I haven't been able to get my hands on the ICE Industry Status Report yet along with some other independent information survey sources. From memory (the last Status report and some recent periodicals) it looks as if sub 0.50u processing is crossing into the high 50% to low 60% threshold. It will take some time to get a real good handle on this but at a cursory glance:
DRAMs-SRAMs-EPROMs-FLASH-along with other Memory type products are mainstream at 0.35u or lower with the major providers in 0.25u production and ironing out their 0.18u in Process Development. This is a cutthroat business and I would guess that those not at 0.25u are scurrying to get there.
Microprocessors - The INTC and AMDs are at 0.25u now from all the articles I have read. AMD just pulled out of numerous Investment technical seminars recently and will do so in the future, as they ramp up their 0.25u process. They are tired of the microprocessor harassment they are getting from the analysts. In the article a few posts ago they stated they were not just a microprocessor company and not to fixate on that since they have other product lines. We can therefore assume that the other CPU manufacturers like NSM and IDTI are ramping 0.25u for their advanced devices.
LOGIC and ASICS fall into the same category. Right now I can safely say that 0.35-0.50u technology is mainstream with a heavy emphasis on the 0.35u process with the lead companies like LSI moving rapidly toward 0.25u processing. Right now I can name an ASIC supplier that is shipping 1.0u, 0.75u, 0.50, and trying to get 0.35u into high production while working on the 0.25u libraries to get into production by the end of this year.
The above is really the big rub. Non Memory devices used to be 1-2 process generations behind Memory and CPU device technology. As the life cycles of design rules (critical feature sizes) get more and more compressed, it seems that we are about to reach parity in the industry. This is good and bad. It used to be that an outmoded DRAM fab made an excellent ASIC factory relative to extending the life of these expensive fabs. As we reach parity, this will no longer be the case. However, if properly integrated, there are certain synergies between the two types of processes that an integrated manufacturing site might be able to handle both types of devices for a more stable manufacturing production environment such that they weather certain downturns a little better. you do give up certain efficiencies as you move away from low mix high volume to higher mix and lower volumes. It remains to be seen if this can be adequately done. I know the ASIC producers can easily adopt this strategy. It is the DRAM and other high volume producers that may not have the "talent" to make it worthwhile ( a little plug for Engineers like me that have worked both sides of the fence and know how to do it efficiently).
Finally, the rush to advanced technology is done for three major reasons: cost efficiencies(more die per wafer), higher yields due to smaller die and lower defects per square centimeter (more die per wafer) AND for technological performance functionality (higher ASPs and margins).
However, you still have the old products and customers to take care of that do not need to have the advanced technologies for the markets they serve nor do they want to bear the costs involved with the advanced technologies. This is what is affecting the mix and shift to 0.25u at many diversified fabs. True the CPU and DRAM guys have a fast trailing off of their products but they still need to make low end devices for the under $1000 PCs or the WebTV boxes. But what about the industrial controls industry that designs a product and its life cycle is 5-10 years. These guys do not want to re-design and requalify new design rules for their stable products. A redesign and requalification could take up to 6 months and costs could run close to $250K or more. That is pure bottom line waste for many companies. Each technology life cycle is getting shorter but you still have these older technologies that have to be dealt with.
As these twilight technologies and devices become a burden for companies, they foundry them out, end of life them, or give the technology away to second tier companies whose fabs cannot do the advanced technologies but are well suited for these twilight devices. These companies will also buy used equipment cast aside by the big boys that were upgrading their fabs. These twilight technologies are usually a good business since the new providers jack up the price of the parts to higher ASPs and margins since they have the end users over a barrel. I do not know what percentage of the business Dallas Semiconductor has that deals with twilight technologies but I do know where some of their chips reside on PC products. Some of their products are not feature size driven so they do not need to rush to the advanced technologies. Companies like Orbit and SGS Thomson deal with the automotive industry where devices are designed in for long car life cycles of 3-5 years. Automotive is a tough business. They are all looking for yearly price breaks but you are locked into a device for the life of a model generation.
When I get more visibility, I will report back. Hope this helps you get a perspective on things. Smaller is not necessarily better or needed.
Andrew |
