About WTS Prof. Clair Brown Faculty, Students and Affiliates Research Areas Online Research Reports Working Papers


Second Interim Report
Clair Brown, Editor

10. How does Knowledge Flow? Inter-Firm Patterns in the Semiconductor Industry
Melissa M. Appleyard

10.2 Why Would Rivals Share?

This section provides a framework for thinking about the costs and benefits derived from knowledge-sharing. When discussing inter-firm knowledge flows, I prefer the broad concept of knowledge-sharing, which I define as the transfer of useful know-how or information across company lines. Figure 10-1 depicts the knowledge-sharing decision faced by a firm. If the firm decides to share a piece of technical knowledge, it can do so publicly or privately, and either place legal restrictions on its use or permit unrestricted use. If the firm determines that the gains to exclusive use of the technical information outweigh the gains to sharing, then the firm can resort to secrecy.

The primary mechanisms of inter-firm knowledge exchange examined in this paper are arranged in Figure 10-2 in terms of access to and use of the shared knowledge. Access to knowledge can occur either through public channels : patents, reverse engineering, newsletters, popular press, trade journals, and conference presentations; or through private channels: e-mail, the telephone, face-to-face meetings, visits to other companies' fabrication plants (fabs), consortia or benchmarking studies. Even if access to knowledge is public, its use may be restricted by legal constructs such as patents and non-disclosure agreements.

In constructing Figure 10-2, I assume that the use and dissemination of knowledge obtained by visiting other companies' fabs or participating in a consortium or benchmarking study face legal restrictions such as non-disclosure agreements to a greater degree than "informal" interactions via email, the telephone or face-to-face conversations. Private-unrestricted knowledge-sharing is similar to private-restricted in that costs are incurred to search for appropriate partners and forge relationships, but the two modes differ in important ways. First, there is limited legal recourse if the private-unrestricted sharing relationship faces a problem such as opportunism. But the minimal transactions costs associated with private-unrestricted transfers compared to the costs of drafting a formal agreement, may more than offset the costs of intermittent opportunism.

From the viewpoint of the firm, the decision whether or not to share knowledge with another company depends on whether the expected benefits from relinquishing the monopoly over the knowledge outweigh the expected costs. That is, a unique piece of useful knowledge confers monopoly rents upon the possessor by allowing her to lower her costs of production relative to others in the industry. When she divulges the information to a competitor, she gives up her monopoly over the knowledge, and must share the rents, suffering "'competitive backlash'" (Carter, p.156). As long as she expects her sharing partner to reciprocate with knowledge that is as least as useful or some other form of compensation, such as a licensing fee, she can justify inter-firm disclosure.

For firm i at time t, the cost and benefit comparison can be represented by the following,

Eit [B9L,R,A0] >= Eit [C(D,T)]

The firm's expected benefit from sharing the piece of knowledge, EitB, is an increasing function of the following: revenue from licensing, L, legal rights to use the recipient company's technology, R, and/or knowledge from the recipient company, A. Firm i's expected cost of knowledge-sharing at time t, EitC, increases with the decline in profitability, D, and the transaction costs, T, associated with the knowledge transfer.

Drawing on previous empirical studies, the cases listed in Table 10-1 capture the potential net benefits from knowledge-sharing. When the knowledge-owner patents and then licenses the knowledge to a market rival, Case III, L increases while the sharer's cost advantage may fall, since the rival can legally use the knowledge. Depending on pricing conditions, a fall in the sharer's cost advantage may translate into a decline in profitability, an increase in D, which will at least partially offset the benefits of L. A similar outcome obtains when the knowledge-owner patents and then cross-licenses the knowledge, Case IV, but instead of licensing revenue the sharer receives rights of use, R, or, as many people in the semiconductor industry claim, a promise not to be sued. Finally, in the know-how trading case, Case V, the knowledge-owner benefits from acquiring new knowledge, A, but must give up the monopoly rents associated with the knowledge that it shares. As for the transaction costs associated with the three cases, the cost of undertaking know-how trading is, in general, much lower than the other two especially in the presence of legal fees charged to draft and maintain licensing agreements.

10.3 Cross-Industry Comparison of Knowledge-Sharing: Semiconductors vs. Steel

To explain why different magnitudes of private knowledge-sharing occur across industries, one needs to look more closely at the process of innovation. The examination of both the pace and nature of technological change and how it translates into increased performance in different industries can illuminate why distinctive patterns coexist. I hypothesize that if an industry experiences a rapid pace of technological change, reflected by the average time between new product or process introductions, private knowledge-sharing in that industry will be less likely.

Previous studies of knowledge-sharing behavior have focused largely on manufacturing intensive industries with a slow pace of technological change, in particular, the steel industry. However, when considering the net benefit of knowledge-sharing in a research and development intensive industry, such as the semiconductor industry, the net benefits may exhibit much higher variances. In the semiconductor industry, uncertainty surrounds the payoff to a particular piece of knowledge due to difficulties in predicting: its useful life; the breadth of its applicability across the industry; the ease with which it an be incorporated into another company's process flow; or whether it can be reverse engineered. Imperfect information may lead a knowledge-owner to engage in knowledge-sharing when, in fact, the ex post net benefit is negative. For example, if firm i licenses a processing technique to a rival, it earns L. But if the rival incorporates the technique much more quickly into its production process than firm i had anticipated, firm i may face a steeper decline in its profits, D, than initially anticipated.

In support of the above hypothesis, von Hippel (1988) finds ample evidence that private-unrestricted know-how trading occurs in the U.S. steel minimill industry, an industry with a slow pace of technological change. In contrast, he finds such activity almost non-existent among powdered metals fabricators and producers of the biological enzyme klenow, which are industries that exhibit more rapid technological change (p.83). As reported below, I have found evidence that knowledge is shared on private channels linking semiconductor engineers, but not to the degree reported by the studies of the steel industry.

Inventive activity in a fast paced industry, such as the semiconductor industry, will occur in the top range of Table 10-1, with the opportunity to patent new ideas much more frequent. In slower paced industries, such as the steel industry, technological change is characterized by the accumulation of numerous incremental improvements over a long time horizon, and so patent opportunities are fewer. Patenting benefits a firm by establishing an enforceable claim to rents that accrue to an innovation. Patents also give the patent owners defined "bargaining chips" for technology swapping, e.g., cross-licensing, represented by R in the expected benefits function above. Also, if an industry exhibits rapid technological change, it would behoove firms in that industry to share little with their competitors so that they can maximize the size of the "technical cushion" separating them from their competitors; thus insulating themselves from a marked decline in profits, captured by a rise in D in the expected cost function, if they were to engage in "real-time," private and unrestricted knowledge-sharing.

Apparently refuting the hypothesis, though, is von Hippel's finding that private trading is present in the U.S. aerospace industry, a research intensive industry (von Hippel, p.83). Two reasons may explain this finding: (1) much of the research is "basic," and thus falls in the bottom category of Table 10-1; or (2) the nature of demand, which is primarily fueled by U.S. government procurement. As von Hippel observes, the aerospace companies maintain frequent contact until government contracts come up for bid. After contracts are granted, informal knowledge-sharing resumes (von Hippel, p.87). That is, when significant profits are at stake, private knowledge exchanges taper off, but recommence after the spike in demand dissipates.

The following statistics help to characterize the semiconductor industry as knowledge intensive and rapidly evolving when compared to the steel minimill industry.
1. The average number of patents/year/firm—Table 10-2. This chart illustrates the distinctive patterns of knowledge creation in the semiconductor industry versus the steel industry. For example, in 1992, Nucor received less than five patents whereas LSI Logic received twenty. This difference is striking given the relative size of the two companies: In 1992, LSI employed 20% fewer workers than Nucor, with total sales less than half of Nucor's.
2. R&D expenditures as a percentage of total sales—Figure 10-3. Research, development and integration consumes a large portion of the total costs associated with a new semiconductor process. R&D expenditures consistently run over 10% of sales in many integrated circuit (IC) companies in the industry. This emphasis translates into a higher incidence of patenting and a higher degree of secrecy in the semiconductor industry relative to the steel industry.
3. Rate of price decline—Figure 10-4. When the 16 megabit DRAM chips were introduced in late 1991, they ran nearly $300 per chip. By mid-1994, their price had plummeted to just under $50. Their current price is under $20 (The Economist, p.66). Since the penalty of late entry into a product market is extremely high, semiconductor developers often are reluctant to provide specific technical information to their peers at competing firms.

Other industry characteristics, such as the nature of product markets would also assist in distinguishing the two industries. In a few major semiconductor product markets, such as microprocessors, where standards are based on firm-specific architectures, one would predict a higher level of secrecy. This is consistent with Rogers' findings in his study of microprocessor producers in Silicon Valley (Rogers, 1982).

These characteristics of the semiconductor industry—active patenting, large R&D expenditures, rapid price declines—reflect the frequent introduction of new production processes and products. Given this high frequency, the hypothesis proposed above would suggest that private knowledge-sharing would be less likely in the semiconductor industry than in the steel industry. Section 10.5 provides preliminary findings that support such a conclusion.

Go to Next Section of this Chapter
Go to Table of Contents for this Chapter
Go to Table of Contents for the CSM-HR Interim Report

© 2005 Institute for Research on Labor and Employment. 
2521 Channing Way # 5555 
Berkeley, CA 94720-5555