Editorial: Diverging paths ahead for process tools?
01/01/2000
The process tool requirements for different types of semiconductor fabs are increasingly diverging. The result may be that the "one size fits all" approach fades in the new millennium, as new tool options emerge to meet varying needs. We can see one aspect of the divergence in the new International Technology Roadmap for Semiconductors (ITRS) (see p. 31). The tables specify different requirements on the timelines for memory, logic, and system-on-a-chip technology, and there are numerous interim nodes defined.
At one time, memory technology was a year or more ahead of that for processor/logic ICs, so that some older DRAM tools could be shifted to logic processing when a new memory generation came along, particularly in Japanese companies that made both. This was a convenient way to amortize the high cost of tools. Recently, with the quest for higher speed in processors, the technology has not only equalized, but advanced processor vendors have even pushed ahead of the traditional cycles. In addition, the leading foundries have been upgrading their toolsets to match advanced fab capabilities. It might seem as if this rough equalization would unify the tool market, with all types of fabs looking for the same kinds of tools for each generation change. Instead, a much more complex picture is emerging, as the 1999 Roadmap indicates.
One problem is the high cost of masks, as resolution enhancement techniques are used to stretch optical lithography to subwavelength features. This is not a big problem for high-volume DRAM fabs, because they want to keep their steppers going as long as they can, and mask costs can be spread over millions of the same devices. The same is true for mainstream microprocessors for PCs.
But this is not true for all the other types of IC chips made by most semiconductor fabs. Shorter runs mean that complex masks costing $50,000 or more will have a severe impact on product costs in markets where margins are often razor-thin. These fabs would like to see next-generation lithography come quickly to avoid tricky and expensive subwavelength maskmaking (as long as the chosen NGL does not itself involve costly masks).
This is only one aspect of the growing divergence. While the mass production DRAM and PC microprocessor fabs want tools that give them high throughput and short processing cycles, different needs are emerging. Foundries, which are becoming a major factor in the market, want an emphasis on flexible tooling. They have a wide assortment of customers with diversified circuit needs, sometimes incorporating memory and analog circuitry, as well as a variety of intellectual property (IP) cores on the same chip. Because short cycle times are critical for many of these specialized chips, short set-up times and fast learning cycles are vital. Many European fabs making chips for telecommunications and a wide variety of embedded systems have similar needs.
Minilines are being developed in Japan for production of specialized system LSI (system-on-a-chip or SoC) designs for consumer equipment such as computer games. These chipmakers are not as concerned with high throughput as they are with tools offering lower prices and smaller footprints. They want to be able to build smaller fabs for 2000 or 3000 wafers/month, costing perhaps $200-300 million, rather than spending $1.5 billion or more for a fab to churn out tens of thousands of wafers/month. Flexibility will also be critical, since these specialized chips will have shorter runs.
While a good market will remain for high-throughput tools with all the bells and whistles, these emerging needs are creating opportunities for more specialized, flexible, fast-changeover tools targeted to the particular needs of those users who do not want to churn out millions of the same chips for three years, with some shrinks along the way. These niches in the tool market may help spread out the vendor base, rather than going toward the intense consolidation some are predicting. The days of the garage-shop start-up tool company may not be gone after all!
 |
Robert Haavind
Editor in Chief
[email protected]