Intel say fabless model collapsing... really?

[John Moor]

Hello colleagues on Semiwiki

I'm a newbie here but keen to extend my network out from the UK. I came across this headline reported in EETimes (Intel exec says fabless model ) and I wondered what everyone else thought about it. If you follow it to a logical conclusion it suggests only Intel will be able to make chips, the IDM model wins out yet its full of paradox and contradiction (of course). Any comments?

John
NMI

<script src="//platform.linkedin.com/in.js" type="text/javascript"></script>
<script type="IN/Share"></script>

 

[Daniel Payne]

John,

Intel is full of hutzpah, the foundry model is not broken now nor in the foreseeable future.

Intel has three public customers, so they are taking a very different approach to the foundry business compared to TSMC, GLOBALFOUNDRIES and UMC.

 

[Daniel Nenni]

A guy who has spent his ENTIRE 33+ year career at Intel thinks the foundry model is broken? What a shock! Lets not forget, the origin or the foundry/fabless industry was due to the fact that the IDM model was broken. Think seriously about where we are today and tell me if not for the foundry/fabless model would we have smartphones? Would we have the modern electronics we do today? I really do not think so.

Now Mark Bohr/Intel says design is too complicated for the modern day semiconductor ecosystem to succeed? A guy who has spent his entire life at Intel is qualified to say that?

I understand Intel's interest in the foundry business. Atom failed at the foundries so how else can Intel get Atom into the mobile ecosystem? Become an Atom specialized foundry, great strategy. I do not understand however Intel's strategy of insulting the entire semiconductor ecosystem. Renaming FinFets Tri-Gate is one thing, but they are really pissing in the well on this one, and believe me this type of hyperbole will come back to haunt, absolutely!

Cheers,

D.A.N.

 

[Paul McLellan]

Intel seems to think the foundry model is broken based on the fact that Qualcomm is unable to get enough 28nm silicon from TSMC and they would not be in poor shape if they had their own fab. Or something like that. I'd be interested to know what Qualcomm forecast to TSMC a couple of years ago would be their wafer demand now. Before Qualcomm won the Apple socket. And whether they put up any money to guarantee capacity by paying ahead for wafers. I have no inside knowledge at either TSMC or Qualcomm but I bet Qualcomm are asking for a lot more wafers than they told TSMC they expected to, and that TSMC built capacity to provide. And if they had built their own fab based on their own forecast they would be in the same situation.

 

[Daniel Nenni]

Intel seems to think the foundry model is broken based on the fact that Qualcomm is unable to get enough 28nm silicon from TSMC and they would not be in poor shape if they had their own fab. Or something like that. I'd be interested to know what Qualcomm forecast to TSMC a couple of years ago would be their wafer demand now. Before Qualcomm won the Apple socket. And whether they put up any money to guarantee capacity by paying ahead for wafers. I have no inside knowledge at either TSMC or Qualcomm but I bet Qualcomm are asking for a lot more wafers than they told TSMC they expected to, and that TSMC built capacity to provide. And if they had built their own fab based on their own forecast they would be in the same situation.

Here is the difference between IDMs and Fabless companies. IDMs wait to ship silicon until production is in full swing since they have no one to blame for shortages. Fabless companies ship silicon before production is in full swing to gain a PR advantage over competitors. Not one of the fabless CEO's have said that TSMC did not meet 28nm contractual commitments. The 28nm ramp is better than expected so either these CEO's do not understand semiconductor design and manufacturing or their comments have been grossly taken out of context by industry pundits to get clicks for their ads. My guess is the later.

D.A.N.

 

[Paul McLellan]

There is certainly some of that Dan. But having worked in an ASIC company for many years, I have seen no end of these forecast issues. For example, at one point a consumer electronics company forecast that they needed 0 wafers of their ASIC Nicam (stereo sound for TV) chips since they would have their own internal standard part development completed. But then they didn't tapeout and expected to suddenly go back to their previous (high) run rate. I think we managed to deliver in the end. But I think the sales team enjoyed the price negotiation.

 

[Bill Martin]

I tend to see various aspects in our ecosystem that go in cycles. Cycles can be short or long lived.

Custom/ASIC/SOC/FPGA/ASSP/etc continually rotate around depending on what is needed. At some point, the 'std products' get saturated and to differentiate yourself, you need some type of customization not afforded by std products (which could be FPGAs). This differentiation allows easier selling until imitated by someone else. The more difficult to imitate, the longer you can differentiate yourself. What would be harder to imitate: HW, SW design or some unique mfg technology (silicon or packaging)? My bet: manufacturing an easy winner. The original Harris process patent caused lots of royalty payments on anyone fabrication plant back in the 1970/80s: a basic implant technique that if anyone wanted to compete, they needed to use it. It was very easy to catch violators and cause them to pay royalties on any of their revenue stream that used this technique. For them to compete, they used it on all of their products.

There will be a time in the future where owning your own fab will have huge advantages vs. foundry model. This might be caused by silicon availability (remember, we make money when end products are sold...if we are on allocation, we limit $s that can be reaped) OR specific capabilities that are not available in the foundry model (pick any silicon mfg capability that might be very unique) OR...... I would never use the term 'never'.

Many may question if Intel is a well run company, I do not. Do they make mistakes? Sure, but they are not afraid to try something and if it fails move on to something else. They perform counter cyclical, capital investments understanding that investments in infrastructure can take years to get fully operational and producing yields that are desirable. They are trying to be prepared to optimize their return in the 'good' times. They also try to minimize risks to their revenue stream and always remove the weakest link to protect their revenue. I believe Forbes wrote an article about Intel and Chipsets years ago discussing how Intel saw a major weakness that threatened their CPU sales: the 5->3->1 chipsets that were required for a mother board design. They solved that problem. Paul probably remembers this pretty well. Once solved, they looked for the next weak link.

At some point, the scale will tip back in owning a foundry might be more beneficial and removing part of the supply chain that has most risk for success.

I think it is a question of 'when' rather than 'if'....just a matter of time/circumstances.

 

[wruby]

I think the Intel article is a reflection of the on-going consolidation trend in the semiconductor industry. Larger companies are buying up the smaller ones, and fabless startups are not getting funded due to large investments and risks involved. I agree that differentiation in this space fundamentally boils down to HW, SW, and/or manufacturing, but disagree that manufacturing is an easy winner. Part of HW differentiation is architecture - e.g. ARM RISC vs. Intel x86. It seems to me that a RISC architecture is better suited for low-power applications. Part of SW differentiation is the operating system - e.g. Apple O/S vs. Windows.

If Intel is choosing to differentiate through manufacturing, that's fine - they have the capability to do so. But in doing so, to declare the impending death of the fabless semi model is, in my opinion, myopic. Consolidation will continue, but I do not think that it will lead to every semi company also being an IDM and investing in billion-dollar fabs.

 

[Daniel Nenni]

“Bohr claims TSMC’s recent announcement it will serve just one flavor of 20 nm process technology is an admission of failure. The Taiwan fab giant apparently cannot make at its next major node the kind of 3-D transistors needed mitigate leakage current, Bohr said.”

Not true of course. TSMC has been working with FinFets and has 20nm FinFet silicon up and correlated. Morris mentioned it in the most recent conference call and I know this to be true from my work with Virage Logic (leading SRAM company bought by Synopsys) and Solido Design (Solido owns the market for SRAM high sigma verification). SRAM is used to ramp new processes and I know SRAM people, believe it. So TSMC will have 20nm planar and FinFet (my opinion). Intel will only have FinFet.

“Now FinFET for significant performance case, we’re going to introduce FinFET after the 20-nanometer planar. We’ve been working on FinFET for more than 10 years. We’re quite confident that we will have a robust FinFET technology.” Morris Chang, Taiwan Semiconductor Manufacturing Company Ltd. (TSM) Q1 2012 Earnings Call April 26, 2012 8:00 AM ET

 

[bhavana]

Hello John and all,
very interesting discussion. Me too is newborn as compared to you all, very limited experienced of working on SRAM with one of the semicon firm in USA. what I feel personally is that this is very generalized statement. There is no silver bullet solution. Intel is pioneer of technology innovation specifically process technology. That is INTEL's competitive advantage over fabless firms and foundaries. The economies of scaling and scope will decide about other fabless firms.. how long can they survive. what if tomorrow similar to open source software technologies, open source fab technology blue ocean emerges? then in that situation fabless firms will be the customers instead of competitors and new blue ocean will be created.

 

[George Storm]

@ Daniel "the origin or the foundry/fabless industry was due to the fact that the IDM model was broken"
Really? It did not look "broken" for Intel, TI, Motorola, National and quite a few others.
It seems to me that the reality is that there were always opportunities that IDMs would not exploit.
However, there was also an issue with initial investment, so it was difficult for early foundries* (was Orbit the first pure-play foundry?) to gain traction.
In fact I suspect it needed a combination of non-western business model and culture for foundry to initially become competitive.

Perhaps Intel are only trying to state the position in their (very large) corners of the semiconductor business: it does not see the fabless model as currently very competitive in Intel's core markets - and they believe they can expand into markets currently dominated by other players (plausible - but remains to be demonstrated).

*Possibly this difficulty was a motivator for Ferranti et al to introduce uncommitted arrays

 

[George Storm]

@Daniel (sorry if I'm being a pain): "one flavor of 20 nm process technology" and "we’re going to introduce FinFET after the 20-nanometer planar".
Does this mean that TSMC will introduce only one flavour initially and then add a new flavour, or that there will be a single flavour with optional added toppings (perhaps a modular approach built upon fully-depleted SOI)?
(Initially a vread base with cheese only?)

 

[Daniel Nenni]

I think the first TSMC 20nm version (planar) will have red sauce followed by a pesto version (FinFet). Okay, now I'm hungry! Remember, TSMC is customer driven so whatever the customer orders will be delivered.

At the TSMC Symposium Morris also mentioned 18nm and 16nm are possible before 14nm. TSMC feels that EUV will be required for 14nm, Intel does not agree.

Either way I do not see this as a zero sum game, both TSMC (foundry) and Intel (IDM) will thrive in the new geometries. The fabless model has brought us many new innovations and a very rich ecosystem which will be very hard to break. To much money is at stake here and Silicon Valley is full of entrepreneurs who thrive on challenge and like doing the impossible. Me for example...... ;-)

D.A.N.

 

[LinkedIn]

Intel may just have wishful thinking in this matter. In fact, TSMC and others have been creating their own technology nodes, processes and such for some time. They have the collective pool of design engineers from many companies (not just one companies (Intel), and the collective CAD suppliers modeling and simulating their processes in their corner. The "IDM" model will survive, but only for a few. I do not believe the fabless model due to collective power is in jeopardy. The "real men have fabs", quote didn't play true and this statement by Intel will not either.
Posted by Perry Denning

 

[LinkedIn]

I think Intel is out of line not the Fabless world they are still pushing X86 when
ARM taking over with anderoid

1.The numbers of brand name chips made in foundry like TSMC UMC and global
foundries is going up not down.

2.The cash return on each process node is going down as the cost of going smaller
and smaller go up at disproportionately .Where moor 's law is becomming more and more irrelevant.The cost outweighs the usefullness of Moor's law.

3.So it makes good buisness to share the cost of the new process nodes by
producing in a foundry.(If you do not have deep pockets like Intel).

4.The Intel foundry model where one designs arround a Intel Atom is one of the most
useless Idea's with Intels rapid change of process nodes any longterm design will
be doomed as Intel goes smaller.look at the cost in shrinking a design from 40nm
to 22nm

Posted by Arie Lashansky

 

[LinkedIn]

I can see it happening, yes. Foundries have been getting bigger to provide the economies of scale demanded by their biggest customers. Those customers have also been demanding that foundries compete with leading edge technologies, all fair enough.

Now if I'm a small to medium sized company and maybe The process I need doesn't fit well with a foundry process or the volume I have is insufficient to interest a foundry, what do I do? There is plenty of older used process equipment on the market. At that point I can see some folks deciding that making it yourself is a better bet than depending on an off-shore foundry. I don't see anything wrong with this if it's happening!
Posted by Geoffrey Rowe

 

[LinkedIn]

I think the first TSMC 20nm version (planar) will have red sauce followed by a pesto version (FinFet). Okay, now I'm hungry! Remember, TSMC is customer driven so whatever the customer orders will be delivered.

At the TSMC Symposium Morris also mentioned 18nm and 16nm are possible before 14nm. TSMC feels that EUV will be required for 14nm, Intel does not agree.

Either way I do not see this as a zero sum game, both TSMC (foundry) and Intel (IDM) will thrive in the new geometries. The fabless model has brought us many new innovations and a very rich ecosystem which will be very hard to break. To much money is at stake here and Silicon Valley is full of entrepreneurs who thrive on challenge and like doing the impossible. Me for example...... ;-)

D.A.N.

Daniel, I agree and share that "Can Do" attitude.
Posted by Perry Denning

 

[LinkedIn]

It depends on where you target a product. When I worked at companies with their own high performance FABs, we would estimate that if our parts were done in a TSMC-like FAB we would loose 1/2 of our clock frequency. INtells FABs are like that. Intel can drop more money on transistor development than all of the foundaries put together. So, at the top end where Intel palys their game, what was stated is true.

At the SmartPhone end of things where power is number one and performance somewhat secondary, the FABless model is working quite well.
Posted by Mitch Alsup

 

[LinkedIn]

Absolutely, just listen to the "giant sucking sound" of Intel, Qualcomm, Broadcom, TI and a few other large Semiconductor and System companies vacuuming the remaining small(ish) Fabless startups. Also, for many years it has been extremely difficult to get funding for a Fabless Semiconductor start-up. At the same time see the explosion of highly-valued SW companies created in the last few years (Facebook, Google, Twitter, Dropbox, etc.) and a trend appears where the innovation is in using every possible transistor placed in front of you by Intel, AMD, Qualcomm, IBM, etc., and not on creating a chip that will solve large needs.
Posted by Raul I Lopez

 

[Daniel Nenni]

“Being an integrated device manufacturer really helps us solve the problems dealing with devices this small and complex,” Bohr said “the foundries and fabless companies won’t be able to follow where Intel is going.”

This is complete nonsense. This is not a David versus Goliath situation, this is hundreds of Davids versus Goliath. This is crowd sourcing, not unlike Twitter and Facebook where millions of people around the world collaborated and toppled ruthless dictators. This is the entire semiconductor ecosystem (Synopsys, Cadence, Mentor, ARM, TSMC, UMC, GlobalFoundries, QCOM, BRCM, NVDA, AMD, and hundreds of other companies) against Intel. Hundreds of billions of dollars in total R&D versus Intel's billions.