Are foundries adding too much capacity?
08/01/2003
George Burns, Strategic Marketing Associates, Santa Cruz, California
Foundries have transformed the way chips are made and because of them, Asia Pacific is fast becoming the center of the industry's manufacturing. With the success of the foundry model, more companies are moving into the neighborhood. It is becoming crowded — perhaps too crowded.
In 1987, TSMC was the industry's only foundry. Now there are almost 80 companies offering foundry services, including some leading integrated device manufacturers (IDMs) who see their foundry services as a significant source of revenue. While many of these companies don't compete directly with TSMC, many do, and their number is growing.
With a growing number of competitors, there is inevitably more and more capacity along with the threat of the debacle of overcapacity, much as the DRAM manufacturers experienced in the mid-1990s.
Foundry capacity growth and capital spending
Foundry capacity (including those IDM fabs that offer foundry services) stood at a little more than 600,000 wafers/month in 1995 (Fig. 1). Since then, 65 foundry fabs have started construction; one in four of all fabs started during that time. Last year, almost one half of all the fabs starting construction were foundry fabs.
 Figure 1. Foundry capacity more than doubled from 1995–2000; by 2005, it will double again. Source: Strategic Marketing Associates (May 2003) |
null
As a result, foundry capacity doubled between 1995–2000. By 2005, it will likely double again. From 1995–2002, foundry capacity grew at a CAGR of 16%; during this same period, chip revenues were flat and nonfoundry capacity grew at a CAGR of 5%. This phenomenal growth was fueled primarily by the fabless chip companies, whose market share has grown from 2% in the early 1990s to 15% of all chip revenues today. Since 1990, however, the growth of foundry revenues has been boosted by the decision of many IDMs, such as Motorola, Agilent, Philips, and others, to outsource a major part of their production.
At the beginning of the 1990s, foundry capital spending was just 1% of all capital spending and by 1995, it had risen to 5%; since then, it has grown spectacularly. Last year, almost one in every four dollars spent was by a foundry (Fig. 2).
 Figure 2. Foundries are a major source of capital spending. Source: Strategic Marketing Associates (May 2003) |
null
While the percentage of foundry spending to total spending continued to rise during the down years of 2001 and 2002, the total amount spent fell by 39% from >$10 billion to a little more than $6 billion.
Here comes the competition
Not all foundries cut their spending during this period, however, and this is the problem the top three foundries face (TSMC, UMC, and Chartered Semiconductor). While the top three cut their spending by 63%, the challenger foundries raised theirs by 48% (Fig. 3). The challenger foundries are basically those that aren't the top three. They include the new foundries 1st Silicon and Silterra in Malaysia; Dongbu Anam Semiconductor in South Korea; ASMC; Grace; Shanghai Hua Hong NEC; SMIC and others in China; Tower Semiconductor in Israel; and Communicant Semiconductor in Europe. They also include IDMs such as IBM and Toshiba, which are offering very advanced processes.
 Figure 3. Challenger foundries are now outspending the top three. Source: Strategic Marketing Associates (May 2003) |
null
Last year, for the first time, the challenger foundries spent more on plant and equipment than the top three foundries. This year, based on current plans, capital spending by the challengers will be >2x that of the top three. The result of this huge increase in spending is that between 2001 and 2005, the challengers will add 44% more capacity than the top three, whose share of total foundry capacity will drop from >70% in 2000 to barely one-half by the end of this year.
Though much of the new capacity coming online is leading edge, not all of it is — some foundries choose to exploit older technologies in which the competition is not as fierce. Since 2000, sub-180nm foundry capacity has increased almost 5x by more than 600,000 wafers/month, an amount equal to 50% of the total capacity online in 2000 (Fig. 4).
 Figure 4. Foundry capacity in 2000 and 2003. Source: Strategic Marketing Associates (May 2003) |
null
This year, Dongbu Anam will begin to offer 130nm capability, as will Silterra and Tower. In China, the new fabs are not being held back by technology restrictions under the former COCOM cold war regime. SMIC, for example, plans to begin offering 110nm devices on 300mm wafers in 2004 using technology from Infineon. It is already making 130nm devices for Texas Instruments. Grace Semiconductor, whose fab is designed to be 300mm capable, has the capability to begin 300mm pilot production as early as next year.
The top three are also feeling the heat from IDMs that see the foundry market as appealing. Toshiba's most advanced fab in Oita is available for foundry work at 130nm and the company plans to offer 90nm SoC processes at Oita by 3Q03.
Without a doubt, IBM's entrance into the foundry market is the most formidable threat to the established foundries, especially to TSMC and UMC. We estimate that IBM's new 300mm fab in East Fishkill, New York, is currently equipped to handle only 25% of the fab's designed capacity, but this fab is ramping fast and could be completely full by year's end.
Already, IBM's foundry offerings and fab process capability are causing problems for TSMC and UMC. AMD and UMC abandoned efforts to jointly develop sub-65nm processes after AMD partnered with IBM for advanced process development. AMD and UMC have also all but given up their planned 300mm joint venture in Singapore (Au Semiconductor) because of AMD's new relationship with IBM.
TSMC is also feeling the heat from IBM. In March, IBM announced that it will make wafers at 130nm for Nvidia, one of TSMC's biggest customers. In addition to taking away business from TSMC and UMC, IBM is helping one of their competitors, Chartered Semiconductor. IBM has entered into a technology development and 300mm capacity-sharing relationship with Chartered — the companies will jointly develop processes for ≤90nm. Chartered will have access to IBM's 300mm capacity and thus be able to delay equipping its own 300mm Fab 7 in Singapore until the end of this year.
Conclusion
The first purpose of foundries is to provide capacity, both for fabless companies and for IDMs. It is estimated that by 2005, foundries will be bringing more capacity online than IDMs. But with the large number of new foundry providers, some of this capacity will be excess. With IDMs like IBM and Toshiba now offering such foundry services, and the emergence of new advanced technology competitors such as Dongbu, 1st Silicon, Silterra, Tower, SMIC, and Grace Semiconductor focusing on the leading edge, it is likely there will be overcapacity even in the leading-edge technologies.
And the leading edge will provide no refuge. The IDMs have not only outsourced their capacity to the foundries, they have outsourced the risk of overcapacity as well. That, too, is the purpose of the foundries.
George Burns is president of Strategic Marketing Associates, a high-technology market research firm specializing in fab information since 1992, and publisher of World Fab Watch (a database), the Quarterly Spot Report on Semiconductor Fab Projects, and other fab-related reports. e-mail [email protected], www.scfab.com.