Why the Valley Went First

17 Why the Valley Went First

Aggregation and Emergence in Regional Inventor Networks

Lee Fleming ■ Lyra Colfer ■ Alexandra Marin ■ Jonathan McPhie

It has become increasingly fashionable to iden- dependent upon networks” but contend that a tify social networks as crucial contributors to re- region’s network(s) will adjust to suit its techno- gional innovative capacity (Marshall 1920; Piore logical competencies over time. Where Saxenian and Sabel 1984; Krugman 1991; Stern and Por- (1994) sees causal differences in Silicon Valley ter 2001). Networks have been argued to offer and Boston networks, Florida and Kenney (1990, improved customer-supplier relations, more effi- 98–118) see indeterminate similarity and pro- cient venture capital and legal infrastructure, and pose that technological trajectories drive regional increased knowledge spillovers between firms advantage. Turning the argument on its head, the and regional institutions. Knowledge spillovers skeptics propose that networks result from—and are thought to be particularly crucial to fast- do not necessarily improve—regional innovative developing technologies such as semiconductors advantage (Feldman 2001). and, more recently, biotechnology. Spillovers These opposing arguments for and against correlate with increased labor mobility (Angel causality immediately raise the suspicion of co- 1989), relaxed enforcement of non-compete cov- evolution. Surely networks influence regional enants (Gilson 1999), and increased labor mo- advantage and are in turn shaped by regional suc- bility and brain drain (Marx et al. 2009, 2011). cess or failure. But much of the current discussion Saxenian (1994) makes the functional argument about networks and regional advantage remains connecting networks and innovative capacity, static (for important exceptions, see Owen-Smith proposing that Silicon Valley’s rapid labor mo- and Powell 2004), implicitly assuming that net- bility, collective learning, interfirm relationships, works differ across regions but remain essentially and informal knowledge exchange gave it a deci- unchanged within them. If this assumption were sive edge in competing against the more secretive untrue—and could be cleanly unpacked—the dis- and autarkic firms of Boston. cussion could be greatly enriched. Nevertheless, there is still skepticism about the With these goals in mind, this chapter has causal influence of networks in regional innova- two objectives: first, to understand how isolated tive productivity. For example, Kenny and Burg clusters of regional inventors become connected, (1999) acknowledge that “all business activity is and in particular, why Silicon Valley aggregated earlier than Boston; and second, to describe the information flows and creative ecologies of such We would like to thank Ivin Baker, Jeff Chen, and Adam Juda for their help with matching algorithms and illustrations; networks. We begin by comparing the structural Christine Gaze for her editing; and the Harvard Business histories of the patented inventor coauthorship School Division of Research for support. Most important, we would like to thank all the inventors who spent a great deal networks of Boston and Silicon Valley from 1975 of time with us discussing their careers. through 1999. Following Fleming, King, and

Padgett_rev-proof.indb 520 6/19/12 11:01 AM Why the Valley Went First ■ 521

Juda (2007), we first demonstrate that the larg- cases less robustly so—than those of Boston. est connected network component in Silicon Val- While our interviews indicate that information ley underwent a dramatic transition in the early flowed more freely between firms in theV alley, 1990s.1 Although small at first—and similar in there were plenty of engineers and scientists in size to Boston’s largest connected component in Boston who were also willing to risk manage- 1989—it grew rapidly from 1990 on, encompass- ment stricture and talk to their colleagues across ing almost half of Silicon Valley’s patenting inven- organizational boundaries. Willingness to share tors by 1999. Boston did not undergo a similar information appears to be more strongly corre- transition until the mid-1990s, and even recently lated with a managerial versus technical profes- its largest connected network component remains sion than with location. proportionally smaller, containing approximately a quarter of its inventors. We investigated this historical divergence by focusing on the actual Data and Methods ties that inventors created—or failed to create— across the dominant network components in To gain empirical traction on the issue of how their regions. To fulfill our second objective, we the social structures of Boston and Silicon Valley asked the inventors about the knowledge flows differed and the effect this had on the develop- in their careers, both within and across the ob- ment of their innovative capacity, we consider all served clusters, and how such flows influenced patented inventors and their coauthorship rela- their creativity. tions in the two regions. For our purposes, there While inventors move from job to job and is a relationship between patented inventors if create collaborative ties for a wide variety of rea- they have coauthored any patent over a five-year sons (Gulati and Gargiulo 1999), our data high- moving window (alternate window sizes also light the importance of “academic” institutions demonstrated a qualitatively similar emergence in the creation of ties across regional organiza- phenomenon). This relational definition results tions. Much of the difference in aggregation2 can in many disconnected components that demon- be traced to Stanford doctoral students taking strate a skewed distribution, with most compo- local employment, in contrast to MIT students nents of small size and fewer and fewer of larger leaving the Boston region. We also find that a size. We refer to the largest and right-most com- single institutional program—the postdoc fel- ponent on this distribution as the “largest com- lowship at IBM’s Almaden Valley Labs—was ponent.” Appendix A contains a description of responsible for 30 percent of the Valley’s initial the matching algorithm we used to identify indi- aggregation. Young inventors play a particularly vidual inventors over time (for a later version of important role in the process of regional aggre- this algorithm, see Lai et al. 2011). gation; while older inventors move to start-ups Figure 17.1 illustrates the proportion of pat- (and are less likely to move in general; see An- ented inventors encompassed within a region’s gel 1989), young inventors move from graduate largest component.3 For example, if there were school through private firm postdoc programs ten inventors in a region and six of them co- and other positions within large network com- authored any patents together in the prior five ponents, bridging technological communities years, then the proportion in that region would and generating new technological combinations. be 0.6 or 60 percent. Note that the relationship is With respect to the competing comparisons transitive—if inventor A and B worked together of the two regions (Saxenian 1994; Florida and on one patent, and B and C on another, then A Kenney 1990; Kenney and Burg 1999), we found and C can trace an indirect coauthorship to one that Silicon Valley’s patenting coauthorship net- another and lie within the same component. If works are indeed more connected—but in some

3 We define a patent as being in a region if at least one 1 We define “component” as a cluster of inventors con- inventor lives within that region, as determined by the home- nected by at least one patenting coauthorship tie in the previ- town listed on the patent. Hometowns are classified within ous five years. Metropolitan Statistical Areas (MSAs) by the U.S. Census 2 We define the coming together of isolated inventor Bureau (Ziplist5 MSA 2003). Note that this definition en- networks into larger networks as a process of network “ag- ables inventors from outside Silicon Valley or Boston to be gregation.” The term is intentionally dissimilar to the word included as regional inventors if they worked with someone agglomeration used in the economics literature, which refers who did live within the region. We analyzed a more restricted to economies that firms gain by clustering together and shar- definition and found only minor qualitative differences in the ing pooled labor availability, infrastructure, suppliers, and processes . All graphs include all 337 U.S. MSA regions for other services. comparison and illustrate five-year moving windows.

Padgett_rev-proof.indb 521 6/19/12 11:01 AM 522 ■ Chapter 17

1.0 Boston Silicon Valley 0.8 t ntors in ve 0.6 mponen

0.4 gest Co tion of MSA In Lar or 0.2 Prop

0 1979 1981 1983 1985 1987 198911991 993 1995 1997

Figure 17.1 Box plots of the size of the largest con- Statistical Area (x-axis indicates last year in five-year nected component relative to the entire network of moving window). patented inventor collaborations by U.S. Metropolitan

four of the ten inventors had coauthored patents in the Valley to form its largest component in and no other group of coauthors was bigger, 1990, while in Boston, only the 3rd, 13th, and then the proportion would be 0.4 or 40 percent. 384th largest merged to form its largest com- The interesting feature of figure 17.1—and the ponent in 1990. By 1992 the largest component original motivation for this chapter—is the ag- in Silicon Valley had over 1,600 inventors, in gregation process in Silicon Valley. It began in contrast to Boston’s approximately 330 inven- 1990, and by 1998 almost 50 percent of the Val- tors. Furthermore, figure 17.2 shows the extent ley’s inventors had aggregated into the largest to which Silicon Valley saw a greater number of component. Boston’s aggregation, by contrast, smaller and distinct components from one time did not begin until 1995, and by 1998 its largest window merging to form its largest component component had only reached 25 percent of the in the subsequent time window. Even though the region’s inventors. process begins with the linkage of larger com- The histograms of figure 17.2 show which of ponents, it reaches a critical mass at which the the prior year’s network components aggregated largest component begins to suck in components to form the following year’s largest component, of all sizes. from 1988 to 1992. Note that the size of any given component is simply the number of inventors it includes. Each region contains thousands Qualitative Methods of components of varying sizes in any given year (most of which contain just twenty or fewer in- We conducted in-depth interviews with key inventors and therefore fall above the frequency ventors in both regions to understand the his- cutoff used for the y-axes in figure 17.2). torical and social mechanics of the aggregation Figure 17.2 illustrates the early similarity in process. We identified these inventors in two the distributions of the two regions’ components. rounds. First, we graphed the largest component In 1988 Boston had a larger largest component of 1990 in both regions to pinpoint the inven- (although figure 17.1 obscures this because it il- tors who provided crucial linkages from the pre- lustrates the proportion of inventors and Boston vious year’s components. For example, drawing had slightly more inventors in that time period). on the histograms in figure 17.2, we identified In 1989 the distributions of the larger compo- the inventors who connected the 1st, 2nd, and nents across the two regions were similar. Yet 6th largest components together in the Valley as the 1989 panels of figure 17.2 illustrate, the and the 3rd, 13th, 384th, and 707th largest in 1st, 2nd, and 6th largest components merged Boston. We then identified similar inventors who

Padgett_rev-proof.indb 522 6/19/12 11:01 AM Why the Valley Went First ■ 523

10 Found in 89 LC Boston 88 Histogram 8 Not found in 89 LC 6 4 equen cy

Fr 2

10 Found in 89 LC Silicon Valley 88 Histogram 8 Not found in 89 LC 6 4 equen cy

Fr 2

10 Found in 90 LC Boston 89 Histogram 8 Not found in 90 LC 6 4 equen cy

Fr 2

10 Found in 90 LC Silicon Valley 89 Histogram 8 Not found in 90 LC 6 4 equen cy

Fr 2

10 Found in 91 LC Boston 90 Histogram 8 Not found in 91 LC 6 4 equen cy

Fr 2

10 Found in 91 LC Silicon Valley 90 Histogram 8 Not found in 91 LC 6 4 equen cy

Fr 2

10 Found in 92 LC Boston 91 Histogram 8 Not found in 92 LC 6 4 equen cy

Fr 2

10 Found in 92 LC Silicon Valley 91 Histogram 8 Not found in 92 LC 6 4 equen cy

Fr 2

10 Found in 93 LC Boston 92 Histogram 8 Not found in 93 LC 6 4 equen cy

Fr 2

10 Found in 93 LC Silicon Valley 92 Histogram 8 Not found in 93 LC 6 4 equen cy

Fr 2

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 Component Size (x103)

Figure 17.2 Time series of histograms of component size red, the number of those components which merged to frequency of Boston and Silicon Valley. The x-axis identi- become a part of the single largest connected component fies the range of possible sizes (in number of inventors) of that region in the following year. Because of space con- for network components (demarcated into bins of 10 for straints and to emphasize the right-skewed outliers, we readability), while the y-axis reflects, in blue, the num- truncated the y-axis of each histogram. Boston generally ber of connected components of a given (bin) size found has a larger number of inventors in the first category— in that region during that year (truncated to 10 to allow that is, its distribution is more left-skewed—over all the for visibility of the red bars described hereafter) and, in time periods.

Padgett_rev-proof.indb 523 6/19/12 11:01 AM 524 ■ Chapter 17

Perlman Koning

Kaufman

Stewart

Digital Equipment Corporation Hewlett-Packard Company

Figure 17.3 Largest component of Boston, 1986–90, by as- did not aggregate into the 1990 largest component. (Previ- signee and importance of inventions. The DEC component ously published in Fleming, King, and Juda 2007.)

did not create such linkages between other large interviewed them mainly during July and August components—for example, the 3rd, 4th, and 5th 2003, presenting the inventors with the histo- largest components in the Valley and the 1st, grams shown here and illustrations of their own 2nd, 4th, and 5th largest in Boston. network components with all of their coauthors We chose this second set of control inventors identified. We asked them about their careers, based on its similarity to the first set of linking what was happening within their component inventors. All inventors from components that during the time period under study (especially did not aggregate into the 1990 largest compo- with regard to job mobility), and where their nent but were similar in size to those that did collaborators were now. We asked specifically were at risk of control selection. We ran a Eu- about the collaborators in their patent networks clidean distance-matching algorithm (the com- and about any other networks, such as social or pare command in STATA) with variables that scientific networks. measured the linking inventor’s patenting his- Follow-up questions probed for inaccura- tory. We included variables to measure the in- cies in our illustrations and name-matching al- ventor’s access to information and likelihood gorithm and for sampling bias caused by failed of career movement opportunities, such as the patent attempts or by technical efforts that were mean degree of collaborations, clustering of the not intended for patenting. None of our inven- inventor’s collaborators (a density measure of tors indicated an inaccurate name match or col- the actual number of ties between alters, divided leagues, and all felt that the illustrated network by the possible number of ties), number of pat- reflected their patent coauthors accurately. For ents by time period (or basic inventive produc- example, Salvador Umatoy of Applied Materi- tivity), and future prior art citations by the time als indicated that a failed project had not been period (since citations have been shown to cor- patented but that his collaborators on success- relate with patent importance; see Albert et al. ful patents were all reflected; Jakob Maya, a 1991). Finally, we interviewed Robert Stewart lighting scientist, noted that some of his proj- because of his compelling position at the center ects concluded with published papers rather of the disintegration of Boston’s largest compo- than patents, as did Radia Perlman and Charles nent, as illustrated in figures 17.3 and 17.4. Kaufman (both computer engineers originally We were able to contact many of the link- with Digital Equipment Corporation [DEC]), ing and control inventors we identified. We but none recalled any patent collaborators who

Padgett_rev-proof.indb 524 6/19/12 11:01 AM Why the Valley Went First ■ 525

Perlman Koning

Kaufman

Stewart

Multiplex Communications Data Processing: Design and Analysis of Circuit or Semiconductor Mask

Figure 17.4 Largest component of Boston, 1986–90, by DEC component did not aggregate into the 1990 largest technology type and usage of scientific literature. The component.

were not represented in his or her network com- around the time of study that might have caused ponent as illustrated. Given evidence from pat- the aggregation processes we observed. Finally, ent citation data that information flows across even though universities as a whole were patent- indirect linkages (Singh 2005) and that aggrega- ing more over the time period than they had be- tion processes improve regional inventive pro- fore, the elite schools such as Stanford and MIT ductivity (Fleming, King, and Juda 2007), we did not change their patenting rates very much also asked the inventors about information flow (Mowery et al. 2001). Also, given that Boston across the illustrated linkages (the second moti- had more university patents than the Valley did, vation for this chapter). Finally, we simply asked this may well have increased aggregation in the them what they thought might have caused the region, as inventors left school and took local aggregation processes we observed. employment. To supplement these detailed analyses of the individual components, we also investigated plausible alternatives. Additional analyses of Qualitative Data the patent data (available from the first author) showed that Boston inventors were slightly more Our interviews with the Valley and Boston inven- likely to work alone, be self-employed and there- tors revealed both common and region-specific fore own their own patents, and work with a reasons for aggregation and non-aggregation. fewer number of collaborators. There were only We organized these reasons by regions and slight differences in tie density over time for the whether the cause was specific by region.T hese two regions, in the age and diversity of technol- reasons are summarized in table 17.1. We did ogy, and in the number of assignees per inven- not hear of any exactly similar aggregation tor for the two regions. Fleming and Frencken processes, although we will discuss the obvi- (2006) explicitly investigated inventor mobility ous similarities among the stories that follow. between the regions and found that mobility was The Silicon Valley–specific reasons for aggre- slightly higher in the Valley in 1975. The differ- gation included an IBM postdoc program and ence in mobility steadily reversed, however, such local hiring of local graduates. Boston-specific that differences were negligible in 1990. reasons included internal collaboration within Most important, none of these poten- DEC. Common non-aggregation reasons be- tial causes demonstrates an abrupt transition tween the regions included the instability of big

Padgett_rev-proof.indb 525 6/19/12 11:01 AM 526 ■ Chapter 17

Table 17.1. Summary of Reasons for Aggregation and Non-aggregation in Silicon Valley and Boston Aggregation Non-aggregation

Silicon Valley Local graduate employment Internal labor markets IBM postdoc program Start-ups Big firm instability Self-employment Boston Internal collaboration Internal labor markets Start-ups Big firm instability Non-local graduate employment Lack of internal collaboration Non-local internal collaboration Patenting/publication policies

firms, internal labor markets, and the movement to the unstable backbone of the Boston aggre- of personnel to start-ups. Valley-specific reasons gation process, a point to which we will return for non-aggregation included the movement of later). Stanford’s Ginzton Applied Physics Lab personnel to self-employment, while Boston- network joined the Valley’s largest component in specific reasons included non-local graduate em- 1989 through the career of William Risk. Upon ployment, lack of internal collaboration, internal graduation from Stanford with a Ph.D. in electri- firm collaboration that was non-local, patenting cal engineering, Risk accepted employment (and policies, and product life cycles. obviously patented) at IBM.5 Further Stanford aggregation occurred in 1990 with William Koz Valley-Specific Reasons for Aggregation lovsky’s graduation and departure from Profes- We identified two aggregation processes unique sor Robert Byer’s lab. Hence these multiple ties to Silicon Valley, both driven by IBM. The com- created a robust conduit of inventors and ideas pany hired local doctoral graduates, connecting from Stanford to IBM. it with Stanford components, and it sponsored William Risk and Professor Gordon Kino6 a postdoctoral fellowship program, connecting elaborated on the mobility of students and it to the large pharmaceutical and biotech com- the resultant knowledge flows.7 Kino reported ponent in the Valley. Figures 17.5 and 17.6 il- that his students of that era had gone on to a lustrate the largest component of the Valley from variety of academic and technical positions, in- 1986 through 1990.4 cluding start-ups in the Valley and in Oregon, IBM’s Almaden Valley Research Lab provided self-employment as an entrepreneur in Wyo- the stable backbone of the 1990 Silicon Valley ming, academic positions at Stanford, UC–Santa aggregation. IBM constituted the largest com- Barbara, and Wisconsin, and employment at ponent in the Valley by 1987 and remained the Tektronix, Bell Labs, AT&T, and IBM New largest component in 1988 and 1989 (in contrast 5 William Risk is still at IBM Almaden Research Labora- 4 Each node corresponds to an inventor and network ties tory and has done research in applied physics, optics, and correspond to coauthorship of at least one patent. A1 colors photonics. the nodes by firm and A2 colors them by technology (only 6 Gordon Kino received his Ph.D. from Stanford Univer- A2, A7, and A14 are in full color; other pictures are grayscale. sity in 1955 and has done research in nondestructive test- Node size in A1 corresponds to future prior art citations to ing, fiber optics, fiber-optic modulators, fiber-optic sensors, the inventor’s patents over the five-year time period and can and optical, acoustic, and photo-acoustic microscopy. He is a be interpreted as the importance of the patent holder’s inven- member of the National Academy of Engineering. tions (Albert et al. 1991). Node size in A2 indicates the num- 7 Technically the agglomeration between Gordon Kino of ber of non-patent (generally scientific) references.T ie strength Stanford and William Risk of IBM occurred one year earlier corresponds to coauthorship strength, as measured by the than the 1986–90 window. Given that we were unable to number of coauthored patents, normalized by the number meet with William Kozlovsky and Robert Byer and given that of inventors on the patents. Tie color corresponds to tie age: the Stanford-IBM inventors knew each other well and cor- green ties were formed in the prior year, blue ties in the sec- roborated the processes described here (Kozlovsky did so in a ond through fourth prior years, and red ties were formed five phone interview), we report from Kino and Risk. Given that years prior. All network diagrams were plotted in Pajek with a very similar process occurred twice over two years, it would a directed force algorithm (Batagelj and Mrvar 1998). appear to be a robust and frequent occurrence.

Padgett_rev-proof.indb 526 6/19/12 11:01 AM Why the Valley Went First ■ 527

Byer

Kozlovsky Risk Kino Scott Ribi Guion Khanna

IBM Corporation Syntex (U.S.A.) Inc. Stanford University Xerox Corporation Biocircuits Corporation

Figure 17.5 Largest component of Silicon Valley, 1986–90, assignee (generally a firm or university). Yellow nodes in by assignee and importance of inventions. Node sizes lower right indicate Abbott Laboratories, to which Pyare indicate the number of future prior art cites to an inven- Khanna (along with Edwin Ullman) moved to late in the tor, normalized by the number of collaborators. Tie width time period. Graphed in Pajek with Kamada-Kawai/Free reflects the number of collaborations, tie color indicates algorithm (Batagelj and Mrvar 1998). (Previously pub- age of tie (red is five years prior, blue is two to four lished in Fleming, King, and Juda 2007.) years prior, and green is prior year), and colors indicate

York. He and his students studied microscopy, of technological applications and industrial acoustics, photonics, and microwave phenom- opportunities.8 ena, and his students went on to work in a wide Kino and Risk renew old ties at conferences variety of industries, including medical instru- and visits (Risk had visited Stanford the week mentation, electronics, optics, and scientific in- prior to the Kino interview). The former stu- strumentation. Professor Kino’s description of dents and their professors discuss technical work local employment for Stanford graduates ap- at conferences even though they work for differ- pears to be the flip side of Professor Richard ent firms. With the exception of Kino’s formal Cohen’s description below of non-local employ- consulting relationships, neither Kino nor Risk ment for MIT graduates. As such, the processes remembers other substantial or formal technical of local and non-local employment of graduates information flows. Both agreed that the techni- surely operate similarly across regions—when cal information only flows through a strong, appropriate local firms are hiring, graduates are more likely to stay, and when they are not, or if 8 Even though Angel (1989) provides some evidence that the region lacks such firms, graduates emigrate. Valley firms are more likely to hire local graduates than are For example, William Risk stressed the impor- firms in other regions, our categorization of such local hiring processes as Silicon Valley–specific is mostly an expositional tance of optics to a wide variety of industries convenience, based on our interview sampling and the eco- and how the Valley provided a great diversity nomic conditions at the time.

Padgett_rev-proof.indb 527 6/19/12 11:01 AM 528 ■ Chapter 17

Byer

Kozlovsky Risk Kino Scott Ribi Guion Khanna

Chemistry: Molecular Biology and Microbiology Drug, Bio-A ecting and Body Treating Compositions Radiation Imagery Chemistry Active Solid-State Devices Dynamic Magnetic Information Storage or Retrieval

Figure 17.6 Largest component of Silicon Valley, 1986–90, indicates number of collaborations, tie color indicates by technology type and usage of scientific literature. Node age of tie (red is five years prior, blue is two to four years sizes indicate the number of references to non-patent prior, and green is prior year), and colors reflect U.S. literature by an inventor (mainly peer reviewed science), Patent Office technology class. Graphed in Pajek with normalized by the number of collaborators. Tie width Kamada-Kawai/Free algorithm (Batagelj and Mrvar 1998).

informal social network. In particular, they felt of John Campbell Scott10 and the (now failed) that graduates from the Ginzton Applied Phys- start-up of Biocircuits. ics Lab at Stanford had maintained particularly Scott described how the Almaden Lab hired close contact since leaving Stanford. postdocs straight from school (generally PhDs The largest aggregation occurred with the but other degrees as well) with the intention that linkage of the second-largest component in the they would leave for employment with another Valley—Syntex (a research-intensive pharma- private firm after one or two years. Modeled af- ceutical firm) and smaller biotech firms—with ter academia and similar programs at Bell Labs, IBM in 1986–90.9 The actual connection was in- the practice was intended to seed the technologi- direct and occurred indirectly through the career cal community with experienced IBM-friendly scientists. Such a process would obviously create observable ties between IBM and a wide variety of other firms. Unlike the departure of se- 9 Even though Silicon Valley is known in this time period nior inventors from large and established firms as a center of semiconductor and computer technologies, only for start-ups (which does not create ties between the 1st and 5th largest components covered such technologies, namely magnetic media (computer disks at IBM) and semiconductor manufacturing equipment (at Applied Mate- rials). The 2nd, 3rd, and 4th largest components consisted 10 John Campbell Scott still works at IBM Almaden Re- of pharmaceutical (Syntex), polymer chemistry (Raychem), search Laboratory. He earned his Ph.D. in solid state physics and optical (Xerox PARC/Spectra Physics, Hewlett Packard) at the University of Pennsylvania and has worked in materials technologies. science for most of his career.

Padgett_rev-proof.indb 528 6/19/12 11:01 AM Why the Valley Went First ■ 529

large components), the postdocs found future argued that patents are used to protect propri- employment across a variety of firms. Hence the etary property and that coauthorship did not in- IBM postdoc program played a crucial role in dicate a higher probability of information flow. the initial and continuing aggregation processes Interestingly, the manager on the other side of in the Valley because it linked large components the IBM-to- biotech/pharma connection, Pyare to other large components. Khanna, also complained about the possibil- While the connection of the Syntex and IBM ity of information flow. Both Ribi and Khanna components relied on the postdoc program, it were managing start-ups at the time of the inter- actually occurred through the career of a young view and felt much more vulnerable to the loss inventor at Biocircuits, an early (and ultimately of proprietary information and key individuals, failed) electronics-biotech start-up that developed in contrast to the IBM scientists who, as “good biosensors.11 Todd Guion, a Stanford graduate in corporate citizens,” felt resigned to the possibil- chemistry, worked for Scott during his postdoc ity of such loss. at IBM and then took a job at Biocircuits. Victor Pan took a similar path from San Jose State and Boston-Specific Reasons for Aggregation Santa Clara University through IBM to Biocir- We identified only one Boston-specific reason for cuits. Biocircuits was attempting to build a bio- aggregation. The largest component in 1990 re- sensor based on polymeric material and wanted sulted from internal collaboration—newly initi- to get a charge through a polymer. Guion thought ated interaction of smaller work groups—within that optical technology might help and recom- Digital Equipment Corporation, as illustrated in mended to Hans Ribi,12 the CEO of Biocircuits, figures 17.3 and 17.4. We describe the integra- that he contact Scott for help. After some initial tion of the DEC component through the careers difficulty, Scott secured permission from IBM of Charles Kaufman, Paul Koning, Radia Perl- management to act as a scientific adviser, given man, and Robert Stewart.14 that there were no apparent conflicts of interest. Charles Kaufman, discussing his own role Scott spent many days at Biocircuits and inter- as a “point of connection” in these processes, acted with most of its employees. He suggested noted that he was particularly likely to be re- the use of bio-refringence associated with specific sponsible for information flow across multiple binding to solve the problem. He reported that departments of DEC for two reasons. First, he “definitely learned a lot of interesting things” he was one of “the gang of four” chosen from that he is now, many years later, applying as IBM four distinct working groups to design DEC’s moves into biological technologies. He had no “next generation of security.” Second, while he interaction with Pyare Khanna,13 however, the was a software engineer by trade, he often so- prominent pharmaceutical inventor on the other cialized with those working in hardware. Paul side of the Biocircuits bridge. Koning, addressing the same question, noted Hans Ribi, a Stanford graduate in biochem- that his shifting collaborators usually corre- istry and the owner/CEO of Biocircuits, had sponded to shifting task assignments but that a much less positive view of information flow two exceptional features of working at DEC across collaborative linkages, believing that it could explain some of his more interesting col- should not and generally does not occur. He laborations. First, his working group’s manager

11 The start-up might be described as a forerunner of to- 14 Charles Kaufman attended Dartmouth for mathemat- day’s combinations of biological and digital technologies, ics and worked with a Dartmouth-related technology ven- seen in products such as Affymatrix’s combination of assay ture prior to accepting a position in the network architecture and semiconductor technology into a gene array chip; in pub- group at DEC. Paul Koning worked with Charles Kaufman lications, such as BIO IT World, that focus on the application and Radia Perlman at DEC before moving to smaller start-up of computing power to biological and genomic problems; and ventures. He is currently the founder and CTO of a success- in research laboratories, such as Stanford’s BIO-X, that hope ful VC-backed start-up just outside the Boston area. Radia to encourage collaboration between chemistry, engineering, Perlman earned her Ph.D. from MIT while employed by biological, and medical research. Pyare Khanna felt that Bio- DEC. She is currently a Distinguished Engineer at Sun Micro circuits failed because it was too early and the integration systems and serves on the Internet Architecture Board of was too difficult. the Internet Engineering Task Force. Robert Stewart earned 12 Hans Ribi received his Ph.D. in biochemistry at Stanford undergraduate and master’s degrees in electrical engineering University in 1988 and was the CEO of Biocircuits at the time from MIT. He took employment with DEC upon graduation that Todd Guion suggested that John Campbell Scott work and remained with the firm until its purchase by Compaq. We with the firm. interviewed Stewart because he was so central to the disinte- 13 Pyare Khanna worked at Syntex as a senior scientist gration of DEC in 1990. He did not meet our typical criteria during the period of the study. He is currently the CEO of of being either a bridging node that caused aggregation or a Discoverx, a drug target company in Fremont, California. bridging control node.

Padgett_rev-proof.indb 529 6/19/12 11:01 AM 530 ■ Chapter 17

routinely sought brainstorming solutions from inventors from established firms generally go to a wide distribution of engineers. Second, co- start-ups rather than to other large established inventor Radia Perlman’s collaborative style of firms.T his implies that they will link established brainstorming made her a particularly strong firms with large components to start-up firms candidate for generating information flow dur- with small or nonexistent components rather ing this process (as much with him as with other than linking large components to other large individuals), as did her tendency to prefer top- components. Finally, when established firms be- ics and projects “at the boundary of academic come unstable, they do not hire and their current research and engineering.” On the other hand, inventors often spend more time protecting their Koning also noted that Perlman was probably jobs or seeking new ones than they do inventing. unable to patent much of this work when the This will be reflected in a decreased rate of pat- participants spanned company boundaries. Both enting and thus in smaller components. Kaufman and Perlman independently confirmed Salvator Umatoy’s18 career matched the ex- this viewpoint, enumerating several bureaucratic planation for Applied Materials’ (the fourth obstacles they have had to surmount in order to largest component in the Valley) failure to aggre- work together since leaving DEC. One particu- gate. The firm’s business boomed during the time larly interesting example required both parties period under study and there were many inter- to persuade their respective employers that their nal technical and managerial opportunities for joint invention, while worthy of patenting, was its employees. (Even now, during much tougher not worthy of commercial sale.15 Like Scott in times, it retains a strong internal labor market the Valley, these Boston inventors overcame le- and hires mostly new college graduates.) Applied gal, managerial, and strategic obstacles to col- Materials provided its employees with generous laboration across organizational boundaries. incentives, such as stock options, to stay within Koning and Kaufman both reported switch- the firm. Most of the colleagues in Umatoy’s ing job functions within DEC several times,16 network there (figure 17.7) remained with the typically to new technologies where the knowl- firm and—at the time of our interviews—were edge of earlier collaborators proved less useful. still technical contributors or had become senior Koning often maintained loose ties with prior managers, working in close proximity to each collaborators throughout this process, occasion- other (“he works down that aisle . . . he works ally passing back information about old projects in the building next door”). Umatoy commented but rarely requesting help or technical advice that only managers went to other large firms; from his old network. On the other hand, he senior engineers went to start-ups (which fur- also noted that he and Perlman are a significant ther inhibited aggregation). When asked about exception to this trend because they have contin- people in his network with whom he had not ued to collaborate in new ways (e.g., on multiple patented at the time and who had left (part of academic papers and publications) for well over our concern about sampling bias), he mentioned a decade now, despite working for different em- an engineer who left technology and the Valley ployers since 1993.17 altogether and a technology process manager who left for IBM. Umatoy did not work directly Common Reasons for Non-aggregation with this manager (he was not illustrated in the Some of the explanations we heard for non- figures).T his memory serves to bolster Umatoy’s aggregation between components were common earlier conjecture that engineers left for start- to both regions. First, large established firms ups and only managers left for other large firms. with internal labor markets generally retain Umatoy expressed mixed opinions about infor- their employees (Angel 1989). Second, successful mation transfer across firms. He also felt that Applied Materials did not “give you time for any outside life [that would enable knowledge 15 This joint invention was a strong password protocol transfer].” Yet he reported that before starting a they created specifically to serve as a free alternative to two patented protocols. Both of their employers agreed not to pat- project, Applied Materials engineers do call their ent it and they published a paper to place the protocol into the public domain. 16 Koning reported switching firms several times, choosing 18 Salvador Umatoy (control to Glenda Choate, a bridg- one start-up after another, two of which he founded. ing inventor at Biocircuits whom we were unable to locate) 17 Lotus Development acquired Koning’s employer, Iris As- worked in the medical instrumentation industry before com- sociates, in 1994; IBM acquired Lotus Development in 1995. ing to Applied Materials in the early 1980s. He remains there Despite these changes, Kaufman continues to work with the and currently manages mechanical engineers designing wafer same group, now under the IBM umbrella. fabrication equipment.

Padgett_rev-proof.indb 530 6/19/12 11:01 AM Why the Valley Went First ■ 531

Umotoy

Advanced Micro Devices, Inc. Applied Materials, Inc. IBM Corporation

Figure 17.7 Applied Materials component, Silicon Val- importance of inventions. Applied Materials did not ag- ley’s fifth largest component in 1989, by assignee and gregate into the 1990 largest component.

friends (including colleagues at other firms), We heard similar stories about the power of contact professors at universities, and read the internal labor markets from our Boston inven- patent and scientific literature. tors. In addressing why the DEC component In contrast to the seeming lifetime employ- did not remain the largest after 1993,20 Charles ment at Applied Materials, most of the inven- Kaufman observed that DEC was not hiring, tive colleagues of Robert Sprague19 have left the due to its economic concerns,21 and that leaving legendary Xerox PARC. He listed a variety of was considered “kind of ‘traitorous.’” In fact, he destinations for his coauthors during the period noted that DEC had an explicit policy that em- of study, including Spectra Diode Labs (figure ployees who left were not to be rehired and he 17.8), Komag, Exxon Enterprises, Canadian Re- recalled few people leaving before formal layoffs search Corporation, and a variety of start-ups. began in 1991.22 Most became CEOs, CTOs, or chief scientists, Despite the increasingly gloomy economic and they often left with the core technology they climate along Route 128 during the latter half had invented at PARC. He could not remember of the 1980s, the DEC inventors did not recall any colleagues who left for an established firm, perceiving any risk to their own careers at the mainly because the start-ups provided stock time. They recalled many alternative opportuni- opportunities. He divided the movement of ties available to them, both in Silicon Valley and technology out of PARC into three categories: along Route 128, but they preferred staying at disgust, opportunities, and friendly, the last cat- DEC for several reasons. While Kaufman noted egory being sponsored and supported by Xerox. He included Spectra Diode Labs and his own, 20 As mentioned in an earlier footnote, the GTE/Siliconix Michigan-based start-up, Gyricon, in the last component displaced the DEC component to become the category. While Xerox might have done a bet- largest in Boston in 1991. Thereafter, the DEC component ter job in commercializing its PARC technolo- resumed its rank as largest in 1992, only to be displaced a final time in 1993. All three of the bridging inventors at DEC gies, Sprague did not express resentment at the with whom we spoke departed in 1993. mobile inventors and the spillovers they caused. 21 At the same time, he also pointed out that he himself had been hired during a freeze and perceived that such exceptions were not particularly rare. 22 Drawing on the first author’s anecdotal experience at 19 Robert Sprague (control to Pyare Khanna) earned his Hewlett Packard, he remembers many of his lab’s best engi- Ph.D. in physics from the University of Rochester. He has neers leaving for an early pen-computing start-up. They were worked at Xerox PARC since the period of study and is CEO rehired following the start-up’s failure and given a party upon of Gyricon, a Xerox PARC spinout. their return.

Padgett_rev-proof.indb 531 6/19/12 11:01 AM 532 ■ Chapter 17

Sprague

Xerox Corp. Hewlett-Packard Company Spectra-Physics, Inc. Spectra-Physics Lasers, Inc.

Figure 17.8 Xerox PARC and Hewlett Packard component, assignee and importance of inventions. This component Silicon Valley’s fourth largest component in 1989, by did not aggregate into the 1990 largest component.

that DEC had a reputation for treating its en- specific story explaining non-aggregation. gineers particularly well and that no other of- Raychem—neither a semiconductor firm nor a fers he received at the time could match DEC’s computer firm but a large and established poly- compensation, Koning and Perlman also empha- mer chemistry firm—had been theV alley’s larg- sized that their collaborators were still sharp, est component until it was overtaken by IBM in their work was still innovative, and they were 1987. Froix took his first job in theV alley with still being given opportunities with the potential Raychem as a senior scientist in 1979 and left for large-scale impact. In fact, both Koning and in 1985 as a lab director. According to Froix, Perlman specifically described their small work the firm had initially provided an environment groups within DEC as being rather “start-up where inventors could work on anything that like,” explaining that despite suffering its share would lead to a business. This changed in 1983, of bureaucratic dysfunction, DEC had “por- however, when non-technical management as- tions” that were still very successful and excit- sumed control. Without technical foresight from ing, at least technologically speaking, even then. management, Froix felt that politics became ram- All three remained at DEC until 1993, acknowl- pant and this caused many senior inventors and edging that they had stayed on well after the scientists to leave. Destinations included medical headlines on the business pages of the Boston device and fiber-optics firms, small start-ups, and Globe had soured. medium-sized firms such as JDS Uniphase. This was unfortunate for Raychem because it was the Valley-Specific Reasons for Non-aggregation only large company in the Valley with polymer Michael Froix23 provided the most interest- expertise at a time when polymer applications ing career story and uncovered the one Valley- were “exploding” in the medical, optical, and

23 Michael Froix (control to William Risk) earned his Ph.D. at Xerox, Celanese, Raychem, Cooper Vision, and Quanam, in physical chemistry from Howard University. He has worked and has been a very successful independent inventor.

Padgett_rev-proof.indb 532 6/19/12 11:01 AM Why the Valley Went First ■ 533

chip and board fabrication industries. In Froix’s Boston Scientific’s chief technology officer, the opinion, Raychem’s management repeatedly technology has become an important part of the failed to seize these opportunities. For example, firm’s product portfolio (Cohen 2003). Froix is Advanced Cardio Systems asked for help in ap- now working with a molecular biologist on tis- plying Raychem’s electron beam techniques to sue generation with stem cells. the medical pacemaker market—which was un- As can be seen in figure 17.9, Froix did not related to Raychem’s current markets—yet Ray- have many collaborators at Raychem, but he has chem management turned down the request for stayed in touch with them and other former col- the purported fear of losing advantage in their leagues over the years. Although this was mainly current markets. for job searches, he has also discussed techni- Froix left Raychem in 1985 out of frustration cal matters within this network over the years. with no job but a part-time teaching position at Froix’s experience provides a compelling story the University of San Francisco (USF). He de- of inventive tenacity in the interstices of a tech- cided to invent a material that would decrease nological ecosystem. It is difficult to understand the clotting that occurred on the surface of an how representative his experience was, however, artificial heart (recipients of such hearts would without a better understanding of the sampling generally survive the first few weeks, only to distribution of inventors and their likelihood of suffer strokes caused by such clots). He worked their bending corporate and university rules. The after hours in a friend’s corporate lab. He had Valley might be more supportive of such inven- approval, since his friend was the founder, but tors, but Boston inventors may also have had Froix supplied all his own materials, had no after-hours access to corporate and university access to proprietary information, and did not laboratories and there may have been professors interact with the employees. He also worked in at MIT or Harvard who were willing to support the lab of a supportive professor at USF. He then their research. Determining how widespread read about an analytic technique to measure such practices are, in Boston or any other re- the effectiveness of his material, developed by gion, would require inventors to admit to viola- Channing Robertson at Stanford. He contacted tions of corporate and university rules, possibly Professor Robertson in 1986 and asked for help. putting their jobs at risk. Hewlett Packard had Robertson replied that he would leave the dean oft-repeated story (told by the protagonist cision to his best graduate student, Seth Darst in Packard 1995) about the founders coming in (now a professor at Rockefeller University). on the weekend and finding the central lab sup- Darst agreed to help but, like a typical gradu- plies locked. They sought out a security guard, ate student, didn’t begin working until midnight. had the padlock cut, and ordered that lab sup- Undeterred, Froix would sit on the stairs next to plies should never again be locked. They felt that the lab from 6:00 p.m., when the building was supporting an inventor’s creativity outweighed locked, until Darst arrived many hours later. The any employee theft that might occur. Such sto- collaboration worked well and Froix perfected ries remain anecdotal, but they consistently sug- his invention,24 sold his technique to Cooper Vi- gest that strong engineering and science cultures sion, and helped implement its application to (wherever they might be) place creativity before a corneal implant product. He was then intro- financial and proprietary concerns. duced to a Stanford cardiologist, Simon Stertzer, Paul Koning expressed skepticism regarding and began working on a drug-delivery stent in such a generous flow of information or resources his garage in Mountain View and at Stanford. across collaborative linkages; he specifically felt He formed a start-up, Quanam, which has been that Froix’s story was incomplete. In comparing bought by Boston Scientific.25 According to his own more mundane stories of cooperative exchange with accounts of fledgling entrepreneurs 24 Professor Robertson, now a dean in the Stanford School slipping into the offices of established firms to of Engineering, did not recall Froix specifically. “There were borrow slack resources on the late shift, Koning so many people who contacted me over the years,” he ex- doubted the underlying truth of these anecdotes. plained, “I can’t remember them all. I have no reason to believe the story isn’t true.” Darst corroborated Froix’s de- While such stories might be true to a point, he scription via email. contended, surely there was always some form 25 Froix supported other inventors as he had been supported. “When I was running Quanam, I met a physicist on the tennis courts. He had some ideas about a new approach to a surgical cutting device. I made the Quanam labs available creative people around. Neither I nor Quanam had any pro- to him to carry out some of his experiments and to evaluate prietary interest in his technology, nor did we desire any such prototypes of his devices. My view on this was, and still is, it’s interest. Understanding the science of what he was doing and always a lot of fun and it is very stimulating to have bright being in a position to help him was the only consideration.”

Padgett_rev-proof.indb 533 6/19/12 11:01 AM 534 ■ Chapter 17

Froix

Raychem Corp. Raychem Ltd. (no rm)

Figure 17.9 Raychem component, Silicon Valley’s third of inventions. Raychem did not aggregate into the 1990 largest component in 1989, by assignee and importance largest component.

of unseen equity relationship underlying this bility is dramatically illustrated by the career of seemingly informal cooperative behavior. Robert Stewart in figures 17.3 and 17.4. Stewart is the only inventor who integrates the three ma- Boston-Specific Reasons for Non-aggregation jor subcomponents at DEC. He (2004) indicated We found seven reasons for Boston’s non- that his integrating role arose from his popular- aggregation. First, MIT graduates tended to take ity as a design reviewer across different DEC academic and private sector jobs outside the Bos- product lines. While these design reviews did ton area, despite the wide variety of academic not create the observed ties, they made Stewart opportunities available there. Second, MIT gradu- and other technical leaders aware of where the ates went to smaller firms in the medical device in- experts were located in the corporation. When dustry, so their mobility did not link large clusters. Stewart or other smart colleagues had a question Third, continued aggregation of the DEC or problem that might benefit from collabora- component was hampered by management’s tion, they knew whom to contact. These contacts encouragement of internal rivalry and compe- then resulted in the observed ties. As illustrated tition. Fourth, engineers at Honeywell, another by the red color of Stewart’s ties, however, they large component in the time period, only collab- are all five years old.T he abruptness of DEC’s orated with Intel inventors and other Honeywell structural disintegration was caused by the inventors outside the region. Fifth, the pensions product life cycle. DEC’s lawyers generally filed at older firms penalized mobility. Sixth, the heav- all necessary patents the night before a product ily academic focus of the Boston area resulted in shipped. In this case, the upper and right ties had less emphasis on patenting and more on the pub- been created with the shipment of the Nautilus lication of scientific papers. Finally, some firms project in early January 1986. In addition to the patented reluctantly in order to control costs. Nautilus project ties, the lower left tie had been Whereas the IBM component emerged by one of many collaborations between Stewart 1987 to serve as the underlying foundation of the largest Valley component in all subsequent years, the composition of the largest Boston to become the largest in Boston in 1991. Thereafter, the DEC component resumed its rank as largest in 1992, only to be component shifted from one year to the next displaced a final time in 1993 by the merging of one portion until 1993.26 The immediate cause of this insta- of the former 1989 largest component with several other mid- sized components to create a single aggregation of inventors across organizations as diverse as MIT, Polaroid, Reebok, 26 The GTE/Siliconix component, which was 2nd largest Kopin Corp., Motorola, Mobile Oil, and United States Surgi- in 1989 and 1990, actually displaced the DEC component cal Corporation, among many others.

Padgett_rev-proof.indb 534 6/19/12 11:01 AM Why the Valley Went First ■ 535

and the R&D and networking groups and just Moreover, within the larger biotech industry, happened to expire at the same time. Cohen felt that the medical device business was During the 1985–89 window, the largest quite distinct from the pharmaceutical business. component in the Boston network consisted The smaller end market for devices tended to primarily of MIT affiliates. Richard Cohen27 of sustain much smaller, less generously funded, the Division of Health Sciences and Technology and perhaps more insular companies. The served as a key bridging point among these indi- smaller scale of medical device efforts is con- viduals. Reflecting on his involvement on a 1985 sistent with Froix’s Valley experience, where he “cut-patent”—a patent for which collaborator was able to commercialize breakthrough medi- ties were not renewed or reinforced by subse- cal technology without the complete resources quent patenting activity within the next five-year of a large firm. As a result of the typical transfer window—Cohen observed that nearly all of his and development into smaller firms, we would collaborators on patents between 1985 and expect less aggregation. 1990 were graduate students from his lab who Patenting policies also influenced the second- left the Boston region altogether on complet- largest connected component in Boston— ing their degrees and research responsibilities at composed largely of scientists and engineers at MIT. Their employment destinations included General Telephone and Electric (GTE)—during universities, hospitals, and, less frequently, busi- both the 1985–89 and 1986–90 windows. We nesses across the country and abroad. Cohen ac- asked GTE inventors Alfred Bellows and Jakob knowledged that his particular division of MIT Maya28 why the GTE component did not aggre- had not kept many of its own graduates, despite gate to rise in size rank from 1989 to 1990 and, the fact that they often proved to be some of the more significantly, why it did not persist as the most compelling candidates on the job market largest connected component after displacing several years later (when they had become too the DEC component in 1991. They explained senior and well compensated to be drawn back). that people at GTE (and in the lighting technol- Cohen’s comments imply that elite universities ogy field more broadly) typically view patents as might actually have less influence on local ag- very costly (for example, one quarter of a million gregation than non-elite universities, since their dollars to internationally patent a single inven- graduates are more likely to leave the area in tion on an ongoing basis), so the culture of the search of comparably elite positions. industry is to limit them to genuinely innovative Nonetheless, based on his experiences at MIT work for which the protection is thought abso- and as the founder of Cambridge Heart, Inc., lutely necessary. Success in research on lighting Cohen reported that because biotech informa- technology has been carried out with and ben- tion flows quite freely within the academic com- efited from a high level of cross-fertilization munity, it is a particularly fertile environment between scientists in industry and academia (es- for “proof of concept” research. Given that pecially for government-contracted research and Boston technology relies to a much greater ex- development). This work routinely generates tent on university patents and published science, papers, however, rather than patents.29 Maya es- its technical social networks might actually be timated, based on his own patent collaborator more connected than the Valley’s. On the other network graph from 1985–89, that the true size hand, Cohen also believed that academic interest of his portfolio of collaborative relationships at in new ideas tended to shift from the successful the time was about three times what we had de- proof of one concept to another without sus- picted, noting specifically that he had as many taining the creation or exchange of knowledge through the subsequent design or development of commercial products. Compounding this 28 Alfred Bellows (control to Charles Kaufman) is currently working with OSRAM Opto Semiconductors. At GTE, problem, according to Cohen, the businesses Bellows was engaged in R&D projects relating to inorganic that did bring such products to market inhibited chemistry and the properties of materials such as ceramics any information flow specific to their commer- and silicon nitride. Jakob Maya (control to Paul Koning) holds a Ph.D. and is currently leading research in lighting cialization processes. technology at Matsushita Electric Works R&D Lab. Before joining Matsushita, Maya was a director of R&D at GTE. 29 Our patent data support this assertion. Patents also cite non-patent references, and these are mostly peer-reviewed 27 Richard Cohen (control to Radia Perlman) holds an scientific papers (Sorenson and Fleming 2004). Since 1975, M.D. and Ph.D. Dr. Cohen applies physics, mathematics, en- Boston patents cited 30 percent more science papers on gineering, and computer science to problems in medicine and average than Valley patents did. Boston also had a greater health. He helped found Cambridge Heart and is the Whita- proportion of academic patents over the entire time period ker Professor in Biomedical Engineering at MIT. as well.

Padgett_rev-proof.indb 535 6/19/12 11:01 AM 536 ■ Chapter 17

papers with other authors (and at times not in knowledge elsewhere (publicly or otherwise). the same firm) as he did patents. Second, the Our patent data strongly support Joyce’s descrip- GTE component in Boston, already relatively tion of Honeywell’s insularity. Of the eighty-one weak, was probably made even weaker when inventors in the 1986–90 window, eleven had col- GTE Sylvania sold its lighting business to Sie- laborated on one or two of three non-Honeywell mens’s Osram in 1992, though this might have patents, while Honeywell held the ninety-one re- temporarily connected Siemens and GTE. Con- maining patents linking this component. sistent with Froix’s description of Raychem’s im- Kaufman, Koning, and Perlman also empha- plosion, Maya reported that people spent several sized how organizational culture influences the years thereafter worried far more about simply level of patenting, noting that DEC’s explicit keeping their jobs than about the quality, rate, or patenting policies motivated them to identify volume of their inventive work.30 their patentable work proactively. They felt that The trade-offs between public science and these policies implicitly encouraged employees private technology also influenced the collab- to identify other collaborators for each of their orative linkages of Honeywell, the sixth largest patents, partly because DEC awarded the full connected component, though explicit career patent bonus amount of $500 to as many as considerations also mattered. Thomas Joyce,31 three inventors per patent. So those with ideas to a lifetime employee at Honeywell (1960–2000), patent were often inclined to seek out collabora- provided three reasons why the Honeywell com- tors (whether needed or not) in order to “share ponent did not aggregate to rise in size rank from the wealth” and to encourage others to “return 1989 to 1990. (In fact, it dropped from fifth to the favor.” Additionally, DEC granted a steeper sixth in the following year.) First, collaboration set of awards for cumulative patenting ($5,000 at Honeywell tended to be global rather than for 5 patents, $10,000 for 10, up to as much as local; Joyce recalls working with a number of $20,000 for 20, or perhaps even $25,000 for European Honeywell employees at the time but 25), and these awards allowed for any number never exchanging information with anyone out- of collaborators per patent. Kaufman also noted side Honeywell, regardless of region. He attrib- that DEC displayed a cyclical pattern based on uted this fact partly to the nature of Honeywell’s patenting objectives that were established in technology and partly to his own personal situ- response to a cross-licensing relationship with ation, as both his own skill set and Honeywell’s IBM, which would grant a company the use of development opportunities were constrained by all IBM-patented technologies in exchange for the distinctly proprietary nature of the chip de- IBM’s right to use that company’s patented tech- sign work being done there. nologies. Because IBM’s fee for this arrangement Second, Joyce noted that he was linked to a was inversely proportional to the size of the com- comparatively more mature cohort of inventors, pany’s portfolio of patents, DEC business man- “older hangovers from the 1960s and 1970s,” agers recognized a value to patents exceeding many of whom had more pressing family con- licensing revenue or protection from imitation. cerns or were nearing a reasonable age for retire- It would seem that these policies would have ment. Honeywell, like other Boston firms, made increased collaborations and made the DEC its pensions contingent on retirement with the component larger and more robust. Saxenian firm, which certainly would have inhibited these (1994) and others, however, have commented older employees from leaving and thereby served on the less collaborative norms within and as bridges to link the Honeywell component to across Boston firms. Paul Koning confirmed this other Route 128 components.32 reputation, describing how Ken Olsen, DEC’s Third, Joyce added that Honeywell’s chip founder and CEO, routinely created compet- designers found themselves “under the secrecy ing internal groups as a means of fueling rapid cloak of Intel by the early 1990s”; collaborat- progress. Koning went on to note that the prac- ing with Intel prevented Honeywell from sharing tice severely strained internal morale and inter- departmental cooperation. Furthermore, given 30 Maya left GTE just prior to this change because he an- that patent law clearly stipulates that only con- ticipated it; he would have stayed otherwise. tributing inventors be listed on a patent, the 31 Thomas Joyce (control to Radia Perlman) worked as a logic designer and patented repeatedly at Honeywell, Honey- collaborative awards policy may have been of well Bull, and Bull until his retirement. limited effectiveness. This might account for the 32 Preliminary conversations with two Harvard Business persistent fragility in the DEC’s networks and is School accounting professors, Paul Healy and Greg Miller, indicated great plausibility for this argument, although they were consistent with its reputation for fostering com- unaware of any specific citation in the accounting literature. petition between work groups.

Padgett_rev-proof.indb 536 6/19/12 11:01 AM Why the Valley Went First ■ 537

Taken collectively, these inventors’ comments institutions in the aggregation of regional inven- broadly suggest that the corporate policies and tor networks. Consistent with the themes of this strategies of the dominant firms in the Boston volume, universities and postdoc programs play region at the time often served to blunt aggrega- a catalytic role in the initial connections between tion both within and across firms. However, in- components. This catalytic role creates opportu- vention also stagnated at these firms due to more nities for inventors (particularly young inven- sweeping strategic business decisions—pursuing tors) to forge bridging opportunities. Second, we proprietary technologies (at DEC, Data Gen- are not struck by any fundamental differences in eral, and Honeywell) and selling ownership to the network structures of Silicon Valley and Bos- an acquiring firm (at GTE and Honeywell). In ton. To quantify these impressions, we explore the cases of proprietary technology, invention and demonstrate that the micro-level structure suffered as firms struggled with the negative eco- of collaboration in the Valley is on average simi- nomic outcome of their decision, while inventors lar and sometimes less robust than that in Bos- were constrained in their careers by proprietary ton. Finally, we will collect our impressions of skill sets. In the cases of acquisition, it is reported the differences between Boston and the Valley that many inventors left their respective fields, and comment on the Saxenian argument that the retired, or focused more effort on keeping their Valley is more networked. jobs than on inventing. In the Valley, by contrast, IBM’s postdoc program enabled young inven- inventors entered the external labor market with tors to move across inventor components and sellable skills because technologies were less pro- explore new combinations of technologies and prietary (Angel 1989; Fallick et al. 2006). ideas. IBM modeled its program on Bell Lab’s At the same time, the slow pace of intra- postdoc program (which, after the breakup of organizational job movement was certainly not AT&T, no longer exists). When asked why IBM a function of limiting proprietary skill sets or supported such a program, William Risk and organizational upheaval alone. The majority of John Campbell Scott provided a variety of rea- Boston region inventors stressed firmly that their sons and motivations. First, the postdocs pro- decision to remain in the same firms was primar- vided cheap labor. Second, new people with fresh ily due to their satisfaction with both their work ideas were seen as valuable. Third, IBM assumed opportunities in those organizations and the that such people would depart as ambassadors way in which those organizations treated them for the firm. Risk and Scott did not mention the as engineers and scientists. In fact, when these concerns about loss of proprietary information individuals finally left their firms (and any others expressed by Hans Ribi and Pyare Khanna. Part subsequently in their careers), they reported that of this reflects IBM’s academic and admittedly it was almost always because they saw no viable “ivory tower” attitudes at the time. It also re- alternative; the organizations were either chang- flects founder and time period effects for the ing ownership or failing visibly. Naturally, many Almaden Lab in the 1960s. IBM operated as a of these economic failures can be attributed in virtual monopoly then. According to Scott, “the part to these firms’ proprietary technological research division was set up by scientists with strategies. Thus there are two distinct ways in foresight.” Their foresight had an impact well which the decision to remain with proprietary beyond IBM. (IBM has since reduced the post- development hindered the growth of collabora- doc program due to the firm’s financial problems tive inventor networks in the Boston region. At in the early 1990s. Other firms, however, such as the individual level, proprietary technology lim- Hewlett Packard, have begun similar programs ited the job mobility of some, and at the orga- [Fleming et al. 2005]). nizational level, it contributed significantly to The institutional support of mobility by the ultimate failure or disruptive acquisition of young inventors appears to have greatly fostered at least three dominant firms in the area—DEC, their careers and, in turn, knowledge flow across Data General, and Honeywell. firms in theV alley. Modeling at the inventor level of analysis also indicates that brokerage opportunities are most fruitful for young and rela- Discussion tively inexperienced inventors (Fleming, Mingo, and Chen 2007). After an inventor has gained a As with all qualitative data, our presentation breadth of creative experience, she gains greater and analysis remain inseparable. Nonetheless, marginal benefits by collaborating cohesively we wish to highlight three issues in our discus- because she brings non-redundant information sion. First, we are struck by the importance of that offsets the insularity of closed networks. It

Padgett_rev-proof.indb 537 6/19/12 11:01 AM 538 ■ Chapter 17

Cohen

Dana-Farber Cancer Institute, Inc. Massachusetts Institute of Technology Foxboro Company Kopin corporation

Figure 17.10 Boston’s largest component in 1989, by assignee and importance of inventions. The MIT component did not aggregate into the 1990 largest component.

is interesting, though probably non-causal, that components indicate little difference, except that inventors are most mobile early in their careers, the second-largest component is more robust in when they can most benefit from exposure to Boston (GTE) than in the Valley (and, indeed, new ideas and technologies. is by far the most robust of any component we In the course of our interviews and graphical analyzed). Appendix B describes the robustness exploration of collaboration networks, we also analyses in detail. The analyses suggested that perceived that Boston networks were less dense the Valley’s greater degree of aggregation was and robust than Valley networks. Whereas the not caused by a fundamental difference in the IBM component emerged by 1987 to serve as the microsocial structure of its collaborative net- underlying foundation of the largest component work. Indeed, the analyses (and even a visual in all subsequent years in the Valley, the compo- comparison of the figures) indicated that the top sition of Boston’s largest component continued six components of the two regions were quite to shift from one year to the next until 1993 similar, with the exception that the GTE/Silico- when the Digital component was permanently nix component was more densely networked displaced. Figure 17.10 illustrates another dra- than its Valley counterpart. matic example of this process, the disintegration Finally, we sought to understand whether of the MIT/Foxboro/Dana-Farber component, Boston and the Valley had different information Boston’s largest component in 1985–89. Its flows. We are struck by the bi-modal distribution red ties mark the patents that had expired by of attitudes on the issue, mainly along profes- the following year (basically, patents that had sional lines and independent of the region. Most been applied for in 1985). This illustrates how of the inventors from both regions expressed the component lost important bridging ties and similar laissez-faire, open, and positive atti- completely fell apart. Given that this disinte- tudes toward information flow. Many of their gration process would support the Saxenian stories described an effort to evade efforts by arguments for Silicon Valley’s more densely net- management to contain their boundary-crossing worked social structure, we tested the hypoth- collaborations. The most strident concerns esis that the Valley components were indeed about the leakage of proprietary information more robust. Surprisingly, we found the oppo- through collaborative relationships and extra- site: paired comparisons across similarly ranked firm networks actually came from threeV alley

Padgett_rev-proof.indb 538 6/19/12 11:01 AM Why the Valley Went First ■ 539

interviewees—Hans Ribi, Pyare Khanna, and (to adequately address situations where business a lesser extent) Salvador Umatoy. and engineering interests are at odds. Likewise, Khanna explicitly described spillovers as bad, there is always a delicate balance between the saying that it took one year to train a scientist, desire to rely on public standards to protect after which he preferred to keep the scientist in proprietary decisions and the need to disclose isolation. He felt that the important connections proprietary decisions in order to institute those across the firm boundary were at his level and standards in the first place. As Koning put it, “It that scientists should work in silos. He sends his gets to be a very interesting dance. Sometimes it people to conferences, but only outside the Valley, feels more like diplomacy than engineering.” to prevent poaching by rival Valley firms. At one Taken collectively, these inventors’ comments time his firm had been in Concord, California, suggest that simple characterizations of Bos- outside the traditional commuting distance of ton secrecy and autarky versus Silicon Valley Silicon Valley. He preferred this location because cooperation and interdependence fail to reflect salaries were 20–30 percent lower and people the tension between managers and engineers were less likely to leave. He remained noncom- on both coasts. Both communities struggled as mittal about why he subsequently moved his firm they sought a practical and productive balance to Fremont (a city considered within the confines between making money, promoting public stan- of the Valley), merely commenting: “Here there dards, and collectively solving problems. While is the nucleus of growth.” He opined that Ken- unwanted spillovers certainly detract from the dall Square (a popular public plaza near MIT in value of location in fast-paced technological Cambridge) in contrast, had no industry, only regions like Boston and Silicon Valley, there universities.33 Khanna also remained noncom- are clearly many counterbalancing attractions. mittal about the classic argument for location Managers can identify and attempt to keep their in technologically dynamic regions, namely the firm’s mobile gatekeepers, but ultimately, and availability of technical personnel (Angel 1989). particularly in regions that do not enforce non- The inventors in the Boston region noted a competes or trade secret law, their options re- similar tension between managers and engineers main limited (Fleming and Marx 2006). regarding the decision to share information. “At Digital,” Kaufman explained, “management thought we had all these great secrets to con- Conclusion ceal; the engineers knew that the value was in collaboration.” Koning felt that the core of the Why do regional inventor networks aggregate or issue could be found in the underlying multiplic- disintegrate? And what influence does such ag- ity of purposes for patenting. For example, an gregation have upon knowledge flows and cre- inventor might wish to patent a technology as ativity? We found many mechanisms that hamper a means to block its development by others in aggregation, including the breakup of firms and order to monopolize its sale or licensing. Alter- the related uncertainty that saps morale and pro- natively, an inventor might patent as a means to ductivity; the dispersal of graduates to jobs out- steer the technology’s subsequent development side the region; the departure of senior inventors by others via “licensing on very generous terms” to start-ups and self-employment rather than to in order to acquire a first-mover/first- to-market other established firms; company policies that advantage. (The latter motive is far more com- discourage collaboration; discrete product life mon for inventions that lend themselves to open cycles; and proprietary strategies that make col- standards and/or enjoy network effects, such as laboration unproductive. We found fewer influ- the computer networking hardware and soft- ences that enhance aggregation. These include ware with which Koning is most familiar.) As collaboration across academic and firm bound- both an engineer and an entrepreneur himself, aries; collaboration within large firms; hiring Koning believed that in most cases, both motiva- local university graduates; and postdoc fellow- tions reflect the same basic principle: “You dis- ships that seed local businesses with technically close x or license y because you make a business trained personnel. In the particular case at hand, or engineering decision that the gain is greater Silicon Valley aggregated before Boston because than the loss.” Naturally, this heuristic may not Stanford graduates took employment at IBM’s Almaden Valley Labs and because IBM spon- 33 An observation that is out-of- date, as any stroll through sored a postdoctoral program that seeded the Kendall Square would reveal. Valley with IBM patent coauthors. In contrast,

Padgett_rev-proof.indb 539 6/19/12 11:01 AM 540 ■ Chapter 17

MIT graduates did not take employment at GTE, did not create spurious cutpoints). Given the DEC, Data General, or Honeywell, and none of sensitivity of the measures to cutpoints, generat- those firms sponsored collaborative programs ing false negatives remains preferable to generat- like that at IBM. These differences were reflected ing false positives or to incorrectly matching two in the generative ecologies of the two regions: different inventors. Silicon Valley mobility increased the possibility The analyses presented relied on all patents of knowledge spillovers between firms and tech- with at least one inventor within the region. Thus nologies. We found the attitudes of engineers to- if inventors from inside and outside a region co- ward spillovers to be remarkably similar in the authored a patent, the patent (and both inven- two regions, however. Engineers appear eager to tors) would appear in each region. To explore the share ideas and facilitate creativity, independent sensitivity of this definition, we regraphed all data of their location. with the more exclusive definition that did not include inventors from outside the region. While the graphs and network diagrams were generally Appendix A: Matching Algorithm smaller (as might be expected, since there will be at most the same number of inventors in each), We extracted source data on all granted U.S. pat- the qualitative results were unchanged. ents from 1975 through 2002 from the United States Patent Office (USPTO) Cassis product, and MSA data for 2003 (ZIPList5 MSA 2003). Appendix B: Patent Every patent includes all inventors’ last names Robustness Analysis (with varying degrees of first and middle names or initials), inventors’ hometowns, detailed in- One obvious explanation for the greater ag- formation about the invention’s technology in gregation in the Silicon Valley network is that subclass references (there are over 100,000 sub- its components were more robust. We tested classes), and the owner or assignee of the patent this hypothesis at the inventor level of analysis (generally a firm and less often a university, if and then at the patent level of analysis. Figures not owned by the inventor). Since the USPTO in- 17.11 and 17.12 illustrate the inventor level of dexes source data by patent number, we devised analysis for the largest and second largest com- an inventor-matching algorithm to determine ponents in the regions. (Illustrations for the third each inventor’s patents and the other inven- through sixth largest component comparisons tors with whom the focal inventor has coau- looked qualitatively similar to those for the larg- thored at least one patent. The database includes est component and are not shown.) The y-axis 2,058,823 inventors and 2,862,967 patents (for of these illustrations is the proportion of nodes description of more sophisticated algorithms and that remains connected in the largest resulting public accessible database, see Lai et al. 2011). component after a proportion of the original The matching algorithm refines previous ap- nodes have been removed.The x-axis represents proaches (Newman 2000). If last names match, the proportion of original nodes that is removed. first initials and middle initials (if present) must Consider figure 17.12 first, illustrating the sec- then match. Whole first names and whole middle ond largest components. The point 0.05 on the names (if present) are then compared. If all these x-axis indicates that 5 percent of the nodes have comparisons are positive, the algorithm then re- been removed from what were originally the sec- quires an additional non-name similarity: home- ond largest components of Boston and the Valley. town and state, corporation (via assignee codes), At this point, the y-axis indicates that the mini- or technology (via technology subclassifica- mum proportion of nodes that remain connected tions). We also implemented a common name in the reduced largest component is about 30 parameter that ignored the additional match re- percent for the Valley and well over 40 percent quirement if the last name made up less than .05 for Boston. The graphed points are summary sta- percent of the U.S. population, as determined by tistics (minimum, median, and maximum) of 50 the U.S. Census Bureau. samples for each data point. We sampled to avoid For 30 randomly selected inventors, the algo- the combinatorial explosion of exhaustively cal- rithm correctly assigned 215 of their 226 patents culating all possible choice combinations. (as determined by résumé searches and personal Figure 17.11 reveals very similar robustness contact). The 11 incorrectly determined patents for the two regions. Figure 17.12, however, il- were assigned to four isolated nodes (i.e., they lustrates that the Valley component is more

Padgett_rev-proof.indb 540 6/19/12 11:01 AM Why the Valley Went First ■ 541

1.0 max Silicon Valley Boston 0.8

median 0.6

0.4

tion of Nodes Remaining min or 0.2 Prop

0 0 0.2 0.4 0.6 0.8 1.0 Proportion of Nodes Removed

Figure 17.11 Size of component after removal of specified y-axis illustrates the proportion of nodes that remains proportion of component’s nodes, for Boston and Silicon connected in the largest resulting component after a Valley’s largest components in 1989. The x-axis represents proportion of the original nodes have been removed. the proportion of original nodes that is removed. The

1.0 Silicon Valley max Boston 0.8

0.6 median

0.4 tion of Nodes Remaining

or min 0.2 Prop

0 0 0.2 0.4 0.6 0.8 1.0 Proportion of Nodes Removed

Figure 17.12 Size of component after removal of specified connected in the largest resulting component after a proportion of component’s nodes, for Boston and Silicon proportion of the original nodes have been removed. As Valley’s second largest components. The x-axis repre- can be seen on the lower left, Silicon Valley was more sents the proportion of original nodes that is removed. vulnerable to node removal than Boston. The y-axis illustrates the proportion of nodes that remains

vulnerable to the loss of a few nodes. The steep simply more dynamic, breaking and re-forming initial drop in figure 17.12 for Silicon Valley indi- nodes much more quickly, which is probably a cates that the loss of a few key inventors quickly reflection of its greater career mobility. breaks the component up into much smaller To confirm our results, we repeated the analy- pieces—similar to what is illustrated in figures sis at the patent level. For each of the components, 17.3 and 17.4. Silicon Valley appears to be we examine the extent to which the component

Padgett_rev-proof.indb 541 6/19/12 11:01 AM 542 ■ Chapter 17

would be disconnected by the removal of each less damage to the connectivity of the network. patent. We define the extent to which a com- The maximum possible value would exist in a ponent is disconnected by the proportion of in- component where all inventors were coauthors ventor dyads in that component that would no on one patent and no other coauthorships ex- longer be able to reach one another after the pat- isted. In this case the removal of the one shared ent is removed. We find this measure by consid- patent would result in the disconnection of all ering each of these components individually and inventor dyads. then calculating for each patent: We measure the vulnerability of each network by taking the mean proportion of inventor dy- K / (/nN)2 , ads disconnected by each patent. As stated ear- c = 1 lier, the maximum value of this number is 1.0 for individual inventors. Calculating the maximum where N is the number of inventors in the origi- value for the mean of patents in a component nal component, c is a component created by the is considerably more complex and beyond the removal of a patent, n is the number of inven- scope of this chapter. However, since the maxi- tors in a component c existing after a patent is mum possible value will be related to the com- removed, and K is the number of components in ponent size, caution should be exercised when the post-removal network. comparing mean values across components of This measure yields a high value when the different sizes. removal of a patent results in the creation of Table 17.2 illustrates robustness results. As many new components and the inventors are di- the low numbers suggest, most patents within vided equally among components. For example, each component can do only minimal dam- if the removal of a patent divides a component age to the network. What is most striking is into ten smaller components with one-tenth of the lack of systematic difference across the two the inventors in each component, this results regions. The mean vulnerability over all the in 0.9 of dyads being disconnected. However, Boston components is 0.0241; over all Silicon if the removal of a patent results in a similar Valley components it is 0.0272. Consistent with number of components but with inventors less the inventor-level analysis, the second compo- evenly spread among them, the value generated nent appears to be much more robust in Boston, by this measure will be smaller. For example, relative to all other components in both Boston given a component of 100 inventors, if the re- and the Valley. moval of a patent breaks the component into 10 Both of these analyses suggest that the Val- components with 9 of these being isolates and ley’s aggregation did not occur because its com- 91 inventors in the remaining component, then ponents were more robust and able to merge 0.171 of dyads are disconnected, indicating far with other components.

Table 17.2. Patent Analysis of Component Robustness Component Component Vulnerability No. of Patents Maximum

Boston 1 .0212 (.0763) 208 .52 Boston 2 .0074 (.0231) 345 .20 Boston 3 .0301 (.0762) 123 .49 Boston 4 .0179 (.0806) 182 .65 Boston 5 .0226 (.0610) 116 .35 Boston 6 .0451 (.0989) 45 .46 Silicon Valley 1 .0311 (.0757) 159 .49 Silicon Valley 2 .0208 (.0552) 161 .45 Silicon Valley 3 .0209 (.0477) 107 .38 Silicon Valley 4 .0330 (.0950) 131 .52 Silicon Valley 5 .0338 (.0729) 60 .49 Silicon Valley 6 .0237 (.0712) 78 .54

Note: Component vulnerability is the mean number of the proportion of inventor dyads disconnected by the removal of each patent within a given component (higher values indicate more vulnerable components). Standard deviations in parentheses.

Padgett_rev-proof.indb 542 6/19/12 11:01 AM Why the Valley Went First ■ 543

References Florida, R., and M. Kenney. 1990. The Breakthrough Illusion. New York: Basic Books. Albert, M. B., F. Narin, D. Avery, and P. McAllister. Gilson, R. 1999. “The Legal Infrastructure of High 1991. “Direct Validation of Citation Counts as Technology Industrial Districts: Silicon Valley, Indicators of Industrially Important Patents.” Re- Route 128, and Covenants Not to Compete.” New search Policy 20:251–59. York University Law Review 74(3):575–629. Almeida, P., and B. Kogut. 1999. “Localization Glaeser, E., H. Kallal, J. Scheinkman, and A. Shleifer. of Knowledge and the Mobility of Engineers 1992. “Growth in Cities.” Journal of Political Econ- in Regional Networks.” Management Science omy 100(6):1126–53. 45:905–16. Graham, R. 1989. “’Miracle’ Proves a Mirage: New Angel, D. 1989. “The Labor Market for Engineers in Arrivals Find Too Little of High-Tech Boom, Too the U.S. Semiconductor Industry.” Economic Geog- Late.” Boston Globe, July 28, p. 13. raphy 65(2):99–112. Granovetter, M. 1973. “The Strength of Weak Ties.” Angel, D., and J. Engstrom. 1995. “Manufacturing American Journal of Sociology 78:1360–80. Systems, and Technological Change: The U.S. Per- Grove, A. 1990. “The Future of the Computer Indus- sonal Computer Industry.” Economic Geography try.” California Management Review 33(1):148–61. 71(1):79–102. Gulati, R., and M. Gargiulo. 1999. “Where Do Inter- Batagelj, V. and A. P. Mrvar. 1998. “A Program for organizational Networks Come From?” American Large Network Analysis.” Connections 21:47–57. Journal of Sociology 104(5):1439–93. Pajek version 0.91, 2003, http://vlado.fmf.uni-lj.si/ Gura, T. 2004. “Joining a Trend, Scientists Increas- pub/networks/pajek/. ingly Say ‘Call My Agent.’” Science 303:303–5. Beam, A. 1989. “Our Technology Graveyard.” Boston Henderson, V. 1997. “Externalities and Industrial Globe, July 26, p. 21. Development.” Journal of Urban Economics 42: Burt, R. 1992. Structural Holes. Cambridge, MA: Har- 449–70. vard University Press. Henderson, V., A. Kuncoro, and M. Turner. 1995. “In- Clark, D. 1991. “Newcomers Put Stamp on Silicon dustrial Development in Cities.” Journal of Political Valley.” San Francisco Chronicle, July 8, p. B1. Economy 103(5):1067–90. Cohen, F. 2003. Personal interview. February 13, Jacobs, J. 1970. The Economy of Cities. New York: Natick, MA. Random House. Cohen, W., and D. Levinthal. 1990. “Absorptive Ca- Kenney, M., and U. von Burg. 1999. “Technology, En- pacity: A New Perspective on Learning and Innova- trepreneurship, and Path Dependence: Industrial tion.” Administrative Science Quarterly 35:128–52. Clustering in Silicon Valley and Route 128.” Indus- Fallick, B., C. Fleischman, and J. Rebitzer. 2006. trial and Corporate Change 8:67–103. “Job-Hopping in Silicon Valley: Some Evidence Krugman, P. 1991. “Increasing Returns and Eco- Concerning the Micro-Foundations of a High Tech- nomic Geography.” Journal of Political Economy nology Cluster.” Review of Economics and Statis- 99(3):483–99. tics 88:472–81. Lai, R., A. D’Amour, A. Yu, Y. Sun, and V. Torvik. Feldman, M. 2001. “The Entrepreneurial Event Revis- 2011. “Disambiguation and Co-author Networks ited: Firm Formation in a Regional Context.” In- of the U.S. Patent Inventor Database.” Working pa- dustrial and Corporate Change 10(4):861–92. per, Harvard Business School. Fleming, L. 2001. “Recombinant Uncertainty in Techno- Leslie, S., and R. Kargon. 1996. “Selling Silicon Val- logical Search.” Management Science 47(1):117–32. ley: Frederick Terman’s Model for Regional Advan- Fleming, L., and K. Frenken. 2006. “The Evolution of tage.” Business History Review 70:435–72. Inventor Networks in the Silicon Valley and Bos- Marshall, A. 1920. Principles of Economics. 8th ed. ton Regions.” Advances in Complex Systems 10(1): London: Macmillan. 53–71. Marx, M., J. Singh, and L. Fleming. 2011. “Regional Fleming, L., and M. Marx. 2006. “Managing Creativ- Disadvantage: Non-Competes and Brain Drain.” ity in a Small World.” California Management Re- Working paper, MIT Sloan School of Management. view 48:6–27. Marx, M., D. Strumsky, and L. Fleming. 2009. “Mo- Fleming, L., D. Chen, and S. Mingo. 2007. “Collaborative bility, Skills, and the Michigan Non-Compete Ex- Brokerage, Generative Creativity, and Creative Suc- periment.” Management Science 55:875–89. cess.” Administration Science Quarterly 52:443–75. McCormick, J. 1999. “Asia’s Best and Brightest Flock to Fleming, L., C. King, and A. Juda. 2007. “Small Silicon Valley.” Electronic Business, http://www.reed Worlds and Regional Innovation.” Organization -electronics.com/eb-mag/index.asp?layout=article Science 18(2):938–54. &articleid=CA65539&rid=0&rme=0&cfd=1. Fleming, L., M. Thursby, and J. Quinn. 2005. Mowery, David C., Richard R. Nelson, Bhaven N. “HP Nanotech: Partnership with CNSI.” Har- Sampat, and Arvids A. Ziedonis. 2001. “The vard Business School case 606-045. http://hbr Growth of Patenting and Licensing by U.S. Uni- .org/product/hp-nanotech-partnership-with-cnsi/ versities: An Assessment of the Effects of the Bayh- an/606045-PDF-ENG. Dole Act of 1980.” Research Policy 30(1):99–119.

Padgett_rev-proof.indb 543 6/19/12 11:01 AM 544 ■ Chapter 17

Newman, M. 2000. “The Structure of ScientificC ol- ———. 1999. “Comment on Kenney and von Burg, laboration Networks.” Proceedings of the National ‘Technology, Entrepreneurship, and Path Depen- Academy of Sciences of the United States of Amer- dence: Industrial Clustering in Silicon Valley and ica 98:404–9. Route 128.’” Industrial and Corporate Change Nohria, N. 1992. “Information and Search in the Cre- 8:105–10. ation of New Business Ventures: The Case of the 128 Schweikart, L. 2000. The Entrepreneurial Adventure: Venture Group.” In Networks and Organizations: A History of Business in the United States. Fort Structure, Form, and Action, ed. N. Nohria and Worth, TX: Harcourt College Publishers. R. Eccles. Boston: Harvard Business School Press. Simon, J. 1989. “Job Chill on 128: With Cutbacks and Owen-Smith, J., and W. Powell. 2004. “Knowledge Slowed Growth, the Hiring Sizzle Is Gone.” Boston Networks as Channels and Conduits: The Effects Globe, April 30, p. A1. of Spillovers in the Boston Biotechnology Commu- Singh, J. 2005. “Collaborative Networks as Determi- nity.” Organization Science 15(1):5–21. nants of Knowledge Diffusion Patterns.” Manage- Packard, D. 1995. The HP Way: How Bill Hewlett and ment Science 51(5):756–70. I Built Our Company. New York: HarperCollins. Sorenson, O., and L. Fleming. 2004. “Science and the Piore, M., and C. Sabel. 1984. The Second Industrial Diffusion of Knowledge.” Research Policy 33(10): Divide: Possibilities for Prosperity. New York: Ba- 1615–34. sic Books. Stein, C. 1989a. “128: Why It Will Never Be the Robertson, P., and R. Langlois. 1995. Firms, Markets, Same.” Boston Globe, January 3, p. 21. and Economic Change: A Dynamic Theory of Busi- ———. 1989b. “Along Route 128, High Tech Hits ness Institutions. New York: Routledge. Lows.” Boston Globe, August 1, p. 1. Rosegrant, S., and D. Lampe. 1992. Route 128: Les- Stern, S., and M. Porter. 2001. “Innovation: Location sons from Boston’s High-Tech Community, New Matters.” Sloan Management Review 42(4). York: Basic Books. Stewart, R. 2004. Personal interview, June 17, Cam- Saxenian, A. 1990. “Regional Networks and the Re- bridge, MA. surgence of Silicon Valley.” California Management Watts, D., and S. Strogatz. 1998. “Collective Dynam- Review 33(1):89–113. ics of Small-World Networks.” Nature 393:440–42. ———. 1994. Regional Advantage: Culture and Com- ZIPList5 MSA. 2003. Downloaded April 24 from petition in Silicon Valley and Route 128. Cam- CD Light, LLC: http://www.zipinfo.com/products/ bridge, MA: Harvard University Press. z5msa/z5msa.htm.

Padgett_rev-proof.indb 544 6/19/12 11:01 AM