The Origin, Nature, and Implications of
The Benchmark of Progress
in Semiconductor Electronics
by Bob Schaller
In fulfillment of
course requirements for
Dr. Roger Stough
September 26, 1996
This study will examine the development and evolution of semiconductor electronics, and in particular attempt to more completely explain "Moore's Law," a phenomenon unique to the rapid innovation cycles of this technology and thus the semiconductor industry as a whole. Gordon E. Moore's simple observation more than three decades ago that circuit densities of semiconductors had and would continue to double on a regular basis has not only been validated, but has since been dubbed, "Moore's Law" and now carries with it enormous influence. It is increasingly referred to as a controlling variable -- some have referred to it as a "self-fulfilling prophecy." The historical regularity and predictability of "Moore's Law" produce organizing and coordinating effects throughout the semiconductor industry that not only set the pace of innovation, but define the rules and very nature of competition. And since semiconductors increasingly comprise a larger portion of electronics components and systems, either used directly by consumers or incorporated into end-use items purchased by consumers, the impact of "Moore's Law" has led users and consumers to come to expect a continuous stream of faster, better, and cheaper high-technology products. The policy implications of "Moore's Law" are significant as evidenced by its use as the baseline assumption in the industry's strategic "roadmap" for the next decade and a half.
Organization of the Paper
This paper attempts to describe the origin, nature, and implications of "Moore's Law" in a comprehensive fashion. It begins with an historical overview of the major developments in semiconductor electronics that led up to Gordon Moore's 1965 observation. The next section examines the "Moore's Law" concept in detail along with some of its broader implications. This is followed by a review of the critical input side of the industry -- semiconductor manufacturing equipment makers. The paper then briefly examines "Moore's Law" analogues along with more general interpretations and policy considerations. Finally, preliminary conclusions are offered.
Genesis: Bell Labs and the Transistor
The invention of the transfer resistor or "transistor" in 1947 by Bell Laboratory researchers heralded in a new era of solid-state electronics. The concept was based on the fact that it is possible to selectively control the flow of electricity through a material such as silicon, a solid material -- thus "solid-state" -- with unique conducting properties, designating some areas as conductors of current and adjacent areas as insulators -- thus the term "semiconductor." Compared with the vacuum tube (known also as the thermoionic valve), which was the dominant technology for this task at the time, the transistor proved significantly more reliable, required much less power, and most importantly, could be miniaturized to almost infinitesimal levels. This paper examines this last point in particular as it is the basis for "Moore's Law."
The 1950s saw significant progress in solid-state research along with the creation of an entire new industry that would design and manufacture semiconductor devices. Although AT&T's Bell Labs is credited with the birth and early development of this new industry, a 1956 Consent Decree ending an anti-trust case prohibited AT&T from marketing commercial solid-state devices and required them to disseminate their patents and technology throughout the industry. Ironically, that very same year three AT&T scientists at Bell Labs won the Nobel prize for their discovery of the transistor. Indeed, AT&T's Western Electric would initially become the largest semiconductor producer to satisfy the device requirements for its own telecommunications systems. Firms with this internal demand became known as "captive" users. Although other systems houses -- most notably IBM -- would also do the same, AT&T's pure and applied scientific contributions were vital to the launching of the industry in such a profound way. Dosi (1984) appropriately refers to this critical role as the industry's 'bridging institution'.
From Science to Technology of Production
There is almost universal acceptance that the discovery of the transistor is the modern era's example of the economic fruits of science and a testament to Vannevar Bush's "Endless Frontier" assertion for continued post-WWII emphasis on basic research. But commercial production of the new device proved very difficult and would take most of the 1950s to iron out. Early devices were hand-made under extremely crude conditions compared with today's "clean rooms." Thus, initial yields (% of good devices manufactured) were very low -- 20% to 30% were common and even lower rates on some sophisticated devices -- while operating characteristics of working devices varied considerably. From the start, improving yields became one of the industry's primary production challenges. It would take advances in technology, specifically process technology, to improve production methods and, in turn, develop a viable semiconductor industry. As Braun states, "It was process technology that determined the winners in the semiconductor race." (Forester 1982) Gordon Moore himself recalls the important and unique role of technology in the early stages, "Indeed, the technology led the science in a sort of inverse linear model." (Moore 1996)
Throughout the 1950s the industry continued to learn the art of semiconductor production, continually refining ad hoc, trial-and-error methods. Improved production methods enabled additional advances in products and processes. The development of the integrated circuit in 1958 represents a major product milestone, made possible by overcoming technological barriers. Jack Kilby, inventor of the integrated circuit (IC), commented on this new device as a prime example of the transition from a science-based enterprise to one increasingly based on technology:
"In contrast to the invention of the transistor, this [integrated circuit] was an invention with relatively few scientific implications. . . Certainly in those years, by and large, you could say it contributed very little to scientific thought." (Braun & Macdonald 1982)
A second major breakthrough of the 1950s is better described as a series of incremental process innovations in the manufacturing of semiconductor devices. Work at Bell Labs and General Electric produced most of these innovations. Bell Labs' sharing of these methods in formal symposia made possible rapid process technology diffusion throughout the industry. The two most noteworthy innovations are the diffusion and oxide masking process, and the planar process, both becoming the permanent basis for production since. The diffusion process allowed the producer to diffuse impurities (dopants) directly into the semiconductor surface, eliminating the tedious practice of adding conducting and insulating material layers on top of the substrate. The addition of sophisticated photographic techniques permitted the laying of intricate mask patterns on the semiconductor so that diffusion took place only in designated areas. This greatly increased the accuracy of production while improving the reliability of devices. With diffusion, production moved from a craft process of individual assembly to batch processing.
The planar process was a logical outgrowth of the diffusion and oxide masking process. Planarization was the creation of physicist Jean Hoerni of newly-formed Fairchild Semiconductor. Hoerni observed the production limitations of conventional 3-dimensional transistor designs (e.g., the "mesa" transistor). Hoerni reasoned that a design based on a "plain" would be superior. Thus, the planar transistor, as the name implies, was flat. Flattening the mesa enabled electrical connections to be made, not laboriously by hand, but by depositing an evaporated metal film on appropriate portions of the semiconductor wafer. Using a lithographic process of a series of etched and plated regions on a thin, flat surface or wafer of silicon, the "chip" was born out of the planar transistor. Like the printing process itself, the planar process allowed for significantly greater rates of production output at even higher yields.
More importantly, the planar process enabled the integration of circuits on a single substrate since electrical connections between circuits could be accomplished internal to the chip. Robert Noyce of Fairchild quickly recognized this. As Gordon Moore recalls:
"When we were patenting this [planar transistor] we recognized it was a significant change, and the patent attorney asked us if we really thought through all the ramifications of it. And we hadn't, so Noyce got a group together to see what they could come up with and right away he saw that this gave us a reason now you could run the metal up over the top without shorting out the junctions, so you could actually connect this one to the next-door neighbor or some other thing."
Fairchild introduced the first planar transistor in 1959 and the first planar IC in 1961. As will be discussed later, Moore views the 1959 innovation of the planar transistor as the origin of "Moore's Law." Perhaps more than any other single process innovation, planarization set the industry on its historical exponential pace of progress. As one early industrial technologist noted, "The planar process is the key to the whole of semiconductor work." George Gilder's account in his 1989 treatise, Microcosm, is more eloquent:
"Known as the planar integrated circuit, Fairchild's concept comprised the essential device and process that dominates the industry today. . . Ultimately it moved the industry deep into the microcosm, and put America on the moon."
With time and experience, ad hoc production methods were replaced with more formalized technology-based processes. To underscore the importance of process innovations, Braun and Macdonald (1982) state that much of the early growth in semiconductor electronics "was not only permitted by new processes, but actually precipitated by them, for batch production in general, and planar in particular, prompted both a rapid increase in the numbers of components produced and an even more rapid decline in their price."
Amazingly, the industry has not veered from this course since then. With time, chip manufacturers improved the lithographic process with more precise photographic methods and "photolithography" thus became the standardized production method for the industry. More pertinent to "Moore's Law," photolithography enabled manufacturers to continue to reduce feature sizes of devices. Commenting on the significance of photolithography within the planar process Malone states, "Thus were planted the seeds of Moore's Law, the very principle that governs the information age." The research focus of the 1950s, moving from laboratory to the production floor, gradually shifted its emphasis from understanding why to learning how. Creating and mastering the art of photolithography is an excellent example of this transition from science to technology.
"The use of photolithography is yet another example of interdependence of technologies and cross-fertilisation. The method had been developed for printing purposes and had been in use in this area for some time. It is but one outstanding example of the adoption and adaptation of extraneous technologies to improve the manufacture and design of electronic devices." (Braun and Macdonald 1982)
A New Industry from a New Technology
The secondary literature on the development of the semiconductor industry -- including the phenomenon called "Silicon Valley" -- is extensive and need not be reviewed here. One common theme worth noting is that this industry is qualitatively different as characterized by its base technology which seems to provide a limitless source of performance advancement. From the beginning this was recognized primarily by new firms, not existing electronics device firms.
Harvard's Theodore Levitt's "Marketing Myopia" (1960) noted that the once-dominant railroad industry had completely missed the opportunities brought about by technological advances in other modes of transportation. The railroad industry's narrow definition of its market as the "railroad" business, as opposed to the broader "transportation" business excluded its participation in whole new automobile, truck, and airplane/airline industries. A similar parallel can be drawn regarding the creation of the semiconductor industry -- none of the major semiconductor players today bears the name of dominant electronics firms of the 1950s (e.g., General Electric, RCA, Raytheon, Sylvania, Philco-Ford, and Westinghouse). These firms, all heavily engaged in the production of vacuum tubes, did make substantial early investments in semiconductor electronics. But the semiconductor industry that emerged by 1960 is represented by a whole new breed of firms, some from other seemingly unrelated industries, some entirely new. Texas Instruments, Shockley Laboratories, and Fairchild Semiconductor are three of the many new firms that had emerged. Each had a traceable connection to Bell Labs.
Texas Instruments, a geophysical company that provided oil well services, was one of the first to purchase a license from AT&T and begin semiconductor design and manufacturing operations. Texas Instruments' Gordon Teal, a former Bell Labs researcher, successfully produced the first silicon transistor that would prove significantly easier to manufacture while possessing much improved operating characteristics over the germanium transistor in use at the time. Robert Shockley, also formerly of Bell Labs and Nobel laureate for the co-discovery of the transistor, established Shockley Transistor Laboratories, gathering together some of the best minds at the time, including a young engineer named Gordon Moore. Within a few years, Moore and others at Shockley Labs convinced Fairchild Camera and Instrument, an aerial survey company, to finance a new enterprise, so they left Shockley and formed Fairchild Semiconductor. Moore would head up the research department at Fairchild, where he would later make his circuit density doubling observation. The innovative breakthrough of the IC in the late-1950s as previously discussed actually involved both firms. Jack Kilby at Texas Instruments produced the first germanium IC while Robert Noyce at Fairchild quickly made the concept technically and economically feasible -- thus commercially viable -- by developing the planar process. Moore recalls the significance of the planar process at Fairchild, "In the planar structure, Fairchild struck a rich vein of technology." (Moore 1996)
The story of Fairchild Semiconductor is a fascinating one and is illustrative of the dynamic nature of this industry, especially in its early days. Fairchild is the subject of much industry lore. The young founders (including Moore), seeking to make good on commercial success in semiconductor production, left Shockley Laboratories in 1957 calling themselves the "Fairchild Eight" and founded Fairchild Semiconductor. Fairchild is thought to have spawned no fewer than 150 companies, including Moore's and Noyce's Intel in 1968 -- these spin-offs have come to be referred to as "Fairchildren." It was at Fairchild that Gordon Moore, Director of Research, made his profound density-doubling observation and extrapolation.
Gordon Moore's Observation
The April 19, 1965 Electronics magazine was the 35th anniversary issue of the publication. Located obscurely between an article on the future of consumer electronics by an executive at Motorola, and one on advances in space technologies by a NASA official is a less than four page (with graphics) article entitled, "Cramming more components onto integrated circuits," by Gordon E. Moore, Director, Research and Development Laboratories, Fairchild Semiconductor. Moore had been asked by Electronics to predict what was going to happen in the semiconductor components industry over the next 10 years -- to 1975. He speculated that by 1975 it was possible to squeeze as many as 65,000 components on a single silicon chip occupying an area of only about one-fourth a square inch. His reasoning was a log-linear relationship between device complexity (higher circuit density at reduced cost) and time:
"The complexity for minimum component costs has increased at a rate of roughly a factor of two per year. Certainly over the short term this rate can be expected to continue, if not to increase. Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will remain nearly constant for at least 10 years." (Moore 1965)
This was an empirical assertion, although surprisingly it was based on only three data points.
Ten years later, Moore delivered a paper at the 1975 IEEE International Electron Devices Meeting in which he reexamined the annual rate of density-doubling. Amazingly the plot had held through a scatter of different complex bipolar and MOS device types (see Product and Technology Overview) introduced over the 1969-1974 period. A new device to be introduced in 1975, a 16k charge-coupled-device (CCD) memory, indeed contained almost 65,000 components. In this paper, Moore also offered his analysis of the major contributions or causes of the exponential behavior. He cited three reasons. First, die sizes were increasing at an exponential rate -- chip dice were getting bigger. As defect densities decreased, chip manufacturers could work with larger areas without sacrificing reductions in yields. Many process changes contributed to this, not the least of which was moving to optical projection rather than contact printing of the patterns on the wafers.
The second reason was a simultaneous evolution to finer minimum dimensions (i.e., feature sizes or line widths). This variable also approximated an exponential rate. Combining the contributions of larger die sizes and finer dimensions clearly helped explain increased chip complexity, but when plotted against the original plot by Moore, roughly one-third of the exponential remained unexplained.
Moore attributed the remaining third to what he calls "circuit and device cleverness." He notes that several features had been added. Newer approaches for device isolation, for example, had squeezed out much of the unused area. The advent of metal oxide semiconductor (MOS) technology in the late-1960s and early-1970s had allowed even tighter packing of components per chip. Interestingly, he also concluded that the end of "cleverness" had arrived with the CCD memory device:
"There is no room left to squeeze anything out by being clever. Going forward from here we have to depend on the two size factors - bigger dice and finer dimensions."
So Moore revised upward his annual rate of circuit density-doubling. Every eighteen months seemed to be a reasonable rate and was supported by his analysis. He redrew the plot from 1975 forward with a less steep slope reflecting a slowdown in the rate, but still behaving in a log-linear fashion. Shortly thereafter someone (not Moore) dubbed this curve, "Moore's Law." Officially, Moore's Law states that circuit density or capacity of semiconductors doubles every eighteen months or quadruples every three years. It even appears in mathematical form:
(Circuits per chip) = 2(year-1975)/1.5
In 1995 Moore compared the actual performance of two device categories (DRAMs and microprocessors) against his revised projection of 1975. Amazingly, both device types tracked the slope of the exponential curve fairly closely, with DRAMs consistently achieving higher densities than microprocessors over the 25 year period since the early-1970s. Die sizes had continued to increase while line widths had continued to decrease at exponential rates consistent with his 1975 analysis.
Moore's early prediction was based on the shared observations by many in the fledgling semiconductor industry. The invention of the transistor had started a miniaturization trajectory in semiconductors which had produced the integrated circuit in the late-1950s, soon followed by medium scale integration (MSI) of the mid-1960s, then large scale integration (LSI) of the early-1970s, very large scale integration (VLSI) of the 1980s, and ULSI (ultra) of the 1990s. Today's Intel PentiumJ microprocessor contains more than three million transistors, the Motorola PowerPCJ microprocessor contains almost seven million transistors, and Digital's 64-bit AlphaJ microprocessor contains almost 10 million transistors on a thin wafer "chip" barely the size of a fingernail. In early-1996 IBM claimed that a gigabit (billion bits) memory chip was actively under development and would be commercially available within a few years. Papers presented at a 1995 IEEE International Solid-State Circuits Conference contend that terachips (capable of handling a trillion bits or instructions) will arrive by the end of the next decade. (Stix 1995)
Implications: Technological Barometer?
The implications of Moore's Law are quite obvious and profound. It is increasingly referred to as a ruler, gauge, benchmark (see subtitle), barometer, or some other form of definitive measurement of innovation and progress within the semiconductor industry. As one industry watcher has recently put it:
"Moore's Law is important because it is the only stable ruler we have today, . . It's a sort of technological barometer. It very clearly tells you that if you take the information processing power you have today and multiply by two, that will be what your competition will be doing 18 months from now. And that is where you too will have to be." (Malone 1996)
Since semiconductor cost is measured in size and complexity, unit cost is directly related with size -- as circuit size has been reduced, so has cost. As a result, virtually all electronics used today incorporate semiconductors. These devices perform a wide range of functions in a variety of end-use products -- everything from children's toys, to antilock brakes in automobiles, to satellite and weapon systems, to a variety of sophisticated computer applications. The fact that all these products (and many, many more) are now so accessible to so many users is due in large part to continually declining costs of the core microelectronics made possible by the innovation of the semiconductor.
Perpetuum Mobile, Self-Fulfilling Prophecy, or Both?
Perhaps the broadest implication of Moore's Law is that it has become an almost universal guide for an entire industry that has not broken stride in exponential growth rates for almost four decades now. The repeated predictability and regularity of Moore's Law are characteristics of the elusive perpetuum mobile for this industry. Some have referred to Moore's Law as self-reinforcing or a "self-fulfilling prophecy." Moore himself recently stated:
"More than anything, once something like this gets established, it becomes more or less a self-fulfilling prophecy. The Semiconductor Industry Association puts out a technology road map, which continues this generation [turnover] every three years. Everyone in the industry recognizes that if you don't stay on essentially that curve they will fall behind. So it sort of drives itself." (Moore 1996)
There is intuitive merit to this view. The inherent characteristics of the technology contribute significantly to this "drive itself" tendency. Chip makers have long recognized the combined benefits of miniaturization. As Moore summarizes:
"By making things smaller, everything gets better simultaneously. There is little need for tradeoffs. The speed of our products goes up, the power consumption goes down, system reliability, as we put more of the system on a chip, improves by leaps and bounds, but especially the cost of doing things electronically drops as a result of the technology." (Moore 1995)
Braun and Macdonald (1982) also refer to the "self-sustaining" nature of miniaturization in semiconductors as "tradeoffs," in Moore's words, don't really enter into the equation. In economic parlance, this is the proverbial "free lunch."
From a different angle, George Gilder (1989) argues that the technology itself possesses an almost natural "microcosmic" force toward integration in smaller and smaller spaces. He refers to this as the "law of the microcosm" and suggests that users and other institutions affected by the technology understand and follow its direction:
"Rather than pushing decisions up through the hierarchy, the power of microelectronics pulls them remorselessly down to the individual. This is the law of the microcosm. . . The very physics of computing dictates that complexity and interconnections -- and thus computational power -- be pushed down from the system into single chips . . . Above all, the law of the microcosm means the computer will remain chiefly a personal appliance . . . Integration will be downward onto the chip, not upward from the chip."
He draws some fairly broad implications by stating that the evolution of chip-related industries "will remorselessly imitate the evolution of the chip." That is smaller, thus cheaper, yet more powerful chip capabilities will redefine entire industries away from larger, oligopolistic structures to smaller structures, more conducive to an entrepreneurial environment. He makes a convincing case with telecommunications, referencing Peter Huber's post-AT&T break-up analysis of the "geodesic" or horizontal network that evolved within the telephone system. Huber asserts that the traditional pyramidal network model, where all switching was done in the central office, had been made obsolete by decentralized switching systems such as private branch exchanges, local area networks, and related systems. Thus it was only natural to accord an industry a more horizontal competitive landscape consistent with its redefined geodesic network structure.
Yet another dimension, involving non-technical or non-physical variables such as user expectations contribute to the dynamic of fulfilling this law. In this view, Moore's Law is not based on the physics and chemical properties of semiconductors and their respective production processes, but on other non-technical factors. One hypothesis is that a more complete explanation of Moore's Law has to do with the confluence and aggregation of individuals' expectations manifested in organizational and social systems which serve to self-reinforce the fulfillment of Moore's prediction.
A brief examination of the interplay among only three components of the personal computer (PC) (i.e., microprocessor chip, semiconductor memory, and system software) helps reveal this point. A very common scenario using the IBM-compatible PC equipped with an Intel microprocessor and running Microsoft's WindowsJ software goes something like this. As the Intel microprocessor has evolved from the 8086/88 chip in 1979 to the 286 in 1982, to the 386 in 1985, to the 486 in 1989, to the PentiumJ in 1993, and the Pentium ProJ in 1996, each incremental product has been markedly faster, more powerful, and less costly as a direct result of Moore's Law. At the same time, dynamic random access memory (DRAM) and derivative forms of semiconductor memory have followed a more regular Moore's Law pattern to the present where a new PC comes standard with 8Meg (million bits) to 16Meg of memory as compared to the 480k (thousand bits) standard of a decade ago. Both of these cases reflect the physical or technical aspects of Moore's Law.
However, system software, the third piece of this puzzle, begins to reveal the non-technical dimension of Moore's Law. In the early days of computing when internal memory was costly and scarce, system software practices had to fit this limitation -- limited memory meant efficient use of it or "tight" code. With the advent of semiconductor memory -- especially with metal oxide semiconductor (MOS) technology -- internal memory now obeyed Moore's Law and average PC memory sizes grew at an exponential rate. Thus, system software was no longer constrained to "tight spaces" and the proliferation of thousands, then many thousands, and now millions of "lines of code" have become the norm for complex system software.
Nathan Myhrvold, Director of Microsoft's Advanced Technology Group, conducted a study of a variety of Microsoft products by counting the lines of code for successive releases of the same software package. (Brand 1995) Basic had 4,000 lines of code in 1975 -- 20 years later it had roughly half a million. Microsoft Word consisted of 27,000 lines of code in the first version in 1982 -- over the past 20 years it has grown to about 2 million. Myhrvold draws a parallel with Moore's Law:
"So we have increased the size and complexity of software even faster than Moore's Law. In fact, this is why there is a market for faster processors -- software people have always consumed new capability as fast or faster than the chip people could make it available."
As the marginal cost of additional semiconductor processing power and memory literally approaches zero, system software has exponentially evolved to a much larger part of the "system." More complex software requires yet even more memory and more processing capacity, and presumably software designers and programmers have come to expect that this will indeed be the case. Within this scenario a kind of reinforcement multiplier effect is at work.
A "Slipstream" to Software Development?
This network reinforcement multiplier effect is most noticeable in computers and related products. This point is further emphasized since computers represent the single largest user category of semiconductor devices at 60% of the entire industry demand, primarily for microprocessors and DRAMs. A very distant second is telephones at 10%. After that, nothing else comes close. (Hutcheson and Hutcheson 1996, Economist 1996) Arguably in computer software -- as in semiconductors -- complexity has also been rising exponentially. As just discussed, though, the rate of increased software complexity appears to be outpacing that of the chips that comprise the hardware that drives the software. One noted software programmer has propounded two new Parkinson's Laws for software: "Software expands to fill the available memory," and "Software is getting slower more rapidly than hardware is getting faster." (Gilder 1995) Indeed, newer programs seem to run more slowly on most systems than their previous releases (e.g., compare WordPerfect 6.0 for Windows with WordPerfect 5.1 for DOS). Microsoft, especially with its Windows development and emergent "Wintel" (Windows-Intel) de facto standard, owes much of its success to shrewdly exploiting the advances of microcosmic hardware. (Gilder 1995, 1989)
"[Bill] Gates travels in the slipstream behind Moore's Law, following a key rule of the microcosm: Waste transistors. . . 'Every time Andy [Grove] makes a faster chip, Bill uses all of it.' Wasting transistors is the law of thrift in the microcosm, and Gates has been its most brilliant and resourceful exponent." (Gilder 1995)
When asked recently about his view of "Wintel," Moore, Chairman of the Board at Intel, quips, "Our legal department doesn't like it at all." He then expands on the strategic importance of the hardware/software technological alliance, but also acknowledges the independence of architectures made possible by an ever-changing industry.
"We certainly will try to keep it [Wintel] going that way. We have a tremendous asset in all the compatible software that's out there, so any new processor we introduce has to be able to run that stuff and as long as we keep a very large fraction of the processors, I think Microsoft will be sure that they write things that run well on our processors. And, we both have ideas of being somewhat independent. We're happy to have Java applications, and then Netscape, UNIX and everything else, and also Microsoft ports NT to [Digital's] Alpha, but in fact there is a tremendous advantage to the volume centers business."
It is clear that a type of lock-in (Arthur 1994) has occurred with respect to PC system hardware and software architecture. The history of this particular alliance beginning with IBM's early selection of both Intel and Microsoft as critical component suppliers (microprocessor and operating system, respectively) of their revolutionary PC has ultimately contributed to a decade and a half of mutual learning by the two firms, ironically now without IBM. Had the choice been different, who knows what system architecture would have evolved? What's important is that one has, involving -- and producing -- two of the most important players in the PC industry today in Intel and Microsoft. And there is no doubt that this alliance affects development and innovation cycles of both firms, thus the industry at large. Whether Moore's Law is the slipstream to software development as George Gilder asserts, or the other way around, may be a kind of "chicken and egg" question. There is little doubt that a significant expectations feedback loop involving Moore's Law is at play. This feedback mechanism is illustrated in Figure 1.
Scaling from J-Shaped to S-Shaped Curves
As stated earlier, the exponential pace of innovation was generally understood within the semiconductor community by the mid-1960s. Erich Bloch credits Gordon Moore as being "the most articulate" of the early group of technologists in communicating this phenomenon. Carver Mead, noted computer scientist at Caltech, did a series of early calculations to determine the precise scaling effects of the technology. This work intensified with the introduction of MOS technology in the late-1960s and by 1972 the first comprehensive scaling analysis was published. Mead's analysis confirmed that Moore's extrapolation was not only possible, but probable, and added academic credence to the assertion. In a more recent study, Mead (1994) reexamined his earlier scaling estimates and then looked ahead:
"Over the ensuing 22 years, feature sizes have evolved from 6 to 0.6F, and the trend shows no sign of abating. . . I shall conclude that we can safely count on at least one more order of magnitude of scaling, with a concomitant increase in both density and performance."
This form of analysis is consistent with technology development along exponential "J-shaped" or "S-shaped" curves. (Rothschild 1990, Foster 1986, and Klein 1984, 1977) In the field of economics, particularly its evolutionary strain in the field of Complexity Science, this phenomenon is known as increasing returns. (Arthur 1994, Waldrop 1992) Whatever label is applied, there is little question that there is still considerable "learning" occurring in exploiting the potential physical properties of semiconductors along with the associated production processes. At some point this technology -- like all technologies -- will reach its limit of exponential growth and begin to experience diminishing marginal returns (at the top hump of the "S").
So When Will Moore's Law End? Is This The Right Question?
A 1995 article in the Economist is titled, "The End of the Line" and discusses the impending fate of Moore's Law. A similar Forbes article is titled, "Whither Moore's Law?" while a recent editorial's headline in a Unix Users Group's Internet home page (CUUG 1996) reads, "The End of Moore's Law: Thank God!" Numerous other forecasts have come to similar conclusions. But as discussed earlier, Moore's Law started out as a simple observation and extrapolation. Actual performance and experience have validated Moore's original plot, proving him quite prophetic. An intriguing point about Moore's Law is that throughout its existence, forecasts of its demise have consistently been wrong. For example, in an exhaustive study on the history and impact of the semiconductor, Braun and Macdonald (1982) came to a similar conclusion as Gordon Moore had in his original 1965 article:
"Unlike the consequences arising from the future use of semiconductor electronics, the technical future of the technology, though still uncertain, can be forecast with a degree of confidence over the very short term. For the time being, trends of increased circuit densities will continue, although no-one expects Moore's Law to hold for very much longer."
The authors go on to say that the microprocessor would probably reach its zenith with a 32-bit architecture and questioned whether the 256k DRAM would become the "ultimate single chip memor[y]." In a decade and a half since, Digital's 64-bit Alpha 21164 microprocessor chip contains almost 10 million transistors, operating at more than 300 MHZ, and the 16 Meg DRAM are now the contemporary state-of-the-art chips, with advanced designs soon to eclipse these capabilities.
In late-1994, The Semiconductor Industry Association (SIA) published the National Technology Roadmap for Semiconductors. The Roadmap is a consensus view of the industry's technical vision and forecast over the next decade and a half -- through 2010. The second paragraph of the document contains the statement, "A central assumption of the Roadmap is an extension of industry history according to Moore's law." (SIA 1994)
A recent survey of some of the best thinkers in the high-tech industry revealed a wide range of responses to the question, "How many more years will Moore's Law play out?" including:
"With conventional lithography, another three to five [years], max. . . 10 to 15 years max. . . At least another 20 years or more. . . Moore's Law has worked in the past 25 years or so. There's no doubt that it will continue. . . We'll all be dead when Moore's Law is played out." (Malone 1996)
A very revealing follow-up question, "What will stop it [Moore's Law] -- design limits, manufacturing limits or fabrication costs?" has several predictable answers such as, "The fundamental physics of silicon will become a limiting factor." However, one respondent, Dan Lynch, President and CEO of CyberCash, offers a starkly different view by answering, "Moore's Law is about human ingenuity progress, not physics." (Malone 1996)
Along similar lines, Carver Mead (now Gordon and Betty Moore Professor of Engineering and Applied Science at Caltech) states that Moore's Law "is really about people's belief system, it's not a law of physics, it's about human belief, and when people believe in something, they'll put energy behind it to make it come to pass." Mead offers a retrospective, yet philosophical explanation of how Moore's Law has been reinforced within the semiconductor community through "living it":
"After it's [Moore's Law] happened long enough, people begin to talk about it in retrospect, and in retrospect it's really a curve that goes through some points and so it looks like a physical law and people talk about it that way. But actually if you're living it, which I am, then it doesn't feel like a physical law. It's really a thing about human activity, it's about vision, it's about what you're allowed to believe. Because people are really limited by their beliefs, they limit themselves by what they allow themselves to believe what is possible. So here's an example where Gordon [Moore], when he made this observation early on, he really gave us permission to believe that it would keep going. And so some of us went off and did some calculations about it and said, 'Yes, it can keep going'. And that then gave other people permission to believe it could keep going. And [after believing it] for the last two or three generations, 'maybe I can believe it for a couple more, even though I can't see how to get there'. . . The wonderful thing about [Moore's Law] is that it is not a static law, it forces everyone to live in a dynamic, evolving world." (UVC 1992)
The historical literature reveals a pattern -- beginning with Moore's original 1965 prediction -- that the longer-term predictions (greater than 10 years) of diminishing marginal complexity increases simply have not yet come to pass. In fact, the latest set of "predictions" in 1996, although collectively more optimistic than previous samples, still peg a future longer-term limit at less than 15 years. (Malone 1996) In a very recent interview, Moore himself seems to stick to the "about another decade" prediction he originally made in 1965:
"I think much of the rate of progress can be expected to continue for at least a few more generations. Three generations of technology at three years per generation is about a decade. So I can see us staying on roughly the same curve that long." (Moore 1996)
At the same time, Moore recognizes that history has proven him and mostly everyone else wrong about past predictions. His closing remarks at a Microlithography Symposium in February 1995 challenged the participants to continue to "think smaller":
"Semiconductor technology made its great strides as a result of ever increasing complexity of the products exploiting higher and higher density to a considerable extent the result of progress in lithography. As you leave this meeting I want to encourage each of you to think smaller. The barriers to staying on our exponential are really formidable, but I continue to be amazed that we can either design or build the products we producing today. I expect you to continue to amaze me for several years to come." (Moore 1995)
Internal and External Sources of Innovation
The transistor and its extensive lineage of semiconductor products are arguably the result of much technology push, intrinsic to the nature of these devices. Arguing against the conventional wisdom that product innovations are typically developed solely by product manufacturers, von Hippel (1986) used the title, The Sources of Innovation to explain that:
". . . the sources of innovation vary greatly. In some fields, innovation users develop most innovations. In others, suppliers of innovation-related components and materials are the typical sources. In still other fields, conventional wisdom holds and product manufacturers are indeed the typical innovators."
In an exhaustive case study of the semiconductor industry, Dosi (1984) concluded that U.S. public (military and space) policies initially performed an essential external role of selection and guidance of the directions of technical progress, but noted that this role has since decreased. Moore (1996) agrees with this view, noting that defense R&D and particularly the space program of the 1960s had a "negligible impact on the semiconductor industry."
Dosi then poses the important question, "What are the factors which shape the directions of the innovative activity when powerful external factors cease to exert their 'pulling' and 'pushing' influence?" He goes on to argue three major factors. First, 'normal' technical progress maintains a momentum of its own which defines the broad orientation of the innovative activities. He refers to this "in-built heuristic" in so many words as Moore's Law:
"Take, for example, the fundamental trend in the industry towards increasing density of the circuits: the doubling of the number of components per chip every year (in the late 1970s every two-three years) is almost a 'natural law' of the industry. After 1K memories one progressed to 4K, 16K, 64K and further increases in integration are expected. The same applies to microprocessors, from 4 to 8, 16, 32 bit devices. This cumulative process has an important role in the competitive process of the industry, by continuously creating asymmetries between firms and countries in their relative technological success."
The second factor stems from the mutual relationship between innovation in semiconductors and end-use applications. Technical change in semiconductors defines one of the boundary conditions of possible technical advances in 'downstream' sectors. At the same time, both technological problems and commercial opportunities in these downstream sectors focus and lead the direction of technological advances in semiconductors. As previously discussed, the interplay of the "Wintel" architecture is most evident here. Moore acknowledges this, but continues to emphasize the "pushing" force of semiconductor electronics:
"There's still a lot of push [going on], we work it on both ends. You look at what Intel does, for example, our thrust in video conferencing. That is driven principally as an application that requires higher performance processing to support, so our business depends on continuing that model where everybody needs more computing power every year, so we're trying to drive as much push as we can."
A third factor Dosi cites is the more traditional economic "market-pull" influence from changes in relative prices and distributive shares. Dosi emphasizes that market factors operate particularly with respect to 'normal' technical progress, and second, that technical progress occurs within the boundaries defined by the basic technological trajectory. This suggests that user feedback can be self-reinforcing within the parameters of the technological trajectory of semiconductors.
Finally, Hutcheson and Hutcheson (1996) offer a more critical view of the regularity typically associated with Moore's Law. They point out that underlying production limitations are becoming increasingly difficult to overcome.
"In stark contrast to what would seem to be implied by the dependable doubling of transistor densities, the route that led to today's chips was anything but smooth. It was more like a harrowing obstacle course that repeatedly required chipmakers to overcome significant limitations in their equipment and production processes. None of these problems turned out to be the dreaded showstopper whose solution would be so costly that it would slow or even halt the pace of advances in semiconductors and, therefore, the growth of the industry. Successive roadblocks, however, have become increasingly imposing, for reasons tied to the underlying technologies of semiconductor manufacturing."
The physics underlying semiconductor manufacturing steps suggests several potential obstacles to continued technical progress and density doubling. For example, the gigabit chip generation may finally force technologists up against the limits of optical lithography. Lithographers confront the formidable task of building structures smaller than the wavelength of light (see Figure 2). "Think of it as trying to paint a line that is smaller than the width of the paintbrush," says a researcher at Bell Labs. (Stix 1995) He goes on to say that there are ways of doing it, but the cost involved may be prohibitive. Economics may constrain Moore's Law before the limits of physics. The reality is that both are closely intertwined which brings us to "Moore's Second Law."
Moore's Second Law: Economics
In 1977, Robert Noyce, then Chairman of the Board at Intel, wrote:
"Today, with circuits containing 218 (262,144) elements available, we have not yet seen any significant departure from Moore's law. Nor are there any signs that the process is slowing down, although a deviation from exponential growth is ultimately inevitable. The technology is still far from the fundamental laws of physics: further miniaturization is less likely to be limited by the laws of physics than by the laws of economics."
Almost two decades later, Noyce's foresight of economic limitations has brought about what has been referred to Moore's Second Law. (Ross 1995) "What has come to worry me most recently is the increasing cost. . . This is another exponential," writes Moore (Economist 1995). In today's dollars, the cost of a new "fab" (fabrication plant) has risen from $14M in 1966 to $1.5B in 1995. By 1998 work will begin on the first $3B fabrication plant. Between 1984 and 1990, the cost of a fab doubled, but chip makers were able to triple the performance of a chip. In contrast, the next generation of fabs will see cost double again by 1998, but this is likely to produce only a 50% improvement in performance. The economic law of diminishing marginal returns appears to be setting in. If this exponential trend continues, by 2005 the cost of a single fab will pass the $10B mark (in 1995 dollars) or 80% of Intel's current net worth. According to Dan Hutcheson, President of VLSI Research, Moore's Law will fall victim to economics before it reaches whatever limitations exist in physics:
"The price per transistor will bottom out sometime between 2003 and 2005. From that point on, there will be no economic point to making transistors smaller. So Moore's Law ends in seven years." (Ross 1995)
State-of-the-art fabs become obsolete in three to five years; staying ahead in such a business requires a large chip maker to spend vast sums simply to keep up with technology. In 1995 the industry spent $30B in new fab capacity; Intel's share alone was $3B. To recoup its investment, a semiconductor firm will want to run the plant as near to full capacity as possible. When levels of demand change (and supply remains fixed), then wide price swings -- much like in farming, another commodity business -- cause supply to adjust stickily to demand. The result is an historical pattern of volatile market cycles producing mass surpluses and shortages. These industry-unique cycles are further aggravated by normal business cycle behavior at the macroeconomic level. The U.S. industry crisis of the early-mid 1980s is a vivid reminder of this economic impact.
So what are firms to do? Hutcheson and Hutcheson (1996) suggest that firms collaborate -- team up. They cite that increasingly, chip makers are sharing the costs of a new fab with customers, competitors, even countries. IBM and Toshiba are building a plant together, as are Motorola and Siemens. Also state-organized consortia appear to be on the rise in the newer participant countries in semiconductor manufacturing such as Korea and Singapore. Another point is that the role of the suppliers of materials and especially, manufacturing equipment, has become even more vital to the overall success of the industry. The next section describes the evolving role of the semiconductor manufacturing equipment (SME) industry.
The Semiconductor Manufacturing Equipment (SME) Industry
The invention of the point contact transistor by Bell Laboratories signaled the emergence of a new industry in the early-1950s -- semiconductor manufacturing. During this industry's infancy (and prior to the introduction of ICs), the companies producing semiconductors generally developed the equipment required to manufacture these new electronic components. The semiconductor equipment manufacturing industry, in fact, evolved out of the chip manufacturing firms. Von Hippel (1988) provides a detailed historical account of fifteen major innovations in silicon semiconductors (among other technologies) and strongly argues the dominant role of the user (chip-maker) in the innovation process. In many cases (e.g., mask alignment using split field optics), Fairchild or another user firm initially developed the technique in-house which was later offered commercially by an equipment manufacturer.
Gordon Moore recalls one experience of his early days at Fairchild involving a technician who was paid for work on nights and weekends at home to make capillary tubes used in a critical gold-bonding process:
"Pretty soon that business got so big that he quit and set up Electroglass, which was the first one of these equipment companies that I know of. . . and he'd also been helping me build furnaces -- we had to build our own furnaces in those days. So he took the basic furnace design and started building furnaces, first for us, then for the industry."
Volume production of ICs, starting in the 1960s, intensified this pattern. The increased complexity of ICs necessitated the development of new production equipment capable of much clearer resolution, narrower line widths, and more exacting alignment specifications than those used to produce discreet electronic components. Printing complex circuit designs onto highly polished wafers called for the development of sophisticated photolithographic and other wafer processing techniques such as ion implanting and etching, in addition to appropriate testing and assembly operations.
Since the early days of the industry, the production process has become increasingly -- now almost exclusively -- automated. Today, semiconductor manufacturing technologies can go no farther than the equipment necessary for their manufacture will allow. For example, the minimum line widths and, therefore, the maximum integration levels attainable in pursuit of Moore's Law, are directly influenced by the manufacturing equipment's lithographic capabilities.
The relationship between chip-makers and equipment-makers continues to be very close due to their mutual interdependence. This is in contrast to the traditional American view of "arm's length and cautious" behavior between buyer and seller. Equipment makers have long recognized their role as contributors to, and participants in, this technology-driven industry. Partnership arrangements, both formal and informal, along with technical seminars and publications, trade association meetings, industry conventions, and normal vendor/user relationships reinforce cooperative efforts between equipment producers and device manufacturers.
The next section takes a broader look at various interpretations and uses of Moore's Law, and its implications including those important to public policy.
Other Interpretations and Uses
Moore's Law is increasingly used as a metaphor or label for anticipated rates of rapid change -- not only in semiconductors, but in broader contexts. The source of this change is technological, but the effects of it are economic and social. In this very complex arrangement, Moore acknowledges that Moore's Law "gives us a short-hand to talk about things."
Recently, a software representative was quoted in the New York Times as saying, "The length of eternity is 18 months, the length of a product cycle." In some sense, Moore's Law has taken on a life of its own as it finds its way into the broader community of users and other institutions impacted by the technology. To assess this impact, an Internet keyword search on "Moore's Law" was recently conducted. Out of well over 100 pertinent references, more than two dozen quality references were obtained. Most references came from direct industry application including the front-end component of the SME industry. The majority of the references were from downstream user communities including software, PC users, and network and Internet applications. It is interesting to note that Moore's Law now has many "spin-offs" such as "Metcalf's Law." Surprisingly, the fields of education and even marketing have referred to Moore's Law. The following is a sample of the wide range of uses, interpretations, and applications found. Note that processing power, not circuit density, is increasingly becoming the new basis of Moore's Law.
"Management is not telling a researcher, 'You are the best we could find, here are the tools, please go off and find something that will let us leapfrog the competition.' Instead, the attitude is, 'Either you and your 999 colleagues double the performance of our microprocessors in the next 18 months, to keep up with the competition, or you are fired.'" (Odlyzko 1995)
"'Moore's Law' may one day be as important to marketing as the Four Ps: product, price, place, and promotion. . . If it is borne out in the future the way it has in the past, the powerful Pentium on your desktop will seem as archaic as a 286 PC in a few years." (Koprowski 1996)
"We have become addicted to speed. Gordon Moore is our pusher. Moore's law, which states that processing power will double every year and a half, has thus far held true. CPU designers, always in search of a better fix, drain every possible ounce of fat from processor cores, squeeze clock cycles, and cram components into smaller and smaller dies." (Joch 1996)
"So holding 'Moore's Law' as the constant, the technology in place in classrooms today will not be anything like the classroom of five years from now!" (Wimauna Elementary School 1996)
"The End of Moore's Law: Thank God!. . . The End of Moore's Law will mean the end to certain kinds of daydreaming about amazing possibilities for the Next Big Thing; but it will also be the end of a lot of stress, grief, and unwanted work." (CUUG 1996)
"Computer-related gifts must be the only Christmas presents that follow Moore's Law." (Sydney Morning Herald 1995)
"Moore's Law is why . . . smart people start saving for the next computer the day after they buy the one they have. . . Things are changing so fast that everyone's knowledge gets retreaded almost yearly. Thank you, Mr. Moore. . . [for] the internet, a creature of Moore's Law. . ." (Hettinga 1996)
Are There Any Good Analogues?
The examination of Moore's Law would not be complete without drawing analogues to other technologies. This has been done often for various reasons. For example, in arguing the uniqueness of the million-fold cost reductions and performance improvements in semiconductors, Gordon Moore jokingly cites that if similar progress were made in transportation technologies such as air travel, a modern day commercial aircraft would cost $500, circle the earth in 20 minutes, and only use five gallons of fuel. However, it may only be the size of a shoebox. Stephen Kline of Stanford has suggested a bit more appropriate use of the aircraft analogy, suggesting that the earlier era of rapid advances in aircraft speed and performance may offer additional insight.
Carver Mead suggests that magnetic storage, specifically disk drive technology, has followed a similar scaling path as semiconductors. He cites that PC hard drives in particular have evolved from megabyte (million bytes) to gigabyte (billion bytes) capacity in roughly a decade. This thousand-fold capacity improvement approaches Moore's original extrapolation. Mead has done some scaling calculations and continues to be amazed with the phenomenon. He acknowledges, "I still don't understand that."
Mead and Erich Bloch have also suggested the field of biotechnology beginning with Watson's and Crick's discovery of DNA. While there are others that could be examined, some that have been used really miss the point. Take, for example, the following Moore's Law analogy to railways recently offered by the Economist (1996).
"Consider the development of America's railways as an example. In 1830, the industry boasted a mere 37 kilometers (23 miles) of track. Ten years later it had twice as much. Then twice that, and twice again -- every decade for 60 years. At that rate 19th-century train buffs might have predicted that the country would have millions of kilometers of track by 1995. In fact there are fewer than 400 km. Laying rails were too expensive to justify connecting smaller towns; people simply did not need track everywhere. Exponential growth gave way to something more usual -- a leveling off around a stable value at which economic pressures were balanced. . . Americans stopped building railways, but they did not stop becoming more mobile. As rail's S-curve tailed off, Americans took to driving cars and built roads."
Used as an analogue to describe the limitations of Moore's Law, the railroad analogy is limited in its application. Increasing railroad track area (or roads, sea routes, bandwidth, etc.) really deals with implementation or diffusion of technology -- transportation infrastructure in this case -- not technological innovation. Moore's Law is about the pace of innovation (i.e., ideas).
The next section attempts to summarize and draw together the major findings of this examination. In doing so, implications for future research are discussed.
Moore's Law Reconsidered
Beginning as a simple observation of trends in semiconductor device complexity, Moore's Law has become many things. It is an explanatory variable for the qualitative uniqueness of the semiconductor as a base technology. It is now recognized as a benchmark of progress for the entire semiconductor industry. And increasingly it is becoming a metaphor for technological progress on a broader scale. As to explaining the real "causes" of Moore's Law, this examination has just begun. For example, the hypothesis that semiconductor device users' expectations feed back and self-reinforce the attainment of Moore's Law (see Figure 1) is still far from being validated or disproved. There does appear to be support for this notion primarily in the software industry (e.g., "Wintel" de facto architecture). Further research, including survey research and additional interviews, is required to address this possible relationship.
What has been learned from this early investigation is the critical role that process innovations in general, and manufacturing equipment innovations in particular play in providing the technological capability to fabricate smaller and smaller semiconductor devices. The most notable of process innovations was the planar diffusion process in 1959 -- the origin of Moore's Law. Consistent with Thomas Kuhn's (1962) paradigm-shifting view of "scientific revolution," many have described the semiconductor era as a "microelectronics revolution." (Forester 1982, Braun and Macdonald 1982, Gilder 1989, Malone 1996, and others) Indeed, the broad applications and pervasive technological, economic, and social impacts that continue to come forth from "that astonishing microchip" (Economist 1996) seem almost endless. However, this phenomenon has also been aptly described by Bessant and Dickson (1982) as evolutionary, albeit at an exponential rate.
"In a definite technical sense there has been no revolution (save, perhaps, for the invention of the transistor in 1947) but rather a steady evolution since the first invention."
Moore's Law is one measure of the pace of this "steady evolution." Its regularity is daunting. The invention of the transistor, and to a lesser degree the integrated circuit a decade later, represented significant scientific and technological breakthroughs, and are both classic examples of the Schumpeterian view of "creative destruction" effects of innovation. This is evidenced by the literal creation of an entire new semiconductor industry at the expense of the large electronics firms that dominated the preceding vacuum tube technological era. This period of transition from old technology to new technology is characterized by instability, and factors that underpin very irregular performance. This would be considered a shift in the economic and technological paradigm (Dosi 1984, 1988) similar to Constant's (1980) account of the "Turbojet Revolution" where the invention of the turbojet, along with co-evolutionary developments including advancements in airframe design and materials, enabled significant performance improvements in air speed and altitude. The turbojet produced a whole new "jet engine" industry and helped redefine both military and commercial aircraft industries and their users (e.g., airlines). Following the early experimental years of the turbojet, these industries settled in on a new technological trajectory (Dosi 1984, 1988) toward the frontier of the "jet age."
Innovations within the boundary limits of this new frontier occurred at a rapid, but more regular rate. The role of accumulated knowledge -- both tacit and explicit (Freeman 1994) -- and standards (e.g., the role of the Proney brake as the benchmark for performance measurement and testing) are emphasized. Similarly, semiconductor development since the planar process has followed Klein's (1977) description of "fast history," but is more in line with Pavitt's (1986) application of "creative accumulation" (i.e., the new technology builds on the old). The "new" technology in this case is the accumulated incremental -- particularly process-oriented -- advancements indicative of the Moore's Law semiconductor "era." As for standards, indeed Moore's Law itself is used throughout the industry as the benchmark of progress, evidenced most strikingly by the kilo- to mega- to giga-bit density DRAM chips. Increasingly, regular advances in microprocessor performance measures such as MIPS (millions of instructions per second) and MHZ processing speeds follow -- and become part of -- Moore's Law.
Preliminary Conclusions and Future Research
Based on a review of the literature (academic, popular business, and computer trade), an Internet keyword search, and a few personal interviews with major semiconductor players including Gordon Moore and Carver Mead, much has been learned. But no firm conclusions can yet be drawn about what "causes" or what is "caused by" Moore's Law. This examination has revealed that there are two major lines of pursuit from this point. The first is based on the user or "downstream" point of view. This analysis would address the "Wintel" and other "demand-pull" innovation arguments including the expectations feedback hypothesis, but requires more extensive and direct research methods. The second avenue is the from the supplier or "upstream" perspective. Since much of the literature is concerned with process limitations (e.g., is it possible to reach 'Point One') -- reflecting the reality of the industry's everyday challenge -- there is a tendency to examine the "physics" limitations of photolithography and other essential fabrication aspects. At this point it is not clear whether this is just another example of the endless technological pursuit of increasing capabilities and performance similar to earlier advances in turbojet technology. Or is Moore's Law, in Carver Mead's terms, "permission to believe that it will keep going," reinforced by human belief systems? (UVC 1992) Or is it some altogether different variable, yet to be determined?
The answers to these questions are probably all "yes, to some degree." Future research will attempt to better answer these and related questions with more specificity.
Product and Technology Overview
This appendix provides a brief explanation of some of the terminology associated with semiconductor electronic devices, products, and related technologies. A semiconductor is a solid-state electronic device which can be switched to conduct or block electric currents (thus, the term semiconductor). A defining characteristic of the semiconductor since the invention of the solid state transistor almost half a century ago has been the continuous miniaturization of these devices. Because of this, the popular term, "microelectronics" has become synonymous with "semiconductors." Most semiconductors are made with silicon, which has the unique "semiconducting" chemical property, although other materials, such as gallium arsenide, can also be used. Semiconductors can be divided into two main groups: integrated circuits and discrete devices. Integrated circuits (ICs) consist of many active and passive elements that are fabricated and connected together in a single chip. Discrete devices, by contrast, consist of a single switching element such as a diode, rectifier, or transistor.
ICs can be further broken down into digital or linear (analog) devices. Digital ICs store and process information expressed in binary numbers or "bits" (i.e., "1"s or "0"s). They perform arithmetical operations or logical functions by manipulating binary signals (on-off switches of constant voltages). Digital devices form the basis of modern computing and telecommunications technologies. In contrast, linear or analog devices deal with continuous scales in which each point merges imperceptibly into the next. Real-world phenomena, such as heat and pressure, are analog in nature. These devices measure input analog signals, amplify input to output analog signals, or convert analog signals to digital data or vice versa. The largest subcategory of analog devices are special consumer circuits for specific consumer applications such as radio or television receivers.
ICs are, by far, the major and fastest growing semiconductor product. The two most important IC product categories are the microprocessor (and related "micro" products) and memory devices.
A microprocessor is an IC which provides on one chip functions equivalent to those contained in the processing unit or "brains" of a computer. The popular Intel Pentium(tm), used in most IBM-compatible PC systems, is the most recent in a successful lineage of Intel microprocessors which date back to 1971 (Adams, Kash, and Rycroft 1996). A microcontroller is a microprocessor and memory integrated on the same chip -- a computer on a chip -- and is used in dedicated applications such as traffic signals, laser printers, or antilock brakes in automobiles. A Digital Signal Processor (DSP) is a type of microcontroller. Microperipheral devices accompany microprocessors to handle a computer's related function such as graphics. Logic chips handle the mathematical treatment of formal logic by translating AND, OR, and NOT functions into a switching circuit, or gate. The basic logic functions obtained from gate circuits form the foundation of computing machines. This category includes application-specific ICs (ASICs) including gate arrays, standard cells, and programmable logic devices (PLDs). ASICs are semi-custom devices, with a customer connecting standard elements in a customized fashion.
Semiconductor memory devices store information in the form of electrical charges. In terms of sales volume, memories represent the largest single product category. Memory devices can be subdivided into volatile and non-volatile families. The Random-Access Memory (RAM) is a volatile memory product, which means it will lose stored information (charges) once power is turned off. This product sub-category includes the popular DRAM (Dynamic RAM) and, to a lesser degree, the SRAM (Static RAM). In contrast, non-volatile memory products retain information even after power is turned off. They are used in applications requiring repeatedly used information. Non-volatile memory includes the ROM (Read Only Memory) and its "erasable" derivatives, the EPROM (Erasable Programmable ROM) and EEPROM (Electrically EPROM). In an EPROM, stored information is erased by exposure to ultraviolet light, whereas the EEPROM has the convenience of selective erasure of information through electrical impulses rather than exposure to ultraviolet light. "Flash Memory" is an IC which has the ability to bulk erase its entire contents simultaneously. It shares the advantage of other non-volatile memory in that it retains information when power is turned off. Its ability to repeatedly erase and re-program information makes it competitive with DRAMs or disk drives for storing data, although flash is presently more expensive. Flash memory is a rapidly growing market.
Finally, semiconductor devices can also be classified according to the technology used in the fabrication process. Digital devices can be manufactured by two different process variations: metal oxide silicon (MOS) or digital bipolar. Traditionally, digital bipolar devices are faster, but require more power and generate more heat, while MOS products consume less power. Modern MOS processes are making this distinction obsolete, however. Both MOS and digital bipolar manufacturing processes can be used to create logic parts to perform arithmetical operations, as well as memory devices for the storage of data.
Chips are made by creating and interconnecting transistors to form complex electronic systems on a sliver of silicon. The fabrication process is based on a series of steps, called mask layers, in which films of various material -- some sensitive to light -- are placed on the silicon and exposed to light. After these deposition and lithographic procedures, the layers are processed to "etch" the patterns that, when precisely aligned and combined with those on successive layers, produce the transistors and connections. Typically, 200 or more chips are fabricated simultaneously on a thin disk, or wafer, of silicon.
In the first set of mask layers, insulating oxide films are deposited to make the transistors. Then a photosensitive coating, called the photoresist, is spun over these films. The photoresist is exposed with a step and repeat device (i.e., stepper), which is similar to an enlarger used to make photographic prints. Instead of a negative, however, the stepper uses a reticule, or mask, to project a pattern onto the photoresist. After being exposed, the photoresist is developed, which delineates the spaces, known as contact windows, where the different conducting layers interconnect. An etcher then cuts through the oxide film so that electrical contacts to transistors can be made, and the photoresist is removed. More sets of mask layers, based on much the same deposition, lithography and etching steps, create the conducting films of metal or polysilicon needed to link transistors. All told, about 19 mask layers are required to make a chip.
In present-day chips, the physical separation between semiconducting regions is less than 1 micron, and the entire transistor is invisible to the naked eye. Furthermore, the number of impurities must be controlled to within a few parts per billion in some regions of the device. All of these processing steps must be carried out in an environment completely free of particles -- clean-room facilities. Thus, the trend in chip fabrication has consistently been towards greater automation of the processes, which reduces the chances of contamination or human error. (OTA,1986).
Scale of Approximate Sizes
(each step in the scale is a factor of 10)
Audible sound (in air)
micron (micro meter)
DNA coil, virus
small organic molecule
Adams, Richard C., Don E. Kash, and Robert W. Rycroft. 1996. Innovation of the Intel Microprocessor (GMU TIPP Working Paper no. 96:2).
Agres, Ted. 1996. "IC Density Growth Is Key Issue for Industry," R&D Magazine, June, pp. 29-32.
Allen, Ian. 1996. "Chip Power," Online. http://www.abc.net.au/http/pipe/chipower.htm
Angel, David. 1994. Restructuring for Innovation: The Remaking of the U.S. Semiconductor Industry (New York: The Guilford Press).
Arthur, W. Brian. 1994. Increasing Returns and Path Dependency in the Economy (Ann Arbor, MI: University of Michigan Press).
Bajarin, Tim. 1996. "Technology For the 21st Century: Roundtable in Multimedia," Online. http://fantasia.brel.com./acw/news/na049611.html
Barlow, John Perry. 1996. "It's a Poor Man Who Blames His Tools: What Does Technology Threaten? What Is Human?" Wired Online. http://www.hotwired.com.scenarios/workman.html
Bell, C. Gordon. 1991. High-Tech Ventures (Reading, MA: Addison-Wesley).
Bessant, John, and Keith Dickson. 1982. Issues in the Adoption of Microelectronics (London: Frances Pinter).
Bessant, J.R., J.A.E. Bowen, K.E. Dickson, and J. Marsh. 1981. The Impact of Microelectronics: A Review of the Literature (New York: Pica Press).
Brand, Stewart. 1995. "The Physicist" (interview with Nathan Myhrvold), Online. http://www.hotwired.net/wired/3.09/features/myhrvold.html
Braun, Ernest, and Stuart Macdonald. 1982. Revolution in Miniature: The History and Impact of Semiconductor Electronics (Cambridge: Cambridge University Press).
Constant, II, Edward W. 1980. The Origins of the Turbojet Revolution (Baltimore: Johns Hopkins University Press).
Conway, Lynn. 1981. The MPC Adventures: Experiences with the Generation of VLSI Design and Implementation Design and Implementation Methodologies (Palo Alto: Xerox PARC).
Calgary Unix User's Group. 1996. "The End of Moore's Law: Thank God!" Online. http://www.cuug.ab.ca:8001/CUUGer/9605/editorial/html
Davey, Tom. 1996. "Industry Buzzes About 1,400 Mhz, 64-bit Chip," PC Week On Line, July 24.
Davis, Jim. 1996. "Researchers Compete to Speed Up Chips," Online. http://www.news.com/News/Item/0,4,1748,00.html
Dorfman, Nancy S. 1987. Innovation and Market Structure: Lessons from the Computer and Semiconductor Industries (Cambridge: Ballinger).
Dosi, Giovanni. 1988. "Sources, Procedures, and Microeconomic Effects of Innovation," Journal of Economic Literature, XXVI:3, September, pp. 1120-1171.
__________. 1984. Technical Change And Industrial Transformation: The Theory and an Application to the Semiconductor Industry (London: The MacMillan Press).
The Economist. 1996. "That Astonishing Microchip," and "When the Chips Are Down," March 23, pp. 13-14 and pp. 19-21.
__________. 1995. "The End of the Line," July 15, pp. 61-62.
Edwards, Owen. 1995. "ASAP Legends: Gordon Moore," Forbes ASAP, p. 158.
Electronics Industries Association. 1995. The U.S. Consumer Electronics Industry: In Review '95 Edition.
Fallows, James. 1993. "Looking at the Sun," The Atlantic Monthly, November, pp. 69-100.
Foster, Richard N. 1986. Innovation: The Attacker's Advantage (New York: Summit Books).
Freeman, Christopher. 1994. "The Economics of Technical Change," Cambridge Journal of Economics, 18, pp. 463-514.
Gemini C4 Lab. 1996. "Context: Convergence, Multimedia and the Information Highway." Online. http://www.gemconsult.com/internet/contcomp.html
Gilder, George. 1996. "Feasting on the Giant Peach," Forbes ASAP, August 26, 19 pp.
__________. 1995. "The Coming Software Shift: Telecosm," Forbes ASAP, August 28, pp. 147-162.
__________. 1994. "The Bandwidth Tidal Wave," Forbes ASAP, December 5, 1994, 16 pp.
__________. 1993. "Metcalf's Law and Legacy," Forbes ASAP, September 13, pp. 158-166.
__________. 1989. Microcosm: The Quantum Revolution in Economics and Technology (New York: Simon and Schuster).
__________. 1988. "You Ain't Seen Nothing Yet." Forbes, April 4, pp. 88-93.
Hanson, Dirk. 1982 The New Alchemists: Silicon Valley and the Microelectronics Revolution (Boston: Little, Brown and Co.).
Hazewindus, Nico, with John Tooker. 1982. The U.S. Microelectronics Industry: Technical Change, Industry Growth and Social Impact (New York: Pergamon Press).
Hettinga, Robert. 1996. "The Geodesic Network, OpenDoc, and CyberDog," Online. http://www.partmerchant.com/odz/issues/96/03/hettinga.html
Hoeneisen, B., and C.A. Mead. 1972. "Fundamental Limitations in Microelectronics I: MOS Technology," Solid-State Electronics Vol. 15, pp. 819-829.
Howell, Thomas R., William A. Noellert, Janet H. MacLaughlin, and Alan Wm. Wolff. 1988. Microelectronics Race: The Impact of Government Policy on International Competition (Boulder: Westview Press).
Huber, Peter W., and U.S. Department of Justice Antitrust Division. 1987. The Geodesic Network: 1987 Report on Competition in the Telephone Industry (Washington D.C.: U.S. Government Printing Office, January).
Hutcheson, G. Dan and Jerry D. Hutcheson. 1996. "Technology and Economics in the Semiconductor Industry," Scientific American, January, pp. 54-62.
Institute of Electrical and Electronics Engineers. 1995. 1995 University/Government/Industry Microelectronics Symposium (proceedings).
Joch, Alan. 1995. "Heart Throbs: CPU Designers Beg, Borrow, and Steal for Every Ounce of Performance They Can Get," Byte, Online. http://www.byte.com/art/9511/sec7/art1.htm
Johnston, Stuart J., and David Needle. 1996. "The PC Still Matters," Information Week, June 3, pp. 48-50.
Karlgaard, Karl. 1995. "Waste Is Good," Forbes ASAP, October 9, p. 9.
Kash, Don E., and Robert Rycroft. 1996. Technology Policy in a Complex World (unpublished manuscript.
Kerridge, Charles. 1983. Microchip Technology: The Past and the Future (Chichester: John Wiley & Sons).
Kilpatrick, Jr., Henry E. 1995. Competition Policy, Trade Policy and Learning By Doing in the Japanese Semiconductor Industry (unpublished paper).
Klein, Burton H. 1984. Prices, Wages and Business Cycles (New York: Pergamon Press).
__________. 1977. Dynamic Economics (Cambridge: Harvard University Press).
Kline, Stephen J., and Nathan Rosenberg. 1986. "An Overview of Innovation," in Ralph Landau and Nathan Rosenberg, eds., The Positive Sum Strategy (Washington, DC: National Academy Press), pp. 275-305.
Koprowski, Gene. 1996. "The Next Big Wave: Marketing Tools," Online. http://www.marketingtools.com/mt_current/mt416.htm
Kuhn, Thomas S. 1962. The Structure of Scientific Revolutions (Chicago: University of Chicago Press).
Lenzner, Robert. 1995. "The Reluctant Entrepreneur," and "Whither Moore's Law?" Forbes, September 11, pp. 162-168.
Lazowska, Edward. 1996. "20th Annual Faculty Lecture." University of Washington, Department of Computer Science and Engineering, Online. http://cs.washington.com/faculty.lecture/index.html
Levitt, Theodore. 1960. "Marketing Myopia," Harvard Business Review, July-August, pp. 45-56.
Malone, Michael S. 1996. "Chips Triumphant," Forbes ASAP, February 26, pp. 53-82.
McCormick, Jim. 1995. "A Brief History of Silicon Valley." Online. http:www.silvaonline.com/silhist.html
Mead, Carver A. 1994 "Scaling of MOS Technology to Submicrometer Feature Sizes," Journal of VLSI Signal Processing, 8, pp. 9-25.
__________. 1985. Mead's Four Laws of the Economics of Innovation (brief).
Moore, Gordon E. 1996. "Some Personal Perspectives on Research in the Semiconductor Industry," in Rosenbloom, Richard S., and William J. Spencer (Eds.). Engines of Innovation (Boston: Harvard Business School Press), pp. 165-174.
__________. 1995. "Lithography and the Future of Moore's Law." Paper presented to the Microlithography Symposium, February 20.
__________. 1965. "Cramming More Components Onto Integrated Circuits," Electronics (Volume 38, Number 8), April 19, pp. 114-117.
Murphy, Michael. 1996. "Betting on Moore's Law," Wired, June, p. 88.
Norton, R.D. 1996. The Westward Rebirth of American Computing (unpublished paper).
Noyce, Robert N. 1977. Microelectronics (San Francisco: W.H. Freeman and Scientific American).
Odlyzko, Andrew. 1995. "The Decline of Unfettered Research." Online. http://www.math.washington.edu/Special/comment/science.html, October 4.
Organization for Economic Co-operation and Development (OECD). 1985. The Semiconductor Industry: Trade Related Issues (Paris: OCDE).
Parkinson, G. Northcote. 1957. Parkinson's Law and Other Studies in Administration (Boston: Houghton Mifflin).
Pavitt, Keith. 1986. "'Chips' and 'Trajectories': How Does the Semiconductor Influence the Sources and Directions of Technical Change?" in Roy M. MacLeod, ed., Technology and the Human Prospect (London: Frances Pinter).
Pennings, Johannes M. and Arend Buitendam (Eds.). 1987. New Technology As Organizational Innovation: The Development and Diffusion of Microelectronics (Cambridge: Ballinger).
Popper, Karl R. 1986. Objective Knowledge: An Evolutionary Approach (Oxford: Clarendon Press).
Rayner, Bruce E. 1996. "Can Moore's Law Continue Indefinitely?" Computerworld, Online. http://www.computerworld.com/...722LEADSL9607lead.html, (excerpts from a lecture by and interview with Gordon E. Moore), 4 pp.
Robertson Stephens & Co. 1996. "Highlights of the Robertson Stephens Semiconductor Conference." August 1, 16 pp.
Ross, Philip E. 1995. "Moore's Second Law," Forbes, March 25, pp. 116-117.
Rothschild, Michael. 1990. Bionomics: The Inevitability of Capitalism (New York: Henry Holt).
Semiconductor Industry Association. 1996. Academics Information Package 1996.
__________. 1994. The National Technology Roadmap for Semiconductors.
__________ (Thomas R. Howell, Brent L. Bartlett, and Warren Davis). 1992. Creating Advantage: Semiconductors and Government Industrial Policy in the 1990s (San Jose: Dewey Ballantine).
Siewiorek, Daniel P., C. Gordon Bell, and Allen Newell. 1982. Computer Structures: Principles and Examples (New York: McGraw-Hill).
SpinDoctor (a). 1996. "The Coming Software Economy." Online. http://www.spindoczine.com/dose/3.3.html
__________ (b). 1996. "Daily Dose -- The Next Big Thing 3: Fry Pilots." Online. http://www.spindoczine.com/dose/3.3.html
Steinmueller, William E. 1987. Microeconomics and Microelectronics: Economic Studies of Integrated Circuit Technology (doctoral dissertation), Stanford University.
Stix, Gary. 1995. "Toward 'Point One'," Scientific American, February, pp. 90-95.
Sydney Morning Herald. 1995. "Light Years Ahead," Online. http://www.smh.com.au/lead/lead-951205.html
Time-Life Books. 1990. The Chipmakers (Alexandria, VA: Time-Life Books).
Toda, Bobby. 1996. "Papken S. Der Torossian Speaks at the May Meeting," Online. http://www.3wc.com.aama.papkenspeaks.html (Asian American Manufacturers Association).
Turton, Richard. 1995. The Quantum Dot (New York: Oxford University Press).
Tyson, Laura D'Andrea. 1992. Who's Bashing Whom? Trade Conflict in High-Technology Industries (Washington, DC: Institute for International Economics).
University Video Communications. 1996. Gordon Moore: Nanometers and Gigabucks [videotape] (Stanford, CA: UVC).
__________. 1993. Tracking the Teraflop [video recording] (Stanford, CA: UVC).
__________. 1992. How Things Really Work: Two Inventors on Innovation, Gordon Bell and Carver Mead [video recording] (Stanford, CA: UVC).
U.S. Congress, Office of Technology Assessment. 1986. Microelectronics Research and Development -- A Background Paper, OTA-BP-CIT-40 (Washington, DC: U.S. Government Printing Office, March).
U.S. Department of Commerce, International Trade Administration. 1985. A Competitive Assessment of the U.S. Semiconductor Manufacturing Equipment Industry (Washington DC: U.S. Government Printing Office).
Vvedensky, Dimitri. 1996. "Size Is Everything." Online. http://go2.guardian.com/...960606coonoffline.html
von Hippel, Eric. 1988. The Sources of Innovation (New York: Oxford University Press).
Waldrop, M. Mitchell. 1992. Complexity: The Emerging Science at the Edge of Order and Chaos (New York: Simon and Schuster).