Ten trends shaping the next 10 years

05/01/1997

Ten trends shaping the next 10 years

G. Dan Hutcheson, VLSIResearch Inc., San Jose, California

Solid State Technology has built a strong track record for being one of the most forward-looking publications in the world. Many of today`s key manufacturing technologies first appeared in this publication 10 or 20 years ago as ideas in their infancy. In preparation for their 40th anniversary, the editors of Solid State Technology have asked me look 10 years into the future, and write about the likely effect of ten now hotly debated trends, the more controversial issues, with views that would make readers think about the world differently.

1. Human resources

Shortages in human resources will not be solved as we move into the future. They have been a continual problem since the invention of the IC and will continue to be an endemic one due to the industry`s rapid growth. There is no equivalent to Moore`s Law to drive the labor supply, which, growing at a mere 1% to the chip industry`s 15-20%, will not even make the starting grids at any of hi-tech`s races. This problem will be aggravated in Silicon Valley, Austin, Oregon, Japan, and other technology centers as growth in industries offering semiconductors, software, Internet services, PCs, and other technologies fight over an ever-scarcer supply of technically capable labor.

Automation will not solve these problems because it has proved a poor substitute for people. In most cases, it has worsened technical labor shortages, because automation often replaces low-skilled labor with a need for more highly trained, educated, and skilled individuals. The high fixed cost of an automated factory demands that it be operated at the highest possible levels of efficiency. People make the difference. These factors will induce both companies and nations to put more emphasis on training and education. Companies will do this because they need to ensure a steady flow of human resources. Countries will do this in order to increase the incomes of their constituent populations.

The hi-tech business cycle will continue to exacerbate labor issues. Employment opportunities cycle between an employer`s market and an employee`s market for human resources. The employees with the most stable careers will be those who treat their skill sets as the strategic components of a personal business. They too will view training and education as the only way to be competitive.

2. Equipment productivity

Equipment productivity will continue to be an important cost issue, but progress will be limited by product diversification trends in the chip industry. Chipmakers and equipment suppliers have improved equipment reliability and uptime since the late 1980s. Yet, equipment productivity not only remains dismally low, it has worsened. Today, fab planners calculating equipment needs divide the theoretical throughput of a system by four. In the `80s, they divided by two. So even though uptime has improved, the overall effectiveness of equipment has been cut in half.

There have been two culprits driving this productivity loss. The first is the Balkinization of the chip market. The second is the natural consequence of Moore`s Law pushing the industry so close to the physical limitations of even the latest equipment.

The rapid improvements in design tool capability have given semiconductor manufacturers the ability to rapidly and easily differentiate their products. As new designs are easier to make, new products are designed faster than ever, reducing typical time-to-market for most products. The cost to design a product has dropped on a per transistor basis, making it possible to address smaller markets profitably. The result of chip companies pursuing these strategies has been a Balkanization of the chip market, which in turn, means that fab managers have had to deal with increasing numbers of product designs on their lines. More part numbers and process variations mean more set-ups, robbing equipment of its productivity.

Some companies have tried a strategy of simplifying their sets of products in order to improve equipment productivity. While this lowers costs, it also pushes a company into commodity products - where silicon is sold by the pound - thus lowering sales revenue. Efforts to reach local optimums by increasing unit productivity, while lowering revenue productivity, were doomed from the start. Equipment needs lower set-up times to be flexible enough to meet today`s production demands.

3. Geographic barriers

Geography presents barriers in the forms of labor immobility, commerce restrictions, and intellectual property (IP) diffusion. Each of these barriers are being lowered by rapid developments in communications technology and by the deregulation of the airlines. It is far easier for workers to move around the globe than in any time in previous history. Products can be transported anywhere in two days. While legal barriers to IP diffusion continue to exist, geographic barriers are almost nonexistent. IP can be e-mailed over the Internet at stunning speeds. Key people can hop on an airplane and educate competitors around the globe. The global repercussions of these trends can be seen in the more than two decades it took for the Japanese to gain a dominant position in the memory market; the Koreans shortened this to one decade; while the Taiwanese may do the same in half a decade (Fig. 1).

Figure 1. Regional distribution of semiconductor sales, by headquarters; a) 1987 total: $43.6 billion, b) 1992 total: $69.6 billion, and c) 1997 projected total: $156.1 billion.

Even still, some geographic areas are able to keep their technical centers due to infrastructure, historical precedent, and labor immobility. One example of this can be seen in Silicon Valley, which has remained a dominant force in technology despite many predictions of its demise. It remains strong because of its three international airports, deep water ports, excellent universities, mild weather, technical labor supply, capital markets, and unique ability to reinvent itself. Tokyo and Silicon Glen are also solidly ensconced as technical centers of excellence. Europe has developed a significant labor advantage with its excellent education system. One reason it was able to attract so much foreign business in the last upturn was its copious supply of engineers. Communities seeking to tap into high-tech`s wealth can learn many lessons from these areas.

4. The globalization of business

Business is driven to go global as geographic barriers fall, because that is where the money is. Meanwhile, the fall of the communist block has opened enormous market potential that is mostly untapped, while the end of the Cold War has lessened military tensions. In this environment, business is continuing to grow in importance, as it is the fountainhead of all jobs and wealth. National borders are becoming less important as business becomes more global, and can apply pressure to governments on all sides to lower trade barriers. The effects of this can be seen in the unification of the European Common Market and NAFTA.

5. Consolidation

Consolidation has always played a role in this industry, as seen in the recent wave of mergers. The common wisdom is that the small equipment companies will not be able to survive in the future. The argument behind this reasoning is strong: large companies have both the resources and global reach to satisfy today`s large chip makers. But while these arguments have strong logical appeal, this is an industry where the underdog often wins.

In the early `80s, there was a similar acquisition phase (see table). Concentration levels were at their highest, with the top 10 equipment companies holding just over half (53%) of the world`s equipment market. At the time, Eaton and General Signal were rapidly acquiring equipment companies, each attempting to become a general store. Perkin-Elmer and Fairchild were the world`s leading equipment suppliers who were also trying to be general stores in their respective areas. Meanwhile, there was this small maker of epi reactors whose name was often mentioned as an example of the type of company that could not make it in this environment. Its name: Applied Materials.

In the end, the general store strategy failed to take hold because a piece of fab equipment is like a link in a chain: no fab manager will buy from one vendor if the equipment set has weak links. In the early `80s, the strongest links were being offered by smaller, highly focused vendors. By 1986, the share held by the top 10 had fallen to 32%. Nearly 40% of the acquisitions made in this period actually hurt the combined revenue potential of the companies involved. Just over 50% of the acquisitions had no discernible effect on the combined companies` revenue potential, and a mere 8% showed a positive effect.

Only two companies who entered the `80s in the top 10 remain there today. The industry is highly concentrated once again. Last year, the share held by the top 10 was 46%, and talk of the general store has begun to re-emerge. But at the beginning of the last decade, fully six of the current top 10 equipment companies were minor players (Tokyo Electron, Nikon, Canon, Advantest, Hitachi, and Dainippon Screen) and two did not exist (Lam Research and ASM Lithography). These companies captured their positions through good execution and good products. The large companies that fell, failed because they could not perform these simple objectives. The success or failure of today`s giants will lie in their ability to satisfy customers.

6. Integrated process cells

Over the last 20 years, there has been a marked trend for equipment suppliers to link several process steps together. In the early days of the industry, equipment suppliers were little more than sophisticated machine shops offering custom-made equipment. The first true equipment suppliers began to market their systems to multiple customers, but their focus was inevitably on the hardware. By the early `80s, several equipment companies had begun to break away from the pack by offering equipment with fully developed processes. Soon these companies were developing processes for specific steps. In the early `90s, savvy equipment companies began to integrate process steps. Now, equipment companies talk about taking ownership of all the steps between certain masking layers. They call these integrated process cells, and see them as a way to capture market share. They hope to violate the "weak link in the chain" theory of competition by optimizing the process for the overall cell (i.e. by building bigger links). So far, the equipment industry has mainly taken on the responsibility for technical tasks, but as chipmakers shift their focus from technology to productivity, it is likely that equipment makers will do the same. It is quite conceivable that over the next 10 years we will see equipment companies that lease processes and even output - charging on a per wafer basis - and even equipment companies who offer foundry services. Mirroring this are the chip companies taking on more responsibility for system design and production for their customers. Both are simply moving down the food chain together.

7. 3D transistor structures

3D transistor structures will continue to attract attention at technical conferences, but will make little progress in the market. The limitations to these structures will be the lack of a low-cost method for growing single-crystal silicon films over amorphous films, and the fact that scaling interconnect will continue to be the greater part of the problem - not packing transistors.

8. Low permittivity dielectrics and low resistivity conductors

Low permittivity dielectrics and low resistivity conductors will come to pervade chip structures over the next 10 years. Speed has become one of the most important factors driving the continual scaling of chip CDs. Faster speed translates into more revenue for the chipmaker, which in turn, makes it possible to afford the ever rising cost of semiconductor fablines. The problem is that as devices continue to scale, speed becomes less dependent on transistor switching speeds and more dependent on interconnect conductivity. Below quarter micron, interconnect conductivity will become the most critical factor limiting device speed. The only solution for staying on Moore`s curve will be to lower the dielectric constant of insulating films to lessen capacitive coupling, and to decrease the resistivity of conducting metals. SiO2 is being replaced with SiOF, which will itself be replaced with more exotic materials as chipmakers try to push the dielectric constant below 3.0. Interconnect metals will change from AlCu to Cu, once the process bugs can be fixed.

9. Optical lithography

In microlithography, the safe bet has always been optical. For at least 10 years, x-ray has been waiting patiently, only to see optical steppers snatched up. Although several new technologies hope to grab attention from optical, it will remain the safe bet for the foreseeable future: it has the most funding, resources, and customer interest in its extension. It is already clear that 248 nm will be pushed to 0.18 micron and maybe even 0.15; 193 nm will first fall into place at 0.15 and then will be driven down to 0.13 micron and possibly even 0.1; while 157 nm or 126 nm likely will be the technology of choice for 0.1 and even 0.08 micron. Beyond that, lithography choice issues may lose their importance, since interconnect density becomes the limiting speed factor, and nonoptical methods would be too costly to implement if speed were the limiting factor.

10. Understanding Moore`s Law

The key to understanding Moore`s Law is in understanding the impact that equipment, and in particular lithography, has on the industry. Moore said that his law would not have been possible without the gains made by the equipment industry (Fig. 2). So as equipment has slowed in its ability to deliver yield gains and is now showing signs of not being able to deliver reductions in linewidth and speed, fears have arisen that Moore`s Law might actually slow and lose its ability to deliver market growth. But there are still many opportunities to exploit: improving overall equipment effectiveness and increasing integration scale using multichip modules (as done in Intel`s Pentium Pro) are two new ones that have emerged in recent years. I am reasonably certain that more will emerge.

Figure 2. Transistor counts showing Moore`s Law.

Since Moore`s Law was discovered in the mid-`60s, its end has been believed to be between 10 and 20 years out. We have not only met the challenge of the Law, but have consistently pushed its predicted end out into the future, so that it remains 10-20 years out. Even if we cannot do this in the future, I cannot imagine that the world would stop using chips because they cost more per bit. We all still buy cars even though they hit the bottom of their learning curves over 60 years ago. We will continue to buy chips after they hit bottom because, like cars, they are an essential part of everyday life.

G. Dan Hutcheson is president of VLSI Research Inc., 1754 Technology Dr., Suite 117, San Jose, CA 95110-1308; ph 408/453-8844, fax 408/437-0608, e-mail [email protected].