An Analysis of Technology Trends within the Electronics Industry

Transcription

1 An Analysis of Technology Trends within the Electronics Industry Summary P J Palmer and D J Williams Prime Faraday Partnership Dept Manufacturing Engineering Loughborough University Leicestershire LE 3TU UK tel: fax: p.j.palmer@lboro.ac.uk; d.j.williams@lboro.ac.uk This paper begins with a brief review of previous work in the field of technology forecasting and discusses the Fisher Pry model in particular, before concluding with an attempt at forecasting the rate of development of smart technology. The Fisher Pry model is shown to fit well with an assumed ultimate microprocessor clock speed of.2 GHz. By combining data from multiple sources support is found for the hypothesis that rates of development can be correlated with development activity, as measured by the number of US patents filed in specific technology areas. A forecast for the development of smart technology is made to demonstrate how the models may be used to generate forecasts of future developments, where limited data is available. Review of Previous Work Porter () and Twiss (2), both review the full complement of forecasting techniques available for the management of technological change. Hills (3) review of technology forecasting methods focuses particularly on techniques that may be of particular interest to product designers. The value of using forecasting techniques for assisting in product design is particularly stressed by Twiss. Dubash and Chakravati (4) review and summarise the results of many recent technology trend forecasts within the electronic product domain, with a view of what product capability will be over the next few years. As a review, the results may be broadly classified as a Delphi forecast, and present a vision of product developments and technological improvements that may be seen in the late 990 s. Although the Delphi forecast is useful, it lacks an obvious quantitative relationship to measured trends. Of particular interest here, are those models that attempt to fit predictive equations to log linear plots of trends, as technological trends of this type are common. Moore s Law, illustrated in Figure, associated with the number of transistors in a single integrated circuit (here plotted for microprocessors) is particularly well known. The data shown has been collected from many public domain sources. Figure is a simple log/linear plot of the data, and the reader is implicitly asked to ignore the trend line at

2 Figure Moore s Law Total Transistors General Trend Leading Edge Trend its extremities. Intuitively, a complete model must include a static zero value before the trend starts, and a constant value at the end of the trend when all developments have been exhausted. There are several classic trend models that have just such characteristics and we review the best known of these, the Fisher Pry model (5). Before considering this model, it is helpful to review a well-established trend. Moore s Law: A Classic Trend Moore s Law was proposed by G Moore in 965 (6) after he noticed that the data storage capacity of memory devices doubled every 2 months. Moore s observation has now become enshrined as Moore s law and is usually cited as the number of transistors in an IC will double every year to eighteen months. Using data from microprocessor announcements since 97 (7), Figure shows that Moore s Law follows a very convincing exponential trend line albeit with a doubling time of more than 29 months - rather longer than usually stated. Post 990 the plot is far more scattered. This is due in part to the wide range of microprocessor types included within the plot, many of which fall well below the current state of the art in terms of the number of transistors on a chip. It may be argued that it is the leading edge capability that defines Moore s Law and it is therefore more appropriate to plot the leading edge announcements. This yields a doubling time of approximately 22 months a result that corresponds with the 8 to 24 months version of Moore s law published by Intel (Dr. Moore s employer) on their web pages (6). Although there is clearly room for debate with regard to the actual slope of Moore s Law, the evidence supporting the existence of exponential trend line is compelling. A Newly Observed Trend For well publicised trends such as Moore s Law the suggestion that trends are self fulfilling prophecies is plausible, as each new development leapfrogs over past capability in a race for technological supremacy. This argument is much less convincing for trends that occur over longer time scales - Figure 2 illustrates the development of magnetic Figure 2 Magnetic Material Development Magnetic Material Development Energy product KJ m

3 material capability over more than a century (data from (8)). The trend line started in the 9 th Century and has not been publicised as a major technological trend. Despite the lack of publicity, the trend has continued apparently unaffected by economic cycles and world wars. Whilst not definitively falsifying the hypothesis that technology trends are self fulfilling prophecies, the longevity of the trend suggests that the underlying drivers are a natural consequence of human Figure 3 Trends in Data Communication Rates ETDM (Electronic Time Domain Multiplexing); WDM (Wavelength Division Multiplexing) Capacity Gb/s Trends in Data Communication Rates of Announcement copper cable commercial ETDM experimental WDM commercial WDM endeavour rather being than forced by directed research and development, and that trends are not easily diverted from their path. The Effect of Legacy Technology on the Development of Trends The development of data communication rates is illustrated in Figure 3. Copper technology has had the longest history of development, but in terms of capability it has been overtaken by two types of optical fibre technology. These illustrate "jumps" in technological capability that are probably impossible to forecast. However, Figure 3 also illustrates that each trend line continues to develop, despite being superseded in capability. Of particular interest is the observation that copper technology has continued to improve in capability, despite the appearance of two superior (and now well-established) fibre optic systems. Indeed, copper cable has actually been able to exceed the initial capability of ETDM (Electronic Time Domain Multiplexed) fibre optics. The time differential in capability represents about a ten year gap. Because there is already an extensive access infrastructure of copper in the telecommunications network into domestic premises, the net effect has been that the data carrying capacity of the existing copper cable has increased over time. As a consequence, early forecasts that an optical access infrastructure would be required to support new data services have been flawed. For example, domestic consumers are now offered ISDN lines by BT for high speed access to the Internet through their existing copper connection, without the need to upgrade the connection to the telephone exchange to optical fibre (9). Installation of fibre optic connections are largely confined to applications that require the higher data transfer rate that optical fibre can support, such as main trunk connections, or feeds to large commercial users. Because of the continued improvement in the data carrying capability of copper, it is still likely to remain the main connection method of connecting telecom services to the home in the UK for the near future.

4 Figure 4 Fisher Pry Trend Line for Maximum Announced Microprocessor Clock Speeds MHz The Fisher Pry Model of Trends In 97 Fisher and Pry (5) published a paper describing a simple model of technological change. Despite the simplicity of the model, it has been shown extremely effective in modelling the competitive substituting of one technology by another. With little modification, the same model may also be fitted to data sets defining the improvement in capability of a technological process. Equation is the preferred form as it allows the use of linear regression to fit the data (t 0 represents the mid-point of the trend where f = 0.5 and α is an empirical constant). Note that in the Fisher-Pry model f, the measure of the market held by the technology, is constrained to 0<f<. In practice this constraint is met by dividing a measured parameter Y, say the total number of units sold, by L, the maximum market size. In their original paper Fisher and Pry presented 7 cases of substitution of technology that fit the model, these include: synthetic rubber for natural rubber; plastic for natural leather; and plastic for metal in automotive applications. It is important to note that when f << Equation approximates to Equation 2 an exponential trend line. Therefore the limit to a trend L cannot be interpolated from a set of data points unless the trend has progressed beyond f > 0.5. ln ln f ( f ) Maximum Announced Microprocessor Clock Speed for Each (Fisher Pry Trend Line) = 2 [] f 2 ( t t ) α ( t t ) Equation 0 = α 0 Equation 2 Making a Forecast The Fisher Pry model is demonstrated by using a data set of microprocessor clock speeds, and a forecast extrapolated from the data. The derived trend line is illustrated in Figure 4 - note that the early part Fisher Pry model approximates to exponential growth, and therefore would appear linear on a logarithmic scale. The rate of growth slows as the limit of.2ghz is reached causing the trend line to curve. Note the sigmoid form of the curve and that 997 corresponds to approximately ½ the total development of this trend i.e. t 0 in Equation. Thus, the forecast trend line does not deviate significantly from a simple exponential regression until after 998. It will therefore be several years before this particular trend line can be verified. However, if it is correct then it suggests that microprocessor designers will have to turn increasingly to other design features to increase the performance of each new generation of microprocessor. The technology roadmaps published by Intel and other companies list a large number of innovations other than clock speed in each new microprocessor (7).

5 The relationship between effort and rate of change One method of measuring progress is to count the number of patents filed within a particular technological field. Although the method has drawbacks due to differing patent laws and practice throughout the world, and patents may not necessarily be filed for inventions with military applications, the method does have the advantage of simplicity. In the case of the American patent database, searches have been free via the Internet for some time. Free Web based searches of the British and European Patent databases are likely to become available during 999. We chose to plot technology trend data from published literature and from the Internet sources and obtained the data (0) to produce Figure 5 Rate of Technological change Patents Publisher per Annum 0 Rates of Progress for Technological Trends Slope of Trend Line Figure 5. This figure gives the slope of the exponential trend line for nine technological trends of interest and the number of patents in each technical area issued between 976 and 997 This data suggests that there is a useable relationship between technological progress and development activity as shown in Figure 5. However, selecting a suitable metric for use within the model is difficult, as is collecting suitable data. Given that these hurdles can be overcome, then perhaps a quantitative methodology of technology forecasting may be developed given today s technologies. As a cautionary note, Helmer () made very similar comments about the development of the science of technology forecasting. It is questionable how much progress has actually been made since those comments were made. One major change in favour of improved quantitative forecasts is the availability of personal computers for carrying out calculations and the availability of data via the Internet. Given the availability of data sets, the equations of Fisher Pry model are easily manipulated within standard spreadsheet application packages. As such the methodology is far more accessible today than 30 years ago, and is worth revisiting in a modern context. The data sets used in this paper have been collated by Loughborough University- see the Prime Faraday website for detailed data It seems reasonable to apply this data to forecasting the development of a technology that relies, at least in part, on the progress and application of electronic devices. The following forecast of the development of smart technology is offered as an example to illustrate the potential application of the technique. A Technological Forecast The development of smart technology is a current area of research activity and product development (2). It is of major commercial interest to identify when this technology will become commercially significant at a large scale. Smart materials can be defined as materials that can perform two or more of the primary functions of sensing, processing, actuation or structure in response to external stimuli. However, it is now clear that despite the early optimism, and the development of a few smart

6 components, such as self-dimming mirrors for automotive applications, progress is slow. In fact, only five patents are currently filed with the American patent database that include smart technology in their keywords in this context. (There may, of course, be other related patents without any smart technology identifiers). It is reasonable to assume that the trend on article publication will continue unchanged, and that the number of patents will also grow, but at what rate? Crowe (2) states that smart technology is currently entering a period of slow progress and the rate of publications and patents shown in Figure 5 would support this opinion. The results suggest that for a slow growth scenario, a slope of 0. would be appropriate. Given that since 994 patents have been filed at about one per year, then a suitable value for the exponential growth and an appropriate trend line can be extrapolated. Figure 6 illustrates attempts at providing a forecast for the number of smart patents filed each year using constants derived from the conditions outlined above. Of the models, the simple exponential extrapolation is the easiest to apply and justify since there is no easy way in which a limit to the growth in the number of patents can be defined. The trend extrapolation suggests that by the year 2020 American patents on smart technology will be granted at a rate of about 8 per year. If we take this level of patent activity to mean that there Figure 6 A Smart Technology Forecast is widespread commercial interest in its application, then it will be Smart Technology several decades before smart technology is commonplace. Number Patents Papers The relative simplicity of the exponential and Fisher-Pry models, suggest that developing a relationship between development activity and the slope would be a worthwhile refinement for predictive purposes. Conclusions There is an apparent correlation between development activity and rate of technological progress, however, there is insufficient data in this paper to quantify the relationship with any precision, a result that correlates to that obtained by other authors (3;4;5). When models of technological growth were first proposed, the number and complexity of the calculations involved were major impediments to their implementation. The problem of calculation may now be overcome with the general availability of computers, and the advent of the Internet has helped by making many public domain sources of information easily available for research purposes. The simpler exponential and Fisher-Pry models have been shown to offer useful results when used with a value for the slope derived from Figure 5. Further research is required to validate the relationship between patents and rates of progress, however, in concept it clearly has potential as a forecasting tool. In this work, we have used patents as a measure of effort. It could be asserted that patents record the results of work that is deemed commercially exploitable and as

7 such demand protection. Such work is likely to be relatively mature and predominantly arise from work within industry. More speculative work is frequently carried out with the academic system and predominantly recorded as published papers. Such publication data may provide indicators similar to patents for less mature innovations. This could be explored as an indicator of research and development effort. References. Porter A. L. and others. Forecasting and the Management of Technology. New York: John Wiley and Sons, Inc; 99; ISBN: Twiss B. Forecasting for Technologists and Engineers. London: IEE; 992; ISBN: Hills P. C. Market Information and the Designer [Master of Philosophy Thesis]: University of Southampton; Dubash R. K. and Chakravati A. K. Technology Trends Monitor. Electronics Information and Planning. 996; 9: Fisher J. C. and Pry R. H. A Simple Model of Technological Change. Technological Forecasting and Social Change. 97; 3: Intel Corporation. Moore s Law - Intel Website [Web Page] Available at: 7. Burd, T. CPU Info Center [Web Page]. Accessed 998 Aug 3. Available at: 8. Hoon S.R. (s.r.hoon@mmu.ac.uk). Magnetic Materials. to: Palmer P.J. (p.j.palmer@lboro.ac.uk). 997 Sep. 9. British Telecom. BT Home Highway [Web Page] Available at: 0. Palmer P. J.; Hughes C., and Williams D. J. Electronic Technology Trends: An Update. 997; 97/.4. (Loughborough University Processes Group Technical Report).. Helmer O. Technological Forecasting for Industry and Government: Analysis of the Future - The Delphi Method. in: Bright J. R.NJ: Prentice-Hall Inc; 968; pp Crowe C. R. Research to Products. SPIE. 996; 272: Godin B. The Relationship between Science an Technology: Bibliometric Analysis of Papers and Patents in Innovative Firms [D Phil Thesis]: Sussex University; Katz J. S. and Hicks D. Desktop Scientometrics. Scientometrics. 997; 38. : Anonymous. Scientometrics: Discovering Science and Technology. Software Magazine. 998; 8. 7:34. ISSN: