Are “finishing fabs” the answer to 450mm wafers?
06/01/2006
The 2005 International Technology Roadmap for Semiconductors (ITRS, p. 75) has generated a lot of press by designating the 450mm wafer size for production in 2012. Some of the issues being raised include the cost of developing the conversion tools-estimated at US$20B-and the ongoing business costs of improving the existing technology. According to a recent VLSI Research report, the total semiconductor equipment market was $25B in 2004. If one applies the rule-of-thumb that 15% of sales revenues are applied for R&D, and that one-third of that is for new product development, then the total annual funds available for new developments (including 450mm) are ~$1B. Since 2012 is six years from now, having the technology ready by then would require an investment in 450mm equipment development of $2.3B/year, beyond the ~$1B available.
The advantages of employing larger wafers accrue to high-volume manufacturers (i.e., those running tens of thousands of wafers per design). IDMs producing smaller lots must carry the cost of the mask set over fewer wafers so they have little advantage in moving to the larger wafer size.
So where is the industry headed? Not counting system-on-a-chip or 3D circuitry, but only considering a continuation of existing processes, there are at least two viable scenarios. Neither will involve business as usual, except for the very largest manufacturers.
Segmentation of manufacturers. The large-volume IC makers will continue to progress along the larger and larger wafer path, with 675mm wafers emerging after 2020. The lesser manufacturers will remain with smaller wafer sizes. The argument against this is that the suppliers will not support multiple wafer sizes, but this ignores the availability of bridge tools. With about one-third of the total market a potential player in this smaller size arena, the equipment suppliers, out of business sense, must support the smaller manufacturers. But then, the recovery of investment in the larger wafer-size tools will be spread over fewer tools. The consequence is that the cost of 450mm tools will increase even more.
Collaboration and co-ownership. The second scenario is based on the number of collaborations among potential competitors that are being, or have been, formed. The Crolles Alliance is an example of where the future could lead. The cooperation of Philips, STMicroelectronics, and Freescale Semiconductor to establish a facility to develop future processes was understandable because it provides a base for sharing research, tool, and consumable costs. However, when the participating companies are in production, their products could be competing for the same markets. How can this make business sense for them, and provide for the opportunity to retain their own IP?
Consider a possible future scenario where wafers are 450mm or larger, and the cost of the mask set is US$10M or more. The only means of covering that investment is to produce high volumes of a few designs. Assume that the designers from the companies in the alliance meet and agree on a small number of basic design structures. If this future facility produces 500,000 wafers/year and has only 10 designs, the contribution per wafer of the $10M mask would be $20; and with hundreds of devices per wafer, the cost is very manageable.
Now, if the centralized wafer fab (or foundry) only built the wafers up to the early metal levels, customization by the individual companies at the higher levels could provide IP advantages. This finishing fab (or mini-fab) co-located with the foundry would handle the finalization of the devices according to their individual specifications. Pushing the high-cost equipment and masks to a centralized facility provides for a large amortization base. It then makes business sense to co-own the facility.
Without financial support from the companies that will benefit most-the high-volume manufacturers-the implementation date of 2012 will not be met for 450mm wafer manufacturing. The funds are simply not available, and even if they were, the investment recovery might take a long time. Thus, one can make a strong case for 450mm missing 2012 by a number of years.
 |
Contact Walter Trybula, PhD, IEEE fellow, and a director and senior policy fellow at The Trybula Foundation Inc., 4621A Pinehurst Drive South, Austin, TX 78747; e-mail [email protected] or at http://www.tryb.org.