Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
4 A Unique Innovation Engine HOW AND WHY THE ENGINE HAS WO:E21lED WELL The technologies and infrastructure discussed in the preceding sections are essential tools for meeting the national challenge in computer science and technology. They are based on the unparalleled record of achievement and innovation in U.S. computer science and technology during the past three decades. If we are to face the challenge before us, we must continue that record. To do so we must maintain the innovation engine that we have built and operated so successfully. In simple terms, what we call the U.S. innovation engine consists of three components: universities, venture companies, and mature companies. The federal government primarily through research and development funding from DARPA and NSF and to a lesser extent from the Office of Naval Research (ONR), the Air Force Office of Scientific Research (AFOSR), the Department of Energy (DOE), the National Institutes of Health (NIH), and the National Aeronautics and Space Administration (NASA)plays an essential role in Ju- bricating and tuning the engine and In deterrn~ning its long-term future. Government has a direct interest in computer science and technology as a customer and as a user. Moreover, by influencing the health of the overall economy, government affects the environment for privately funded research and development as well as private-sector 22
23 market development. Finally, government has also advanced com- puter technology by underwriting large projects that link companies and universities to produce novel systems and prototypes. The universities perform most of the basic research that fuels the engine, and they supply the talent required to pursue research and development activities across the economy. Because they are organized specifically to seek new knowledge and disseminate it, uni- versities are where novel and innovative ideas have been and are most likely to be hatched. Over the last three decades, the Defense Advanced Research Projects Agency has played a leading role in funding basic research in computer science at universities. DARPA's strategy has stressed the funding of a few high-risk visionary projects and the building of a critical mass of proven research talent at rela- tively few locations. And through most of its history, DARPA has invested in technologies that have proved to have both military and civilian applications, thereby using research dollars to stimulate com- mercial development. DARPA's approach has produced landmark innovations that include time-sharing, artificial intelligence and ex- pert systems, computer graphics, VEST design tools, packet-switched networks, and, more recently, new architectures for multiprocessor and distributed systems. NSF has also been a major contributor to university-based basic research and is credited with funding sev- eral important advances in theoretical computer science as well as supporting experimental computer science and upgraded educational facilities. Smaller yet significant contributions to basic research have been made under funds provided by the military services, NASA, NTH, and the Department of Energy. These innovations represent some of the most important thrusts in computer science and tech- nology over the past 30 years, and they account for the successful operation of the university component of the U.S. innovation engine. Successful world leadership in computer science and technology would not have been possible without venture and mature compa- nies. Landmark industrial innovations include modern semiconduc- tor technology, the microprocessor, the personal computer, rotating and solid state memories, supercomputer architectures, as well as several materials, packaging, and manufacturing breakthroughs. For most companies, the pressure to sustain and increase profits makes it difficult to justify long-term research. Moreover, companies are loathe to invest in acquiring new knowledge that might accrue to the benefit of outsiders, particularly competitors. This is especially the case with the fast-paced computer technology, where the head
24 start advantage of the innovator may be short-lived. Venture com- panies, which are often spin-offs from university computer science departments or from large companies, translate the results of re- search into leading-ecige products and get technological innovations into the marketplace relatively quickly. Venture companies do vir- tuaBy no research, lacking the time or the resources to do anything other than get their new products to market. Mature companies, which in many cases began as smaller venture companies two or three decades ago, also contribute new technologies. Only a few of the largest mature companies conduct basic research, often under the pressure of product development needs. Their primary emphasis lies in anticipating and meeting the worId's demand for large num- bers of innovative, reliable, and affordable computer products of high quality. The U.S. innovation engine works because its components com- plement one another. People and ideas flow among the three compo- nents of the engine, helping to integrate the intellectual curiosity of academia, the vigor and flexibility of the entrepreneur, and the re- sources and dependability of the giant corporations. This union is not perfect: companies and, in particular, computer science and tech- nology graduate programs have complained of shortages in skilled computer science and technology personnel; college programs have suffered from obsolete equipment; university researchers are often slow to explore the real-life problems facing companies; the capital markets are too impatient, with a quick profit orientation that dim courages risk taking and disparages long-term research; shakeouts among the venture companies often sweep away good ideas before they have any chance to pay off and too frequently reward the iniita- tor rather than the innovator; large companies are often bureaucratic and reluctant to adopt new ideas. But, at least until now, the en- gine's strengths have clearly outweighed these weaknesses, and the ability of the United States to generate and bring to market a steady stream of innovations has been unparalleled. T== "~. · "~1' TN~:IT~ · OmT~TT^~TT== ~ V ~:JO~:J^~J" An ~ ^~O ~ ~ U ~ ~ U "= An important element responsible for the successes of the inno- vation engine in both academia and industry has been the presence of an experimental infrastructure consisting of advanced research tools. Advanced research took are as essential to the work of many computer scientists as particle accelerators have been to the work of high-energy physicists. Their importance is directly related to the
25 largely experunental character of the discipline. Advanced technolog- ical systems will continue to be required by experimental computer scientists who are attempting to address research problems at the cutting edge of the field. Three important elements of this research infrastructure are discussed below. Advanced Computer Resources for Research Providing researchers with machines of the greatest possible speed and memory capacity and with the most advanced software systems has proven to be a sound investment in the future. In the past, dedicated large computers and forefront workstations enabled researchers to write larger programs, express them better, run them more rapidly, and advance the state of the art faster than they could have if they had had to rely on more limited personal computers or on the keyboards of dumb terminals attached to time-shared mainframe computers. Network connections have enabled researchers to share their results and gain access to important sources of information. In the future, local computational resources for individual re- searchers will grow in number and improve in individual perfor- mance. We expect that massive multiprocessors and many of the other promising technological innovations already mentioned and further described in Part IT will find their way into tomorrow's ad- vanced workstations. The trend must continue if the U.S. arsenal of research tools is to be the best worldwide, enabling U.S. researchers to be among the first in making and consequently In exploiting new · - c .lscoverles. Prot otyping Throw at ion and S mmlation A second important element of the research environment has consisted of powerful emulators and simulators tools used to proto- type ambitious software and hardware systems before they are built. Typically, these tools take a relatively long time (perhaps hours) to iniit ate how the system being analyzed would behave in a very short tune interval (perhaps a fraction of a second). Nevertheless, they save time and money compared to the alternative of building and testing systems based on new and uncertain ideas. The increasingly complex architectures of contemplated systems, such as multiproces- sors, speech and vision systems, and supercomputers, make this kind of pre-production modeling mandatory.
26 Unproved V[S] Design and ]?abr~cation The third important component of the computer science research infrastructure has been the design and prototyping of new sol-id state circuits (VESI chips) that are the building blocks of aD computer systems. Improvements In VEST architectures, i.e., in the design of these circuits ~ terms of more elementary components, and in the processes that translate these designs into silicon unplementations are unportant because they lead to more powerful VEST functions that, in turn, make ever more sophisticated computer applications possible. For example, a key innovation at the chip level giving rise to crucial and novel components of a special-purpose multiproces- sor might, in turn, lead to higher-level systems capable of speech understanding, vision, and learning. VEST improvements also have strong commercial implications. For example, the relatively low cost of the many home appliances that utilize microelectronics is a direct consequence of more powerful and less expensive VESI components. An important issue related to VEST design has been the ability of the research community to convert its ideas into silicon circuitry as rapidly as possible. To date, researchers have used private com- panies along with the government-sponsore<1 Metal Oxide Se~con- ductor Implementation System (MOSIS) foundry to that end vnth turnaround tunes from design to silicon prototype ranging from a few weeks to several months. Increasing dependence of computer systems on their components, the fast pace of change in computer technology, and growing foreign competition in semiconductor technology make it important to strengthen these foundry processes, speed up their turnaround times, make them more widely available, and maintain them at the cutting edge of research frontiers as a crucial component of the research infrastructure. WHY TEE ENGINE MAY NOT RUN SMOOTHLY IN THE FUTURE The U.S. system has proved uniquely successful on a global scale. Other nations, including our strongest competitors, have different institutional structures, which, by and large, have not been as con- ducive to computer science innovation. That is, national differences have inhibited the duplication of our innovation engine. Japan, for example, has healthy mature companies that have generated impres- sive advances in the unplementation of computer technologies, but its universities are weaker innovators and the country has virtually
27 no venture sector. Western European companies, mature or venture, lack U.S. strength or consistency in bringing innovations to market, while the venerable Western European universities do not contribute as extensively to innovation in computer science as do their U.S. counterparts. This has not stopped the Europeans, however, from becoming very successful at innovating programming languages (such as Ada and Pascal). In the absence of comparably effective institutions, other coun- tries have tended to depend on innovations generated by U.S. re- search. This dependence, as well as widespread appreciation for the global scale of competition in computer-related markets, has led in- dustrialized and newly industrializing nations to embark on programs involving government, industry, and/or academia to strengthen lo- cal computer science and technology capabilities. With consistent funding and commitment among the parties, these programs may eventually give rise to robust innovation engines tailored to different local societies. This development may take time, but many nations look more favorably than the United States does on investments with long payback periods. In the meantime, competitors from other nations who may follow U.S. firms in introducing new products ben- efit from lower expenditures on research, development, and market building. In this copy-cat environment, productivity and the many factors that give rise to competitiveness (such as product design, marketing, pricing, and quality) determine which competitors and which countries- will be market leaders over the long run. Foreign nations are becoming increasingly competent in com- puter science and technology, a development that may increase their productivity and competitiveness compared to those of the United States. In the aggregate, the U.S. trade surplus in computer and business equipment peaked in 1981 at just under $7 billion; by 1986, it had fallen to about $2.2 billion (CBEMA 1987; see Figure 2~. If the pattern in computer science ~d technology were to follow that of textiles, steel, automobiles, and machine tools, we would lose a key source of competitive advantage in the world economy and a key source of national security. Already the Far East has achieved superior market share in commodity semiconductors (DRAMs) and personal computers (clones of U.S. machines) manufactured in the Pacific Basin (see Figure 3~. While these products may not be state-of-the-art, Japan has moved aggressively into supercomputer production (it already produces some of the fastest machines avail- able), and the West Europeans have made substantial advances in
28 c u R 50- . R o 7~, _~s2.2 BILLION B 1 .50 L o -1 00 N o -1 50 ~ . L L A -200 - R 1960 1965 1970 1975 1980 1986 S t.. GOODS AND SERVICES ~ CPU&BE INDUSTRY ~ MERCHANDISE VS105.7 BILLION SOURCE: U.S. Depa.l'..ent of Commerce -S1 53.3 BILLION FIGURE 2 U.S. trade balance versus computer and business equipment in- dustry trade balance, 1960-1986. Reprinted, by permission, from CBEMA, 1987. Copyright (A by the Computer and Business Equipment Manufacturers Association. c u T 1 2000 1 0000 M 8000 L L 1 o N 6000 4000 D O 2000 L L A 0 sR ~. ~ Am NORTH AMERICA FAR EAST OTHERS · 1970 ~ 1980 O 1986 FIGURE 3 Computer and business equipment industry imports by geographic area, 1970-1986. Reprinted, by permission, from CBEMA, 1987.
29 software and systems integration. Further, Japanese strengths in consumer electronics are expected to facilitate computer technology development in such areas as optics, which is a focus of emerging storage technologies.* The necessary development of the field and maintenance of the U.S. innovation engine are impeded by underrecognition of the need for basic research in computer science and technology. The situation is symptomatic of the more general problem in U.S. industry of not adopting a sufficiently long-term perspective, as has been revealed in the more established industrial sectors of chemicals and aircraft (positively) and steel, automobiles, and consumer electronics (nega- tively) (U.S. Congress, Office of Technology Assessment 1983; OECD 1985~. While far-sighted companies may invest in the tools, facili- ties, and training required or their product strategies, U.S. industry generally does not tend to invest significantly in either basic research at universities or basic education. Such investments have long been the province of government, to which we direct our concerns. Because basic computer science research depends so heavily on funding from two organizationsDARPA and NSFit is particu- larly sensitive to changes in policy and funding behaviors at those organizations. For example, DARPA has recently begun to retreat from its traditional support of long-term basic research. Funcling of basic computer science research has declined since 1983, while fund- ing for applied research has more than doubled. DARPA has also begun to seek research results with more immediate military rele- Vance and has instituted more bureaucratic procedures for funding and managing new projects. In pant because the experimental nature of computer science research makes tight management difficult, this trend may undermine progress In the computer field. Recently, NSF has attracted attention through its programs to launch supercomputer research centers and provide computer net- working for researchers. These programs operate out of the same unit responsible for most of NSF's computer research funding; they use computer technology and employ computer scientists to assist researchers in the physical sciences, but they do not support basic research in computer science and technology. In contemplating the *Competitiveness in computer science and technology depends on many factors and is sensitive to many influences. In recognition of the importance and complexity of the topic, the board is planning to focus projects on key issues pertaining to the competitiveness of the U.S. computer sector.
30 future of this area, it is important for policymakers to recognize that increasing the computer sophistication of the physical (and other) sciences, although extremely important, ~ not the same as conduct- ing computer science research. Consequently, funding for scientific computing should be evaluated separately from funding for computer science research. The NSF contribution to basic computer science research ~ less than half that of DARPA approximately $60 million versus approx- imately $155 million. A major component of the NSF effort the computer and computational research programwas funded at $19 million in 1987, grew only to $20 million in 1988, and ~ likely to grow by led than $2 million In 1989 in a highly constrained budget environment. Coming from the two principal federal supporters of this field, this order of magnitude for federal basic computer science research support, approxanately $215 mullion, is alarmingly low. To- tal federal investment in basic computer science research, including high-performance computing research, ~ estimated by OSTP at only $300 million (OSTP 1987~. Both figures may also be overestimates, since some applied research tends to be labeled basic research. The board may study computer science and technology research patterns to derive more insight into current trends. But available ev- idence arouses its concern since, as noted below, pursuing even one of the grand challenges in the field may require more than $100 million dollars. In a budget environment that ~ driven by the immediate mnssions of government agencies and characterized by fragmentation of support among multiple agencies with multiple missions, computer science and technology appears to be at risk of losing support for ba- sic research at a time when increased funding is needed more than ever before. As a relatively young field, with only the beginnings of the theoretical and empirical bases needed to achieve the substan- tial advances described elsewhere, computer science and technology is particularly vulnerable to the increasing politicization of federal research support.