so you’ve got to be working on the next step.

Bohr= For Intel, you’re right. For other companies, it’s many years away. For 10nm, which is where I’m spending most of my time these days, I know we have a solution. I can’t elaborate at this point.

Many years away for competitors huh, if you're going to go with III-V only at 7nm? Spoiler: Samsung etc. will go to SiGe at 7nm. Their next step at 5nm will probably be the horizontal/lateral GAA (not the superior vertical one; http://www.extremetech.com/wp-content/uploads/2013/07/different-transistor-topologies.jpg).

It's certainly interesting to hear cautious remarks from Intel, but I wouldn't underestimate them (unless you're talking mobile 😉).

 

[witeken]

But sure, the future looked much better in the past, when 10nm was coming in 2015, was going to have EUV, was going to have 450mm, was going to have Germanium and III-V, etc.

At least we got air gaps in 14nm 😉. Who would be not excited about that?

Compare:

It is one of those things that I think is just an unavoidable consequence of the math when you start looking at what Intel must be planning if they are intending on having three or four 450mm 10nm fabs loaded.

versus

Intel's 10nm is finfet.

http://forums.anandtech.com/showpost.php?p=34204125&postcount=9 and http://forums.anandtech.com/showpost.php?p=37735021&postcount=105

(Just imagine, though, how the world (and INTC's stock) what have looked like if they delivered on all these things...)

 

[krumme]

Thanx

The 2012 interview with Bohr is still really interesting and good thread back then btw.

Bohrs comment on the need for foundry business to become more like IDM is perhaps more serious than he thought. He focus was the soc and integration between design and process here.

The example i gave about Samsung back then have shown what an IDM actually should be. The soc is just the way to a higher purpose eg the phone. The need for integration is on a far higher level and spans far more than engineering tasks.

Go look at where Atom is today. The product is on 14nm. Executed as planned. Prioritised even before servers.
And yet mobile cost aprox 4b a year.

 

[witeken]

At another investor conference a few days ago, Brian Krzanich had some nice remarks about the Altera acquisition with relation to Intel's acquisition history (McAfee).

I want to say one other thing because you and I, we are kind of talking about it outside. I think it's a lot of skepticism on Intel’s ability to do an acquisition and I think within this room it should be rightfully so. I’d say that one of the other things that I’m proud of the team that we have at Intel over the last two years -- two and half years is really improve the performance in Mcafee which was the other big investment we did which I’ll be completely transparent wasn't performing as well as it should.

We hired a new leader with Chris Young, great guy, hired him out of Cisco, just a strong leader knows this business, very accountable. We fully integrated the organizations so they're part of Intel now, they’ve been held separate. If you take a look at the last couple of quarters, our costs are down. Our performance in revenue is up and we’ve actually shown growth over the last few quarters. And we believe as we go into 2016, we continue to -- can continuously see it as a growth company now. So we've made these improvements.

So we're taking that same learnings and applying it right up upfront to Altera. So I can tell you my Board is holding me accountable, that this will not be another bad acquisition.

http://seekingalpha.com/article/372...com-brokers-conference-transcript?part=single

Why do I also want to be a part of this? One of the things that frustrated me as an engineer in Intel and this is one of those long-term investments that you guys have to trust me on as shareholders. I look out at the universities, I look out at these Maker Faire and these are the people that are going to be the inventors of the technology in 10 years, 15 years and none of them are working on Intel x86 architecture. They were all working on ARM.

We've lost touch with the community that was going to be the next people who invent the next great machine that we all love and use. I have to go back and get those people in touch with Intel, see that there is a technology that there is a capability that they cannot match anywhere else on the planet. That is why we developed Quark. We developed Quark, several of us development it about 4 years ago quietly, it was a secret project that nobody knew at Intel was funded. I actually funded that out of the factory budget, not the architectural budget for 2 years.

Sure, so would tell you that if all of the business, if the future of that business was purely NAND, whether it's 3D NAND or 2D NAND, I wouldn't have changed the business model. The business model works quite nicely between us and Micron right now and we have restructured it such that the cyclicality is much-much less for us as players in that market. What significantly changed it for me was SXP [ph] or you guys know it as 3D Xpoint. By the way we always change the codenames as we go through these things, to confuse the engineers and the public.

This is a very unique architecture, it truly shifts the way memory is done. It has 80% to 90% of the speed of DRAM, while having the non-volatility and cost structure much more like NAND and it allows us to architect a whole new type of memory storage hierarchy and so remember I said, my whole theme around Intel is IPs, they can be shared across these virtuous cycle. Well, if such go to things like Big Data, Big Data is huge amounts of data, hence its name Big Data and the more of that processing, the more closely you can attach the data to the processing, then much faster those Big Data analytics occurred.

We believe that we can get multiple orders of magnitude improvement in Big Data applications by taking 3D Xpoint where we can now take terabytes of data and move it right next to the CPU and have it act much more like memory rather than storage, and so we're building an architecture that we believe will shift how datacenters are built in the future and that's why it's a unique technology that links directly with the CPU, we believe we have a significant leadership over the competition and that's why we're going to go and investing there.

I need to grow in that market, will we be a long-term player in that, I think we have to let it play out over the next five to seven years and see whether others come in, whether somebody else has another ability to copy the technology, we’ll let it play out and we'll be careful about how big the investment is, but for right now we believe we can accelerate the growth of this and it will have a major impact in how things like Big Data Analytics occur in the datacenter.

LOL

Edit: Interesting article. Since 2006, Intel has beefed up its own HPC performance by 50X. http://www.nextplatform.com/2015/11/26/intel-supercomputer-powers-moores-law-life-support/

 

[krumme]

Intel dcg will really be able to get most out of 3dx. That advantage is still there wether its copied or not.
Mcaffee is not remotely so potent and never was. Top management simple decision like thinking safety = money. Shouldnt even be mentioned. An outright failure of epic dimensions.

 

[witeken]

Chip Execs Hear of Bin Laden, Moore

The SIA gave its annual Robert Noyce award to Bill Holt, general manager of Intel’s technology and manufacturing group.

“He is one of the smartest engineers I know, the team he leads has been defining the future of Moore’s Law and reinventing everything we know about computing every two years,” said Intel chief executive Brian Krzanich in giving the award to the 42-year Intel veteran.

 

[ksec]

TSMC has a clear roadmap to 7nm. Ok it won't be as good as Intel's 7nm. But the point is, Intel has no place in Mobile and Tablet Market. The top end are completely dominated by Apple and Samsung, both has their own SoC. The Mid Range is a battle between Mediatek and Qualcomm. While lower end is MediaTek and many smaller Chinese Players.

As some body has pointed out Intel don't like to be in the low margin business. That is only half true, another point is Atom isn't even competitive in the Sub 5W TDP. May be Core-M, but then Intel isn't interested making very little money.

PC is declining, Laptop are declining as well. Which means Intel is only left with their Cloud business, which is still growing massively YoY.
I don't think ARM stand a chance in the Cloud / Datacenter Market. At least not in the next 3 - 4 years. But AMD is making a come back, what if ZEN is pretty good and offer to support More DRAM, and Networking by default? Intel often try to segment the Server Market with Core Count and Memory Support, both point could easily be targeted by AMD.

What about Intel's modem business. They bought Infineon and have yet to make anything out of it.

 

[dark zero]

TSMC has a clear roadmap to 7nm. Ok it won't be as good as Intel's 7nm. But the point is, Intel has no place in Mobile and Tablet Market. The top end are completely dominated by Apple and Samsung, both has their own SoC. The Mid Range is a battle between Mediatek and Qualcomm. While lower end is MediaTek and many smaller Chinese Players.

As some body has pointed out Intel don't like to be in the low margin business. That is only half true, another point is Atom isn't even competitive in the Sub 5W TDP. May be Core-M, but then Intel isn't interested making very little money.

PC is declining, Laptop are declining as well. Which means Intel is only left with their Cloud business, which is still growing massively YoY.
I don't think ARM stand a chance in the Cloud / Datacenter Market. At least not in the next 3 - 4 years. But AMD is making a come back, what if ZEN is pretty good and offer to support More DRAM, and Networking by default? Intel often try to segment the Server Market with Core Count and Memory Support, both point could easily be targeted by AMD.

What about Intel's modem business. They bought Infineon and have yet to make anything out of it.

The only way Intel wins is that they starts to make software too... seems that the developers aren't as good enough as expecting... that's why Apple is wining the Mobile battle at least.

 

[Hugo Drax]

I heard Intel is having a 20-22% defect rate on the high end i7 skylakes. This would make sense with the backlog and price hikes on whatever i7 67xx cpus can be found.

I wonder if those numbers are typical or real bad.

 

[frozentundra123456]

TSMC has a clear roadmap to 7nm. Ok it won't be as good as Intel's 7nm. But the point is, Intel has no place in Mobile and Tablet Market. The top end are completely dominated by Apple and Samsung, both has their own SoC. The Mid Range is a battle between Mediatek and Qualcomm. While lower end is MediaTek and many smaller Chinese Players.

As some body has pointed out Intel don't like to be in the low margin business. That is only half true, another point is Atom isn't even competitive in the Sub 5W TDP. May be Core-M, but then Intel isn't interested making very little money.

PC is declining, Laptop are declining as well. Which means Intel is only left with their Cloud business, which is still growing massively YoY.
I don't think ARM stand a chance in the Cloud / Datacenter Market. At least not in the next 3 - 4 years. But AMD is making a come back, what if ZEN is pretty good and offer to support More DRAM, and Networking by default? Intel often try to segment the Server Market with Core Count and Memory Support, both point could easily be targeted by AMD.

What about Intel's modem business. They bought Infineon and have yet to make anything out of it.

Having a "roadmap" and executing it are two different things. TSMC does seem to be doing very well, but I anticipate a lot of problems with every foundary with each new node now.

I agree basically with you about intel in mobile, although maybe you stated the case a bit strongly. But 14 nm atom needed to be a killer product performance wise, which it is not, and the integration is slow to come also.

As for Zen in servers, AMD has "targeted" a lot of things and not hit them. Rumors are that Zen will be a good product, but it is already basically 2016, and AMD is being rather quiet about any specifics. So I remain skeptical until we see benchmarks on a shipping product.

 

[Arachnotronic]

I heard Intel is having a 20-22% defect rate on the high end i7 skylakes. This would make sense with the backlog and price hikes on whatever i7 67xx cpus can be found.

I wonder if those numbers are typical or real bad.

Where did you hear this? Given the lousy availability, I'd say those would actually be pretty good yields...

 

[ShintaiDK]

I heard Intel is having a 20-22% defect rate on the high end i7 skylakes. This would make sense with the backlog and price hikes on whatever i7 67xx cpus can be found.

I wonder if those numbers are typical or real bad.

Assuming 80% yield, that's relatively low for Intel. But its beating the rest of the industry by a long shot. A9 for example may still be 50-60% yield.

 

[Hugo Drax]

Okay so if the yields are not an issue the. It must be much higher than ever demand if skylake i7

 

[Arachnotronic]

Okay so if the yields are not an issue the. It must be much higher than ever demand if skylake i7

Nah, Intel is currently having yield issues on 14nm.

 

[ShintaiDK]

Yield or production limitations. Same same as result.

With that said, there is higher pressure on the higher bins due to the upsale the dGPUs saw.

 

[witeken]

Instead of producing armchair theories about Intel yields and demand, why not go listen to what Intel itself has actually to say about these things?!

http://intelstudios.edgesuite.net/im/2015/archive/wh/archive.html

TL;DW:

- Yields still below 22nm (at same point in development)
- All production targets met (but of course those targets probably were lower than if yields had been higher; for example BDW-E and Xeon Phi probably got delayed because of that)
- But higher demand than expected

 

[mrmt]

Instead of producing armchair theories about Intel yields and demand, why not go listen to what Intel itself has actually to say about these things?!

http://intelstudios.edgesuite.net/im/2015/archive/wh/archive.html

TL;DW:

- Yields still below 22nm (at same point in development)
- All production targets met (but of course those targets probably were lower than if yields had been higher; for example BDW-E and Xeon Phi probably got delayed because of that)
- But higher demand than expected

Not sure what you are trying to say here. From the slides you posted Intel is having yields issues, because at the same point of the deployment of the 22nm the yields on 14nm are worse than on 22nm, so you are just confirming Intel's yields problems.

In fact, by the graphics you posted Intel is taking 6 quarters more to reach optimum yields on 14nm than it took on 22nm, and 6 quarters was supposed to be 3/4 of the useful life of a node for mainstream products.

 

[Abwx]

I heard Intel is having a 20-22% defect rate on the high end i7 skylakes. This would make sense with the backlog and price hikes on whatever i7 67xx cpus can be found.

I wonder if those numbers are typical or real bad.

That s the %age of a mature process so it s not bad at all, the 80% yield is surely related to the usable chips including the die harvested ones, that is the i5s.
In this respect a 100% yield could be good financiarly speaking but awfull technicaly if only 20% can be qualified as i7.

 

[Khato]

Unfortunately that 22nm vs 14nm yield graph has no indication whatsoever as to the format/scale of the Y axis - it could just as easily be an absolute percentage as it could be a logarithmic scale. The difference being 14nm yield being, say, 10% behind 22nm in one case and 1/10 of 22nm in the other. aka, useless chart meant to placate investors.

Personally, my guess is that Skylake availability is a combination of issues as witekin implies. With the additional factor of the number of fabs they've converted over to 14nm possibly? Which I don't believe they've said much about lately?

 

[mrmt]

The difference being 14nm yield being, say, 10% behind 22nm in one case and 1/10 of 22nm in the other. aka, useless chart meant to placate investors.

I would be very surprise if Intel was feeding their investors with this kind of useless information, especially when it is already clear that they have issues with 14nm.

 

[Exophase]

This is a very unique architecture, it truly shifts the way memory is done. It has 80% to 90% of the speed of DRAM, while having the non-volatility and cost structure much more like NAND and it allows us to architect a whole new type of memory storage hierarchy and so remember I said, my whole theme around Intel is IPs, they can be shared across these virtuous cycle. Well, if such go to things like Big Data, Big Data is huge amounts of data, hence its name Big Data and the more of that processing, the more closely you can attach the data to the processing, then much faster those Big Data analytics occurred.

If this is accurate it'll be revolutionary. Almost all DRAM applications would be better off with this technology. Just having much cheaper and denser RAM that's only moderately higher latency is already a win without even really exploiting the non-volatility. Fast entry to and exit from full system hibernation is an easy additional bonus.

Unfortunately that 22nm vs 14nm yield graph has no indication whatsoever as to the format/scale of the Y axis - it could just as easily be an absolute percentage as it could be a logarithmic scale. The difference being 14nm yield being, say, 10% behind 22nm in one case and 1/10 of 22nm in the other. aka, useless chart meant to placate investors.

Yeah, those were my thoughts too. I really don't like these unlabeled graphs that Intel seems to keep using.

 

[Arachnotronic]

Unfortunately that 22nm vs 14nm yield graph has no indication whatsoever as to the format/scale of the Y axis - it could just as easily be an absolute percentage as it could be a logarithmic scale. The difference being 14nm yield being, say, 10% behind 22nm in one case and 1/10 of 22nm in the other. aka, useless chart meant to placate investors.

Personally, my guess is that Skylake availability is a combination of issues as witekin implies. With the additional factor of the number of fabs they've converted over to 14nm possibly? Which I don't believe they've said much about lately?

They actually said that they ramped the Lexlip, Ireland plant earlier than originally planned last quarter. The fact that they brought on additional capacity slightly earlier than expected and they're still facing 14nm product "shortages" is a pretty bad sign.

 

[witeken]

If this is accurate it'll be revolutionary. Almost all DRAM applications would be better off with this technology. Just having much cheaper and denser RAM that's only moderately higher latency is already a win without even really exploiting the non-volatility. Fast entry to and exit from full system hibernation is an easy additional bonus.

It will be 2x cheaper, initially, and allows 4x the amount of memory.

And don't forget that they can scale it down pretty aggressively. More layers (although I think cost structure isn't as good for 3D as NAND), smaller geometry, more bits per cell (although not focus in near future).

 

[witeken]

They actually said that they ramped the Lexlip, Ireland plant earlier than originally planned last quarter. The fact that they brought on additional capacity slightly earlier than expected and they're still facing 14nm product "shortages" is a pretty bad sign.

A bad sign of what? That they will only match yields in H2'16? Sure, Stacy Smith gave us some graphs, nothing new. That producing 8B x'tor Xeon Phis isn't easy with subpar cost structure? Yeah, RIP gross margins.

Those 90mm² 2+2 dies aren't the problem, I'd guess: they sell for enough even factoring in the defect density.