Issue



Economics will dictate the future


08/01/2001







Click here to enlarge image

It takes more and more investment to produce each generation of microchips in the march along the Moore's Law semi-log curve. While chip functions get steadily cheaper, and shrinking features allow more functionality and performance on each chip, the tools and fabs to manufacture them escalate in price. Looking ahead to the 50nm era and beyond, it may become necessary to shift from the workhorse optical lithography to EUV (x-rays) or projection electron beams to print circuit patterns on wafers. There are rumbles that EUV steppers might cost in the neighborhood of $40 million each. Building and equipping the fabs for them may cost $10 billion or more within this decade.

Who will be able to afford these escalating investments? The answer: Only integrated device manufacturers with huge markets for their chips, or major foundries producing chips for many vendors. Smaller players won't have the capital to build their own dedicated wafer fabs. The industry will go through profound changes as a result, if we continue on the same path.

We recently heard from an industry veteran that he attended a meeting where an Intel speaker actually seemed pleased that steppers were heading toward the $40 million range, probably because few if any competitors could afford to equip fabs with such expensive tools. This would narrow the number of players in chip markets, whether for logic and mixed signal, processors, or memory. Only the richest could stay in the game. The major foundry players are planning multiple fabs at their 300mm plant sites, all sharing resources and support facilities, thus helping to spread overhead costs across a number of factories rather than just one. A large proportion of the world's chips might be made in a few monstrous "fab farms," in Taiwan,

Singapore, and perhaps China.

Intel remains the giant in the US, although it needs to become a leader in areas like workstations, telecom, and portable devices, with standard chips that can be made in the millions, to go along with its entrenched PC microprocessors. IBM has become the leading US foundry, but it is more of a boutique fab serving only those needing large numbers of the most advanced chips. Collaboration is growing in Europe, as Philips engineers design chips within an STMicroelectronics fab. European chipmakers have focused on mixed signal processes and designs, avoiding the commodity processors and memory chips more common in Asia and the US. Meanwhile, fabless chip design companies are proliferating because of the broadening range of chips needed for special functions in telecom, portable devices, wireless, the Internet, and consumer gear.

There is also a fast-rising need for embedded processor-based systems throughout industry as well as for appliances and all types of electronic controls. Many of these devices do not require the most advanced processes, so there may be a continuing market for these in less cutting-edge fabs far into the future.

This suggests the world of chipmaking may evolve toward a few giant fabs for commodity chipmakers and major foundries, using the most advanced tools, and a lower tier of less capable fabs for churning out devices that don't need the highest performance.

The problem with this scenario is that more and more functionality is being put onto single chips, while the diversification of applications means that many more high-level specialized devices will be needed. The old model of combining scores of standard commodity chips on boards to build systems is fading. Putting more functions on a few chips cuts space and power requirements, and assembly costs. Smaller, lighter and less power drain are requirements for the fast-growing mobile device sector.

Japanese companies have recognized this shift and are collaborating to redo processes to provide more flexibility and to lower costs for equipping fabs (see "Japanese companies design multifunction minifab tools," SST, June 2001, p. 83). Huge, rigid fabs oriented toward high throughput of standard commodity chips may not be the most desirable manufacturing plants of the future. Equipment designers may be looking instead for fabs that can combine many functions, including some specially designed ones, onto chips that also contain dozens of previously used circuit cores. Doing this easily and smoothly, with fast time to market and at affordable costs, will be the goals for many such projects. Some Japanese companies envision minifabs using lower-cost tools being added onto existing fabs, aimed at producing specialized chips for particular needs.

Another possibility as fab investment requirements escalate is the emergence of disruptive technologies — ways to produce working devices at lower cost. Intensive research is now going into areas such as nanotechnology, and molecular structures capable of carrying and switching signals that can be built by much less complex processing. The push toward optical communications is accelerating the evolution of optical functions, and some of these might be combined with electronics in future systems. So far these remain laboratory exercises, and nanodevices will have to be developed that can amplify as well as switch signals. But there is promise, and as the stakes get higher, the impetus toward disruptive technologies will increase.

One thing is certain: economics will rule. If there's a viable way to do the same thing, or something similar, at far less cost, that way will win out in the end.

Robert Haavind
Editor in Chief