cited different trends for two different periods. For the period 1984–1992 the rising curve in computer ownership was beginning to flatten. Then for the most recent three years, from 1997 to 2000, home computer buying rose again as more people used the computers to access the Internet. He also said that people who use computers at work are more likely to use them at home.
The Role of Semiconductors in Driving Growth
David Morgenthaler commented on the degree to which semiconductors have driven recent economic growth, citing a regional study he helped to fund. The question is what were the economic drivers of 100 to 130 years ago when north-eastern Ohio was the Silicon Valley of the country? Early indications are that a combination of petroleum and the internal combustion engine were the drivers that made the region a center of innovation for the country. He suggested that the semiconductor is probably the counterpart today. Dr. Gomory agreed, with the caveat that other advances were probably playing a role as well, such as the ability to price a color printer at less than $100. He then introduced the next speaker, David Mowery.
SEMICONDUCTORS: ECONOMICS OF THE NEW ECONOMY
University of California, Berkeley
Dr. Mowery addressed the economic issues raised by Dr. Spencer’s remarks by discussing how and where the gains in productivity that have resulted from the application of computing and information technology in the semiconductor industry have been realized. He noted that much of the progress in the performance of computers and related information technology products reflects progress in semiconductor components. Conversely, significant advances in manufacturing and design performance within the semiconductor industry are associated with the application of computing and information technology.
The “Copy Exactly” Process Technology
In recent research with colleagues at the University of California, Berkeley, and Georgetown University, Dr. Mowery and his group13 found that innovation in semiconductor manufacturing processes, particularly the development and introduction of new process technology in this industry, retains a strong element of
trial and error. It relies to a remarkably small extent on simulations or reliable, robust, predictive models.
A good example of this approach to process innovation is the so-called “copy exactly” philosophy that the Intel Corporation relies on in transferring process technologies among manufacturing plants. This approach to managing the development and transfer of process technologies concedes that in the absence of a complete understanding of the process technology it is necessary to copy the entire system—the length of pipes, the placement of equipment—exactly, rather than in a way that is based on engineering or scientific knowledge. This is a remarkable example of pre-scientific problem solving and trial-and-error experimentation in this industry. At the same time, however, semiconductor manufacturing process technology and manufacturing performance have benefited enormously from the application of computing technology to manufacturing process control, data capture, and analysis.
In some respects the process technology side of the semiconductor industry resembles that of the chemicals industry in the 1940s and 1950s. During this period in the chemicals industry, process technologies frequently were developed and initially applied in small pilot plants, and subsequently scaled up and applied in commercial-scale plants. This approach to process development was time consuming and tied up expensive facilities for long periods while people worked on incremental improvements in process technologies. The application of computers to the design and simulation of chemical reactions and the use of minicomputers to monitor plant operations have changed the development and management of manufacturing processes in the chemicals industry. The lack of a complete scientific understanding of semiconductor manufacturing processes is likely to constrain the scope for increased application of computer technologies to process simulation and manufacturing management. With advances in understanding, however, such expanded applications could have a transformative effect on industry structure and performance.
New Specialty Firms in Design
Computer-aided tools have been widely applied to the design of semiconductor devices, yet design remains a constraining factor in technical progress. Innovation in this industry has been a horse race between design and manufacturing technologies, and design has been the key constraining factor. This is revealed to some extent in the successive formulations of Moore’s Law. Initially the doubling in transistor density on semiconductor devices took only one year; subsequently the time required for such an advance was extended to a year and a half. These revisions have been driven by constraints in design rather than in manufacturing.
This design bottleneck has had interesting implications for the evolution of the industry’s structure. Integrated firms that combine design and manufacturing
now face increased competition from specialists in design and manufacturing. The semiconductor industry is undergoing structural change and some vertical disintegration with the entry of a large number of specialized design and marketing firms—called fabless semiconductor firms. Specialized manufacturers, the so-called “foundries,” now play a more important role in the semiconductor industry. In addition, semiconductor equipment firms are now more important as developers of new process technologies than simply as suppliers of new equipment. All of this progressive vertical specialization is facilitated by advances in Web-based and other information and communication technologies.
New Models That Rely on Vertical Specialization
What are the implications of specialized entry and the focus of these specialized performers for the long-term productivity of the industry? Research by the University of California, Berkeley, team suggests that the specialized manufacturers of chips adopt a more incremental approach to management of the introduction of new process technologies. Process technologies are being upgraded more frequently and continuously within these foundries, which develop capability in the frequent introduction of process technologies. The pace of these applications may influence the extent to which they create bottlenecks and these bottlenecks may influence the evolution of the industry structure.
The future of this industry will include numerous experiments in new business models that rely on vertical specialization. Design firms increasingly are specializing in individual pieces of real estate on wafers and chips. There is a growing trade among these design firms in design components, or IP blocks, facilitated by the Internet. An interesting question is whether such trends in the semiconductor industry will be replicated elsewhere in the New Economy.
Does New Knowledge Affect Process Development?
One characteristic of the New Economy is its reliance on the application of new knowledge to economic life. How are information and communications technology affecting the processes and perhaps the productivity of knowledge production itself? For example, how has the infusion of information and communications technology affected the trial-and-error processes that historically have governed process development and introduction in the semiconductor industry? Will this industry’s future development follow a path that resembles that of chemical manufacturing during the 1970s and 1980s? Other scholars have argued, with limited evidence, that general-purpose technologies, such as information technology, should raise the productivity of R&D or of knowledge production. This or future workshops might cast some light on this issue through case studies and discussions.
The Contribution of Semiconductors to Other Sectors
Finally, some evidence on the effects of technical progress in semiconductors on the larger economy may be available from an examination of trends in markets for semiconductors. Slightly less than half the end-user markets for semiconductors currently are in the computing industry. The other leading users are automobiles, consumer electronics, and communications. Because measures of quality improvement in the output of these sectors are almost entirely lacking, it is difficult to assess the contributions of semiconductors to these and other sectors of the U.S. economy. Without better measures of quality improvements in the output of various sectors it will be difficult to trace the contributions of the semiconductor industry or other industries that produce critical inputs to the New Economy. This is not a trivial task, given the diverse array of outputs in an industry like consumer electronics.
The Importance of Better Data for Other Technologies
Dr. Jorgenson disagreed that there is a gap in our knowledge about the change in the price of semiconductors. He said that thanks to the work of people at the Department of Commerce, the Federal Reserve Board, and other places we know a good deal about semiconductor prices. In addition, there is a long tradition of measuring computer prices. For telecommunications he noted that, as Dr. Flamm pointed out more than 10 years ago, we have made little progress. However, we do know that semiconductors directly affect those technologies that behave in ways similar to computer prices but whose rates of improvement are not captured in routine statistics. These technologies carry increasing importance and measurement efforts should receive top priority.
A second priority should be to learn more about software. Dr. Jorgenson noted competing views in this respect. Some people think that software is basically something that is a hand industry that changes little; others, including Drs. Spencer and Raduchel, argue that it has changed and that it is affected by the progress of semiconductors because of the impact of computers in the production of software. We, however, do not know much about this effect and it is important to learn more. Dr. Jorgenson placed lower priority on understanding processes that are less closely related to the progress of semiconductors.
David Mowery replied that he had not addressed the organization of the innovation process, but he did stress innovation in the form of entry by different kinds of firms—non-integrated firms in manufacturing, the fabless firms, and the extension upstream of equipment firms into process development.
The Importance of Flexible Organizations
Shane Greenstein offered an historical note, observing that several decades ago Silicon Valley was known for more than just innovations in an engineering sense; it was also known for its inventiveness in the creative organizational design and for its generation of knowledge in these new organizational forms. These had not been seen before and were quite dynamic. People moved across organizational boundaries and took knowledge with them. That was one of the factors that made this industry so dynamic. He said that this might be an important source of bottlenecks and an important place to look for the creation of value and new organizational forms.
Dr. Mowery agreed with Dr. Greenstein that the new approaches to organizing and managing R&D in many of these pioneering firms are important. He suggested, however, that such structures might be even more difficult to understand and quantify than the other factors he had discussed.
Measuring the Activity of “User” Industries
Dr. Mowery also addressed Dr. Jorgenson’s point about semiconductor knowledge gaps, saying that unless we can measure the output of the “user” industries more effectively it is hard to see the contributions of the semiconductor industry. He suggested starting with the communications and consumer electronics industries, given their complexity. He added that the rest of the economy— outside the computer industry—has become “a bit of a dark planet” in terms of understanding quality improvements in their products.
The Complex Interactions of Technology and Society
Dr. Gomory said he would exercise his privilege as moderator to say he agreed with those remarks. He said that it was overly simplistic to try to describe the progress of the semiconductor industry solely in terms of whether finer lines could be etched onto semiconductors. He said that it would be equally wrong to focus on the flying height of disks. This is not a one-parameter phenomenon, he said; much more is going on.
He praised the remarks of Dr. Mowery, saying that even scientific fads had to be reproduced exactly to understand what is happening and how long the pipeline is. We understand some things, he said, that are in narrow areas, but we are also groping and are having to do things empirically.
Alluding to Adam Smith, Dr. Gomory raised the example of the pin factory, where gains arise from specialization. When an industry grows, specialization becomes possible. In the early days of software there was just one undifferentiated bundle that retrieved files. Today the part that used to retrieve files has become the whole database software industry. He compared an industry to the