Post a Comment
Semiconductor Industry Immortality
Daniel Queyssac -- Electronic News, 10/4/1999
The chip industry in recent years has heard much lamenting about the soaring cost of capital equipment for semiconductor manufacturing. However, an analysis of SEMI figures from 1991 to 1999 indicates that the size of the wafer equipment market has increased only one half percentage point relative to the size of the overall semiconductor market. Meanwhile, the productivity and capability of equipment has improved and producers are getting higher revenues for the wafer they make. The result is an increase in the percentage of wafer equipment cost to the semiconductor market of less than one tenth of a percent per year in absolute value.
|
At the same time, total capital requirements are all increasing due to the technology requirements.
However, the wafer equipment cost is certainly not the main culprit and even if one considers the total investment cost of a new fab for the same unit of production, the change in cost is minimal.
A cost of $2.5 billion for 25,000 300mm wafers per month seems a very high number, but it is the equivalent of 59,000 8-inch wafers per month in terms of unit output at equal device complexity. This is approximately $42 million per 1,000 8-inch wafer equivalent per month. A similar number for a straight 8-inch fab is $48 million ($1.2 billion divided by 25); so what is the beef – or, maybe more accurately, where is the beef in the claim that equipment prices are out of control? Considering the yield increase provided by the new generation of machines, the depreciation cost per good die is very competitive between the two generations. This is amazing considering that the number of layers to be created will probably increase by 20 percent between the two generations of wafer size.
When talking about a $5 billion investment as the future step for the next-generation fab, possibly 400mm, the number looks huge compared to the already big, $1.2 billion for a 25,000 wafer per month 8-inch fab of today. What should be kept in mind, however, is that it is equivalent to four 8-inch fabs in output and nobody today will question a $5 billion investment to build four 8-inch fabs.
The question to ask is: "Who will be left to take such large gamble on production increase with a reasonable assurance to be able to sell the output?" So, the dilemma facing the industry is not that the cost of fabs is out of control but that a new fab decision involves a larger and larger amount of output to bet on.
Some of the consequences of this trend are visible today already; since a lot of players are not comfortable forecasting the sale of higher and higher units of output, they choose to go the foundry route or get out. Foundries, by pooling the demand of many IC companies, can afford the output of the new generation of fabs.
Some companies in the electronic segment are electing to get out by severing all ties to their semiconductor unit. After being let loose, those new companies are not very likely to build additional fabs but instead will go with foundries. Recent examples of this trend are Rockwell, Motorola, Harris and National.
The only people left around the multibillion-dollar fabs' poker table will be the memory producers and the foundries. Of course, such a radical pronouncement is a little bit extreme and some other semiconductor manufacturers besides memory and foundry will build multibillion-dollar fabs, microprocessor producers for one and few generalists, but that will be it.
A way around this conundrum for medium-volume manufacturers could be to build smaller output units but most likely the loss of productivity will bring their cost at or above the foundry-selling price, so unless proprietary technology is involved, subcontracting the work will become the best way. From the above reasoning, it seems that the trend toward foundries is going to continue and amplify, in my opinion (shared, by the way, by many others), for some very fundamental reasons.
Daniel Queyssac is president and COO of ASM Front End Operations. This is the second in series of articles by Queyssac examining capital equipment cost issues.
