How will Intel keep its process tech lead in the future?

[cbn]

First off, let me say that I believe Intel will keep its process tech lead....I just wonder from what segment the money will come to fuel this endeavor?

Fabs are expensive and competition appears to be heating up in the consumer space. This may lead to lowering profit margins. In fact, I have read a lot of criticism (from smart people on the internet forum/News sites) regarding the profit margins associated with smartphone SOC...yet Intel appears to have no plans on pulling the plug in this development.

Therefore it is my belief that the money to cover the leading edge fabs cost must come from some new type of innovation strategy brewing within Intel.

But what will this innovation or (more likely) group of innovations be?

 

[pelov]

I think we'll see smaller fabs merge with larger fabs in a consolidation process at the 20nm level -- 20nm because the 20nm-and-below nodes are exponentially more expensive than the higher ones. That 20nm hurdle will gobble up a few.

Intel will likely open up its fabs even more. We've already seen they've made some steps here but because people won't pay Intel margins and x86 tax on phones and tablets, they'll be forced to split the costs by offering WSA's to non-competing companies in need of wafers.

How long it'll last, though... that depends almost solely on demand. Will consumers pay higher prices that Intel requires to keep their fabs to themselves? Will chips keep selling like hotcakes? (this one is unlikely as the projected forecast for 2012 PC and laptop sales has slumped)

 

[ShintaiDK]

Intel is actually estimated to be the last company standing on the process node area in terms of ROI. If Intel cant afford it, nobody can. At 14nm only TSMC and Intel is expected to turn a profit.

You basicly just need volume. And ofcourse margins on the products shipped. x86 holds a huge margin. And Intel is expanding into smartphones and tablets, not to mention the newly Xeon Phi.

 

[ninaholic37]

But what will this innovation or (more likely) group of innovations be?

Haswell will be the last consumer CPU produced, then everyone will have to switch to super wi-fi Chromebook/dumb-terminals where all processing to your display will come from the collective Xeons on the cloud (that charge per clock usage). :sneaky:

 

[dealcorn]

Today, you have to buy different transistors from a bunch of different folks to build a phone. The whole industry understands you need to consolidate everything kinda like Intel did with the PC (where's my sound card?). Intel figures that (i) it knows more about every aspect of consolidating logic functions on a piece of silicone than anyone else, (ii) it's transistors are the world's best and the world's lowest cost to produce in category. and (iii) it appreciates the need to assemble or create the required ip and it is willing to fund same. If everything works according to plan, at the end of the day Intel will produce the best phone chips at the best margins. The phone market is turning to what Intel does well. Time will tell whether Intel executes well.

Moore's Law gives you twice the transistors per wafer with each node shift. You either figure out how to sell the increased volume or you die. Cell phone give Intel room to hold onto Moore's Law a little longer than anyone else.

 

[Borealis7]

@OP:
easily, the aliens in Intel's basement still have a lot they can teach them,
the reverse engineers still haven't gotten through dismantling their ship.
how do you think they came up with "3d transistors"?

just like Einstein got the theory of relativity from aliens. do you honestly think any human could come up with such a wacky idea 100 years ago?

 

[Lonbjerg]

@OP:
easily, the aliens in Intel's basement still have a lot they can teach them,
the reverse engineers still haven't gotten through dismantling their ship.
how do you think they came up with "3d transistors"?

just like Einstein got the theory of relativity from aliens. do you honestly think any human could come up with such a wacky idea 100 years ago?

Aristotle (384-322 BC) made three important observations:

• Every portion of the Earth tends toward the center until by compression and convergence they form a sphere.
• Travelers going south see southern constellations rise higher above the horizon; and
• The shadow of Earth on the Moon during a lunar eclipse is round

That something occured before you memory start dosn't negate the fact...and in case you were trying to be funny....you failed.


You come back from a 1week vacation for personal attacks and you lead with yet-another-insult?

Re: "and in case you were trying to be funny....you failed."

Clearly the last vacation wasn't long enough, lets try 2 weeks this time and see if you understand: No personal attacks or insults.

Administrator Idontcare

 

[Borealis7]

"It's okay, not all humans get my artificial sense of humor." -- M. Romney

in any case, process superiority may not be the deciding factor in the future, the shift may be towards a whole new direction the same way we thought we'd have 10GHz CPUs by now, and instead we have 4-8 cores at 3-4 GHz

 

[Idontcare]

...process superiority may not be the deciding factor in the future...

These companies live and die by finding ways to offer value to the consumer in ways that no one else can.

Intel's process superiority is their way of doing that. It isn't really anyone else's way though because the gap is just so large.

So everyone but Intel needs to figure out how to differentiate their products while competing against a company who can offer superior power-consumption and performance/watt numbers.

The only way process superiority would no longer be a deciding factor is if those who have it decide to not leverage it.

Intel could build the world's most fantastic 14nm room-temp superconducting process node only to then decide they will only use it in making toasters and coffee makers or something...but if they build it with the intent of using it against their competitors then they will find ways to make sure the customer knows there is a difference so they buy the Intel products.

 

[rickon66]

Haswell will be the last consumer CPU produced, then everyone will have to switch to super wi-fi Chromebook/dumb-terminals where all processing to your display will come from the collective Xeons on the cloud (that charge per clock usage). :sneaky:

I pray not, what will we do for fun? If I did not have my semi-annual build to plan for and then execute, woe is me.

 

[cbn]

http://www.reuters.com/article/2012/05/10/us-intel-idUSBRE84914H20120510

With the industry preparing to increase the size of the silicon wafers it uses, letting manufacturers fit more chips on each, future leading-edge factories will cost more than $10 billion each to build, compared with about $5 billion now, Otellini said.

Intel's first fab to exceed $5 billion dollar cost:

http://www.tomshardware.co.uk/intel-fab42-14nm-cpu-factory,news-37599.html

A First Look at Intel's 14nm Fab 42 Manufacturing Facility

11:40 - Wednesday 25 January 2012 by Douglas Perry - source: VLSI Research

Intel is adding a new gem to its network of manufacturing facilities.

Construction of Fab 42, the company's first volume 14 nm factory, has begun in Chandler, Arizona and has been documented in an article in the Financial Times and a slideshow published by analysts at VLSI Research. The massive new fab will be Intel's first factory to exceed a construction cost of $5 billion and somewhat follows the concept of the praised D1X development Fab in Hillsboro, Oregon. Fab 42 will also use the "copy exactly" approach, in which the company aims to recreate the conditions of the D1X development fab in a volume production facility in extreme detail, including interior temperature and air quality, to achieve the production yields delivered by D1X.

What makes Fab 42 special is that it is a modular fab like D1X (which is separated in a manufacturing, development and research portions).However, Fab 42 is more advanced and substantially larger than D1X. The new plant is also the first volume production facility that is compatible with 450 mm wafers, which offer a substantial economic advantage over the current 300 mm generation that Intel launched with its 130 nm processor generation in 2001.

Fab 42 is due to go online sometime next year.

I still have lots of questions on the topic of fabs.....but from what I gather 450mm is a big step up from 300mm.

Now about the $10 billion dollar fabs mentioned in the Reuters link....how soon will these be coming? 10nm? (I believe this is when EUV debuts)

 

[pm]

Since no one else posted it, it's worth mentioning this announcement from earlier today:

http://newsroom.intel.com/community...tion-semiconductor-manufacturing-technologies
http://www.businessweek.com/news/20...rcent-stake-in-asml-for-about-2-dot-1-billion

 

[pelov]

Now about the $10 billion dollar fabs mentioned in the Reuters link....how soon will these be coming? 10nm? (I believe this is when EUV debuts)

Sub 20nm fabs are much much more expensive. After crossing that 20nm hurdle it is smooth sailing (comparatively speaking) until you get to the sub 10nm level.

How soon they'll come is a separate matter. If you go by Intel's projected timeline then it should be 2 years, tick-to-tock, but even for them it may be a case of fine tuning and growing pains and consequent delays. Remember Intel ran into a bit of trouble with 22nm Ivy and it was delayed a couple of months (and subsequently the desktop i3 was still further delayed).

 

[cbn]

Since no one else posted it, it's worth mentioning this announcement from earlier today:

http://newsroom.intel.com/community...tion-semiconductor-manufacturing-technologies
http://www.businessweek.com/news/20...rcent-stake-in-asml-for-about-2-dot-1-billion

I don't understand how EUV is set-up in the fabs....but I am guessing it has something to do with this:

http://www.semi.org/en/node/38986

“"450mm: It's All about Economics," and discussed how periodic wafer size increases are needed to compensate for more expensive nodes, and that the transition to 450mm is expected to save 30% on production costs. Johnson noted that 450mm has “nothing to do with Moore’s Law— it’s purely a cost play.” He continued, “When you look at it from the perspective of the semiconductor manufacturer, it’s really quite simple, cheaper equipment and production costs for the same incremental output. They save money…primarily in capital equipment. When you look at it from the equipment manufacturer side, you’ve got a really good question. We’ve got the opportunity to spend billions of dollars on R&D that could go for other projects and end up cutting our market by about 30% going forward.” He said, “450mm creates a fundamental conflict of interest between the semiconductor and equipment manufacturers and this conflict of interest has to be resolved if we’re going to go forward.”

The 450mm wafers mean less EUV machines?

This means lower sales for the companies developing EUV?

Intel's response to this fundamental problem is to buy ASML? Am I getting the jiste of this correctly? (Someone please enlighten me)

 

[pelov]

Yes. Essentially the billions from Intel are incentive to speed up the transition to 450mm wafers.

Bigger wafers = more chips per wafer and therefore you'll likely need fewer EUV machines.

 

[Khato]

The 450mm wafers mean less EUV machines?

This means lower sales for the companies developing EUV?

Intel's response to this fundamental problem is to buy ASML? Am I getting the jiste of this correctly? (Someone please enlighten me)

Assuming my understanding is correct, basically. The difference in cost between a line of machines for 300mm wafer processing and a line of machines for 450mm processing is small compared to the cost of the machines. For that slight increase in initial investment however you get a 30-40% increase in production.

It sounds like Intel's intention is to basically pay for the majority of ASML's R&D costs just in order to get access to 450mm machines. It likely makes sense for them given their sheer volume - the decreased cost of equipment purchase for a certain volume of silicon likely more than offsets this investment. Not to mention, I wouldn't be surprised if the roadmap for 450mm machine availability didn't meet up with Intel's process roadmap... specifically I'd bet that the machine roadmap introduced EUV first followed by 450mm a year or two later to try and get everyone to buy 300mm EUV 'cause they need it and then 450mm EUV later to lower manufacturing costs.

 

[Abwx]

Assuming my understanding is correct, basically. The difference in cost between a line of machines for 300mm wafer processing and a line of machines for 450mm processing is small compared to the cost of the machines. For that slight increase in initial investment however you get a 30-40% increase in production.

Should read 125% in production and 30/40% in costs...

It sounds like Intel's intention is to basically pay for the majority of ASML's R&D costs just in order to get access to 450mm machines.

A sizeable part of the industry pull for a slower transition ,
so ASML has not enough interestezd customers , hence Intel s investement.

 

[Martimus]

@OP:
easily, the aliens in Intel's basement still have a lot they can teach them,
the reverse engineers still haven't gotten through dismantling their ship.
how do you think they came up with "3d transistors"?

just like Einstein got the theory of relativity from aliens. do you honestly think any human could come up with such a wacky idea 100 years ago?

The theory of relativity is a very basic and simple idea. It is VERY straight forward and logical, that a lot of people could have thought it up. I know that I came up with the exact same equations from the reverse direction when I was 16 trying to model how time behaves when I postulated that time was only the movement of particles in relation to other particles. Once I theorized that the movement was at an atomic level that affected time, I was able to formulate how the movement was related. the equations ended up being exactly the same as various formulas Einstein used in the Theory of Relativity, with the exception that I did not assume there was an absolute "speed limit". Such a limit would invalidate my initial assumption that time was just the movement of particles in relation to other particles, and was not a base element in itself. At the time, I built this up from scratch without even understanding what the theory of relativity was. I don't claim to be smart, I am only claiming that the theory itself is very straight forward and logical when you think of the physics on a base level. It is based on the fact that for every action there is a reaction, but no reaction can actually happen until that initial action happened.

Anyway, I am a simple electrical engineer, and I think process node shrinks are far more advanced than the theory of relativity. In fact, most of Einsteins published theories were more advanced than the theory of relativity.

 

[Ferzerp]

Just no. Nothing you've typed suggests any understanding of either general or special relativity. That's some pseudo-scientific-sounding babbling that you think sounds interesting. The immutability of the universal speed limit of the speed of light forms the foundation for what appears to at first glance to be throwing the laws of reality out the window. An understanding of the nature of the speed of light is requisite for even beginning to understand the concepts of einstein's relativity theories (specifically, special relativity, which, ironically, is much easier to grasp than general relativity, despite the physical oddities it suggests).

It's very likely, you could have easily made the assumptions that form newtonian relativity though.

 

[IntelUser2000]

In fact, I have read a lot of criticism (from smart people on the internet forum/News sites) regarding the profit margins associated with smartphone SOC...yet Intel appears to have no plans on pulling the plug in this development.

They can't avoid competing in the Smartphone market. Even forgetting margins and just looking at revenue though, its hard to see how it can sustain Intel's earnings.

ASPs of Smartphone SoCs: $20-30

Let's assume $30, Intel sells $35 billion Desktop/Laptop CPUs annually. That would require Intel selling 1.2 billion $30 Smartphone chips/year. Since the entire celluar phone(Smartphones/Featurephones etc) market is estimated to be ~1 billion, that's quite inconceivable, since they need to sell low cost chips too.

So Smartphones are not a replacement strategy, rather an expansion of their current market. They want to grow the next 5 years, not stay the same.

In the topic about margins: Margins can be increased with features that differentiate from competitors, and by having a leadership product. You pay extra for Intel chips in the desktop not so much that its x86, but because they have the top chips in the market.

 

[happysmiles]

in the end. I hope we get cheaper chips.

 

[epidemis]

The Chinese and Indians better grow fast lest the volumes doesn't grow fast enough to support this pace of improvements.

 

[ShintaiDK]

The Chinese and Indians better grow fast lest the volumes doesn't grow fast enough to support this pace of improvements.

Consolidation. The weaker will just die of. Essentially only Intel and TSMC got a spot. The rest is just waiting to either close business or get bought up.

Just look at memory makers. One less now.

 

[carop]

The 450mm wafers mean less EUV machines?

This means lower sales for the companies developing EUV?

EUV is late to the party because there are still several major issues with the power sources, mask infrastructure and other technologies. Because of the lack of adequate power sources, EUV throughputs are running about 4 wafers an hour right now. The industry requires throughputs of around 100 wafers an hour to make EUV economically viable in production foundries.

Intel inserted its initial 193nm immersion scanners (deep ultraviolet sources based on argon fluoride (ArF) lasers) at its 32nm node. Nikon Precision was the sole supplier of 193nm immersion scanners for the "critical layers" at Intel’s 32nm node.

At 22nm, Intel rolled out its TriGate transistor technology. The company is still using 193nm immersion scanners for the critical layers at that node. The critical layers are reportedly being split between ASML and Nikon.

Then, at 14nm, Intel will continue to use 193nm immersion for the critical layers. And it will implement an undisclosed type of "pitch-halving" (double patterning) technology at that node. Intel’s 14nm process is expected to be ready by 2013. Like 22nm, Intel will probably use both ASML and Nikon for the critical layers.

At 10nm Intel is keeping its options open:

http://semimd.com/blog/2012/02/12/intel-wants-euv-but-firm-keeps-options-open/

Even though EUV is still a technical challenge, 450mm needs only money and vendor support. It is interesting to note that ASML was one of the reluctant vendors to start 450mm. They kept saying that ASML would only do 300mm EUV first. It was expected that tool vendors would ask semiconductor companies to share the cost of 450mm development. Intel, Samsung, and TSMC were already pushing for 450mm in order to maintain a reasonable cost structure for the semiconductor industry:

http://www.eetimes.com/electronics-news/4076946/Intel-Samsung-TSMC-push-for-450-mm

There is summary of the Intel and ASML agreement at the following URL:

http://www.electronicsweekly.com/Articles/10/07/2012/54092/intel-to-pay-4bn-for-15-stake-in-asml-to-accelerate-450mm-and-euv-development.htm

Most of the money will go to 450mm lithography. Intel has mastered "pitch-halving". They will probably use quadruple patterning at 10nm if EUV is not ready. This would, however, increase the costs. They are pushing for 450mm to maintain the cost structure.

 

[Idontcare]

He said, “450mm creates a fundamental conflict of interest between the semiconductor and equipment manufacturers and this conflict of interest has to be resolved if we’re going to go forward.”

The 450mm wafers mean less EUV machines?

This means lower sales for the companies developing EUV?

Intel's response to this fundamental problem is to buy ASML? Am I getting the jiste of this correctly? (Someone please enlighten me)

The conflict of interest is very real. For example at one point I was developing a wet-clean process that was to be deployed on a tool that cost roughly $4m.

Now to scale up the wafer volume to fab levels we were looking at buying at least 5 tools, possibly 6 to manage the expected preventative maintenance downtime, so the wet-clean process itself was going to represent a capital expenditure of ~$25m.

Naturally I was incentived to improve on the process time of the wet-clean so that it could become fast enough that the fab managers could hit their wafer throughput targets while only needing to buy 3 or 4 tools instead of 5 or 6.

So my project had a potential ROI of ~$8m/fab if the wet-clean could be optimized and sped up as projected.

But the tools themselves are so complex, and in my case required some specializations to the toolset itself (software tweaks plus the addition of a few hardware features).

So the project I was managing, created because of the potential to realize an $8m/fab reduction in capital expenditures, required the active participation of the tool supplier themselves (for software/hardware augmentation)...but their active participation meant they were spending money on salaries for their employees in the pursuit of an end-goal that was going to decrease their revenues by $8m/fab.

Complete and total conflict of interest.

So how did we ever come to an agreement (IDM and tool supplier) to work on the project? Basically it came down to fear and competition.

The tool supplier in question knew that if they didn't agree to support such a JV (joint venture) for tool and process optimization then we (Texas Instruments) would simply pursue taking up the same project with another tool vendor that was willing to sell us 3-4 tools versus the zero tools they were going to sell to us at that point in the node capacity buildout forecast.

So the fear of losing 100% of the revenue from tool sales was the motivation for the tool supplier to spend money to work with us to reduce their own revenue some 20-30%.

Looking at EUV, the problem for the tool buyer (be it Intel or TSMC, etc) is that the tool suppliers have little competition and they know it. You either buy ASML or you wait another 12 months to buy from the competition because the competition is at least 12 months behind ASML in terms of delivery timelines. As an IDM, 12 months means you slip your EUV production release by a whole node, that gets expensive.

So Intel had to incentive ASML to keep up the pressure on EUV and 450mm development.