mannerisms

Ruminations on the electronics industry from David Manners, Senior Components Editor on Electronics Weekly.

The Myth Of Commodity CMOS

A few years back you couldn’t go anywhere without hearing someone say: ‘Advanced digital CMOS is now a commodity – available in any foundry’.

Usually this pronouncement came from financial people – bankers’ analysts and the like. These guys, befuddled by numbers, pushed IDMs to go fab-lite.

 

The financial community regarded it as a sine qua non that fab-liteness or fablessness was the only sane strategy for a semiconductor company.

 

The debacle at GloFo has now made it clear that the financial community was – not for the first time – talking bollox.

 

The GloFo saga shows that manufacturing advanced digital CMOS is bloody difficult – as everyone in the industry always knew.

 

Intel, TI , Toshiba, Samsung and a few other a downy old birds didn’t fall for the financial community’s bollox.

 

Now, excellence in digital CMOS manufacturing is seen for what it is – a crucial competitive differentiator.

 

TSMC has already gone into business to make LEDs and solar cells on its own account, and it has a pretty well captive SOC business in Global Unichip, so what if it capitalises  on its almost unique excellence in manufacturing advanced digital CMOS and starts making proprietary ICs?

 

The fabless and fab-lite would be well and truly caught with their trousers down.

Tags: commodity, digital cmos, myth, pronouncement, semiconductor company

Related Tech News

26 Comments

  1. Robert
    December 07, 2011 03:19

    For me TSMC’s clear lead over all their traditional competitors, beg the question, How will Intel respond?
    When TSMC becomes the defacto leading edge Fab for Everyone, then it will be in TSMC’s business interests to somehow regulate the plethora of Fabless companies using its fabs. What I’m saying is that it does not make business sense for TSMC to loose the fab volume of a major consumer sector, just because(for instance BRCM and QCOM want to fight). They may not care whether BRCM or QCOM is their wafer customer, but they certainly do not want MTK (traditionally a UMC customer) taking market share because, their fabless customers cant agree.
    I cant see QCOM accepting a finger waging from TSMC, so it makes one wonder what other options they will explore? INTC?

  2. Mike Bryant
    December 06, 2011 22:36

    I think terroists droping a nuke anywhere would change the world dramatically !
    Much more likely is an earthquake and tsunami. Have a look on google maps to see just how close TSMC is to the sea and how little elevation there is.

  3. Keith
    December 06, 2011 19:39

    Perish the thought of it, but if some bunch of terrorists dropped a nuke on TSMC’s fabs in Hsinchu, the picture would change dramatically.
    Advanced semi processing == National security.

  4. David Manners
    December 06, 2011 18:06

    I would never have thought of that, Mike, but you’re absolutely spot on. If the japanese had tried to compete with TSMC and failed it would be a massive facer for Japan. That would explain why the Advanced Process Semiconductor Foundry never got off the ground despite all the high-level backing for it.

  5. David Manners
    December 06, 2011 18:01

    An excellent point, HJ, and point taken

  6. HJ
    December 06, 2011 17:58

    David – Yes, exactly.
    Intel indulged in “anti-competitive” and “unfair” conduct – not just “aggressive conduct on the merits”.
    As I said, were it easy to maintain a monopoly by fair means, why would Intel use foul?

  7. David Manners
    December 06, 2011 17:47

    I can only remind you, HJ, of the words of the Chairman of the FTC Jon Liebowitz in announcing that Intel would be operating under a Consent Decree after the investigation of its monopoly. “”We believe Intel stepped well over the line of aggressive competition on the merits, and engaged in unfair, deceptive and anti-competitive conduct. The sum total of all this anti-competitive conduct unfairly prevented companies from competing, bolstered Intel’s monopoly, and harmed consumers by stunting innovation, diminishing quality, and keeping prices higher than they would otherwise be.” And how long has Intel enjoyed its monopoly position?

  8. Mike Bryant
    December 06, 2011 17:30

    I do wonder if the Japanese reluctance to have a JV fab is that it would have highlighted just how far TSMC have moved ahead of Japan, not just in process technology but more importantly in manufacturing. It is the efficiency of TSMC’s fabs more than their technology that is the juggernaut that is currently squashing alternatives. By not competing directly it could be that no face is lost.
    The Common Alliance have made a valiant attempt to offer an alternative and one could have expected to see a Japanese member there. The lack of one possibly highlights the same problem as being in the IBM club moreorless forces you to adopt ARM’s cell libraries, again maybe causing a loss of face.

  9. HJ
    December 06, 2011 17:21

    David – the Sherman act was really about preventing cartels and monopolistic practices, designed to illegally maintain a monopoly/pricing position.
    Provided companies don’t act illegally, it really is very difficult to attain and maintain a price-inflating monopoly. There is a huge body of work by economists on this subject.
    Governments are the biggest factor in causing, or enabling, monopolies, through regulation, licensure, etc..

  10. David Manners
    December 06, 2011 17:07

    But as history and the 1890 Sherman Act tell us, HJ, sometimes markets don’t self-correct and governments have to step in and stop monopoly abuses e.g: Standard Oil, AT&T, IBM, Microsoft and Intel.

  11. HJ
    December 06, 2011 17:03

    Ian Dedic – but, as economists will tell you, it is very difficult to maintain a monopoly in a truly free market. Once someone has, or looks like attaining, a monopoly, the incentives for others to attack that monopoly rise.

  12. David Manners
    December 06, 2011 17:03

    Absolutely Ian, when Hashimoto-san was putting that project together, with luminaries like Makimoto-san of Hitachi and Sasaki-san of NEC lobbying the government and the industry, and a government that seemed to be willing to put a lot of money up for it, there seemed there could be nothing to stop it. Then 65nm came and went and it didn’t materialise, and they tried again at 45nm and it still didn’t materialise, and it began to be apparent that a huge opportunity had been lost – as if Japan was handing over its manufacturing mantle to Taiwan.

  13. Ian Dedic
    December 06, 2011 16:52

    A couple of years ago I was convinced that the Japanese government would back (or force…) the founding of a Japanese IDM fab consortium, as opposed to having almost all their domestic electronics industry reliant on offshore silicon suppliers (like TSMC) within a few years.
    It would have almost certainly been for the good of the industry as a whole not just in Japan, but unfortunately it didn’t happen — as you say, due to inability to reach agreement on technology, I suspect nobody wanted to see such a consortium adopting a competitor’s process so instead they all ended up with nothing (or TSMC).
    The issue with TSMC isn’t their technology or their marketing savvy or their business model, it’s that they’re moving closer and closer to a monopolistic position with each new process (unless you’re Intel or Samsung), especially with GloFo’s latest reported problems — anyway GloFo are AMD to TSMC’s Intel, a much smaller second source with less to invest in R&D and (arguably) inferior technology as a result.

  14. David Manners
    December 06, 2011 15:59

    I assumed as much, HJ.

  15. David Manners
    December 06, 2011 15:57

    Maybe I’m getting my knickers in a twist over this, HJ but, with only Intel and TSMC in volume production of 28nm (or 32nm in Intel’s case) then the idea that advanced CMOS is a commodity is bollox.

  16. HJ
    December 06, 2011 15:53

    That last comment was by me, incidentally.

  17. Anonymous
    December 06, 2011 15:51

    David – I didn’t say that the alternative of joint-owned fabs was unviable. I said that you haven’t demonstrated that they are a a viable alternative – especially for most semiconductor companies.
    What I am questioning is your assertion that “it is clear the the financial community has been talking bollox”. You may disagree, but that doesn’t mean that their view is “bollox”.
    You are, in any case, conflating two separate things. Whether CMOS has or has not become a commodity is an entirely different issue to whether or not a going fabless is the only viable strategy for most semiconductor companies.

  18. David Manners
    December 06, 2011 15:41

    Well of course if GloFo, Samsung (for logic), UMC and even maybe SMIC can get 28nm up and running with good yields next year then I’m being unreasonable, HJ, and the financial community wasn’t talking bollox. But at the moment it doesn’t look like those companies will get their 28nm act together and, if they don’t, the financiers will be definitively shown to have been talking bollox. As for the alternative of jointly-owned fabs – Crolles worked for a while, and the Japanese went a long way with their joint-ownership initiative before dropping it, while the Europeans are still discussing the 450mm line idea – so maybe the alternative of jointly owned fabs can’t be definitively adjudged unviable.

  19. HJ
    December 06, 2011 15:20

    You can’t reasonably assert that “the financial community was talking bollox” just because the Japanese looked at an alternative in 2005 and because something similar has been “discussed” in Europe. They haven’t happened and so you can’t say that they are definitely a viable alternative
    If I remember correctly, it wasn’t all that long ago since European semiconductor companies pulled out of their joint venture Crolles facility…

  20. David Manners
    December 06, 2011 15:08

    Reading beyond the headline, HJ, the first line of the post states: ‘Advanced digital CMOS is now a commodity – available in any foundry’. CMOS has been ubiquitous for 25 years. It is advanced CMOS which is in short supply. And as I pointed out to Aristotle in an earleir comment the difiiculty and expense of advanced CMOS didn’t make fab-lite inevitable – the Japanese looked at combining forces for a jointly-owned government-backed fab with a project called Advanced Process Semiconductor Foundry Planning in 2005, and something similar has been discussed in Europe and, with the 450mm pilot line project, is still being discussed in Europe. So the idea that fab-lite was an inevitabilty is wrong – there was another way to go – jointly-owned fabs.

  21. HJ
    December 06, 2011 14:56

    CMOS, David, or Advanced CMOS?
    The headline just said CMOS.
    Clearly, if we are talking about the most advanced CMOS, someone has to be first. That doesn’t mean that it isn’t, or won’t very shortly become, a commodity
    The point is that your assertion that it is not right that semiconductor companies should go fab-lite really doesn’t stand up to scrutiny. Very few semiconductor companies have the resources to invest in the process development or capacity necessary to keep competitive in this area – and especially not for it to be a “competitive differentiator” ahead of what you can buy from huge foundries like TSMC (which is a market leader).
    In any case, nobody ever suggested that Intel should go fabless.

  22. David Manners
    December 06, 2011 14:12

    I did say, HJ, that the myth was that CMOS is a commodity widely available from any fab. Clearly leading edge advanced (28nm) CMOS foundry is not widely available.

  23. HJ
    December 06, 2011 12:43

    David Manners is wrong.
    Just because something is very difficult to produce, doesn’t mean that it isn’t a commodity.
    DRAM is very difficult to produce, but nobody disputes that DRAM is sold and bought as a commodity.
    Oil is pretty difficult to extract from the North Sea, but it’s still a commodity.

  24. Robert
    December 06, 2011 03:02

    In many ways TSMC already has the full service one stop shop of an IDM, its just that they don’t control product marketing and distribution. What I mean is that TSMC already selects and anoints the winners in many sectors. As a business they do not blindly enable any design company but rather they focus on sectors where enabling a new fabless player will take product share from a competing fab or IDM. Their business model is far more sophisticated than you give them credit for.

  25. David Manners
    December 05, 2011 18:33

    You’re absolutely right, Aristotle, the expense is too great for all but a handful. But that didn’t mean that fablite/fabless was the only way to go. Jointly owned fabs could have been another way. In 2005 the Japanese government set up a company called Advanced Process Semiconductor Foundry Planning, which aimed to raise $1bn for a 10,000 wafers-per-month fab and another $5bn to extend it to 60,000 wafers-per-month. The CEO was a NEC guy called Hirokazu Hashimoto, who told me at the time: “Japan has 11 integrated device manufacturers, but their fab capacity is not big, and they can’t compete with TSMC which has 60,000 wafers-per-month fabs.” The Japanese thing fell through because of disagreements over processes. In view of the current state of the Japanese semi industry I expect they now wish they had agreed. In Europe there have similar proposals for the European manufacturers. As in Japan, you get the feeling the authorities are/were willing to help fund such a thing. I don’t think fablite was an inevitability.

  26. Aristotle
    December 05, 2011 17:29

    Leading edge CMOS manufacturing is very difficult and also VERY VERY expensive. Only a tiny handful of IDMs have the revenue to build/support a state of the art fab – moving to fabless model has been the only option for most.
    You pose a good question about the likes of TSMC up-integrating to sell its own ICs. There is certainly higher margin to be gained but will need to take care wrt concerns with competing with its own customers.

Share your knowledge - Leave a comment