INNOVATION ECOSYSTEMS AND THE PACE OF …d1c25a6gwz7q5e.cloudfront.net/reports/2016-02-03... · S-CURVES RON ADNER1 and RAHUL KAPOOR2* 1 Strategy and Management, Tuck School of Business,

Strategic Management JournalStrat. Mgmt. J. (2015)

Published online EarlyView in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.2363Received 16 September 2013; Final revision received 9 September 2014

INNOVATION ECOSYSTEMS AND THE PACE OFSUBSTITUTION: RE-EXAMINING TECHNOLOGYS-CURVES

RON ADNER1 and RAHUL KAPOOR2*1 Strategy and Management, Tuck School of Business, Dartmouth College, Hanover,New Hampshire, U.S.A.2 Management Department, The Wharton School, University of Pennsylvania,Philadelphia, Pennsylvania, U.S.A.

Why do some new technologies emerge and quickly supplant incumbent technologies while otherstake years or decades to take off? We explore this question by presenting a framework thatconsiders both the focal competing technologies as well as the ecosystems in which they areembedded. Within our framework, each episode of technology transition is characterized by theecosystem emergence challenge that confronts the new technology and the ecosystem extensionopportunity that is available to the old technology. We identify four qualitatively distinct regimeswith clear predictions for the pace of substitution. Evidence from 10 episodes of technologytransitions in the semiconductor lithography equipment industry from 1972 to 2009 offers strongsupport for our framework. We discuss the implication of our approach for firm strategy. Copyright© 2015 John Wiley & Sons, Ltd.

INTRODUCTIONWhile the waves of creative destruction regularlycrash on the shores of markets, the pace of sub-stitution varies markedly across different contextsand episodes (e.g., Anderson and Tushman, 1990).Why do some new technologies immediately sup-plant incumbent technologies while others takedecades to take off? Despite the centrality of thephenomenon in the literature, the pace of technol-ogy substitution has been under-explored.

At the level of technologies, the strategy literaturehas focused on whether new technologies will riseto dominate old technologies (e.g., Adner, 2002;Christensen, 1997; Foster, 1986) but largely ignored

Keywords: innovation; substitution; ecosystems; technol-ogy evolution; technology competition*Correspondence to: Rahul Kapoor, The Wharton School, Uni-versity of Pennsylvania, Philadelphia, PA 19104, U.S.A. E-mail:[email protected]

Copyright © 2015 John Wiley & Sons, Ltd.

the question of when dominance will be achieved.Conversely, the diffusion of innovation literaturehas focused on the rate of adoption of new technolo-gies (e.g., Hall, 2004; Rogers, 2003), but has takena static view of the diffusing innovations (i.e., over-looking the continued evolution of both the new andold technologies). At the level of firms, the strategyliterature has focused on how firms’ capabilitiesand experience shape their entry decisions and per-formance in new technologies (e.g., Franco et al.,2009; King and Tucci, 2002; Mitchell, 1989, 1991)but has tended to ignore the pace with which thenew technology substitutes the old. Hence, whilethe emphasis has been on firm-level competition ina new technology, the technology-level competitionbetween the new and old has been overlooked.Absent such a technology-level consideration,the literature is handicapped in offering guidanceregarding how firms should manage the transitionfrom old to new technologies, and in clarifying

R. Adner and R. Kapoor

when persisting with the old technology mayrepresent a more viable strategy than aggressivelypursuing the new technology.

In this paper we argue that understanding thepace of substitution requires joint considerationof the evolution of both the new and the oldtechnologies. In turn, understanding this evolutionrequires an examination of the interdependenciesin the broader ecosystem of components andcomplements in which the focal technologiesare embedded (Adner, 2006, 2012; Adner andKapoor, 2010; Christensen and Rosenbloom, 1995;Hughes, 1983; Iansiti and Levien, 2004; Moore,1993; Rosenberg, 1976, 1982). Developing suchan understanding has important implications forhow firms pursue new and old technology oppor-tunities (e.g., Christensen, 1997; Foster, 1986), andhow firms manage interdependencies within theirecosystems as technologies evolve (e.g., Ethiraj,2007; Kapoor and McGrath, 2014).

We present a structured framework to analyzethe pace of technology substitution that considersthe differential impact of ecosystem on the newand old technologies. On the one hand, bottlenecksthat arise as other ecosystem elements struggle toemerge can act to constrain the competitivenessof the rising new technology (Hughes, 1983;Rosenberg, 1976, 1982)–they create ecosystememergence challenges. On the other hand, advancesin other ecosystem elements that arise after theold technology has matured can act to enhancethe competitiveness of the old technology (Harley,1971; Tripsas, 2008; Utterback, 1994)–they createecosystem extension opportunities. Our joint con-sideration of ecosystem emergence challenges forthe new technology and ecosystem extension oppor-tunities for the old technology allows us to identifyfour qualitatively distinct substitution regimes that,in turn, give rise to a testable prediction regardingthe pace of substitution across these regimes.

We test our theory in the context of the semicon-ductor lithography equipment industry from 1972to 2009, a period during which 10 new technologygenerations were introduced into the market. Thesetransitions offer an interesting puzzle: across the10 generations there was remarkable absence ofvariance across the key factors that the literaturehas identified as driving substitution. In each case,the new technology generation was introducedinto the marketplace on a commercial basis (i.e.,it had overcome its development challenges,e.g., Henderson and Clark, 1990); at the time of

market introduction, the new generation offeredunambiguously superior performance relative tothe predecessor generation on both an absoluteperformance and price-adjusted-performance basis(e.g., Foster, 1986); customers and the semiconduc-tor manufacturing firms were well informed aboutthe availability of the new generation (e.g., Bass,2004; Rogers, 2003); and customers were eagerto adopt higher performance technologies (e.g.,Christensen, 1997). Moreover, each of the new gen-erations preserved the core competences (Tushmanand Anderson, 1986) and complementary assets(Tripsas, 1997) of the lithography equipment firms.Yet despite this constancy in conditions, the paceof substitution varied dramatically across the 10technology generations, ranging from cases ofrapid substitution (market dominance achievedafter a single year), to slow substitution (domi-nance achieved after 10 years), to non-substitution(dominance never achieved).

We draw on a unique array of qualitative andquantitative data collected during two years of field-work to explore this puzzle. We find that ourapproach of explicitly considering the interactionbetween the old and new technology ecosystemsexplains significant variance in the observed paceof substitution in semiconductor lithography. Ourtheory and our observations are at the level of tech-nology. However, through our fieldwork, we arealso able to shed new light on how heterogeneityin firms’ choices and investments shaped the pathof technology transitions. We found that at times,the pace of substitution was slowed due to “lastgasp” efforts by some firms to extend and maxi-mize the value that they could capture from the oldtechnology while other firms shifted aggressively tothe new technology. In other instances, the discov-ery of solutions to the emergence challenges in thenew technology allowed some firms to benefit fromspillovers that extended the performance of the oldtechnology. Finally, we observed a third mechanismthat has not been previously characterized in theliterature: homogeneous actions by heterogeneousfirms as actors across the ecosystem–competitors,suppliers, complementors, users–engaged in a “lastresort” effort to extend the old technology whenconfronted by a collective inability to overcome theemergence challenges of the new technology in theexpected timeframe.

Our study illustrates the importance of consider-ing ecosystem dynamics of the new technology’semergence and the old technology’s extension to

Copyright © 2015 John Wiley & Sons, Ltd. Strat. Mgmt. J. (2015)DOI: 10.1002/smj

Re-examining Technology S-Curves

explain the pace of substitution. In doing so, itinforms the literatures on technology, strategy andinnovation diffusion by showing how the interac-tions between old and new technology ecosystemsshape the context for both competition and marketadoption. Our perspective is not one of technologi-cal determinism but rather one of understanding thecontext and constraints that impact firm choices andoutcomes during technology transitions. In the con-cluding section, we consider the implications of ourapproach and framework for managers and policymakers.

TECHNOLOGY SUBSTITUTION

Explaining the dynamics of substitution is an impor-tant goal of the technology strategy literature.S-curves have become the canonical representationsof both the technology life cycle and of the com-petition between technologies (e.g., Christensen,1997; Foster, 1986; Utterback, 1994). The S-curveapproach holds that the magnitude of performanceimprovement in a given technology for a fixed unitof effort or time is relatively low during the earlydevelopment stages. As the technology is betterunderstood, the rate of progress increases until thestage of maturity, at which point the technologyapproaches its limits and the performance impactof additional effort is subject to decreasing returns.In the context of competing technologies, Foster’sinfluential work posited that the substitution threatfrom the new technology becomes salient once ithas superior performance–that is, whether a newtechnology will dominate the market depends onwhether the new S-curve crosses the old.

The S-curve characterization of technologyevolution has been critiqued and refined along anumber of dimensions. Sood and Tellis (2005)show that the profile of technology improvementover time can vary markedly from the smooth Sshape. Christensen (1997) showed that substitutioncan take place even when the new technologyis inferior to the old, arguing that if users areover-served along the main performance dimensionthey may switch to the new technology if it offerssuperior performance on a different dimension.Adner (2002) found that such disruptive dynam-ics can be explained in terms of price and costasymmetries between producers of the old and newtechnologies. Other studies have shown how newtechnologies can be incubated in distant markets,

emerging as substitute threats when they reach asufficient performance level (Adner and Levinthal,2001; Levinthal, 1998) or when triggered by a dis-continuous shift in the preferences of mainstreamconsumers (Tripsas, 2008).

Notice, however, that studies in this vein implic-itly frame substitution as an event that is governedby the rise of the new technology. They pay the bulkof their attention to understanding whether the newtechnology arises in a market domain that is neareror farther from the current customer base; whetherit draws on knowledge domains that are nearer orfarther from the current competence base; whetherit favors established firms or new entrants. Through-out, the focus is on whether substitution will occur,leaving the question of how fast substitution willunfold unaddressed.

In contrast, the diffusion of innovation literatureexplicitly considers the factors that determine therate of market adoption. This literature takes theavailability of a given innovation as its startingpoint and then considers how characteristics ofthe new technology interact with its users andtheir social context to impact its rate of adop-tion. Rogers’ comprehensive treatment (2003)identifies five canonical attributes of innovationsthat determine the rate of their adoption (relativeadvantage; compatibility; complexity; trialability;observability) as well as contextual factors suchas the nature of communication channels, socialsystems, and promotion efforts. The diffusion liter-ature also considers factors such as network effects(Stremersch et al., 2007), standards (Dranove andGandal, 2003), and information contagion (Iyengar,Van den Bulte, and Valente, 2011).

However, while the diffusion literature has beenarticulate about the factors that impact adoptiondynamics, it has tended to take a static view of thetechnology itself. As Hall notes in her review, thediffusion literature “implicitly assumes that neitherthe new innovation nor the technology it replaceschanges during the diffusion process.” (2004: 5).New technology improvements, however, are atthe heart of the technology strategy literature (e.g.,Christensen, 1997; Foster, 1986). And althoughthe canonical representation posits a flat-liningof the old technology’s performance trajectory,even mature technologies can be reinvigorated.Cases of old technology extension have been welldocumented in settings ranging from sailing ships(Harley, 1971) to pond ice harvesting (Utterback,1994) to typesetters (Tripsas, 2008) to carburetors



(Snow, 2008) to mobile telephony (Ansari andGarud, 2009). Thus, while the diffusion literaturefocuses on the question of how fast the newtechnology gains market share, it does so withoutconsideration of the evolution of the new and theold technologies.

Linking substitution and adoption throughsystems

The technology strategy and the diffusion litera-tures approach the question of substitution fromdifferent directions. Whereas the former focuseson the supply-side concerns of firms developinghigher performance technologies, the latter focuseson the demand-side concerns of users adoptinghigher value technologies. A deeper understandingof substitution dynamics requires linking technol-ogy evolution to technology adoption. The systemsapproach to technology offers just such a link.1

When a technology is consumed as part ofa system, the performance that matters to theconsumers’ assessment of value–the realizedperformance–is not the performance of the focaltechnology on its own, but is rather a function of itsinteraction with the other elements of the system.Here we distinguish between component elementsthat are integrated by the focal firm into its offerand complement elements that are integrated bythe user in conjunction with the focal firm’s offer(Adner and Kapoor, 2010).

The system can have two distinct effects on therealized performance of a technology. First, the

1 Our systems-based approach to technology transitions is closestto Adner and Kapoor (2010) and Ansari and Garud (2009). Adnerand Kapoor (2010) examined the role of ecosystem challenges inaffecting the extent of (firm-level) early mover advantage withina new technology generation. We complement that study byconsidering the effects of ecosystem dynamics, of both the oldand new technologies, in governing the pace of (technology-level)transitions. Ansari and Garud (2009) examined the influenceof the socio-technical system on the transition between secondgeneration (2G) and third generation (3G) mobile communicationnetworks. Whereas their study presents conjectures about thedirect effects of emergence and extension in the context of asingle technology transition, we take the interaction between theseforces as our central construct and show why it is critical toexamine them jointly. We complement their work by developinga framework that considers the degree of emergence challengesand extension opportunities across the new and old technologies,and identify how and why their interaction can give rise toqualitatively distinct substitution regimes. Moreover, we offer asystematic examination of pace of substitution over the courseof 10 generational transitions that helps to account for severalalternative explanations.

realized performance of a technology can be hin-dered by technical bottlenecks within the system.As Rosenberg (1976) notes, “… a given invention,however promising, often cannot fulfill anythinglike its potential unless other [emphasis in orig-inal] inventions are made relaxing or bypassingconstraints which would otherwise hamper its dif-fusion and expansion.” (p. 201). His account of theAmerican machine tool industry points to numer-ous episodes of “technical imbalance” in whichimprovements in new generations of machine toolscould not be realized by users because of bottle-necks in the development of complements suchas drills and cutting materials. It was only afterthe emergence challenges of these complementswere resolved that the new generations of toolscould finally deliver performance that matchedtheir potential, and market adoption could beginin earnest. Similarly, Hughes (1983) rich descrip-tion of the evolution of the electrical power systemhighlights the impact of technological interdepen-dencies and performance bottlenecks on adoption.

Second, the realized performance of a tech-nology can be enhanced when improvements inecosystem elements offset maturity in the focaltechnology. Numerous accounts have shown thatunderlying a technology’s advance are not onlyefforts by producers of the focal technology, butalso systemic efforts by component and comple-ment providers from a range of interdependentindustries. For example, Constant’s (1980) studyof the rise of aircraft piston-engine technologypointed to the critical role played by improvementsin engine components by suppliers such as GeneralElectric, as well as improvements in the comple-ments of fuels and lubricants by oil companiessuch as Royal Dutch Shell. Similarly, Henderson’spioneering (1995) study identified the key role ofimprovements in components and complementsin extending the performance trajectory of opticallithography.

The pace of technology substitution: emergencechallenges vs. extension opportunities

Recognizing the role of the system in constrainingand extending the realized performance of thetechnology allows us to recast the question ofsubstitution not as a competition between a newtechnology and an old technology, but ratheras a competition between a new technology’secosystem and an old technology’s ecosystem.



Eco

syst

em E

mer

genc

eC

halle

nge

(New

Tec

hnol

ogy)

Low

High

Quadrant 1

Baseline pace of substitution

Quadrant 3

Low High

Ecosystem Extension Opportunity (Old Technology)

Intermediate pace of substitution

Quadrant 4

Quadrant 2

Slowest pace of substitution Intermediate pace ofsubstitution

Figure 1. A framework for analyzing technology substitution

Accordingly, we must consider not only theperformance of the new technology, but also theextent to which technology bottlenecks elsewherein the system constrain the new technology’srealized performance from matching its potentialperformance–the ecosystem emergence chal-lenge.2 Similarly, for the old technology, wemust expand our consideration to include theextent to which improvements elsewhere in thesystem enhance the realized performance attainablewith the old technology system–the ecosystemextension opportunity. Figure 1 presents ourorganizing framework and characterizes the jointeffects of these two distinct constructs on thepace of substitution. Each quadrant representsthe conditions that characterize the extent of newtechnology’s ecosystem emergence challenge andold technology’s ecosystem extension opportunity.The specific balance of these two forces gives riseto four different substitution regimes.

2 Our characterization of ecosystem emergence challenges asrooted in technical bottlenecks is qualitatively distinct but fullycomplementary with insights regarding the impact of standardsand direct and indirect network effects (e.g., Katz and Shapiro,1985; Farrell and Saloner, 1986; Dranove and Gandal, 2003) onadoption and substitution. The distinction lies in the differencebetween a partner’s challenge to provide a complement and a part-ner’s choice to provide the complement. In addition, standardswars, direct network externalities, and indirect network external-ities focus on the impact of the number of adopters (installedbase) and complementors who choose to support a given technol-ogy. In contrast, the question of emergence challenges focuses onthe very possibility of participating -- the extent to which techni-cal bottlenecks prevent eager partners from delivering a workingcomplement that would allow the new technology to translate itssuperior technical performance to superior value creation. Here,the constraint to the availability of the complement is not rootedin issues of adoption incentive and critical mass, but rather in thecomplementors’ own capabilities.

Performance

TechnologyOld

Time

QQ1

Q2

Q3

Emergence Challenge

Extension Opportunity

New Technology

Q4

TechnicalPerformanceRealizedPerformance

Figure 2. Technology competition between an old tech-nology with ecosystem extension opportunity and a new

technology with ecosystem emergence challenge

Figure 2 offers a graphical intuition for the impli-cation of these system-level technology dynamicsfor the relative pace of substitution. The base-line case (the two solid S-curves, corresponding toQuadrant 1 in the framework) is the one in whichneither technology’s performance is impacted byinteraction with ecosystem elements, such that tech-nical performance fully determines the realizedperformance: the new technology does not faceecosystem emergence challenges and the old tech-nology does not benefit from ecosystem extensionopportunities. Q1 marks the point at which the per-formance of the new technology exceeds that of theold; and therefore, the time at which we can expectadoption to accelerate with market dominance tofollow.3

3 The figure illustrates the classic pattern of competing S-curvesin which the performance of new technology is initially inferior tothat of the old technology. This is clearly not always the case (Soodand Tellis, 2005). Indeed across the 10 generational transitions



When the extension-emergence balance shifts,however, we can expect very different dynamics. Indiametric opposition to the prior case, is the casewhen the ecosystem emergence challenge for thenew technology is high and the ecosystem exten-sion opportunity for the old technology is high aswell (the two dashed S-curves, corresponding toQuadrant 4 in the framework). Here, despite thepromise of the new technology, its realized perfor-mance is constrained by bottlenecks elsewhere inthe ecosystem. Moreover, the old technology standsto benefit from improvement in its own compo-nents and complements that enhance its realizedperformance, even if its technical performance isstagnant. Q4 marks the point at which the real-ized performance of the new technology exceedsthat of the old; thus, the simultaneous presenceof emergence challenge and extension opportunityshifts out the time at which we can expect adoptionto accelerate.

When the emergence challenge for the newtechnology is low and the extension opportunity forthe old is high (Quadrant 2), competition betweenthe old and new technology will be robust. Thenew technology will make inroads into the market,but improvements in the incumbent technology’secosystem allow it to compete effectively. Weexpect a prolonged period of coexistence beforesubstitution takes place (e.g., Furr and Snow,2014; McGahan, 2004). Although high extensionopportunity is unlikely to reverse the rise of thenew technology, it will materially delay the newtechnology’s market dominance and slow therate of substitution. Thus Q2, which marks thetransition point, occurs later than Q1 but earlierthan Q4.

In contrast, when the emergence challenge ishigh for the new technology and the extensionopportunity is low for the old technology (Quadrant3), substitution will be delayed until emergencechallenges are resolved. We therefore expect Q3 tooccur later than Q1 but earlier than Q4. We thusexpect the pace of substitution in both Quadrants 2and 3 to fall between Quadrants 1 and 4.

that we study in the semiconductor lithography equipment indus-try every new technology is introduced with an initial performancelevel that is substantially superior to that of the old technology.Regardless of initial technology positions, however, the implica-tions of emergence challenges and extension opportunities remainthe same, hindering and enhancing the new and old technologiesrespectively.

The joint consideration of the system-leveltechnology dynamics for the competing old andnew technologies allows us offer the followingprediction:

Hypothesis: The relative pace at which the newtechnology will substitute the old technologywill depend on the joint levels of ecosystememergence challenge for the new technology andthe ecosystem extension opportunity for the oldtechnology: fastest when both are low, slowestwhen both are high, and intermediate in themixed case where one is high and one is low (i.e.,Q1<Q2, Q3<Q4).

Although we are agnostic about the specificordering for Quadrants 2 and 3, we do expectthe underlying technology dynamics within thesequadrants to be qualitatively different: In the case ofQuadrant 2, substitution is slowed due to continuedimprovements in the old technology ecosystem.Here we would expect a gradual shift in marketshares from the old to the new technology. In con-trast, in the case of Quadrant 3, the old technology’sdominance is a function not of improvements inits ecosystem, but rather of the new technology’secosystem emergence challenge. Substitution willbe delayed until emergence challenges are resolved,but once they are resolved the transition in marketshare will be rapid. Here, an analysis of marketshare will likely show that the old technologymaintains high market share, but that marketgrowth has stalled. Because rapid market shareinversion is to be expected once the new technologyfulfills its performance potential, dominance bythe incumbent technology is fragile, in the sensethat it is maintained not by its own progress, but bythe new technology’s emergence challenge. Thus,while our formal hypothesis focuses on the pace ofsubstitution, we expect that our logic can informthe market-share profile of substitution as well.

We note that much of the literature on technologysubstitution has tended to focus on the performanceof competing focal technologies without con-sidering the ecosystem emergence challenges ofnew technologies and the ecosystem extensionopportunities of old technologies (e.g., Andersonand Tushman, 1990; Bass, 2004; Christensen,1997; Fisher and Pry, 1972; Foster, 1986; Soodand Tellis, 2005). Thus, Quadrants 2, 3 and 4 falloutside the scope of mainstream analyses. The



sub-strand of the literature that studies systemsof technology has tended to focus only on theemergence challenges of the new technology(Hughes, 1983; Rosenberg, 1976). Similarly, theliterature on competing systems (e.g., Cusumanoet al., 1992; Katz and Shapiro, 1985) explores therelative emergence challenges faced by competingnew technologies, but overlooks the role of theold technology ecosystem in shaping competitionbetween old and new technologies. Thus, it neglectsQuadrants 2 and 4 of our framework. Conversely,the sub-strand that has focused on the improvementin the old technology (e.g., Henderson, 1995;Utterback, 1994) has tended to overlook the role ofnew technology’s emergence challenge in shapingtechnology competition (i.e., Quadrants 3 and 4).It is our explicit consideration of the joint effectsof the old and new technology ecosystems thatallows us to characterize four distinct substitutionregimes in a single framework giving rise to testablepredictions and distinct strategic implications.

METHODOLOGY AND DATA

We apply our framework to analyze technologytransitions in the semiconductor lithography equip-ment industry from 1972 to 2009. The industryoffers a particularly rich setting in which to explorethe interaction between technology ecosystemsand the dynamics of substitution. It is a context ofrobust technological change. Between 1972 and2009, 10 new generations of lithography equipmenttechnology were commercialized, each enablingsemiconductor manufacturers, the customers, toachieve exponential improvements in chip perfor-mance at lower marginal production cost alongthe trajectory referred to as Moore’s law. Whilelithography equipment firms played an importantrole in pushing the technology envelope forward,progress within each of the technology generationswas critically shaped by components and comple-ments in the ecosystem (Adner and Kapoor, 2010;Henderson, 1995; Iansiti, 1998). Simultaneously,the context was characterized by continuity interms of user preferences (Christensen, 1997;Tripsas, 2008), nature of equipment manufacturers’core competences (Anderson and Tushman, 1990)and complementary assets (Teece, 1986), and lownetwork effects (Katz and Shapiro, 1985).

The research setting also provides natural con-trols for key factors considered in the innovation

of diffusion literature that affect the pace of sub-stitution (Rogers, 2003). Throughout the industry’shistory, the new technology generations were char-acterized by a high degree of observability and tri-alability by the users. This is because lithographyequipment firms aggressively promoted their newgeneration “tools” through trade shows, industryconferences as well as through formal and infor-mal interactions with semiconductor manufacturerswell before the new generation was commercial-ized. Similarly, customers saw the new technologyas key to their own competitive advantage and there-fore dedicated personnel within their organizationsto actively scan and evaluate new technology devel-opments. Moreover, given the high scale of invest-ment and the criticality of the lithography tool totheir manufacturing process, semiconductor manu-facturers conducted extensive experimentation priorto purchasing and implementing the new technol-ogy generations into their processes. In our analysis,we account for the additional diffusion factors ofrelative advantage, compatibility and complexity ofthe new technology generation.

The nature of this study required us to develop athorough understanding of the lithography technol-ogy, its historical evolution as well as the nature ofemergence challenges and extension opportunitieswithin the ecosystems of the different technologygenerations. We interviewed over 30 industryexperts, most of whom have been associated withthe industry for more than 20 years. The interviewswere semi-structured and lasted two hours onaverage. The interviewees came from a varietyof roles within the semiconductor lithographyecosystem: lithography equipment manufacturers,their customers, their complementors, their suppli-ers, research consortia, industry associations, andconsultants.4 This allowed us to understand theemergence challenges and extension opportunitiesacross the ecosystem for each of the 10 technologytransitions and to triangulate perspectives acrossdifferent roles.

For the quantitative analysis, we obtained dataon detailed product specifications and sales foreach of the technology generations from VLSIResearch, a prominent industry consulting firm.

4 Twenty-seven percent of the interviewees were affiliated withlithography equipment manufacturers, 24 percent with customers,17 percent with complementors, 12 percent with suppliers andthe remainder with SEMATECH, Industry associations andconsultants.



Energy Source

Lens

Mask with IC pattern on Top

Semiconductor Wafer with Resist Layer

Figure 3. Basic principle of the semiconductor lithography process

VLSI Research also provided us with historicaldata on prices of lithography equipment and cap-ital expenditures by semiconductor manufacturersthat we used to rule out some important alternativeexplanations. For two of the earliest generations, wesupplemented the data from VLSI Research withproduct specifications reported in the HendersonPhotolithography FIVE Data (Helfat and Klepper,2007; Henderson, 1993). Through our fieldwork,we also developed a list of keywords that we usedto codify technical articles published between 1961and 2010 in Solid State Technology, the leadingindustry trade journal, and we used this informa-tion to develop an objective measure of ecosystememergence challenge for each of the technologygenerations.

SEMICONDUCTOR LITHOGRAPHYTECHNOLOGY

Semiconductor lithography is the process by whicha circuit design is imprinted on a semiconductorsubstrate called a wafer. The basic principle oflithography is illustrated in Figure 3. After thedesign of an integrated circuit (IC) is finalized (i.e.,the wiring, the gates and the junctions), the circuitblueprint is transferred to a “mask.” The lithographyprocess takes place when beams of energy originat-ing from an “energy source” are directed onto themask. The pattern on the mask allows a portion ofthe energy beams to pass through, with or withoutan optical “lens” system, onto the wafer. The waferis coated with an energy sensitive “resist.” Theresist undergoes a chemical reaction wherever themask has allowed the energy to pass through. Thischemical reaction changes the structure of the resistand allows its selective removal from the wafer.Another chemical process is then initiated in which

the exposed parts of the wafer are etched. Finally,the remaining resist is removed, creating a finalcircuit that replicates the initial design.

The process is carried out using lithographyalignment equipment referred to in the industry asthe lithography tool. The energy source and the lensare the two key components that are integrated byequipment manufacturers into the lithography tool.A semiconductor manufacturer may use scores oflithography tools in a single production line. Withmodern tools costing over twenty million dollarseach, investments in lithography equipment repre-sent a substantial portion of the cost of a fabricationfacility. To maintain competitiveness, semiconduc-tor manufacturers continuously reinvest in theirfacilities and look to new tool generations to allowthem to offer products with higher performance atlower marginal cost. The mask and the resist are thetwo main complements that semiconductor manu-facturers must integrate with the lithography toolto carry out the lithography process. Hence, the keyelements of the lithography technology ecosystemare the focal technology (the lithography tool), theenergy source and the lens as the key components,and the mask and the resist as the key complements(Figure 4).5

The main measure of performance in semi-conductor lithography is resolution, the smallestfeature size that can be “printed” on the semicon-ductor wafer. Resolution determines the extent ofminiaturization that can be achieved by the semi-conductor manufacturer. Smaller feature size allowsthe semiconductor manufacturers to pack morecircuits onto a chip and more chips onto a wafer.What this means is that for both the lithography

5 Three of the earliest technology generations transmitted theenergy directly from the source to the wafer without using a lens.These were the Contact, Proximity, and X-ray generations.



Lens(e.g., Zeiss)

Energy Source(e.g., Cymer)

Resist(e.g., Shipley)

LithographyTool (e.g., ASML)

Semiconductor Manufacturing (e.g., Samsung)

Mask(e.g., Photronics)

Figure 4. Semiconductor lithography ecosystem. Firms mentioned in parenthesis are only representative of the firmsparticipating in the lithography ecosystem

tool manufacturers and their customers, the semi-conductor manufacturers, improving resolution isthe key to maintaining competitive advantage and istherefore a top priority. Although other dimensionsof performance matter (e.g., throughput, servicereliability), resolution is by far the dominantconsideration.6

TECHNOLOGY SUBSTITUTION IN THESEMICONDUCTOR LITHOGRAPHYEQUIPMENT INDUSTRY

The semiconductor lithography equipment industrydistinguishes among generations according to thechanges in the tool architecture and in the wave-length of light. From 1972 to 2009, 10 distinctgenerations of lithography equipment have beenintroduced.7 We provide a detailed description of

6 In File S1 that is available as an online supplement, we presenta hedonic regression analysis of lithography tools and find thatresolution on its own explains 81 percent of observed variance inprice.7 This definition of lithography generation is consistent with theindustry norm as well as Henderson (1993) and Henderson andClark (1990). It offers a finer grained level of analysis thanthe general category of “optical lithography” used in Henderson(1995), which would group eight of the 10 generations examinedhere (all but E-beam and X-ray) under the single heading. Asdiscussed below, focusing on the transitions between individualgenerations offers a more nuanced view of substitution. It alsoleads us to characterize the failure of X-ray lithography, discussedin Henderson (1995) as driven in large part by its own emergencechallenges rather than purely by the successful extension ofoptical lithography.

each of the generations in the Appendix I includ-ing information on the nature of ecosystem emer-gence challenge and the years it took to achievea dominant market share. Judged according to theestablished criteria that have been identified in theliterature, the conditions of their launches werenearly identical: At the time of first commercialsale, the resolution performance of each new tech-nology generation was superior to the performanceoffered by the old technology. Each of the gen-erations were also sustaining (Christensen, 1997)in that they were developed to meet the needsof firms’ existing customers who were demandingimproved resolution so as to manufacture integratedcircuits (ICs) with higher performance and lowercost. Finally, the composition of key customer seg-ments, the principle dimensions of performance,as well as the modes through which producersinteracted with customers (Abernathy and Clark,1985) were all consistent throughout the industry’shistory.

Given these structural similarities, conventionalwisdom would yield a prediction that the technol-ogy generations should follow similar substitutionpatterns. This, however, is not the case. Indeed,there is significant variance in the time it took for thenew generation to achieve a dominant market share.There are cases of fast substitution in which thenew technology takes off almost immediately uponits introduction and achieves the dominant mar-ket share. These include the Proximity, Projection,G-line and DUV 193-immersion generations. How-ever, there are also cases of slow transitions in which



the new technology takes much longer to achievemarket dominance. These include the I-line, DUV248 and DUV 193 generations. Finally, there werecases of non-transitions in which despite signifi-cant development efforts by the industry over manyyears, the new technology failed to substitute the oldtechnology. These include the E-beam, X-ray andDUV 157 generations.8

Emergence challenge in the new technologygeneration

Each of the new generations presented significantdevelopment challenges for the lithography equip-ment manufacturers. These include challenges indesigning the new tool architecture as well as inintegrating the different components into the archi-tecture. At the time of first commercial sale, eachnew generation offered superior technical perfor-mance than did the dominant generation. However,these generations differed significantly in the extentto which innovation in complements was requiredin order for the new generation to achieve itscommercialization potential; that is, for its realizedperformance to match its technical performance forthe mass market. As detailed in the Appendix I, 5of the 10 generations faced high emergence chal-lenges, requiring major complement innovations inthe mask and resist. These were the E-beam, X-ray,DUV 248, DUV 193 and DUV 157 generations.The other five generations–Proximity, Projection,G-line, I-line and DUV 193-immersion–facedmuch lower emergence challenges as they wereable to reuse the mask and resist technologiesthat had been developed for the old technologygenerations.

For an objective measure of ecosystem emer-gence challenge we follow Adner and Kapoor(2010) and use a count of the number of technicalarticles published in Solid State Technology, theleading industry journal. We counted articles thatreported on technical challenges regarding the maskand resist for each of the new technology genera-tions.9 A representative discussion of emergence

8 Other technology generations such as Ion Beam and ElectronProjection were also pursued in the 1970s but did not move beyondR&D to the commercialization stage and, hence, are not part ofour study.9 We used a number of keywords to identify lithography relatedarticles that were published between 1961 and 2010. We catego-rized articles as lithography related if any of the following keywords appeared in the title: lithography, microphotograph, mask,

challenge is Hibbs, Kunz, and Rothschild(1995: 69):

“The most difficult challenge in bringing193-nm lithography to full manufacturinguse is the development of a robust photoresistprocess. The resins that are typically used fori-line and 248-nm photoresists (novolac, poly-hydroxystyrene, etc.) are far too opaque at193 nm to be used in single-layer resists at thatwavelength. Other materials, such as acrylatepolymers, have been identified as suitablefor 193-nm resists. Using these materials,prototype resists with good imaging proper-ties have been developed in a collaborativeeffort between IBM/Almaden and LincolnLaboratory. However, the dry etch resistanceof these prototype resists is relatively poor,and modifications of the formulation arebeing developed to provide higher etchresistance.”

As this excerpt, which was published a year inadvance of the first commercial sale of a DUV193 tool, shows that the resist challenge wasa genuine technology bottleneck, rather than aproblem of insufficient incentives for comple-mentors or a lack of a critical installed baseof users.

The count of articles for each new generation isplotted in Figure 5. The average number of articlesacross the generations is 4.6. We categorized emer-gence challenges as “High” and “Low” accordingto whether a given generation was above or belowthis average. The characterization of ecosystem

photomask, resist, laser, UV, DUV, Deep UV, optical, lens, step-per, aligner, mercury, illuminator, exposure, printer, and the namesof the different generations. We then used the article titles to createa match between the technology generation and whether the arti-cle referred to technical problems in mask and resist. If insufficientinformation was available in the title, we read the abstract and theconclusion to ascertain if the article addressed the innovation chal-lenges in the ecosystem for a given generation. A small subset ofarticles discussed the mask and resist challenges for multiple gen-erations, and for these articles we read the relevant sections foreach generation in order to create a match. Our industry sourcesconfirmed that a count of published articles is a good proxy for thelevel of technology challenges that surrounded the development ofdifferent lithography generations. Since our primary concern wasto assess challenges during the emergence of the new technologygeneration, we excluded articles that were published five yearsafter the generation was commercialized. As a robustness check,we also used a 10 year window and found very similar patterns asthose reported here.



0

2

4

6

8

10

12

Proximity Projection E-beam X-ray G-Line DUV 157 DUV 193immersion

Ecosyst

em Em

erge

nce

Cha

lleng

e

I-Line DUV 248 DUV 193

Figure 5. Magnitude of ecosystem emergence challenge for the different technology generations as measured by SolidState Technology article count

emergence challenge based on the article countexhibited very high consistency with the qualitativeaccount of ecosystem emergence challenges that wedocument in the Appendix I.

Extension opportunity in the old technologygeneration

The different lithography technologies faced verydifferent opportunities for extending their resolu-tion performance at the time of the introduction ofthe new technology generation. The resolution ofa technology generation is based on the theoreticalrelationship known as the Rayleigh criterion:

Resolution = k1 × 𝜆∕NA,

where 𝜆 is the wavelength of energy being trans-mitted by the source, NA is a numerical apertureof the lens that is a function of the tool architectureand the size of the lens, and k1 is a process-specificconstant that is a function of the mask and theresist complements. Within a given generation, 𝜆 isfixed. Therefore, the primary means of extendingthe performance of a technology generation are toimprove the mask and the resist complements so asto reduce k1 or to improve the lens component soas to increase the NA. The mode of improvement inthe mask and the resist complements are limited bythe properties of the materials and the associatedmanufacturing techniques. The mode of improve-ment in the lens component is limited by thecomplexity of the lens design and manufacturing,as well as the reduction in the allowable margin oferror for the user.10

10 The technical term corresponding to the user’s margin of erroris the depth of focus (DOF). It is the maximum distance between

In four cases of technology competition the oldtechnology had low extension opportunities. TheContact and Proximity tool architectures did notinclude a physical lens component (i.e., the “lens”that the energy passed through was air, not glass), soimproving NA was not a possible extension mech-anism. The Projection generation used a reflec-tive lens that made it very difficult to increase NAbeyond 0.167 (Henderson, 1995). The extensionopportunity for these generations was also limitedby the fact that they required the use of maskswhose feature size was in a 1-to-1 ratio to thesize of the features to be printed on the wafer.This created very high challenges for the maskproduction process and limited the potential formask-based extension opportunities. By the timeDUV 193-immersion was introduced in 2005 theextension opportunity for DUV 193 was severelylimited by the physical limits imposed by the lens,mask and resist materials that had already been“stretched” for over two decades. In contrast, fourgenerations had high extension opportunities. TheG-line, I-line, DUV 248, and DUV 193 (at the timeof the DUV 157 introduction) generations all bene-fitted enormously from extension opportunities dueto improvements in lens design, lens manufactur-ing, in combination with new techniques for mask

two planes over which there are clear optical images. It can simplybe considered as the range of focus errors the lithography processcan tolerate and still provide acceptable results. Its relationshipto energy wavelength and NA is given by DOF= k2 × 𝜆/(NA)2

where k2 is a constant. An increase in NA, while improvingresolution, lowers the DOF and makes the lithography processless robust. As compared to changes in wavelength, the squaredNA term has a greater effect on the deterioration in the DOFand hence, limits the tool manufacturers’ ability to improve theperformance of the lithography technology by increasing the NA(cf. Henderson, 1995).



10

20

30

50

0

40

Contact DUV 193(DUV 157)

DUV 193(DUV 193immersion)

% im

prov

emen

t in

old

gen

erat

ion af

ter

the

intr

oduc

tion

of

new

gen

erat

ion

Proximity Projection G-Line I-Line DUV 248

k1 improvement NA improvement

Figure 6. Magnitude of ecosystem extension opportunity for the different technology generations

design, mask manufacturing, and from the introduc-tion and refinements of new resist chemistries.

To provide an objective measure of the exten-sion opportunity for the different technologygenerations, we collected the published technicalspecifications for lithography tools sold by everymajor tool manufacturer between 1973 and 2009.Specifically, we collected data on resolution,numerical aperture (NA), wavelength (𝜆), and usedthese to derive the value of k1 for every tool modelbased on the Raleigh criterion. To quantify theextension opportunity in each generation that wasdue to improvements in the ecosystem, we com-puted the percentage increase in NA (extension dueto lens component) and the percentage decrease ink1 (extension due to mask and resist complements)between the best old technology tool sold in theyear that the new technology was introduced andthe best old technology tool ever sold after thatyear. Figure 6 plots the percentage change inNA and k1 for the different technology genera-tions.11 Consistent with our understanding from thefieldwork, the characterization of old technologygenerations show an unambiguous distinctionbetween generations with low extension opportu-nities (Contact, Proximity, Projection, DUV 193following the introduction of DUV 193-immersion)

11 We note that improvements in k1 were not entirely confinedto innovations in mask and resist. Tool makers also contributedto the lowering of k1 by changing the angle of light falling onthe mask and hence, enhancing the contrast of the image. Thisinnovation was known as Off-Axis Illumination (OAI). Althoughits introduction late in the G-Line generation was a non-negligiblecontributor to k1 improvement, our interviews suggest that itscontribution was not as large as that from improvements in resistand mask. Because OAI was incorporated as a standard designelement in every subsequent generation launch, it does not affectour measure of extension opportunity for any other generation.

and generations with high extension opportunities(G-line, I-line, DUV 248, DUV 193 following theintroduction of DUV 157).12

Substitution dynamics: the joint considerationof emergence challenges and extensionopportunities

We have characterized each episode of tech-nology competition according to the degree ofecosystem emergence challenge confronting thenew technology and the ecosystem extensionopportunity available to the old technology. Wenow consider these characterizations jointly toposition each case of technology competition in itscorresponding quadrant within our framework.

Four transitions were characterized by lowecosystem emergence challenge for the new tech-nology and low ecosystem extension opportunityfor the old technology (Quadrant 1 within ourframework). Consistent with our arguments andas noted in the Appendix I, in all these cases thenew technology generations (Proximity, Projection,G-line and DUV 193-immersion) achieved veryrapid market adoption and were the fastest to sub-stitute the old generation. Here, the average time tomarket dominance–the number of years from theyear the new generation was commercialized to theyear its annual market share surpassed that of theold generation–is three years.

12 We note that the operationalization of extension opportunity isbased on the realized extension of the old technology, not theexpected opportunity for extension. In our interviews, industryexperts were in unanimous agreement in characterizing the tech-nology constraints that limited extension opportunities. After pre-senting our results, we explore the different modes through whichthis realization process has unfolded in the section titled “The Roleof Heterogeneity and Choice in Pacing Technology Substitution.”



0%

20%

40%

60%

80%

100%

1 2 3 4 5 60%

20%

40%

60%

80%

100%

1 2 3 4 5 6

G-Line I-Line

Mar

ket

Shar

e

New Generation Year New Generation Year

ProjectionProximity

Mar

ket

Shar

e

Figure 7. Difference in the pattern of substitution between low emergence challenge/low extension opportunitytransition (proximity→ projection) and low emergence challenge/high extension opportunity transition (G-line→ I-line)

Three transitions were characterized by highecosystem emergence challenge for the new tech-nology and high ecosystem extension opportunityfor the old technology (Quadrant 4). Consistentwith our arguments, these new technology gener-ations (DUV 248, DUV 193 and DUV 157) werethe slowest to substitute the old generation.13 Acase in point for the high emergence challengeand high extension opportunity transition is thetransition from I-line to DUV 248. It took DUV248 11 years to achieve market dominance. Thereduction in the wavelength from I-line’s 365 to248 nm resulted in technology bottlenecks requiringchanges in the mask and resist complements. Whiledeveloping a new mask material was difficult, themost significant emergence challenge in the ecosys-tem was with the resist. New resist chemistries hadto be developed that were sensitive enough to absorbsufficient energy from the lower wavelength tocause the chemical reaction, yet specific enough soas not to react when energy impacted an adjoiningmolecule. It took several years before a new type ofresist chemistry–chemically amplified resist–wasdeveloped to meet the commercialization require-ments of the DUV 248 technology. In addition tothe ecosystem emergence challenge faced by DUV248 generation, the transition was further slowed byhigh extension opportunity for the I-line generation.This extension was achieved through increases inthe NA of the lens component and through an inno-vation in the mask called Phase-Shift. Phase-Shift

13 In fact, the ecosystem emergence challenge for DUV 157 was sosignificant that, as of the time of this writing, it has yet to achieveany meaningful market penetration.

improved the resolution of the lithography technol-ogy by lowering k1 and extending the I-line gen-eration from its initial resolution limit of 500 nmto 350 nm (Terasawa et al., 1989). This extensionraised the performance threshold that the DUV 248generation needed to cross in order to unlock itsmarket potential and achieve dominance.

The transition from G-line to I-line was charac-terized by low ecosystem emergence challenge andhigh ecosystem extension opportunity (Quadrant2). It took I-line seven years to dominate G-line.The intermediate pace of substitution for thistransition is consistent with our arguments that thesubstitution dynamics within this quadrant wouldentail a robust competition between the old and thenew technology until the old technology reachesits limits. Figure 7 illustrates this pattern of sub-stitution and contrasts it with a typical case of fastsubstitution (Proximity substituted by Projection)observed for the low emergence challenge–lowextension opportunity category. While Proximity’ssubstitution by Projection is characterized by arapid increase in Projection’s share followed bymarket dominance and a rapid erosion of Prox-imity’s share, G-line’s substitution by I-Line ischaracterized by a much more gradual increasein I-line’s share and a gradual decline in G-line’sshare. This case of the transition from G-line toI-line illustrates that low emergence challengeis not a sufficient condition for fast substitution.Specifically, the pace of substitution criticallydepends on the extension opportunity for the oldtechnology.

Finally, the competition between Projectionand E-beam in 1976, and then X-ray in 1978,



was characterized by high ecosystem emergencechallenge and low ecosystem extension opportunity(Quadrant 3). We expected that in this quadrant theold technology, despite having limited extensionopportunity, would continue its market dominanceuntil the emergence challenge with the new tech-nology was resolved. However, once the emergencechallenges are overcome, the new technology isexpected to have a steep sales takeoff. Hence, in thisregime, the old technology’s dominance is fragilebecause its market position is based not in its ownstrength, but in the rival technology’s weakness.The emergence challenges for E-beam and X-raywere both extremely high. In the absence of suitableresist and mask complements, neither generationcould realize its technical performance advantage.14

Consistent with our expectations, Projection domi-nated the market in the early years that followed theintroduction of the new technologies. However, nei-ther E-beam nor X-ray replaced Projection in lateryears. In this regard, the intermediate pace of substi-tution that we expected was not observed. This wasdue to the fact that the competition became a multitechnology race. Specifically, the X-ray generationwas commercialized two years after E-beam, andthen the G-line generation (which had low emer-gence challenge) was commercialized before eitherE-beam’s or the X-ray’s emergence challenge hadbeen overcome. As such, before they had had achance to displace Projection, both technologieswere themselves displaced. The technology com-petition for dominance rapidly shifted from themode of old vs. new technology (Projection vs.E-beam) to old vs. newer technology (Projectionvs. X-ray) to old vs. newest technology (Projectionvs. G-line). One reason for the parallel efforts inthese different technology generations is that theywere pursued by different firms trying to exploittheir own unique capabilities and positions.

In order to test whether the observed yearsto market dominance for the new technologywithin each of the quadrants follows the pre-dicted ordering (years to dominance in Quadrant1<Quadrant 2<Quadrant 4), we performedthe Jonckheere-Terpstra non-parametric test forordered alternatives. We were able to reject the null

14 We note that while E-beam generation faced high ecosystememergence challenge, it also suffered from lower throughput thatmade it less attractive for large scale mass production. We discussthis aspect of E-beam in detail in the Appendix I.

hypothesis of random ordering and our predictionwas supported with the p-value of 0.009.

Regression analysis

To offer a more systematic analysis that also takesinto account the substitution dynamics in Quadrant3, we performed regression analysis. Our dependentvariable is the new technology generation’s annualshare of total industry sales (in dollars) duringthe period of technology transition. The higher thenew generation’s annual market share during thisperiod, the faster it substitutes the old generation.On average, it took a new generation six yearsto displace the old generation from its industryleadership. Hence, we perform our main analysisfor the first six years of sales for each of the newtechnology generations. As a robustness check, weperformed this analysis for the first four years andeight years as well.

In our analysis, we controlled for a number ofgeneration- and industry-level effects that mayinfluence the annual market share of the newtechnology generation. In order to account for therelative advantage of the new technology gener-ation (Foster, 1986; Rogers, 2003), we controlfor the price adjusted performance difference(performance/price) between the new technologygeneration and the old technology generation. Thismeasure was calculated based on average price andresolution capability of lithography tools sold inthe given year. The primary form of technologycompetition existed between the new technol-ogy generation and the dominant old generation.However, some periods in the industry were charac-terized by the presence of multiple old technologygenerations that may have influenced the new gen-eration’s market share. We controlled for this effectwith a count of the number of generations that cap-tured at least 10 percent market share in the industryin a given year. We set 10 percent as the thresholdto exclude cases of old technologies being used insmall niche applications, such as Proximity genera-tion tools being used in university labs 15 years afterthe generation has lost its dominance. The capabil-ities of the firm that pioneered the new technologygeneration, and the number of firms producing thenew generation, may have also impacted marketshare. We accounted for these effects by catego-rizing whether the pioneering firm was an industryincumbent (incumbent pioneer), and by countingthe number of firms that sold the new generation



Table 1. OLS regression estimates; dependent variable= generation share

(1) (2) (3) (4) (5) (6)

High emergence high extension −0.328*** −0.262*** −0.316*** −0.453*** −0.283***(0.031) (0.034) (0.039) (0.046) (0.035)

Low emergence high extension −0.235*** −0.174*** −0.159** −0.386*** −0.161**(0.047) (0.043) (0.055) (0.065) (0.064)

High emergence low extension −0.207*** −0.239*** −0.259*** −0.237***(0.036) (0.055) (0.043) (0.063)

Price adjusted performancedifference

2.338*** 1.013* 1.868** 1.141** 0.340 2.139***(0.669) (0.526) (0.693) (0.493) (0.702) (0.589)

Number of generations −0.151*** −0.137*** −0.120*** −0.108*** −0.152*** −0.100**(0.031) (0.024) (0.033) (0.025) (0.038) (0.038)

Incumbent pioneer −0.086 −0.006 0.027 −0.049 ∗ 0.040 −0.017(0.090) (0.018) (0.020) (0.024) (0.032) (0.019)

Number of firms 0.019 0.007 ∗ 0.008 0.009 0.014 ∗ 0.001(0.016) (0.004) (0.005) (0.006) (0.007) (0.005)

Old generation tenure −0.002 −0.010*** −0.009** −0.006 −0.016** −0.009**(0.021) (0.003) (0.003) (0.004) (0.006) (0.003)

Newer generation availability −0.037 −0.145** −0.196*** −0.161*** −0.014 −0.211***(0.112) (0.062) (0.055) (0.049) (0.085) (0.046)

Year trend 0.019 0.039** 0.052* 0.023 0.050** 0.053***(0.023) (0.014) (0.026) (0.014) (0.021) (0.015)

Constant 0.316* 0.554*** 0.375*** 0.497*** 0.736*** 0.356**(0.148) (0.069) (0.092) (0.078) (0.103) (0.119)

Observations 59 59 40 75 59 47R-squared 0.48 0.82 0.79 0.75 0.81 0.84

Omitted category is low emergence & low extension. Robust standard errors in parentheses, clustered by technology generation.*Significant at 10 percent; **significant at 5 percent; ***significant at 1 percent.

in a given year (number of firms). The pace of sub-stitution may also be impacted by the differencesin the semiconductor manufacturers’ switchingcosts across the different technology transitions.The longer the generation had been used by semi-conductor manufacturers, the greater may be theirswitching cost. We controlled for this possibilitythrough the variable, old generation tenure, whichis the number of years the old generation has beensold in the industry. The new generation’s marketshare may also be impacted if the period of obser-vation is characterized by the availability of a newertechnology generation with low emergence chal-lenge. This effect is controlled for by the dummyvariable, newer generation availability. Finally, wecontrol for the yearly trend in the market share of agiven generation during the period of substitution.

Table 1 reports the estimates from the OLSregression analysis. The standard errors are clus-tered by technology generation to account for thenon-independence of observations within a gener-ation. Model 1 is the baseline model with controlvariables, and Model 2 includes the effects of thedifferent substitution regimes with the omitted cate-gory of low emergence challenge and low extensionopportunity. The results from the baseline model

are consistent with our expectations. The greater theprice adjusted performance difference between thenew and the old generation, the greater is the newgeneration’s market share. The market share of thenew generation decreases with the number of gen-erations in the industry. The coefficients for numberof firms, old generation tenure and newer generationavailability exhibit expected signs but are signifi-cant only in the full model.

The coefficients for all three substitution regimes(high emergence and high extension; low emer-gence and high extension; high emergence and lowextension) are negative and significant supportingour prediction that the technology transitionscharacterized by low emergence challenge and lowextension opportunity (omitted category) exhibitthe fastest substitution. The magnitudes of thethree coefficients are in the expected order withthe coefficient for the high emergence and highextension category having the highest magnitude.A statistical comparison of the coefficients forthe three regimes reveals that the coefficientfor the high emergence-high extension category issignificantly different from the low emergence-highextension category (p-value of 0.002) and the highemergence-low extension category (p-value of



0.057). Hence, as hypothesized, the technologytransitions characterized by high emergence chal-lenge and high extension opportunity have theslowest pace of substitution. The coefficients forthe low emergence challenge and high extensionopportunity category, and that for the high emer-gence challenge and low extension opportunitycategory supported our prediction of intermediatepace of substitution with the difference betweenthe two coefficients being statistically insignificant(p-value of 0.71).15

We also performed a number of additional robust-ness checks in order to ensure that the regressionestimates are not sensitive to our variable specifi-cations and that our inferences are not subjectedto alternative explanations. In Models 3 and 4, wereport the estimates using data from the first fourand eight years of sales for each of the new tech-nology generations respectively. We also checkedthat our results are not sensitive to the presence ofmultiple technology generations. In Model 5 weused an alternative measure of generation shareby explicitly considering head-to-head technologycompetition between the new and the dominant oldgeneration. Here generation share is based on onlythe sales of the new generation and the old domi-nant generation. Finally, since X-ray and E-beamare non-optical generations and may be charac-terized as having higher levels of complexity andincompatibility with semiconductor manufacturers’past experiences (Rogers, 2003), we excluded themfrom the estimates reported in Model 6. The resultswere robust to these analyses. The only exceptionbeing that the coefficient for the high emergenceand high extension category was not significantlydifferent from that of the high emergence and lowextension category in Models 3 and 4. This is prob-ably because the two new generations in the highemergence and low extension category (X-ray andE-beam) never replaced the old technology. File S1,that is available as an online supplement, examinesadditional alternative explanations for our resultswith regards to users’ capabilities, preferences andheterogeneity and also examines an alternative char-acterization of the emergence challenge measure.

15 An important issue pertaining to the regression analysis is thatthe statistical power of the test, owing to small sample size, may betoo low to yield any meaningful insights. We performed a poweranalysis using the “powerreg” procedure in STATA and found thatbecause of our large effect size, our analysis has very high power(approaching 1.00; almost a 100 percent probability of rejectingthe null hypothesis if it is false).

THE ROLE OF HETEROGENEITY ANDCHOICE IN PACING TECHNOLOGYSUBSTITUTION

Our theory and analysis are at the level of tech-nology. Technological progress, however, does nothappen on its own, but rather is shaped by thechoices and efforts of the heterogeneous partici-pants that compose the ecosystem. In studying thepatterns of technological change in the semiconduc-tor lithography equipment industry we were ableto uncover three different modes by which industryparticipants influenced the pace of substitution.

First, in some cases, the pace of substitution wasslowed by the “last gasp” effort by some incum-bent firms to prolong the life of their old technologygeneration. We observed such a pattern in the caseof the G-line generation after the emergence of theI-line generation. Nikon was the first firm to com-mercialize the I-line generation in 1985. In contrast,Canon, another incumbent, chose to focus its R&Defforts on extending the G-line generation throughan increase in the size of the lens. Canon introducedits I-line tools years after Nikon, only when furtherextending the performance of the G-line generationwas deemed unfeasible. This kind of last gasp effortto reinvigorate performance improvement in maturetechnologies has been observed in other technolo-gies such as sailing ships (Harley, 1971), pond iceharvesting (Utterback, 1994), typesetters (Tripsas,2008). The main driver here is the differences ininnovation incentives and strategies among the dif-ferent focal firms.

Second, there were instances in which R&Defforts initiated to resolve the emergence challengesin the new technology generation triggered the dis-covery of solutions that served to also extend theold technology generation. For example, one of thekey challenges confronting the DUV 157 gener-ation was the development of new lens materialswith high concentration of Calcium Fluoride thatwould allow for transmission of low wavelengthlight. The R&D efforts to resolve this challengealso helped firms to make better lenses for the oldDUV 193 generation, extending its performance.Similarly, the resist development for newer genera-tions consistently helped in improving the chemicalcomposition of the resist used with the older gen-erations. Furr and Snow (2014) document such a“spillback” effect in the automotive industry wherethe development of the new electronic fuel injec-tion technology helped to extend the performance of



the old carburetor technology. They also found thatthe extent of spillback realized by individual firmsvaried according to their specific level of invest-ment in the new technology. Taylor (2010) exploresthe intra-organizational mechanisms through whichsuch spillbacks take place.

Finally, we observed a third mechanism thathas not been previously characterized in the liter-ature. We found instances where participants acrossthe lithography ecosystem (competing firms, users,suppliers and complementors) engaged in a col-lective “last resort” effort to extend the old tech-nology due to a collective inability to overcomethe emergence challenges in the new technology.For example, when it was evident that the emer-gence of DUV 193 in 1996 would be significantlydelayed due to the challenges in the resist, whichwould mean that the generation would not be ableto achieve its planned introduction at 180 nm reso-lution, lithography equipment firms, semiconductormanufacturers, materials and other equipment man-ufacturers collectively revised their developmentemphasis to extend the performance of the incum-bent DUV 248 generation through a combination ofimprovements in the tool, lens, resist and mask.16

Each of the three modes represents a distincttype of firm-level actions in shaping the pace ofsubstitution. In the case of last gasps, some firmsaggressively pursue the new technology whileother firms maximize the performance from the oldtechnology. In the case of spillbacks, R&D effortstoward the new technology help firms to also extendthe old technology. In the case of last resort, firmsacross the ecosystem collectively reallocate their

16 Since the 1990s, progress in semiconductor manufacturing hasbeen coordinated according to collectively defined technologyroadmaps, most recently under the auspices of InternationalTechnology Roadmap of Semiconductors (ITRS). The roadmap isregarded as a very good indicator of the R&D resource allocationacross the different technology generations. The 1991 roadmap,in expectation that the DUV 193 generation would launch in1996 with a resolution of 250 nm, specified that the performanceof DUV 248 would peak at 250 nm and that this limit wouldbe reached by 1996 (Micro Tech 2000, 1991). As emergencechallenges delayed the launch of the DUV 193 generation, the1997 and 1999 roadmaps revised the expected performance limitof DUV 248 to 150 nm by 1999 and 130 nm by 2001 respectively.These revisions signaled a collective agreement by actors acrossthe lithography ecosystem to reallocate substantial developmentresources to extend the performance of DUV 248 (and so awayfrom the DUV 193 effort) as a last resort for sustaining thetrajectory guided by Moore’s law. As a result of these investmentsthe DUV 248 generation was extended to an ultimate resolutionperformance of 130 nm (in 2001), which sustained the industryuntil the emergence challenges of DUV 193 were resolved in2002.

resources to extend the old technology when metwith significant bottlenecks in the new technology.These different modes of firms’ actions illustratehow technology transitions are shaped by a richset of collaborative and competitive interactionsamong heterogeneous firms across the ecosystemthat extend beyond technology constraints andopportunities.

DISCUSSION AND CONCLUSION

This study sheds light on the forces that determinethe pace of technology substitution by presenting aframework that identifies when and how technologycompetition is resolved. It considers the relative bal-ance between the emergence challenges confrontingthe new technology and the extension opportuni-ties available to the old technology as a key gov-ernor of these dynamics. Underlying this approachis a view that, in order to understand progress ina given technology, we need to take into accountprogress in the surrounding ecosystem in whichthe focal technology is embedded (e.g., Hughes,1983; Rosenberg, 1976). We apply this frameworkto study 10 episodes of technology competitionthat have occurred in the semiconductor lithogra-phy equipment industry from 1972 to 2009. We findthat the observed dynamics are fully consistent withour predictions. The framework also helps to clarifythe puzzling variance in transition paths observedin the industry despite the existence of structuralsimilarities across the different cases of technologycompetition.

In addition to these core findings, our in-depthstudy of the semiconductor lithography equipmentindustry sheds light on the different modes offirm-level actions that shaped the observed patternof substitution. We found that at times, the pace ofsubstitution was slowed due to “last gasp” effortsby some firms to maximize the value that theycould capture from the old technology while otherfirms aggressively pursued the new technology. Inother instances, R&D efforts in the new technologycreated a “spillback” effect and helped firms to alsoimprove the performance of the old technology.Finally, we found evidence of heterogeneous actorsacross the ecosystem engaging in a collective “lastresort” effort to extend the old technology when metwith significant emergence challenges in the newtechnology.



Approaching substitution as an outcome of theinterplay between extension of the old technol-ogy and emergence of the new helps clarify obser-vations that highly discontinuous changes at onelevel of a technology hierarchy can be invisibleat other levels (e.g., Funk, 2008). In the contextof semiconductor industry, we observe that despitemany misses and delays in the commercializationof new lithography technologies, semiconductormanufacturers continued to make steady progressalong Moore’s law. Whenever semiconductor man-ufacturers were confronted with the prospect thattheir lithography tool suppliers would be unableto meet the industry roadmap requirements, theyproactively guided resources to find ways of bridg-ing the gap with the old technology until a newtechnology solution became available. This “lastresort” reallocation of efforts had significant impacton the returns to lithography equipment firms’investments in both the old and new technologygenerations.

By focusing on ecosystem dynamics that under-lie technology transitions, we are able to draw a linkbetween two of the most fundamental life-cyclepatterns identified in the innovation community:the S-curve profile of technology improvement andthe S-curve profile of technology adoption. Whilethe primary use of the technology S-curve has beento understand a given technology’s performancetake-off and limit so as to guide resource allocationinto existing and new technologies, the primaryuse of the adoption S-curve has been to understandthe given technology’s market potential and salestake-off so as to guide market expectations andcommercialization strategies for new technologies(e.g., Agarwal and Bayus, 2002; Bass, 2004; Fisherand Pry, 1972). Studies of adoption patterns focuson the dynamics of the demand environment (e.g.,Griliches, 1957; Rogers, 2003), but most oftenassume that both the old and new technologies arestatic (Hall, 2004). Conversely, studies of tech-nology evolution focus almost exclusively on therelative performance position of the new and oldtechnologies, and tend to assume that a technologywith superior performance will automatically takeover a given market. In doing so, they have tendedto overlook the role of demand-side adoption indetermining the pace and extent of technologysubstitution (Adner, 2004).

Our paper informs and clarifies the relationshipbetween these two fundamental life-cycle patterns.Specifically, we note that adoption and substitution

are intimately linked–a user’s decision to adoptthe new technology is, implicitly, a decision notto adopt the incumbent technology. Hence, whatlooks like adoption from the new technology’sperspective looks like substitution from the oldtechnology’s perspective. They are two sides ofthe same coin. For this reason, understanding theadoption of a new technology requires not only adetailed demand side analysis of consumer hetero-geneity and information diffusion mechanism (e.g.,Rogers, 2003) but also consideration of the supplyside factors that govern the technological progresswithin an ecosystem. As with substitution, the paceof adoption will be a function of the dynamicsof emergence in the new technology’s ecosystemand the dynamics of extension in the incumbenttechnology’s ecosystem. The emergence of a newtechnology may require major improvements in thecomplements for the average user to derive the fullbenefits, and this may slow the adoption rate. Sim-ilarly, the extension of a new technology throughimprovements in components, architectures orcomplements may also slow the adoption of thenew technology. The framework that we develop inthis study is able to explain why new superior tech-nologies may face slow market adoption and whyseemingly geriatric technologies may continue todominate.

In this regard, our approach has some pre-dictive implications. Historically, S-curve analy-sis has focused on the performance profile ofa stand-alone technology. One problem with theS-curve is that the time to performance take-off isnotoriously difficult to specify ex-ante. Focusing onthe broader technological ecosystem expands theanalysis beyond the time to take-off for the sin-gle focal technology, to include the performanceprofile of the entire ecosystem on which the focaltechnology’s performance as realized by the userdepends on. This expansion of the consideration setcan improve our ability to make predictions. In aninterdependent system, the realized performance bythe user cannot take off until every element is ready.For this reason, the most challenged element actsas a bottleneck for the whole system. This has animportant implication: the take-off of the focal tech-nology’s technical performance will be realized bythe user only if the focal technology is itself thebottleneck of the system. If the focal technology isnot the bottleneck, overcoming its own developmentchallenge will not resolve the ecosystem’s emer-gence challenge.



Another critique of the S-curve representation oftechnology evolution is the inconsistency betweenthe theorized smooth S-shaped and the observedirregular pattern of technology evolution (Sood andTellis, 2005). By taking an ecosystem view of tech-nology and by considering technology transitions asan interaction between new technology’s ecosystememergence challenge and old technology’s ecosys-tem extension opportunity, our approach helps toprovide some micro-foundations for the observeddifferences in the pattern of technology evolution.

Although this study is conducted at the levelof technologies, its findings have a number ofimplications for firms. First, we propose thatrather than viewing technology substitution asthe race between focal technologies, firms viewthem as the race between technology ecosystems.This view can inform the way in which firmschoose to allocate their valuable R&D resourcesaccording to the likely benefits from investingearly in the new technology or continuing to focuson the old technology (e.g., Adner and Kapoor,2010; Furr and Snow, 2014). Second, the explicitconsideration of technological interdependenciesand asymmetries in the rate of advance amongthe different ecosystem elements can help firmsconsider various modes of coordination rangingfrom vertical integration to collaborative alliances(e.g., Hughes, 1983; Kapoor, 2013).

The logic of our framework can also helpfirms in their resource allocation choices betweentechnology development initiatives, investment inecosystem complements, and investment in com-plementary assets such as those in manufacturing,marketing, sales and distribution. In Quadrant1, with the old technology stagnant and the newtechnology unhampered, firms should aggressivelyinvest in the new technology, following the tradi-tional prescriptions for embracing change to wardoff the winds of creative destruction. In contrast,when emergence challenges for the new technologyare low and the extension opportunities for the oldtechnology are high (Quadrant 2), firms with estab-lished positions in the old technologies can afford tomaintain their investment focus on the old technol-ogy for a longer time, knowing that the technologieswill coexist for an extended period. Alternatively,when emergence challenges for the new technologyare high and the opportunity for extending the oldtechnology is low (Quadrant 3), firms may be betteroff allocating a greater proportion of resourcestowards the development of complementary assets

than those in the development of either the oldor new technologies. Most favored here would bespecialized complementary assets that translatewell across both technologies (e.g., Mitchell, 1989;Rothaermel and Hill, 2005). Finally, in Quadrant 4,firms may be best served in investing in ecosystemcomplements that enhance the performance of theold technology or offer an advantage in unleashingthe potential of the new technology. Here, our paperextends the literature exploring the role of firms’complementary assets during periods of techno-logical change (e.g., Mitchell, 1989; Teece, 1986).For example, Rothaermel and Hill (2005) find thatindustry- and firm- level heterogeneity in the face ofcompetence destroying technologies are impactedby the nature of the complementary assets (genericversus specialized) required to commercialize thenew technology. We complement their work byfocusing on complementary technologies in theecosystem that surround both the old and the newfocal technologies, and by exploring transitions ina setting where technological discontinuities arenot competence destroying.

Finally, our study also provides some usefulinsights for policy. While we suggest that the rateof technology substitution depends on the relativedifference in the emergence challenges and exten-sion opportunities that confront the new and oldtechnologies, we note that three of the four typesof substitution regimes are characterized by sig-nificant technical progress fueled either by the oldor the new technology. However, when emergencechallenges are high and extension opportunities arelow, industries may face stagnation. In Quadrant 3,policy makers may have to create additional incen-tives and mechanisms through which the emergencechallenges with the new technology can be over-come in a timely manner. In addition, a robust tech-nology policy may also consider alternative newtechnologies that may be more likely to overcomeemergence challenges.

Although our focus is on substitution dynam-ics at the level of technologies, it is clear thatthese dynamics result from the aggregation of deci-sions and actions of individual firms. We havetaken the magnitude of the technology challengesand opportunities as relatively exogenous but notethat the resolution of these challenges and realiza-tion of these opportunities are highly endogenous,depending on the resources, capabilities, motiva-tion and strategies of firms across the ecosystem.Our elucidation of how ecosystem dynamics at the



technology-level influence the pace of substitutiongives rise to a new set of research questions at thefirm-level. For example, while scholars have stud-ied the relative advantages and disadvantages ofincumbents during technology transitions accord-ing to differences with respect to competences(Tushman and Anderson, 1986), complementaryassets (Tripsas, 1997), cognition (Henderson andClark, 1990) and customer dependencies (Chris-tensen, 1997), our study raises an additional dimen-sion of difference that is rooted in the interactionbetween the old and new technology ecosystems.We expect that the performance difference betweenincumbents and entrants would vary across the dif-ferent quadrants within our framework. Similarly,while numerous studies have explored how firms’resources and capabilities influence their entry andperformance outcomes in new technologies (e.g.,Franco et al., 2009; King and Tucci, 2002; Mitchell,1989, 1991), our study suggests that a more holis-tic evaluation would consider not only how firmscompete in new technologies but also how theytransition out of the old and into the new with anexplicit recognition of the interdependencies in theecosystem.

As in any investigation of an open system, acritical choice regards where to draw the bound-aries of the analysis (Scott, 2003). In this studywe limit our attention to the technological system(e.g., Constant, 1980; Hughes, 1983; Rosen-berg, 1976). Within this system, we focus on theexternal elements that link directly to the focaltechnology–the components and complements thatcombine with the focal technology to determinerealized performance. Outside the scope of ourstudy, but a promising direction for future work, isthe host of non-technological social and economicfactors (e.g., coordination costs; distribution ofpower among participants; industrial organizationacross sectors; identity; institutional, regulatory,and societal concerns) that may play a criticalrole in determining the timing and outcomes oftechnology transitions. We expect exploration ofthese dynamics to be particularly fruitful and hopeour study has paved the way for scholars to pushthis research agenda forward.

ACKNOWLEDGMENTS

We thank Rajshree Agarwal, Phil Anderson, MaryBenner, Constance Helfat, Peter Golder, Dan

Levinthal, Paul Rosenbaum, Lori Rosenkopf,Dan Snow, Mary Tripsas, Christophe Van denBulte, Peter Zemsky, Arvids Ziedonis, and seminarparticipants at Columbia, HBS, INSEAD, MIT,USC, UNC, Wharton, and the attendees at the 2010West Coast Research Symposium for their helpfulcomments. We are grateful to editor Rich Bettis andthe many anonymous reviewers for their guidanceand feedback. We especially thank VLSI Researchfor their generosity in sharing their data with usand the many industry professionals who providedus with valuable insights.

REFERENCES

Abernathy WJ, Clark KB. 1985. Innovation: mappingthe winds of creative destruction. Research Policy 14:3–22.

Adner R. 2002. When are technologies disruptive: ademand-based view of the emergence of competition.Strategic Management Journal 23: 667–688.

Adner R. 2004. A demand-based perspective on technol-ogy life cycles. In Business Strategy Over the IndustryLifecycle: Advances in Strategic Management (Volume21), Baum J, McGahan A (eds). Elsevier: Oxford, UK;25–43.

Adner R. 2006. Match your innovation strategy to yourinnovation ecosystem. Harvard Business Review84(4): 98–107.

Adner R. 2012. The Wide Lens: A New Strategy forInnovation. Penguin: London, UK.

Adner R, Kapoor R. 2010. Value creation in innovationecosystems: how the structure of technological inter-dependence affects firm performance in new technol-ogy generations. Strategic Management Journal 30(3):306–333.

Adner R, Levinthal DA. 2001. The emergence of emergingtechnologies. California Management Review 45(1):50–66.

Agarwal R, Bayus B. 2002. The market evolution and salestakeoff of product innovations. Management Science48(8): 1024–1041.

Anderson PC, Tushman ML. 1990. Technological discon-tinuities and dominant designs. Administrative ScienceQuarterly 35(4): 604–633.

Ansari S, Garud R. 2009. Inter-generational transitions insocio-technical systems: the case of mobile communi-cations. Research Policy 38(2): 382–392.

Bass FM. 2004. Comments on "A new product growth formodel consumer durables the bass model". Manage-ment Science 50(12S): 1833–1840.

Christensen CM. 1997. Innovator’s Dilemma: When NewTechnologies Cause Great Firms to Fail. Harvard Busi-ness School Press: Boston, MA.

Cusumano MA, Mylonadis Y, Rosenbloom RS. (1992).Strategic maneuvering and mass-market dynamics: Thetriumph of VHS over Beta. Business history review,66(01), 51–94.



Christensen CM, Rosenbloom RS. 1995. Explainingthe attacker’s advantage: technological paradigms,organizational dynamics, and the value network.Research Policy 24(2): 233–257.

Constant EW. 1980. The Origins of the Turbojet Revolu-tion. Johns Hopkins University Press: Baltimore, MD.

Dranove D, Gandal N. 2003. The Dvd-vs.-Divx standardwar: empirical evidence of network effects and prean-nouncement effects. Journal of Economics & Manage-ment Strategy 12(3): 363–386.

Ethiraj SK. 2007. Allocation of inventive effort in complexproduct systems. Strategic Management Journal 28(6):563–584.

Farrel J, Saloner G. 1986. Installed base and compatability:innovation, product preannouncements, and predation.American Economic Review 76(5): 940–955.

Fisher J, Pry R. 1972. A simple substitution model oftechnological change. Technological Forecasting andSocial Change 3: 75–88.

Foster R. 1986. Innovation: The Attacker’s Advantage.Summit Books: New York.

Franco AM, Sarkar MB, Agarwal R, Echambadi R. 2009.Swift and smart: the moderating effects of technologi-cal capabilities on the market pioneering-firm survivalrelationship. Management Science 55(11): 1842–1860.

Funk J. 2008. Components, systems, and technologicaldiscontinuities: lessons from the IT sector. Long RangePlanning 41(5): 555–573.

Furr NR, Snow D. 2014. Inter-generational hybrids: spill-backs, spillforwards, and adapting to technology dis-continuities. Organization Science. http://pubsonline.informs.org/doi/abs/10.1287/orsc.2014.0930 (26 Febru-ary 2015).

Griliches Z. 1957. Hybrid corn: an exploration in theeconomics of technological change.

Hall B. 2004. Innovation and diffusion. In The OxfordHandbook of Innovation, Fagerberg J, Mowery D,Nelson RR (eds). Oxford University Press: Oxford, UK;459–485.

Harley CK. 1971. The shift from sailing ships tosteamships, 1850–1890: a study in technologicalchange and its diffusion. In Essays on a MatureEconomy: Britain after 1840 (Volume xv), McCloskeyDN (ed). Princeton University Press: Princeton, NJ;439 pp.

Helfat CE, Klepper S. 2007. Firm and industry evolutionand entrepreneurship (FIVE project): data overview(November 2, 2007). In 2007 Kauffman Symposiumon Entrepreneurship and Innovation Data. Availableat SSRN: http://ssrn.com/abstract=1022116. (accessed10 January 2008).

Henderson R. 1993. Underinvestment and incompetence asresponses to radical innovation: evidence from the pho-tolithographic alignment equipment industry. RANDJournal of Economics 24(2): 248–270.

Henderson RM. 1995. Of life cycles real and imaginary:the unexpectedly long old age of optical lithography.Research Policy 24(4): 631–643.

Henderson RM, Clark KB. 1990. Architectural innovation:the reconfiguration of existing product technologies andthe failure of established firms. Administrative ScienceQuarterly 35(1): 9–30.

Hibbs M, Kunz R, Rothschild M. 1995. 193-nm lithog-raphy at MIT Lincoln lab. Solid State Technology 38:69–78.

Hughes TP. 1983. Networks of Power: Electrification inWestern Society 1880–1930. Johns Hopkins UniversityPress: Baltimore, MD.

Iansiti M. 1998. Technology Integration: Making CriticalChoices in a Dynamic World. Harvard Business SchoolPress: Boston, MA.

Iansiti M, Levien R. 2004. The Keystone Advantage:What the New Dynamics of Business Ecosystems Meanfor Strategy, Innovation, and Sustainability. HarvardBusiness School Press: Boston, MA.

Iyengar R, Van den Bulte C, Valente T. 2011. Opinion lead-ership and social contagion in new product diffusion.Marketing Science 30(2): 195–212.

Kapoor R. 2013. Collaborating with complementors:what do firms do? Advances in Strategic Management30: 3–25.

Kapoor R, McGrath PJ. 2014. Unmasking the interplaybetween technology evolution and R&D collaboration:evidence from the global semiconductor manufacturingindustry, 1990–2010. Research Policy 43(3): 555–569.

Katz ML, Shapiro C. 1985. Network externalities, compe-tition, and compatibility. American Economic Review75(3): 424–440.

King AA, Tucci CL. 2002. Incumbent entry into new mar-ket niches: the role of experience and managerial choicein the creation of dynamic capabilities. ManagementScience 48(2): 171–186.

Levinthal DA. 1998. The slow pace of rapid technologi-cal change: gradualism and punctuation in technolog-ical change. Industrial and Corporate Change 7(2):217–247.

McGahan AM. 2004. How Industries Evolve: Principlesfor Achieving and Sustaining Superior Performance.Harvard Business School Press: Boston, MA.

Mitchell W. 1989. Whether and when? Probability andtiming of incumbents’ entry into emerging indus-trial subfields. Administrative Science Quarterly 34(2):208–230.

Mitchell W. 1991. Dual clocks: entry order influences onincumbent and newcomer market share and survivalwhen specialized assets retain their value. StrategicManagement Journal 12(2): 85–100.

Moore JF. 1993. Predators and prey: a new ecology ofcompetition. Harvard Business Review 71(3): 75–86.

National Advisory Committee on Semiconductors. 1991.MICROTECH 2000 Workshop Report. U.S. Govern-ment Printing Office: Washington, DC.

Rogers EM. 2003. Diffusion of Innovations. The FreePress: New York.

Rosenberg N. 1976. Perspectives on Technology. Cam-bridge University Press: Cambridge, UK.

Rosenberg N. 1982. Inside the Black Box: Technology &Economics. Cambridge University Press: Cambridge,UK.

Rothaermel FT, Hill CW. 2005. Technological discontinu-ities and complementary assets: a longitudinal study ofindustry and firm performance. Organization Science16(1): 52–70.



Scott WR. 2003. Organizations: Rational, Natural andOpen Systems. Upper Saddle River, NJ: Prentice-Hall.5th ed.

Snow D. 2008. Extraordinary efficiency growth in responseto new technology entry: the carburetor’s “Last Gasp”.Harvard Business School Working Paper #1668643.

Sood A, Tellis GJ. 2005. Technological evolution & radicalinnovation. Journal of Marketing 69(3): 152–168.

Stremersch S, Tellis GJ, Franses PH, Binken JLG. 2007.Indirect network effects in new product growth. Journalof Marketing 71(3): 52–74.

Taylor A. 2010. The next generation: technology adop-tion and integration through internal competition innew product development. Organization Science 21:23–41.

Teece DJ. 1986. Profiting from technological inno-vation: implications for integration, collaboration,licensing and public policy. Research Policy 15(6):285–305.

Terasawa T, Hasegawa N, Kurosaki T, Tanaka T. 1989.0.3-micron optical lithography using a phase-shiftingmask. Proceedings of the International Society forOptical Engineering 1088: 25–33.

Tripsas M. 1997. Unraveling the process of creativedestruction: complementary assets and incumbent sur-vival in the typesetter industry. Strategic ManagementJournal 18(6): 119–142.

Tripsas M. 2008. Customer preference discontinuities: atrigger for radical technological change. Managerialand Decision Economics 29(2–3): 79–97.

Tushman ML, Anderson P. 1986. Technological discon-tinuities and organizational environments. Administra-tive Science Quarterly 31(3): 439–465.

Utterback JM. 1994. Mastering the Dynamics of Inno-vation: How Companies can Seize Opportunities inthe Face of Technological Change. Harvard BusinessSchool Press: Boston, MA.

APPENDIX I: SEMICONDUCTORLITHOGRAPHY EQUIPMENTGENERATIONS AND THEIREMERGENCE CHALLENGES

1. Contact Printers (First Generation)

Contact printing was the earliest and the simplestof the lithography technologies to be commercial-ized in 1962. The modifier “contact” corresponds tothe fact that the mask and the semiconductor waferwere in direct contact with each other. The contactprinters in the 1960s included a stage system forsecuring the mask and the wafer, and an alignmentunit that ensured that the patterns from the maskwere accurately transferred to the wafer. They useda mercury lamp for their light source. Contact print-ers did not incorporate a lens.

2. Proximity Printers (Year of Introduction: 1972;Years to Market Dominance: 2)17

A primary disadvantage of contact printing waslow process yield. This was due to the damageinflicted on the mask and wafers as they wererepeatedly brought in and out of physical contactwith each other during the lithography process.Proximity printing offered a way to overcome thisproblem. With proximity printing the mask andthe wafer were separated by a tiny gap. As dis-cussed in detail by Henderson and Clark (1990),this generational transition can be characterized asan architectural innovation and imposed signifi-cant challenges on the focal lithography equipmentfirms. The transition to proximity printing did not,however, present major challenges to any of thecomplements in the lithography ecosystem.

3. Projection Scanners (Year of Introduction: 1973;Years to Market Dominance: 2)

The continuous need of semiconductor manu-facturers to reduce circuit dimensions and get bet-ter manufacturing yields led to the introduction ofanother architectural innovation. Projection scan-ners introduced the use of lens systems as a com-ponent in lithography tools. The lens system wascomposed of a series of reflective mirrors whichallowed the image of the mask to be transferredto the wafer. While the development of projectionscanners entailed the development of the lens sys-tem, the commercialization of this generation didnot pose any significant ecosystem emergence chal-lenge. Although masks had to be etched with corre-spondingly smaller geometries and greater accuracywhich required that mask makers switch from usingstep and repeat cameras to using the more accurateelectron beam systems in their production process,this change was somewhat incremental.

4. E-beam Writers (Year of Introduction: 1976;Years to Market Dominance: –)

Electron-beam (E-beam) technology involvespatterning the resist on the semiconductor waferdirectly using electron beams that follow apre-programmed pattern, thereby eliminating theneed for a mask. Since a pre-programmed electronbeam travels across the wafer to achieve very low

17 Market dominance is defined as the first year in which thenew generation exceeds the annual market share of the old(previously dominant) generation. In generations marked (–) thenew technology did not achieve dominance.



resolution, the time required to complete a singlewafer can be as high as 10 hours, which is almost anorder of magnitude longer than the alternate opticallithography technologies. The major ecosystememergence challenge posed by E-beam technologywas the development of new resist chemistries thatwork with the emitted electrons and that wouldallow for sufficiently small geometries; that is, aresist in which exposure to electron beam wouldtrigger a chemical reaction only in the moleculesthat were directly exposed to the energy, withoutsetting off a reaction in adjacent molecules.

5. X-ray Printers (Year of Introduction: 1978; Yearsto Market Dominance: –)

The use of X-rays for lithography was proposeddue to their very low wavelength of less than 10 nm.In the early 1970s, X-ray lithography was developedas a simple proximity imaging system. As in prox-imity printing, the radiation from an X-ray source istransmitted through the mask onto the resist. X-raylithography’s very low wavelength created substan-tial ecosystem emergence challenges. The most sig-nificant challenge included the development of asuitable mask that would allow X-ray to accuratelypass through and achieve the desired chemical reac-tion with the resist.

6. G-line (Year of Introduction: 1978; Years toMarket Dominance : 5)

The G-line stepper introduced two key mod-ifications. First, light was projected through themask on to the wafer using a refractive lens system(as opposed to the reflective lens system used inprojection scanners). Second, the light was pro-jected on only a part of the wafer at any one time;the mask was shifted across the wafer in steps,such that multiple exposures are made across thewafer to complete the lithography process. Thissignificantly eased the challenge of mask making asthe circuit patterns on the mask could now be 10×or 5× of the dimensions that need to be printed onto the wafer. While resist suppliers had to developnew novolac-based materials in order to achieve theminiaturization objectives of the G-line generation,it did not present a major ecosystem emergencechallenge.

7. I-line (Year of Introduction: 1985; Years toMarket Dominance: 7)

I-line steppers, introduced in 1985, used lightwith a wavelength of 365 nm to improve over

the resolution achievable with the G-line gen-eration. The generation required development ofa lens system that would transmit light at thelower wavelength. However, the ecosystem comple-ments of mask and resist required only incrementalchanges for the transition from the G-line to theI-line generation.

8. DUV 248 (Year of Introduction: 1988; Years toMarket Dominance: 11)

The DUV 248 generation entailed a furtherreduction in wavelength into the deep ultraviolet(DUV) spectrum at 248 nm. The reduction createda high degree of ecosystem emergence challengerequiring fundamental changes in the mask andthe resist. The challenges imposed on mask mak-ers were overcome by changing the mask materialfrom soda lime glass to quartz in order to provideimproved transmission of the 248 nm wavelength.This, in turn, required major changes to the maskmanufacturing process. The existing novolac resistscould not absorb enough energy from the newwavelength to cause an adequate chemical reac-tion. As a result, new chemically-amplified resistshad to be developed for semiconductor manufactur-ers to create fine circuits using the new lithographytechnology.

9. DUV 193 (Year of Introduction: 1996; Years toMarket Dominance: 11)

The industry’s drive towards finer resolutionscontinued when tools using the 193 nm wavelengthwere introduced in 1996. As was the case for theDUV 248 nm technology, the very low light wave-length required major developments in the resistformulation so that this new generation could cre-ate value for users. With the change to the 193 nmwavelength, the existing resists, which were engi-neered to react to the 248 nm wavelength, wereno longer adequate to the task–a new genera-tion of chemically amplified resist needed to bedeveloped.

10. DUV 157 (Year of Introduction: 1998; Years toMarket Dominance: –)

The subsequent attempts to achieve greaterminiaturization resulted in the development oftools using the 157 nm wavelength. The reductionin wavelength created transmittance problemsand required that the mask substrate and the



pellicle materials to be completely redeveloped toeffectively transmit the low wavelength light.

11. DUV 193-immersion (Year of Introduction:2005; Years to Market Dominance: 4)

The transition to DUV 193-immersion wasenabled by an architectural innovation in whichliquid (instead of air) is used as a medium betweenthe lens and the wafer. The generation requiredmajor efforts by tool producers to manage the flowof liquid in this complex architecture. The tran-sition did not, however, present major innovationchallenges to any other elements in the lithography

ecosystem. As the wavelength of light did notchange, the new generation was able to reuse theexisting resist and mask with relatively minorimprovements.

SUPPORTING INFORMATION

Additional supporting information may be foundin the online version of this article:

File S1. Innovation ecosystems and the pace ofsubstitution: re-examining technology S-curves.


Documents

INNOVATION ECOSYSTEMS AND THE PACE OF …d1c25a6gwz7q5e.cloudfront.net/reports/2016-02-03... · S-CURVES RON ADNER1 and RAHUL KAPOOR2* 1 Strategy and Management, Tuck School of Business,