The four big pure-play foundries have all now revealed their capital spending plans for 2008 with one common theme—a lack of spending! 

TSMC will spend approximately $1.8 billion in 2008, down significantly from $2.6 billion in 2007. UMC will stretch to between $500 and $700 million for 2008, down from $900 million in 2007, while SMIC leaves spending at $700 million, the same as in 2007. Chartered Semiconductor said that capital expenditure for 2008 would be approximately $630 million, compared to $758 million in 2007.

With the sixth year in a row of increased semiconductor sales and another year of growth projected for 2008, coupled to good fab utilization rates, why are the foundries holding back on expansions? Especially at a time when demand remains good and more IDMs pledge asset-lite models or asset-smart strategies that entail greater dependence on foundries in the coming years?

According to Bill McClean at IC Insights, the answer can simply be found in wafer Average Selling Prices (ASPs) at the major foundries. The problem is that they have been falling for several years and the trend was set to continue without some drastic market manipulation!

By reducing the level of capacity expansions, the major foundries are attempting to boost utilization levels above 90 percent and keep them there for as long as possible, regardless of the demand come peak production times.

The end result will in theory enable them to raise wafer ASPs over time to more profitable levels that enable them to make the investments in R&D and capacity in the future.

At a recently held conference near London, McClean ran through the wafer ASP numbers, which didn’t look good!

Looking at wafer capacity over the last several years at the major foundries also shows some interesting trends. Though the vast majority of spending has focused on 300mm fabs, the demand for 200mm capacity has also increased, forcing the likes of TSMC and UMC to seek further 200mm capacity via Chinese operations and de-bottlenecking at existing 200mm facilities.

On the flip side, only TSMC has seen significant 300mm fab expansion over the last five years, while spurts from UMC, SMIC and Chartered have given the impression that 300mm demand is robust, when the reality is something quite different.

It is interesting to see that the major foundries have not been able to persuade many customers to switch to 300mm wafers even when 200mm capacity has been tight for several years. Indeed, the recent CapEx announcements for 2008 are noted for the inclusion of additional spending on 200mm capacity!

One of the manipulations underway is to make the wafer cost transfer to 300mm less significant by forcing ASPs to a higher level. With greater capacity available at 300mm fabs, long lead times coupled to higher prices at foundries’ 200mm fabs may force both fabless and IDM customers to seek 300mm capacity.

The squeeze is most definitely on in 2008 especially when all four major foundries seem to be adopting the same plan.

But will this strategy hold?

Looking at 300mm foundry utilization rates you could be forgiven for thinking that capacity is very, very tight! Leading-edge nodes (90nm and below) remain above 90 percent utilization rates, according to the financial reports from the foundries.

However, such rates are based on utilised tools, not on available tools already installed, nor cleanroom space already available for further capacity expansion.

TSMC, UMC and SMIC all have new 300mm fabs nearing completion or ready for tool install in 2008. None, including Chartered, have fully-used 300mm fabs, though TSMC’s Fab 12 is closest to full capacity than any other.

It will be interesting to observe in 2008 how fabless companies deal with price rises and potentially longer lead times, especially in the second half of the year as demand returns. The bigger fabless companies have spread their manufacturing risk in recent years and will no doubt be less convinced of claimed capacity tightness as they seek wafers from several sources.

The smaller fabless companies look like being hit the hardest as they are more dependent upon a single foundry and lack the clout to demand improved lead times or wafer prices.

They may also suffer most because IDMs are making up a greater percentage percentage of customer type at the foundries. IDMs will leverage everything they have to ensure wafer supply.  

With that in mind, some are playing the foundries already by establishing working relationships, but in times of softening demand, orders are being pulled to fill their own existing fabs where possible. Spansion, Qimonda, AMD et al have all done this recently but with future node reliance on the foundries this will be negated over time.