THE COMPETITIVE SEMICONDUCTOR MANUFACTURING HUMAN RESOURCES PROJECT:

Second Interim Report
CSM-32
Clair Brown, Editor

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

10.4 Knowledge-Sharing in Different Institutional Contexts: Japan vs. the U.S.

The calculation of the net expected benefits to knowledge-sharing not only depends on an industry's characteristics, but also on the institutional environment in which it is taking place. The previous section provided motivation for why inter-firm knowledge-sharing may occur in some industries but not in others. This section extends the analysis by examining the following question: Do knowledge-sharing patterns in a given industry differ across countries? Through a case study of the semiconductor industry, I address this question by comparing the U.S. and Japan. Two institutions, the intellectual property right (IP) regime and the employment system, differ markedly in the two countries and influence the net benefit calculation described in Section 10.2. Considering the features of those two institutions, I hypothesize that workers in the U.S. semiconductor industry are more likely to rely on private channels of communication than their Japanese counterparts. The findings presented in the next section support this hypothesis in general, although cross-country differences do not prove to be statistically significant.

Among many differences in the IP regimes in the U.S. and Japan, their patenting systems are quite distinct. First, Japan follows a first-to-file system, whereas the U.S. is one of the few countries that follows a first-to-invent doctrine. Second, the patent breadth in Japan is, in general, narrower, where the "patenting of individual applications" is possible (Katz and Ordover, p.180 fn. 72). In contrast, the U.S. patent authorities generally require a broader range of application. Finally, in Japan, the system of Kokai, or "laying open," requires that all patent applications be published in the patent gazette by 18 months from the date of filing (Helfand, p.186). In the U.S., only patents that issue are made accessible to the public.

When considering the variables influencing the net benefit to knowledge-sharing discussed above, Japan's patent system would likely discourage private, inter-firm knowledge-sharing to a greater degree than the U.S. system. For example, semiconductor engineers in Japan may be reluctant to discuss ideas in early stages with peers at other companies for fear that they may turn around and try to patent the idea. If the originator of the idea is beaten to the patent office, the originator loses out on the expected benefits to licensing or cross-licensing, L and R respectively. Under the first-to-invent system in the U.S., however, if the originator adequately documents his or her ideas prior to sharing them, the originator can secure property rights over the ideas when they are incorporated into useful applications. Furthermore, since patent breadth in the U.S. is broader, it is unlikely that talking about a "small" idea related to a larger invention would disclose enough to jeopardize the originator's rights over the invention.

Because of both the first-to-file provision and the nature of the breadth of the U.S. patent system, I conjecture that workers in the U.S. semiconductor industry would be more likely to share knowledge privately across company lines than their peers in Japan. The Japanese procedure of "laying open" all patent applications reinforces this conjecture, since instead of having to share "small" innovations, i.e., ones that would not qualify for patent protection in the U.S., over private channels, people in Japan can read patent applications.

The principal difference in labor market characteristics in the two countries that may shape knowledge-sharing behavior is the level of turnover. Of the respondents in our sample described in Section 10.5, none of the Japanese respondents from IC companies had ever changed companies, whereas the U.S. respondents had worked for, on average, 1.1 other semiconductor companies and 0.3 equipment vendors prior to their current employer. Given that turnover of personnel in the semiconductor industry is higher in the U.S. than in Japan, I conjecture that it is more likely that U.S. workers would have relationships with people at other companies, leading to a higher incidence of cross-company communication. To some degree, attending conferences could substitute for turnover in terms of establishing cross-company relationships. Semiconductor engineers from both countries actively engage in professional conferences. In our sample, however, there is not a marked cross-country difference in conference attendance by the IC respondents. On average, the respondents from both countries attend approximately 1.5 conferences per year.

Lynn et al. (1993) found that U.S. engineers do have a greater number of inter-firm contacts on average, but, in contrast to my conjecture, they found "no statistically significant difference (at the 0.05 level) in the frequency with which Japanese and American engineers used personal contacts at other companies for information" (p.68). As detailed in the next section, I also fail to find a statistically significant difference in knowledge-sharing behavior across U.S. and Japanese respondents.

In the cost and benefit framework above, turnover would likely influence the transaction cost, T, and the estimation of the decline in profits, D. I predict that turnover decreases transaction costs associated with knowledge-sharing, since it reduces the time required to find an appropriate contact at another firm. If the relationship with a previous co-worker had been favorable, it is likely that the former employee could call on that person for technical advice. Furthermore, it is likely that the departed employee could estimate how easily his or her former company could incorporate a piece of technical knowledge, thus leading to a more accurate estimate of D. Given this ability, it seems unlikely that the departed employee would receive requests from former co-workers for technical knowledge that would dramatically drive down the profitability of his or her new firm. Such requests would be futile under the assumptions of this paper. Therefore, I predict that the level of private knowledge exchange would be higher in the U.S., since former co-workers would be more likely to fulfill each other's requests.

10.5 Survey Results

To determine whether private knowledge-sharing in the semiconductor industry differs both from patterns in the steel industry and across countries, we distributed the Learning and Communication Survey to people in the industry during 1994 and 1995. As described below, the findings show that public sources of knowledge-sharing are much more prevalent in the semiconductor industry than found by previous studies of the steel industry. The survey data also show that people in the U.S. semiconductor industry rely more heavily on private sources of technical information than their Japanese counterparts. Semiconductor workers in the U.S. are approached more often for technical advice or technological information and are more apt to fulfill at least one request per year. According to ordered logit estimates, however, neither home country nor turnover level of the respondent plays a statistically significant role in the frequency with which an individual is either approached or provides requested technical information.

Of the respondents to the Learning and Communication Survey, the majority, 96, work for U.S. companies, while 27 respondents work for Japanese companies. These respondents are used in the analysis below, and Figure 10-5 shows that they principally work for IC companies, and over 70% of the respondents from both countries have less than 15 years of industry experience.

Since the majority of respondents had either previously agreed to participate in the benchmarking exercises of the CSM study or were attending an industry-related trade show or user meeting when they were handed the survey, it is likely that the respondents are biased towards knowledge-sharing. Even if the findings below overstate knowledge-sharing activity in the semiconductor industry, they provide a useful reference point for both the primary mechanisms of knowledge-sharing and its intensity.

Knowledge-Sharing in Semiconductors versus Steel

Given the greater degree of technological change in the semiconductor industry versus the steel industry as described in Section 10.3, I hypothesize that private, cross-company communication would be less likely in the semiconductor industry. A comparison of the preliminary results from our survey with results from a study of the U.S. specialty steel and mini-mill industry reported in Schrader (1991), supports this hypothesis, but only to a limited degree.

We matched a few questions in our survey to questions contained in The M.I.T. Steel Industry Communication Survey in order to determine industry-specific communication patterns. In both industries, engineers value colleagues at other companies. Our survey asks respondents to rate nine possible sources of technical information on a 7-point scale, where 1 represents "not at all important" and 7 represents "very important." For comparison with the steel study, we limited our sample to the 53 respondents from 21 U.S. integrated circuit companies. The majority of these respondents are engineers who work on circuit design, semiconductor process development or equipment development and modification, although a number work in materials procurement and manufacturing operations. The steel study draws on a much larger sample, 294 respondents, comprised of middle-level managers and engineers. In the semiconductor sample, we found that 57 percent rated technologists at other companies as an important source of technical information, meaning that they scored this source 5 or above on the 7-point scale. This compares with 61 percent of respondents from the steel study (Schrader, p.154). This preliminary finding provides only marginal evidence for the hypothesis that knowledge-sharing is less common in a semiconductor industry than in the steel industry.

Although the average rating of colleagues at other companies as information sources was found to be similar in the two industries, the importance of this source relative to other sources reveals a more striking difference. On average, in the semiconductor industry, technologists at other companies ranked fourth behind colleagues within the respondent's company, journals, books, etc. and presentations at conferences (Figure 10-6). In contrast, the steel study found that colleagues at other companies ranked second, on average, behind colleagues within the same company. This difference highlights the important role that public sources of technical information play in the semiconductor industry, a very R&D intensive industry.

When comparing the frequency with which the respondents in the two industries answer outside requests for specific technical information, the respondents from the semiconductor industry are less likely to engage in knowledge-sharing. A smaller percentage of the semiconductor respondents were approached by someone outside of their company for technical advice or for technological information during the year prior to the survey year: 79 percent of the semiconductor respondents had been approached at least once versus 85 percent in the steel industry (Schrader, p.155). Not only were the semiconductor respondents less likely to be approached, but a higher percentage of those approached fulfilled none of the requests for specific technical information: 10 percent versus only 2 percent in the steel study (Schrader, p.155). One interesting finding to note, however, is that 23 percent of the semiconductor respondents reported being approached 11 or more times, whereas 19 percent of the steel respondents were approached 10 or more times. This may suggest that semiconductor engineers who have established their reputations through journal articles and conference presentations receive a higher number of requests per year relative to "gurus" in the steel industry who do not receive as much public exposure.

Overall, 71 percent of the U.S. semiconductor respondents had participated in cross-company knowledge-sharing compared with 83 percent of the respondents from the steel study. Although this difference is not dramatic, our preliminary data collection does suggest that engineers in the U.S. semiconductor industry rely less on private contacts in other companies for technical advice than in the steel industry.

Knowledge-Sharing in Japan versus the U.S.

This section extends the analysis of knowledge-sharing patterns by examining cross-country differences within the same industry. After limiting the sample to respondents who work for IC companies in the U.S. and Japan, preliminary findings suggest a distinctive pattern of knowledge-sharing in the two countries, supporting my hypothesis presented in Section 10.4: The Japanese respondents rely on public channels for external knowledge acquisition, whereas the respondents from U.S. companies give a higher rank to private channels (Figures 10-6, 10-7, 10-8). This is supported by a similar finding reported by Almeida and Kogut (1994) that, relative to semiconductor engineers in the U.S., Japanese engineers possess fewer "informal contacts" at competitors (p.18).

On average, respondents from both countries scored colleagues in company, journals, books, etc. and presentations at conferences as the three most important source of technical information, but the U.S. respondents rely on technologists at other companies to a greater degree and the Japanese respondents rely on patents to a greater degree (Figure 10-6). In terms of finding out useful technical information from "horizontal" sources, i.e., from fabrication plants at other semiconductor companies, the IC respondents from Japan rated public vehicles as the most important (Figure 10-7). In contrast, the U.S. respondents scored personal channels of communication, namely face-to-face meetings and telephone, in their top four. A similar pattern was found for how IC workers find out useful technical information from equipment vendors, or "vertical" sources (Figure 10-8). Although their most important vehicles of knowledge-sharing overlap to some degree, the Japanese respondents are more reliant on public sources than private sources relative to the respondents from the U.S..

Figures 10-9 and 10-10 present the opportunities for and the incidence of knowledge-sharing involving the IC respondents from the U.S. and Japan. Consistent with the hypothesis that U.S. workers rely more heavily on private knowledge-sharing, nearly 80 percent of the U.S. respondents versus 64 percent of the Japanese respondents were approached at least once in a given year by someone in another semiconductor company or equipment vendor for some technical advice or for technological information (Figure 10-9). Approximately 90 percent of the respondents from both countries fulfilled at least one request (Figure 10-10). Although the Japanese respondents were less likely to be asked and slightly less likely to answer at least one request, those respondents who did fulfill requests were more likely to fulfill over half the requests. This counters the hypothesis posited in Section 10.4, that workers in a labor market with higher turnover (the U.S.) would be more likely to fulfill requests, since they have more information with which to value the costs of knowledge-sharing. This finding likely reflects business etiquette in Japan, whereby one would rarely ask another to share knowledge unless the knowledge owner would feel comfortable providing it (Kato and Kato, p.18).

The Profile of a Knowledge-Sharer

Using the larger sample of 123 respondents from Japanese and U.S. companies, I test whether characteristics of the individual respondents play a role in determining whether they are approached for technical information and whether they provide the information once approached. The characteristics that I include in the ordered logit analyses are presented in Table 10-3, along with their predicted signs.

In addition to testing whether the respondent's home country influences knowledge-sharing behavior, the remaining independent variables fall into two general categories: the respondent's personal characteristics and components of his or her firm's knowledge management system. First, because of institutional differences described in Section 10.4, I predict that home country influences the likelihood of knowledge-sharing, with U.S. residents exhibiting a higher likelihood of sharing. Second, I conjecture that respondents with the following characteristics will be more likely to engage in knowledge-sharing: those who work for equipment vendors, have a high level of education, have a long tenure in the industry, have higher levels of inter-firm turnover, are members of professional societies, consult people outside of their company when solving technical problems, and work on technical issues common to other companies or universities. These characteristics are predicted to heighten the potential for interaction with outsiders.

Finally, the components of a firm's knowledge management system can influence an employee's participation in inter-firm communications. Two components considered are: training in information control and rewards for presenting papers at professional conferences. If the respondent's company provides training on how to control sensitive information, I conjecture that the respondent would be more hesitant to provide requested information. If the respondent's company includes paper presentations at professional conferences in pay or promotion decisions, then the company actively encourages external knowledge diffusion.

After eliminating respondents with missing values, 95 respondents were used in the ordered logit analyses. The categories for the first dependent variable approached, the number of times the respondent was approached for technical information in the year previous to the survey year, are found at the bottom of Figure 10-9: Never, 1-5, 6-10, 11-15, or >=16 times. The categories for provided, the number of times the respondent provided the requested information, are located at the bottom of Figure 10-10: Never, Less than Half, About Half, More than Half, or Always. The signs from the estimated coefficients are reported in Table 10-4.

As described in Section 10.4, I hypothesize that home country and turnover should play a significant role in predicting whether workers in the industry are either approached for technical information or provide such information. The results found in Table 10-4, show, however, that neither working in the U.S. nor the level of turnover confer a statistically significant effect on the likelihood that a respondent would be approached for technical information or would provide requested information.

The only two variables that are statistically significant with the predicted sign are whether the respondent works for an IC company and the respondent's tenure in the industry. The signs reported in Table 10-4 indicate that IC respondents are more likely to fall in the Never categories of both dependent variables, whereas people with long tenure in the industry are more likely to fall in the >=16 and Always categories of being approached and providing information respectively. The results are consistent with my predictions.

A surprising finding is the negative and significant coefficient on "consulted outsider" reported in second set of results in Table 10-4. The negative coefficient suggests that those who asks colleagues at other companies for assistance when problem-solving are less likely to provide technical information when they are asked. This appears to run counter to the intuition that reciprocity drives knowledge-sharing. However, the respondents were not asked to disclose whether they had an established relationship with people who approached them.

To understand the rules of thumb followed by respondents when deciding whether to provide requested technical information, the questionnaire asks: "If you did not [fulfill all requests], how did you decide which requests to answer?" Grouping the answers into broad categories, the respondents replied that they share information across company lines when: 1) the information is of a general sort; that is, non-confidential or non-proprietary; 2) the information does not confer a competitive advantage; or 3) according to guidelines of the employer or a supervisor's assessment, the information may be shared. Two respondents share information only if they think the person asking is likely to reciprocate with information at least as valuable to their employer, implying the existence of trading relationships and countering the negative coefficient on the "consulted outsider" variable.

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