I agree my post was not coherent. The WSJ article covers the views of analysts who see continuing DRAM pricing weakness as a basis to question the long term viability of Fab expansions in Taiwan. For a long term view refer to the following article from Dataquest.
--------------------------------------------------------------------------------
Semiconductor Foundry And Vision 2010: Reshaping the Way Chips Are Manufactured
by Clark J. Fuhs
It is quite clear today to most people familiar with the industry that the semiconductor foundry is here to stay as a part of the semiconductor manufacturing infrastructure. During its origin and emergence in the early-to-mid 1980s the foundry business model, and its corresponding emerging customer model the fabless company, was viewed with skepticism as merely an opportunistic business. This business flourished during the slowdown of 1985-86, as capacity was plentiful as was the availability of creative and risk-taking design engineers. TSMC began the dedicated foundry model in 1987, and during the slowdown of 1990-1992 the foundation for growth in the fabless/foundry industry was solidified.
But Dataquest believes we are only at the beginning of a movement that will dramatically change the complexion of semiconductor manufacturing. These changes will markedly affect the equipment and materials industries, from design of equipment to technology development to business practices.
In this article, we will briefly review why the fabless/foundry model works and then roll ahead to the year 2010 to see how the landscape for semiconductor manufacturing is likely to evolve. Changes we are anticipating include:
growth of share of production done in foundries from today's 10 percent to 35-50 percent over the next 15 years
emergence of the foundry industry as the key technology driver for manufacturing, affecting how equipment is designed
changes in the manufacturing infrastructure including vertical integration of front-end and back-end fabs, the ASIC business as we know it today no longer existing, and equipment companies becoming involved in or owning process design tools selling to fabless companies and chip designers.
Why Do Foundry and Fabless Models Work?
Many presentations over the last few years typically show a chart of how the cost of a fab is getting larger over time, and chip companies use this to apply pressure on equipment companies to improve productivity. These charts are somewhat misleading because in terms of the square inches produced, the fabs are larger. So part of the raw increase in fab cost is because of the complexity of technology, but a sizable portion is also from the sheer size of the fab. If a fab is bigger, it costs more.
As any capital-intensive industry matures, there is typically a movement toward the concentration of capital in order to lower costs from economies of scale. This "concentration of capital" is exactly what is occurring in the semiconductor industry. The result of the increased cost of a fab is not reducing the affordability of fabs, but it is increasing the ante to participate. The natural effect is to promote consolidation and the further concentration of capital.
The emergence of a viable foundry industry is a structural response to the need to concentrate capital. Hundreds of customers who require manufacturing services are concentrated into a few fab sites. Economies of scale are created, and the fabs get bigger again. The average size of foundry fabs currently under construction is roughly double that of the average planned fab today. This means cost structures are lower per square inch of silicon produced.
Concentration of capital also is beneficial to integrated device manufacturers (IDMs), meaning chip companies that both design chips and have their own fabs. An increasing number of IDMs are using foundries as an extension of their own manufacturing infrastructure, off-loading older products or as an outlet for capacity management in general. SGS-Thomson, Motorola, Intel and many Japanese companies are among the larger users. In fact, IDMs continue to represent more than 70 percent of the foundry demand market worldwide.
Further, R&D spending is reduced to process development issues. A foundry has no cost structure involved with chip design or marketing, and can therefore use R&D investments in a much more focused and efficient way. Speed of development also increases because of the focus placed on the effort, as no competing influences from design or "harmonization" of processes across device types are present.
This R&D spending focus is also part of what makes the fabless model work well. No longer are designs dependent upon the processes available from "your fab" but rely upon a standard set of cell designs, tools and a pool of competing manufacturing facilities ready to help tailor the process to the desires of the fabless company. R&D investment can then be concentrated on designing chips and meeting the electronic system design needs.
The increased speed of development by fabless companies because of the focus on R&D also means a shorter time to market and faster response times to changing customer requirements. This advantage is extremely important for the success of the chip company.
Vision 2010: Growth Potential of the Foundry Market is Huge
Today, roughly 10 percent of semiconductors pass through a foundry, and this figure could grow to 35-50 percent over the next 15 years. Why do we believe in this potential? Let's look at the question another way. Why would a company want to both design and manufacture chips? There are two possible reasons.
The first compelling reason is that chip design and technology is a means to an end; that is, the chip is an integral part of the technology in an electronic system. This electronic system is the key focus of the company. A classic example is Motorola, where the technology of the microcontroller is a key driver in their success in the mobile communications business. Intel, while not a PC manufacturer, basically owns and drives the basic PC design, and would also fall into this category. Many vertically integrated Japanese and European companies also fall into this category.
The second compelling reason to both design and manufacture is that the market is large enough to sustain several competitors and achieve equivalent economies of scale to the foundry. The only market that can support this is the DRAM market.
So the semiconductor list in the future may be perhaps 25 companies that will both design and manufacture, and the rest (perhaps as many as 150 companies) will be fabless. Further, these 25 or so IDMs will be using foundries in strategic ways as mentioned earlier. Today's use ranges from 5-15 percent, and it is easy to envision that 20-30 percent of the IDM business could be outsourced to foundries in the future. This structure drives the foundry market to account for as much as 50 percent of the semiconductor production.
Vision 2010: The Foundry as a Technology Driver
The foundry industry is used to migrating to smaller linewidths faster than the overall market. In 1985, when the mainstream market was processing at 1.5 microns, the "sweet spot" of the foundry industry was roughly two generations behind at 2.5 microns. Ten years later through 1995, this gap closed to about half a generation, and today both the leading IC companies and the foundry mainstream are migrating to 0.35 micron. The race is currently on as to which companies will introduce 0.25-micron technologies into production first.
The business model for foundries has been to put in production only well-proven processes, and this has kept R&D spending below what would normally be expected. But there is evidence this is changing, for a couple reasons. First, as is the case for the semiconductor industry in general, there is higher revenue per square inch at smaller linewidth technologies, and we are finding foundries are increasing R&D spending to capture this increased potential to support growth.
But the second, and the most critical reason, is the acceleration of system level integration (SLI). The foundry has become the focus of product and process development for SLI, where DRAM and logic process flows come together in the same fab. The broad customer base of fabless companies, responding to their customer demands for SLI, is the key driving force. This convergence is why we believe Korean companies have recently entered the foundry market in a big way.
So in the long term, the device technology drivers will be the triad of the microprocessor (Intel, etc.), the DRAM and SLI (the foundry). Because of the integration aspects of SLI, it is possible that this area may take the lead in process technology development over the next 10-15 years.
Vision 2010: Ramifications of Foundry For Semiconductor Infrastructure
Changes are already starting to happen. The front-end and back-end are starting to become integrated as a "turnkey" chip manufacturing solution. The entrance into the foundry market by Alphatech and Amkor/Anam, and the back-end line established by Chartered Semiconductor are indications of this trend. Contract manufacturing in the back-end is well-developed, and this will quickly migrate to the front-end and become integrated.
Traditional ASIC companies are being pressured by the foundry offering silicon layout design solutions. The system company is starting to "skip over" the ASIC company and go directly to the foundry. The ASIC company business model may be forced to change to become more system design and software focused, with the actual mechanics of manufacturing migrating to the foundry. SLI will also push this movement.
The tightly managed foundry manufacturing model will continue to apply pressure on equipment and material suppliers to "own" more of the process performance parameters. The ownership of process performance may increase to the point where suppliers will want to have influence in the design aspects of the device. As this movement of ASIC company focus occurs, it is possible that the semiconductor equipment company will find that decisions for design-in of process steps will start to be made earlier in the design process. There are several design tool companies starting to provide tools in this regard to the fabless company or chip designer. It is quite possible that the strategic direction of an equipment company will lead it into this tool business.
Vision 2010: Exciting Times Ahead
Certainly, no one involved in the semiconductor industry would consider it boring. We do not expect that to change over the next 10-15 years. However, we do believe that many companies and industry observers are underestimating the longer term ramifications of the foundry industry. Momentum is building today and changes are starting today. We believe equipment and material companies need to evaluate their strategic business plans to take these potential changes into consideration.
About the Author . . .
Clark J. Fuhs is director and principal analyst for the Semiconductor Manufacturing Group at Dataquest, San Jose. He may be reached by phone at 408.468.8375 or by e-mail at cfuhs@dataquest.com.
|
