Fabless or Fabbed .... the ongoing Jousting

trackback url:
http://www.siliconvalleywatcher.com/mt/mt-tb.cgi/546

sites that mention Real men have fabs. . .chipmakers profit from making their own:

» Headline (and picture) of the day…. from helzerman.com
Real men have fabs.. or so says Tom Foremski in his column today citing a presentation by Gartner chip analyst Bryan Lewis. Basically, the article boils down to the last few lines.. This means that those chipmakers that own and operate fabs will be... ...read more

date September 21, 2005 01:07 PM

Real Men have Fabs is a bit trite.
This is coming from a 22yr veteran process engineer.

The Real Issue - is how to best make money from microdevices and to have your company prosper and grow.

It is not a question of $B fabs, versus nothing, it is a question of given your assets ( human, capital and plant ) how do you maximize profits. Oftentimes it is too easy to give up and take the easy way out.

Given any kind of business model, you can make money if you bother to sweat the details, understand the markets and sell well and profitably.
CLICK READ MORE... for rest of article.

One excellent fab model with no $B fab vain ego, is the superbly run Linear Technology. They sweat the details and there are NO $B capital expeditures in sight. I once ran into an IT Support engineer from LT and he was alternately meekly bemoaning the old computers running the business and taking pride in the fact they were old machines ( DEC Alphas and older PCs ) but that LT's profits ran like clockwork.

The main unspoken weakness of Fabless models is that there is mostly a huge overeliance on PEACE between CHINA and TAIWAN, that is in fact pretty darn shakey at present. That is the unspoken risk hardly ever faced with a degree of reality by any articles .

Fabless IC yields are significantly harder to improve, because the interface between the Fabless engineers and the Foundry has natural Boundaries of significance not to be trifled with. Only weak management or engineers expect the impossible and misplan with abandon.

This makes it for a substantive limiter in rate of yield improvement ( process learning for device specific issues ) but there are also other factors to consider in the Fabless business model.

Fabless devices which use bleeding edge processes MUST have high enough ASP's / Margin, to accomodate the slower rate of Yield Learning. Facts of Life in the Fabless perspective. It is not better nor worse, but different.

Low margin Fabless product types have much less margin for gross yield glitches than High Margin Fabless devices, as margins and asps are typically in a range that is considerably less forgiving of low yield or yield glitches.

Another aspect to this subject is that FABs ( factories really ), to be run well, take a certain strength of stomach so to speak, that many North American microelectronics manufacturers sometimes are weak in.

Case in Point was the mid-80's "Fear" of the vaunted Japanese IC manufacturing skills. Both National Semiconductor and Intel at the time "succumbed" to the exaggerations, and for the time in the mid 80's, there was a modest cry wolf attitude, the sky is falling, etc., about the future of IC Fab Manufacturing in the US.

The then despondent Intel, reeling from DRAM withdrawl, did not accept that it was principally due to specific technical difficulties not being properly addressed by the responsible folks that should have been paying very close attention to their work. Some Intel and other managers of wafer fab domain in the mid-to early 80's - were pretty timid technical manufacturing leaders, as one might call it.

The fear of the Japanese at the time was greatly overblown. Intel, through Barrett's Superb Manufacturing Accumen (with well deserved Fab Kudos to him), blew that one out of the water - with the result that Intel is the biggest FAB business and most successful today.

Despite the credentials of those involved in the mid 80s, the cries of "the sky is falling" - now seems pretty hollow.

The seeds of FAB supremacy of Intel today were laid with the Battle Cry of Gambati in Livermore and elsewhere ( Gambati is Japanese for On-Guard or similar ).

Limits of capability are often in the mind, from weaknesses of discipline of thought. Fabless is best for those who do not have the stomach to manage the huge complexity of FAB manufacturing and Yield Improvement. Quite daunting even to the "best" of engineers, if you don't keep your nose to the grindstone and focus focus focus, and do so strictly pragmatically.

It is too easy to overcomplicate pretty straightforward yield problems, and miss how to do the real yield improvement in fab. Often times yield is held down due to not paying attention to details, and you cannot every take anything for granted ever.

The phrase of the Reagan years - "Trust but Verify" is a mantra appropriate for fab yield improvement. If you let rest technical issues due to bureaucratic ( artificial ) boundaries, and your own FAB's yield suffers - you have no-one else to blame but yourself. It is nice to complain, but better to work hard and sweat the details.

Case in point is that time after time, low yields persist because someone or many, did not pay attention, and misinterpreted data or the limits to an experiment. Often this is subtle to the unsuspecting, but when all hell breaks loose - you better know your stuff cold.

Despite the sometimes seemingly grimy aspect of manufacturing - success at the art requires discipline and insight that some do not have, else you better count on having wages of offshore low wage factories to accomodate your sloth.

The trite phrase "Real Men Have Fabs" is overused and at times inappropriate.

"Real Men" pay attention to the details of their business and do not stray from their responsibilities, FAB or FABLESS, catchy phrases notwithstanding.

wendman nanotechnology nanotech nano mems piezoresistive+Pressure+sensor Noyce Silicon+sensor
problem+solving R&D research+development yield+improvement semiconductor wafer+fab wafer+fabrication wafer+300mm brainstorming six+sigma pdf+solutions carnegie+mellon cmu intel hypothesis nano