Our second major empirical study concerns the population of firms that manufactured and sold semiconductor electronics devices in the United States between 1946 and 1984. The term “semiconductor” refers to the electrical properties of the materials from which microelectronic devices are made. Silicon is the most commonly used semiconducting material. Slices of silicon crystals are coated with other materials that have varying conducting and insulating properties. There are two major categories of products based on this structure. Discrete devices include transistors, diodes, and rectifiers. Integrated circuits, the newer branch of the technology and the one with the largest market, includes analogue devices (used heavily in telecommunications) and logic, microprocessor, and memory devices. These products are themselves components in larger electronic systems of all sizes, such as computers, consumer electronic products, communications switching and transmission equipment, and weapons.
The integral role of semiconductors in other advanced technologies makes the study of this industry interesting for policy reasons. Our interest concerns the ecological dynamics of the industry, including the effects of the environments within which the population operates. The relevant envi- ronments are the markets for semiconductor products and the technology itself.
We study only the firms participating in the merchant market, firms that sell semiconductor products to other firms. This restriction excludes some of the largest semiconductor manufacturers in the world, those that sell only to other branches of the same company and not on the open market. This is often called captive production. An example is IBM, which is probably the largest semiconductor producer in the world. In 1979, it was estimated that 30 percent of the world’s semiconductors were produced by captives (Wilson, Ashton, and Egan 1980). Large corporations design and manufacture their own semiconductor components in order to capture the profits generated by the high value added by manufacture, to ensure a supply of these crucial components, and to ensure that research and development are devoted to applications having the highest payoff for the development of their products. In such cases, the capital allocation, research and development, and production scheduling of the semiconductor division are all subordinated to the requirements of other corporate divisions. Further, the larger corporation buffers the semiconductor division from much of the uncertainty stemming from the market and provides a cushion that lowers the uncertainty generated by change in the technology. Specialization, with appropriate planned lead times for development, protects captive producers against the vagaries of having to design semiconductor devices representing the state of the art. For example, a firm that produces a microprocessor to be used in its own industrial process control system might settle for one that is not the fastest microprocessor available if it performs certain specialized functions better than a generic device available on the merchant market.
Merchant and captive producers are, therefore, quite different kinds of organizations, and there is good reason to expect them to exhibit different life-event dynamics. Rates of entry and exit for captive producers should be functions of the fortunes of their parent companies. Merchant market producers, even those owned by bigger corporations, are likely to be affected more directly by demand for semiconductors. Uncertainty stemming from the technology should affect them more directly as well, because unforeseen changes in technology will be underwritten by the parent and the divisions using semiconductor output.
Data are available on merchant market producers, as we describe below. For captive producers, however, data are difficult to find. When the captives start and cease operations, which products they manufacture, and how they are organized are facts not usually reported and often treated as trade secrets. Consequently, captive producers are extremely difficult to study. So for both practical and theoretical reasons, we confine ourselves to merchant market producers.
Our analysis of the population of semiconductor firms is confined to American firms and operations. One reason for this limitation is availability of data: information on the existence of the firms and on the products they produce is available for the American market but not for the rest of the world. Furthermore, foreign producers operate in quite different financial and regulatory environments. They are included in our study only when they have a subsidiary in the United States.
Semiconductor markets are famous for two qualities: an exponential pattern of growth in sales and a magnified business cycle. Although other semiconductor devices predated transistors, the transistor was the product that began the trend toward miniaturization in electronics. When the transistor was invented in 1947, germanium was used. Germanium transistors could not tolerate the variations in temperature and humidity that most applications would require, and consequently the market for transistors in the early years was confined to specialized products such as hearing aids. The introduction of silicon transistors in 1954 had two important consequences. The first was great expansion in the variety of applications, resulting from better tolerance of fluctuations in temperature and humidity. Second, batch manufacturing techniques were more easily applied to silicon. Batch production not only increased the efficiency of manufacture, it also lowered the risk. Before the development of the most important production process, the planar process, product design was linked to production equipment design. Obsolescence in product design required replacement of production equipment, at great cost.
As the technology advanced, the market expanded. Although prices fell both in absolute terms for a particular product and in cost per function performed, total sales grew at a breathtaking rate. But the business cycle in the semiconductor industry exaggerates the cycle in the economy at large. As indicated earlier, semiconductor devices are generally expensive components of broader systems. As a result of the rapid pace of technological change in the industry, products become obsolete quickly. Consequently, holding large inventories of semiconductor components is both expensive and risky for the system manufacturer; by the time they are needed, alternative designs may be more desirable. When the economy goes into recession, buyers of semiconductor devices look for ways to reduce expenditures and cut their orders for semiconductors. But in order to shrink bloated inventories, they cut orders by a greater percentage than the decline in their own orders. When orders for systems pick up, the process works in reverse. Semiconductors are often in short supply, and when orders for systems begin to grow, a shortage in these critical components can stifle recovery. So customers rebuild inventory, increasing orders by amounts greater than the expansion of orders for systems. Thus the market for semiconductors is characterized by erratic change in the context of long-term exponential growth.
A similar pattern of change characterizes the technology itself because uncertainties stemming from technical change exacerbate the market un- certainties just described. The long-run trend in this technology is toward increasing miniaturization and functional capability. Over time, semiconductor devices have become increasingly complex and able to perform more complicated combinations of functions, to the point that single chips now include all of the capabilities of a computer. In 1965 Gordon Moore, the president of Intel, asserted that the functional density of integrated circuits would double every year. Moore’s Law, as this prediction is called, was remarkably accurate. Technical change involves secular trends, not just inventions occurring randomly in time. Those who follow the technology closely often know what the next major innovation will do. The question is when it will occur, which firm will introduce it first, and what the details will be. So the technology changes in such a way that the functional capabilities of semiconductor devices increase exponentially with time but erratically as well. The pattern of change in the technology exacerbates the change in the market because extensive research and development requires organizational structures that are costly to assemble and disassemble. When cash flow shrinks during slack markets, cutbacks in development threaten a firm’s return to prosperity later. Falling behind in such a rapidly developing technology guarantees failure. So the pressures from market cycles run counter to the pressures from changing technology.
These dual patterns of change in technology and market create opportu- nities for entrepreneurs (Brittain and Freeman 1980). They also create risks because the time periods required to perform various organizationbuilding tasks do not map perfectly onto periods in which resources are available. The dramatically expanding market offers promises of rewards that compensate for taking those risks. Hence the semiconductor business has been a highly volatile one, and it constitutes a particularly interesting laboratory for population studies of organizations.
1. Unit of Analysis
We argued earlier in this chapter that beginning and ending processes are actually composed of subprocesses that occur at varying speeds and in varying orders across organizational forms. One must choose to focus attention on one or the other of these processes as the basis for deciding when life events of interest have occurred. For one kind of organization it may be the legal existence, with filing papers of incorporation and filing under Chapter 13 of the Bankruptcy Act defining the important life events. For another, it may be the start and end of operations. This is the subprocess of interest here because marketing semiconductor products defines the existence of the organization as a semiconductor manufacturer. Initiation procedures, such as seeking the participation of potential founding team members, or resource-gathering activities such as seeking venture capital, are often carried out for companies that never actually are formed. The development of an organization with a division of labor and authority relations is difficult to observe for organizations such as these by methods other than direct observation. Direct observation would confine the number of observations in such a way as to vitiate population-level analysis. Consequently, it seems to make sense to record the time when an organization begins to produce and sell semiconductor devices as the beginning of the organization. Nevertheless, to avoid confusion we refer to “entries” and “exits” in the remainder of this book when discussing semiconductor firms rather than to “foundings” and “disbandings.”
The unit of analysis for the semiconductor study is the part of the organi-zation producing and selling semiconductor devices. It may be a privately held firm, a complete corporation, or a set of corporate divisions. The time of starting for semiconductor manufacturers is defined by market participation. Manufacturing organizations may be bought out by other corporations, but we treat them as intact so long as they market products under their own name. We treat companies that stop making and selling semiconductor devices as ending, even though they may not go through bankruptcy. Employees may come to work and produce other things, and this would be a serious problem were it not for the rather demanding requirements for manufacturing semiconductor devices. One does not blithely acquire the necessary equipment, train the operators, design the products, and commence manufacturing. An organization abandoning all of this to take up some other line of business almost necessarily undergoes a transformation of such magnitude that it would make little sense to call it the same organization even if its name remains the same or its legal ownership is unchanged.
2. Sources of Data
Our key source of information on semiconductor manufacturing companies is the Electronics Buyer s Guide. This is a standard industry source book for purchasing agents and others interested in buying commodity electronic devices, including semiconductor devices. We obtained copies of this source for each year from 1946 to 1984 and coded a three-dimensional data array: firms by devices by years. This produced 6,856 observations of firms in various years. The entry date was defined as the year of entry of each firm in the Guide. There were 1,197 entries. The terminal date was the last year the firm appeared. When a firm was absent for more than one year and then reappeared, it was assumed that a new firm had begun, even if it had the same name as a firm that disappeared. A firm that appeared in only one year but then disappeared for several years was coded as starting at the beginning of the extended string. The single observation was dropped because we assume that the listing in the Guide was in error. There were 302 semiconductor firms still in existence in 1984. Not counting these right-censored cases, there were 895 exits from the industry, our operational definition of mortality for semiconductor firms.
Some large corporations operate multiple divisions that produce and sell semiconductor devices. Sometimes product lines and their production facilities are shifted from one division to another. General Electric, for example, began selling standardized semiconductor products through a sales office in 1965, and GE stopped reporting the names of the corporate divisions actually producing those devices. So GE’s Rectifier Components, Receiving Tubes, and Semiconductor Products divisions all disappeared from the Guide in 1964. Obviously, one would not want to treat these events as exits from the industry. So we aggregated divisional data to the corporate level. The starting date for the organization is then the first year when any of its divisions markets a semiconductor device. In the case of GE this is 1951, when it appeared in the Guide selling discrete devices through its Apparatus Sales division. The exit date attached to any particular firm is the first year in which none of the divisions is still in the semiconductor business. When another company absorbs a free-standing semiconductor firm, a judgment has to be made. If the newly acquired division operates as a subsidiary and maintains its own name, then we do not aggregate the division into the parent company’s observation. However, we treat it as an ending event because the independent company becomes a subsidiary of a bigger firm. Thus when Schlumberger acquired Fairchild, we did not backdate Schlumberger’s observation to the year Fairchild entered the business. On the other hand, if a company gets into the business by acquiring the assets of an existing company but does not sell products under the old company’s name, we use the date when it starts selling semiconductors under its own name as the date it entered.
The Guide listed 85 separate product categories over the 39-year period. Since the purpose of the Guide is to indicate sources of supply, data on the size, organization, and financial performance of firms are not included. We used industry sources, principally data provided by the market research firm Dataquest, to measure the volume of sales in various product families.
We were able to obtain worldwide and North American sales data for nine aggregate families from Dataquest, and from other sources for the earliest years. These families are as follows:
A. Discrete Semiconductor Components
- Diodes and rectifiers
- Transistors and thyristors
B. Integrated Circuits
- Digital integrated circuits
- Analog/linear integrated circuits
- Signal converting integrated circuits
C. Custom and Semi-Custom Devices
- Custom integrated circuits
- Hybrid circuits
D. Optoelectronic Devices
- Light-emitting devices
- Photo-sensitive devices
In the case of the appearance of new products in the Guide, a decision was made to include them within one of the existing product families, or to add to the list of product families. We then matched the sales data to firms so that we could measure characteristics of the markets in which they did business with the dates of their entry or exit.
2. Lack of Precision on Dates
The data just described tell the year in which a firm appeared in the market, the last year in which it was observed, and the years in which various products were sold. We do not know the month or day during the year on which these events occurred. Large numbers of ties in the time spell between entry and exit result. We dealt with this in a number of ways, as we describe in the chapters in Part III.
3. Organizational Forms
One can view the history of the American semiconductor industry as a contest among organizational forms. Some of the first entrants were specialized entrepreneurial ventures; others were established manufacturers of integrated electrical equipment. American Telephone and Telegraph licensed its transistor invention at least partly because antitrust proceedings against it led to a promise not to develop the commercial possibilities. Giants such as RCA and General Electric as well as small companies such as the oil exploration equipment manufacturer Texas Instruments sent engineers to the meetings that laid the groundwork for licensing the technology.
Initially the specialists flourished while the integrated giants backed off. The giants seemed unable to cope with the uncertainties that such an immature technology imposed. In the 1980s foreign firms, especially those from Japan and Korea, made great progress in penetrating the world market, particularly in the mass manufacture of the memory chips so essential to the computer industry. These foreign firms were almost all integrated manufacturers and were active in many of the same industries as were the American giants previously forced to surrender technological leadership in semiconductors. Companies like RCA and Westinghouse had made initial efforts in the industry but lost out competitively to companies like Texas Instruments and Fairchild Semiconductor in the 1960s. The very diversity of the larger companies made it difficult to tolerate the uncertainty of a rapidly changing technology with distant and vague profit potential (Brittain and Freeman 1980). Our data do permit us to examine whether or not the semiconductor firm under study is a division of a larger organization. When it is, we call it a subsidiary; when it is not, we call it independent. The variable SUBSIDIARY assumes a value of one when the firm in question is a subsidiary and zero when it is independent.
In general, we believe subsidiaries are better protected from life-threatening disruption stemming from resource scarcity or temporal variations in resources. Because they are parts of bigger corporations that are usually less driven by technology than semiconductor firms, we expect subsidiaries to be more bureaucratic. Fortunately, we have detailed data on a subsample of semiconductor firms which we can use to describe the difference between these two forms in more detail.
In 1984, we conducted interviews in 47 of the 56 merchant market semi- conductor firms in the area south of San Francisco commonly called Silicon Valley. We gathered data on the structures, business strategies, technologies, and histories of these firms. The respondent was usually the chief executive, but sometimes a subordinate was interviewed. For the larger firms, multiple interviews were conducted with managers responsible for such areas as research and development and manufacturing. The data we collected are summarized in Table 7.1.
Our purpose here is descriptive; we test no hypotheses. Therefore, Table 7.1 does not report tests of significance. We define a subsidiary here as a firm for which, according to our data, more than 50 percent of the equity is held by some parent corporation. Obviously, retrospective data cannot be used to test theories about organizational mortality, and even such descriptive exercises as this must be interpreted with caution because the population in 1984 is a population of survivors.
Some of the variables reported in Table 7.1 are continuously distributed, while others are categorical. Some of the continuously distributed variables are highly skewed. When the absolute value of skewness is greater than 1 for both subsamples being compared, we report medians rather than means. When the size of the combined subsamples is less than 47, there are missing values in the data.
The youth of both kinds of firms is striking; but the subsidiaries are slightly younger than the independents. This should not be too surprising given that we predict a liability of newness in the exiting rate of semicon- ductor firms, and one of the major ways of exiting is to be acquired by a bigger firm.
There are also size differences between the two forms. The independents are approximately three times larger as measured by either sales or number of employees. Again, the smallness of the average firm is striking. The largest of them has sales of well over a billion dollars per year, but the averages are less than ten million dollars.
The subsidiaries sell to a somewhat more concentrated market, and they devote fewer of their efforts to the production and sales of commodity chips. A common business strategy for new firms is serving as a second source of devices invented by other firms. This turned out not to be any more prevalent among the independents than among the subsidiaries; in fact, it is slightly less common among the independents.
The independents do more of their own wafer fabrication than do the subsidiaries. Integrated circuits are made from disks of silicon. Each disk, or wafer, has hundreds of devices on it at the end of the process. After testing, these are cut from the wafer chip by chip. Wafer fabrication is the heart of the manufacturing process and involves the most expensive equipment. We should note that the unit of observation here is the firm or establishment in Silicon Valley. If the same parent has other subsidiaries located elsewhere, wafer fabrication may be located there. It is not necessarily true that the corporations of which the subsidiaries are a part are less likely to do their own wafer fabrication. Indeed, we would be surprised if this were so.
A number of categorical variables seem to support our belief that the subsidiaries are likely to be more bureaucratic than the independents. They are more likely to have produced an organization chart, to have a functional structure, to have a formal profit-sharing plan, and to have written a personnel manual. They also use written job descriptions more often than the independents.
Finally, the independents are more likely to market locally or regionally, whereas the subsidiaries are more likely to market on a worldwide basis. This reflects the greater financial muscle of the latter and allows them to spread out the risks of local booms and busts.
A second difference in form is the level of specialization. Some manu- facturers attempt to produce the full variety of semiconductors existing as of the time in question, while others have chosen to work only within one or two branches of the industry. Specialization allows scale within narrow niches. It suggests risks, however, since the vagaries of a rapidly developing technology may compound business risk with technology risk.
4. Technical Change and Industry Characteristics
Our interest in the ecology of semiconductor firms centers on the patterns of change in the technology and changes in the market. We rely on two sources of information on change in the technology. First, we examine dates of introduction of various technical innovations. In particular, we examine difference in foundings and failure rates over periods of time:
- PI (Early): We begin the study of semiconductor firms with the year For the first few years production was confined to various discrete devices such as metallic rectifiers that preceded the invention of the transistor. The period ends in 1959.
- P2 (Middle): The middle period begins in 1960 with the introduction of the integrated circuit. In the next year, 1961, the planar process was In fact, the early 1960s were a very rich period for technical innovation in this industry; oxide masking, diffusion and epitaxial techniques, and planar techniques were all introduced and broadly disseminated (Tilton 1971, pp. 66-67). The period ends in 1969.
- P3 (Recent): The 1970s saw a surge of developments that made rela- tively inexpensive computing a In 1970 the first dynamic random access memory chip was introduced by Intel and Advanced Memory Systems. The following year the microprocessor was intro- duced by Intel. So the third period begins in 1970 and continues to 1984.
It is obvious that one could define a period for every year, since both large and small innovations have come almost continuously. Also, of course, the innovations do not diffuse instantly. The dates are, therefore, only approximations.
Our data on markets come from industry sources such as the Semicon- ductor Industry Association’s annual reports, as well as various reports supplied by the market research firm, Dataquest. We used these sources to develop the list of product groups presented earlier. We note the date of introduction of the various products in the Electronics Buyer’s Guide and the speed with which firms enter new product markets to measure their innovativeness. We check our dates against various industry sources, in- cluding those produced by the Semiconductor Industry Association as well as reports such as those by Tilton (1971), Braun and MacDonald (1978), and Wilson, Ashton, and Egan (1980).
Finally, we refer to a variety of government reports to deflate data on sales using the Consumer Price Index, the Industrial Producer Price Index (Bureau of Labor Statistics), and the interest rate for Triple A Rated Corporate Bonds (Federal Reserve Bulletin). Variables from these reports were coded as yearly time series. They are, then, constant across firms in a given year.
Source: Hannan Michael T., Freeman John (1993), Organizational Ecology, Harvard University Press; Reprint edition.