An age-old argument in the industry is: Is it affordable? Can anyone carry on paying for the horrendous costs of fabs?
A Synopsys Fellow, Tom Williams, has told a US conference that 45nm fabs will cost $3bn, and 32nm fabs may cost $10bn.
Williams reckons this will mean IDMs will get fewer.
But, as Future Horizons pointed out at their recent IFS Seminar, fabs cost have always cost the same, when measured as a proportion of the semiconductor TAM.
Since 1970, the cost of a state-of-the-art fab has grown at exactly the same rate, 14.1 per cent CAGR, as the semiconductor TAM.
1970 Semi TAM was $2.4bn, fab cost was $30m
1975 Semi TAM was $4.0bn, fab cost was $150m
1980 Semi TAM was $12.8bn, fab cost was $250m
1985 Semi TAM was $21.5bn, fab cost was $330m
1990 Semi TAM was $50.5bn, fab cost was $600m
1995 Semi TAM was $144bn, fab cost was $1.2bn
2000 Semi TAM was $204bn, fab cost was $1.8bn
2005 Semi TAM was $245bn, fab cost was $3bn
And there's always a government somewhere in the world which will put up a substantial proportion of the cost of a fab.
Look at China and India today. India will pay 20 per cent of fab costs. China probably more.
Only a wuzzy pays the full cost of fab.
So, while the cost of fab hasn't changed, what has changed is CEOs' responses. In the old days it was: "We don't care, we're doing it anyway?"
Nowadays CEOs say: "We'll do it, if the accountants let us."
The industry hasn't changed. Only the CEOs.
When Corning Glass owned Signetics, back in the early 70s, one Corning Exec, bothered by the cost of a new epitaxial reactor, wanted to know the useful lifetime. He was horrified that it was only a few years before it would be trashed or given to a University, since a pie-plate press was still going after being installed in the 1920s. Probably why Corning sold their interest to Philips.
Wonderful post , You really hit the
mark with this, I just don't understand why people quite get it.
I don't know how many people I've talked to about this very
thing in the past month, and they just don't grasp it.
Never the less, Excellent post!
Thanks Brandon. The Wall Street people who keep advocating fab-less models don't really have a clue how you differentiate product. It turns out that fabless companies aren't inherently more profitable than IDMs. And, finally, the foundries look as if they're about to jack up prices. How good is the 'fab-lite' model going to look in a few months time?
I really wish semiconductor executives would stop looking at figures that include Qualcomm when they make their decisions. Qualcomm is a hugely successful IP licencing company that just happens to run a semiconductor company as a sideline yet even at the ENIAC forum their numbers were included in the fabless pot to justify Europe going fab-lite. The only European company with IP on that scale is Nokia and last time I looked they didn't make semis.
If a semiconductor CEO doesn't have the confidence they can fill a new fab with products developed in their own R&D there are two options :
a) do a joint venture with another company that has the same lack of confidence
b) retire and find the next Pistorio to take over and motivate his R&D, marketing and sales staff to fill that fab by the time it is built.
Europe is in dire need of a lot more of the option (b) route !
Mike you're very wise. I agree. Fablesslooks good to financial types, but it looks very bad to IC developers wanting to make a uniquely well-performing product.
Also thanks for the spam warning. It seems anyone who says something particularly nice about the blog turns out to be a b.spammer! Action will be taken.