- How do foundries fit into the future of the semiconductor industry?
- Introduction
- At IC Knowledge we have long believed that foundries fit into the semiconductor industry in specific areas, but not in all areas. Our argument is basically that similar size, well run Fabs with the same utilization rates have similar costs anywhere in the world. Since foundries are out to make money and want to add a margin to their manufacturing costs, foundries are really only economical if the product can support the extra cost and margin erosion. We made this comment about a year ago to a foundry analyst at one of the market research firms. The analysts contention was that foundry Fabs have typically had higher utilization rates than IDM Fabs and therefore have lower costs. We argued back that foundry Fabs would only have higher utilization rates until the foundry business started to mature and then the utilization rates at foundry Fabs would move with the industry. Well, we don't know whether to say I told you so or admit that we underestimated how much things would change.
- Effect of the current downturn
- Currently worldwide Fab utilization rates are at 72.7% (see Silicon Strategies, 8/22/01) the lowest level in two decades! But contrast this to the three largest foundries, TSMC, UMC and Chartered, that are reporting utilizations of approximately 50%, 40% and 35% respectively. Apparently the IDMs have all pulled their foundry business back in-house. This suggests that the foundries are now at a cost disadvantage, and while they are not attempting to put through price increases (in fact foundry prices are eroding), eventually foundry margins must improve or they will not be viable companies (no we don't think TSMC, UMC, or Chartered are going out of business).
- Long term trends
- Long term what we believe this means is that commodity products will be made in-house by large IDMs running huge 300mm Fabs, or perhaps by consortiums of companies sharing a large 300mm Fab. Products we consider to clearly fit this commodity mold are, X86 type microprocessors, DRAMs, Flash, probably SRAMs and DSPs. In these areas we expect to see a relatively small number of IDMs or consortia producing product. DSPs in particular looks like a very interesting area to watch. The number one DSP manufacturer, TI, Fabs their own DSPs, number two, Agere Systems Fabs their own DSPs but is rumored to be considering going Fabless, number three, ADI uses foundries. We believe that Agere must continue to manufacture their own DSPs to maintain or gain share and that ADI will not gain significant share on TI unless they bring DSPs in-house as well (of course they also have to design and market great DSPs). DSPs will become more and more of a commodity over time and cost will become more important. Eventually foundries will serve two major roles, one, as a manufacturing source for unique high design value added product up to moderate volumes, and two, as a capacity safety valve for IDMs. If an IDM has four full Fabs but not enough business to justify a fifth Fab, then a foundry will be used to fill the gap until a fifth Fab is clearly justified. It is interesting to note that IDMs already pay higher wafer prices than Fabless companies, probably because the foundries recognize IDMs as fair weather friends.
- Conclusion
- The bottom line to us is that as 300mm Fabs really take off, we expect to see a lot more consolidation and consortia in the semiconductor industry. Foundry use will also grow, but ultimately will be essentially locked out of the largest product types.
- Related articles
Copyright © 2000 - 2004 IC Knowledge - all rights reserved.
