entry diversion in semiconductor manufacturing

[buzunca]

Hi everyone,

I am a PhD student in Strategy at IESE Business School and I use the SEMI world fab watch dataset. I have an expertise question to you all, I would be grateful if you can answer:

Because of the greater efficient scales (economies of scale) obtained in memory and logic submarkets, a new possible entrant to these submarkets can decide to change its mind and enter discrete (LED, power etc.) or MEMS submarkets which have smaller wafer sizes and greater geometries compared to logic and memory submarkets (which makes these submarkets easier to enter). So the new fab entry is not 'deterred', but it is 'diverted' to another submarket in the industry.

Do you confirm this argument? if so, can you come up with one specific firm that did so.

Bilgehan

 

[Daniel Payne]

Bilgehan,

Good morning.

When a company sets up a fab they have to understand their end market first, before building the fab, because the equipment, process steps and skills used in fabrication of Memories is quite different than that for the design of MEMS.

If a market is easy to enter, then you will have more competitors attracted to that market which then makes it more difficult to enter.

I wouldn't say that any IC market is easy to enter, because you have to build awareness for your fab, hire sales and application people, get design wins, build proof and credibility, pay for the fab, then offer something different than your competitors.

In general the number of fabs has been slowing worldwide because with each new smaller process geometry one new fab can produce many more chips per wafer than the previous, larger geometry generation.

 

[buzunca]

Hi Daniel,

Thank you for the quick reply. The information you gave me is very valuable. While you are right in terms of number of competitors in a submarket, I will control for that in my analysis (by putting a variable of fab count in the regression). Let me elaborate a little more to tell more about what I am doing and what I think is the pattern in industry.

Here is what I have been working on so far in a nutshell:

I try to come up with a new theory in my thesis, which I call ‘Entry Diversion’, where entrants are 'diverted' between submarkets rather than deterred completely from the industry. Looking at the industry composing of heterogeneous submarkets (in semiconductor manufacturing these could be analogs, discretes, MEMS, logic, memory chips etc.), I propose that occupancy by incumbents in main focal submarkets conditions new entrants’ incentives from imitating the incumbent to finding new sources of growth, i.e., coming up with new products by opening new submarkets.

As we all know, scaling production either through chip geometry miniaturization, or wafer size enlargement provided improved functionality per chip. Looking at these important characteristics of semiconductor fabrication, i.e., shrinking geometries and increasing wafer sizes, we can map submarkets on a two-dimensional scale: one dimension with 'geometry' or ‘die size' in microns, and another dimension with 'wafer size' in millimeters/inches (see attached graph). As we know, depending on the size of the wafer and the chip geometry, any number from tens to tens of thousands of dies can be processed on a wafer. In such a submarket-map, high capacity production plants (fabs) that focus on commodity chip production, such as memory or logic chips, would be located in the high-wafer size and low geometry section, whereas low capacity ones with more particular product characteristics, such as discrete chips (especially LEDs and Optos), analogs, and even MEMS, would be located towards the other extreme, in the high geometry but lower wafer size part.

The entry pattern observed in recent years is that new fab entry occurs in discretes such as LEDs, optos and MEMS submarkets, whereas memory and logic submarkets have practically 'zero' entry. My question is: are there any fabs/foundries that were previously planning to enter into memory/logic submarkets, however since they cannot catch the scale economies through the wafer size and geometries that TSMC, Intel etc. type of giants have, they changed their mind and 'diverted' their entry into discretes or MEMS submarkets?

Thank you.

Bilgehan Uzunca
PhD Candidate, IESE Business School, University of Navarra,
Strategic Management Department
Avda. Pearson, 21 | 08034 Barcelona, Spain
+34.66.479.97.76 | buzunca@iese.edu
IESE - Bilgehan Uzunca

Bilgehan,

Good morning.

When a company sets up a fab they have to understand their end market first, before building the fab, because the equipment, process steps and skills used in fabrication of Memories is quite different than that for the design of MEMS.

If a market is easy to enter, then you will have more competitors attracted to that market which then makes it more difficult to enter.

I wouldn't say that any IC market is easy to enter, because you have to build awareness for your fab, hire sales and application people, get design wins, build proof and credibility, pay for the fab, then offer something different than your competitors.

In general the number of fabs has been slowing worldwide because with each new smaller process geometry one new fab can produce many more chips per wafer than the previous, larger geometry generation.

 

[Daniel Payne]

Thank you for the chart showing wafer size and geometry.

Only a handful of companies have enough capital ($4B and up) to build a new fab for 28nm and smaller nodes used in SoC design, while the larger nodes are less expensive and easier to enter.

MEMS is a double-digit growth market, so it will naturally attract newer entrants, and the geometry needed for MEMS is large.

I think that economic opportunity is what drives fab building decisions.

 

[Michael Clayton]

Many companies entered the memory market over past three decades, and most failed within one technology generation, or two at most, so the fabs were often diverted to making analog or mixed signal products, and when that business died from newer high tech competition, those fabs turned to MEMS or Power devices OR whomever owned the fab at that point simply sold off all the equipment and real estate. That is not what you are trying to prove, but should be easy trail to follow on the web.

In Qimonda US memory case, they could not sell the fab I was told, but TI bought the equipment, moved it to existing fab building in Texas, and is using it for analog and mixed signal products I hear. So bleeding edge memory fab equipment is useful long term for less litho-sensitive products. The used equipment business seems better than the used fab business since there are often OTHER reasons for fab early failure, such as country tax policy changes, or labor shortages for high tech workers, or distance from customers.

But if you simply focus on the coarser geometry products, you can find entries who buy up older fabs (like microcontroller designers buying older microprocessor or memory fabs). Microchip and On Semi both bought vacant fabs in Gresham Oregon (years apart, but same idea). But we could not find any examples of MEMS designers who decided to make Analog or Power devices instead, or vice versa, to fit your diversion model. Within IDM's in old days, however, many internal older fabs were diverted to MEMS or Power or Analog from earlier use at Memory or Microprocessor fabs. But "post decisional cognitive dissonance" is common after a design team starts up a fab. But usually they hand off to a different design startup, but do not "learn the other design field" due to situation that did not fit first idea.

 

[buzunca]

Hi everybody again,

As far as I can see, the comments here relate to a change in my ‘diversion’ story. Mike, you say that ‘Many companies entered the memory market over past three decades, and most failed within one technology generation, or two at most, so the fabs were often diverted to making analog or mixed signal products, and when that business died from newer high tech, those fabs turned to MEMS or Power devices OR whomever owned the fab at that point simply sold off all the equipment and real estate.’
So the diversion story might go like this: firms that are able to escape from the capital intensity and scale required in memory and logic submarkets divert themselves towards discrete and MEMS.

My question to readers of this thread then becomes: Who are these firms? Do we agree with the view that many companies that were originally in logic/memory evolved towards discretes/analog/mixed signal/MEMS as a strategic choice to focus on those market segments/application spaces and escape the capital intensity and scale required for the former? What’s your opinion and further examples on that?

Thanks
Bilgehan