Globalising innovation through co-inventions–the success case of the Korean lithium-ion battery industry

Radical innovations can shift the global competitiveness of entire nations. While countries typically struggle to absorb knowledge about novel technologies quickly, in which knowledge tends to be spatially sticky, an important exception is the fast catch-up of the Korean Li-ion battery industry from Japan in the early 2000s. In this paper, we conduct an exploratory case study on this surprising success story. Focussing on patent co-inventions between Korea and Japan, we investigate their significance, as well as underlying types of co-inventions and types of transferred knowledge. To this end, we proceed in four steps: (1) a Poisson regression model; (2) social network analyses; (3) patent inventor tracking and (4) patent coding. Our results indicate that Korean–Japanese co-inventions hold significantly greater influence than other cross-country co-inventions, including with patents without cross-country collaboration. We find a pronounced knowledge-transfer intensity during the early 2000s and observe two types of co-inventions: organisation-level and inventor-level. Predominantly, we observe inventor-level co-inventions, i.e. Korean companies hiring experienced Japanese engineers, that proved important to transferring sticky knowledge. Moreover, while most patents target the design of core battery components, the share of manufacturing patents are—contrary to theoretical expectations—highest during the first half of the observation period. We also discuss our findings and draw implications for policy, industrial and academic players, including industry localisation policies, technology-inherent catch-up strategies and directions for future research.


Introduction
Technological change-i.e. the invention, innovation and diffusion of novel technologies-is a key lever for economic development and competitiveness and important to address societal issues such as climate change (Mowery et al 2010, Acemoglu et al 2012).As such, radically novel technologies that alter the foundations of entire global sectors can shift the competitiveness of nations (Kwasnicki 2013) and are often transferred globally (Brewer 2008, Bell 2012, Lema and Lema 2013).International technology transfer commonly is categorised as part of the flow of capital goods (e.g.production equipment), intangible human skills (e.g.operation and maintenance of machines) and intangible human knowledge and experience (e.g.innovation) (Lema and Lema 2012).The first two flows increase production capacity, while human knowledge transfer7 primarily advances local innovative capacity and catalyses local innovation (Bell and Pavitt 1992, Kathuria 2002, Bell 2009, Binz and Truffer 2017).
Analytical explanations of why certain knowledge types (and, thus, technologies) seem easier to transfer than others often draw from the product architecture and complexity literature (Huenteler et al 2016a, Scheifele et al 2022;e.g. Schmidt andHuenteler 2016, Surana et al 2020).Product architecture refers to the organisation and interfaces between a technology's subsystems (Simon 1962, Clark 1985).A commonly used architecture typology is the categorisation of subsystems into core and periphery system groups (Tushman andRosenkopf 1992, Baldwin et al 2014).Technological changes to a core subsystem typically evoke changes to other subsystems, whereas modifying a peripheral subsystem rarely affects other subsystems (Tushman and Rosenkopf 1992).
A concept closely related to product architecture is complexity, i.e. subsystems' interrelatedness within the product architecture (Baldwin et al 2014).In other words, complexity gauges how much overall change is induced by altering a single subsystem.The link between product complexity and knowledge transfer has been established in the literature (Lema and Lema 2012, Balland and Rigby 2016, Schmidt and Huenteler 2016, Wouden, 2019, Malhotra and Schmidt 2020, Surana et al 2020, Meissner et al 2021, Matsuo et al 2022) and often builds on the concept of sticky knowledge.Hippel (1994) coined this term, which refers to certain types of information (or knowledge) that are empirically difficult to transfer.Ultimately, it has been suggested that complex technologies typically come with sticky (inventorspecific) knowledge and are challenging to transfer (Balland andRigby 2016, Wouden andRigby 2019).
Yet, a good understanding of transfer mechanisms is crucial (Bozeman 2000, Stephan et al 2019) given that important technologies are classified as complex, e.g.Lithium-Ion Batteries (LIB) (Stephan et al 2017, Malhotra et al 2019), and faster globalisation rates will support the fight against the climate crisis (IPCC 2022).While a sizable amount of literature has investigated knowledge transfers within or across organisations (e.g.Song et al 2003, Etzkowitz 2008, Dima and Vasilache 2015, Isaksen and Trippl 2017), other studies have focussed on the spatial dimension, such as regional (e.g.Almeida and Kogut 1999) and trans-national (e.g.Jaffe et al 1993, Miguélez et al 2010, Binz et al 2014, Glachant and Dechezleprêtre 2017) transfers.One important angle to knowledge transfer is the co-invention mechanism, i.e. patent collaborations between inventors and/or organisations.Many studies have used co-inventions to investigate the value and determinants of co-inventions (e.g.Lei et al 2013, Cassi and Plunket 2014, Wouden and Rigby 2019, Meissner et al 2021).Amongst these determinants, a prominent one is inventors' mobility (pattern) and properties (e.g.Almeida and Kogut, 1999, Breschi and Lissoni 2009, Kerr and Kerr 2018, Wouden and Rigby 2019).In this context, valuable insights can be derived from using social network analysis, as done by, e.g.Breschi and Lissoni (2009), Gonçalves et al (2023) or Wouden and Rigby (2019).
So far, no study has investigated the dynamics and patterns of international co-inventions, i.e. how coinventions came about, and combined these insights with a content analysis of patents, i.e. what types of knowledge were created through co-inventions.Only through the combination of these two aspects can we learn about the interdependence of technologyinherent characteristics and co-invention dynamics.This study encompasses the following contributions: First, a methodological combination of co-invention dynamic and patent content analysis is proposed.Second, we apply the proposed methodology to the unlikely but successful case of Korean catch-up from Japan in the LIB industry.We contribute to the existing understanding of successful catch-ups for this important technology.Importantly, by leveraging knowledge transfer and product architecture literatures and applying our proposed methodology, we offer a combined insight into the types of coinventions ('how') and types of transferred knowledge ('what').Thirdly, this study discusses the generalisability of our findings and contributes to ongoing questions by closing with implications for policymaking and research.

Case selection
Our study centres on LIB for two primary reasons: its significance from an innovation theory perspective and its global relevance.First, LIBs are dual complex in their design and manufacturing, characterized by highly interdependent product components and manufacturing steps (Schmidt and Huenteler 2016, Stephan et al 2017, Beuse et al 2018).This complexity presents an intriguing theoretical case (Seawright and Gerring 2008), particularly given the rapid market share capture by Korean companies from Japanese industry leaders since the late 1990s (figure 1), a development contradicting theoretical expectations  1), our study focuses on the first phase of the globalisation of innovative activities, where, intriguingly, China does not yet demonstrate a strong presence (figure 3).Instead, Japan and Korea maintain the lead in terms of high-quality patenting activities.Furthermore, as shown in figure 3, the early 2000s witnessed exceptionally high co-invention rates between Korea and Japan, which merit particular attention in our analysis.Thus, our exploratory research design zeroes in on the period between late 1990s and early 2010s of Korean catch-up from Japan, a time when China did not yet emerge as a leader in LIB production.
The second reason for our focus on LIBs is their relevance, underscored by their transformative impact on the global automotive sector.Accelerated by rapid cost reductions (Ziegler andTrancik 2021, IPCC 2022), LIBs are crucial for the widespread adoption of electric vehicles and are, thus, lowering the market entry barriers by altering the merit order of industry capabilities (Christensen 2011, McKinsey&Company 2021).Their importance as a low-carbon technology is further recognized through significant policy attention, including the US Inflation Reduction Act and the EU Battery Regulation (Peiseler et al 2022).Additionally, the strategic shift towards resource independence, particularly from China, underscores LIBs' growing relevance for industry stakeholders and policymakers (Davidson et al 2022).

Methods and data
We prepare the analysis by assembling and imputing a global LIB patent database8 to identify Korea-Japan (KR-JP) co-inventions (Step 0 in table 1).The subsequent analysis encompasses four steps, each addressing separate research questions (see table 1): (1) estimating a Poisson regression model to identify the relative importance of KR-JP patents; (2) selecting patents with KR-JP co-inventions to understand collaboration patterns; (3) tracking the most central Japanese inventors' publication history to shed light on their professional trajectory and (4) coding influential patents' product hierarchy and product share to shed light on the type of transferred knowledge.
As a prerequisite for our analysis (Step 0), we assemble a global LIB patent database following the methodology proposed by (Malhotra et al 2021).By applying LIB-specific keywords and International Patent Classification (IPC) codes, we used the PATSTAT 2021 Spring version to extract a data set comprising 152918 patent families 9 filed between 1915 and 2020.Extensive automated and manual data cleaning was conducted to reduce data impurities, particularly from inconsistent naming and country code conventions (see section SI 1.1).
In the first analytical step, we identify all KR-JP co-inventions in the imputed data set.Viewing a patent's forward citations, i.e. the number of times a patent was cited by other patents within five years of its publication, as a proxy for innovative significance (Trajtenberg 1990, Hall et al 2005, Popp and Newell 2012), we regress them on a dummy indicator of KR-JP co-inventions in a Poisson model with countryyear fixed effects using pseudo-maximum likelihood estimation (Correia et al 2020).As control variables, we include the number of backward citations, patent family size, respective number of applicants and inventors, and dummies indicating whether a patent was granted; whether it was granted in the US, EU and Japan (so-called 'triadic' patents) and whether a patent involved a cross-country collaboration among 9 A patent family bundles all patents from multiple jurisdictions that protect the same invention.To avoid double counting, this study uses simple patent families as the unit of analysis.Please refer to SI section 1.1.2for more information.For simplicity, patent and patent family will be used interchangeably throughout the paper.inventors in general.To avoid distortions through organisation-specific characteristics, we also examine additional fixed effects at the patent applicant level (for more details, see section SI 1.2).
In the second step, we illuminate KR-JP coinvention characteristics via social network analysis.For this, we construct a co-invention KR-JP subset by selecting patents that have Korean and Japanese inventor and/or applicant country codes.table SI 8 provides a breakdown of the different types of coinventions.This query reduces the sample size from 152918-705 patent families and 710 distinct legal persons (individuals and organisations).The resulting data set is materialised as a social network diagram in which nodes represent legal persons, and connecting edges symbolise co-inventions (listing on one patent either as co-inventors, co-applicants or inventor and applicants) between them.
The third step shifts the unit of analysis from patent metadata to individual inventors to understand their professional history.We obtain the 10 most important Japanese inventors from the network created during the previous step using eigenvector centrality as a measure of importance (see section SI 1.4).Once identified, we used the global LIB patent database (Step 1) to track these 10 individuals' patent publications and organisational affiliation histories based on PATSTAT data, Google Scholar and additional desk research.
The fourth and final step focuses on the content of the most influential KR-JP patents cited five or more times within five years of their publication (see section SI 1.5).This step reduces our sample size from 705 to 204 patents.Subsequently, the product architecture is coded manually using the patents' titles and abstracts.We follow the LIB architecture framework that (Malhotra et al 2021) proposed and code the patents' product share.Furthermore, we group patents into core and periphery inventions (see SI section 1.5) to enhance our understanding of each type of transferred knowledge further.To achieve this, we manually categorise each claim of all patents as either a product or process invention, then determine their respective proportions with equal weight assigned to each claim.

Results and discussion
Step 1 provides insights into the relative importance of patent co-inventions for the LIB co-inventor network.Figure 2 presents the estimated regression coefficients, as well as the 95% confidence intervals (see table SI 4 for more details).The figure provides the statistical association (x-axis) of the various regressors (y-axis) on the forward citations for model specifications with country-year fixed effects (blue) and additional first applicant fixed effects (red).We find that cross-country co-inventions are associated significantly only with higher forward citations if we do not control for the first applicant's timeconstant features via fixed effects.However, there is a statistically significant (p < 0.01) difference between general cross-country co-inventions and KR-JP coinventions, with the latter indicating over 25% higher forward citations (see table SI 4).As the point estimate for I (CrossCountry) is positive, this implies that KR-JP co-inventions also exhibit significantly higher forward citations vis-à-vis patent families without any cross-country co-invention.Importantly, these findings represent statistical correlations and not necessarily causal effects.However, they strongly suggest that patent co-inventions involving Korean and Japanese inventors, indeed, are associated with more influential patents than other patents filed in the same country and year by the same organisation.However, the findings on other international co-inventions are less conclusive, indicating a potentially outstanding role by KR-JP co-inventions.
Figure 3 illustrates the evolution of LIB patents over time in Korea, Japan, and China.To account for varying national patenting practices and in line with existing patent work (Nagaoka and Naotoshi 2014, Cheng and Drahos 2018, Tahmooresnejad and Beaudry 2019), we focus on triadic patents, which are high-quality patents protected in Europe, the US, and Japan.The data reveals that Japan initially had and has maintained the highest count of triadic granted patents, with Korea rapidly quickly catching up.In contrast, despite having the largest number of overall patent applications (figure SI 4(A)), China has fewer applied triadic (figure SI 4(C)) and granted triadic patents (figure 3).Additionally, we find that Korea leads in the success ratio of granted to applied triadic patents at 57%, followed by Japan at 42% and China at 26%.
The dashed line in figure 3 represents the share of patents that are KR-JP co-invented as a proportion of all Korean patents.While international co-invention shares for LIB patenting is generally at around 5%, low country-specific co-invention shares for Japan and Korea of around 2%-3% are established by the European Patent Office (EPO) (2020).In light of this, it is all the more striking that we observe KR-JP coinvention shares over 20% for the late 1990s and early 2000s (figure 3).In other words, during this period, more than 20% of Korean high-quality patents were co-invented with Japanese engineers or organisations.A comparison of figures SI 4(A) and (D) reveals that the share of KR-JP co-inventions for high-quality patents was about twice as high as for all patents, emphasizing the significant role of knowledge transfer for high-quality patents. In Step 2, we turn to the types of co-inventions and their dynamics.Figure 4(A) presents a two-mode social network in which the nodes represent legal persons as either inventors (natural person: circle shape) or applicants (organisation: star shape).Edges connect two nodes if (1) two inventors, (2) two applicants or (3) one inventor and one applicant filed at least one patent collaboratively.The network's visual layout, i.e. the nodes' positions, is optimised computationally following a force-directed algorithm to bring connected nodes closer together than unconnected ones (see SI section 1.3), i.e. the resulting clusters form due to co-inventor proximity.Several of these clusters comprise only a handful of nodes and are displayed in the network diagram's outer belt.Concomitantly, they tend to be connected loosely (in some cases completely isolated) to other parts of the network.
More strikingly, however, are the two prominent clusters emerging in the centre of the network.As figure 4(A) indicates, both clusters are pivoting around two large Korean companies, namely Samsung and LG.These two companies account for over 70% of the co-invention patents, confirming not only their uncontested position in the Korean industrial landscape, but also suggesting an active role in co-invention and knowledge transfer.Furthermore, the types of co-inventions seem to differ between the two clusters: while the 'Samsung cluster' features many individual Japanese (red) nodes, only a few Japanese inventors can be found in the 'LG cluster' .The co-invention data indicate that Samsung  maintained a dense, but diversified, network of multiple Japanese entities (mostly inventors), while LG co-invented largely with a single Japanese company (Toray).The two emerging clusters illustratively suggest two types of observed co-inventions.Notably, expert interviews (see SI section 1.6) reference varying levels of in-house knowledge to different company roots for LG (chemical) and Samsung (electronics).Consequently, this diversity in foundational expertise is suggested to influence the respective co-invention needs and strategies.
First, we observe an inventor-level co-invention characterised by high pertinence and asymmetric directionality (figure 4(B)).Early on (until 2010), KR-JP research teams patented inventions for Korean firms, while in later years (starting in 2010), fully Japanese teams worked for Korean companies.The early prevalence of mixed KR-JP inventor teams ties in well with Korean companies' goals to train 'own' engineers as fast as possible through social interactions with Japanese engineers (Howells 1996, Dima andVasilache 2015).Korean firms Samsung (1992) and LG ( 2010   .Nodes represent legal persons as either inventors (natural person: circle shape) or applicants (organisation: star shape).Edges connect two nodes if the nodes filed at least one patent collaboratively.The diagram's layout is optimised using a force-directed Fruchterman-Reingold algorithm (see section SI 1.3).(B) Types of co-inventions over time.Time-aggregated data can be found in table SI 8. 'Appln' refers to applicant(s) and 'inv' to inventor(s).The data presented are based on a three-year rolling average with a minimum rolling window size of two years.activities in Japan starting around 2010, although limitations in patent data prevent the confirmation of a direct causal link.
Second, we observe an organisational-level coinvention involving Korean and Japanese organisations jointly filing and owning patents.Contrary to inventor-level co-inventions, which emphasise the knowledge of individuals, organisational coinventions extend to broader forms of knowledge, e.g.overarching technical understanding or research project structures (Argote et al 2003).In the case of Cluster 2 in figure 4(A), Korean company LG and Japanese company Toray jointly filed patents, yet only Korean engineers were the inventors.This implies that Toray's contribution was not in individual human, but potentially organisational, knowledge.Additional company-specific observations can be found in SI section 3.1.
Figure 4(A) contains all combinations of KR-JP co-inventions, e.g. the network contains patents for inventions by Japanese engineers that Korean companies filed, but also vice versa and every combination in between (see legend of figure 4(B) or table SI 8).For the remainder of the analysis, it is crucial to understand knowledge flows' directionality (figure 4(B)).An analysis, proxied by the types of co-inventions, revealed that the overwhelming majority of patents were filed by Korean companies with at least one Japanese inventor (75%).Furthermore, 38% of all coinventions were filed by Korean companies and invented by only Japanese engineers.However, the reverse scenario, i.e.Japanese applicant and only Korean inventors, occurred in only 1.3% of the patents.
When examining types of co-inventions over time, figure 4(B) visualises that Japanese-filed and Korea-(co)invented patents were emerging only much later (2007) than Korea-filed and Japanese-(co)invented patents (1995).In the early years (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005), almost all patents were filed by Korean applicants with Japanese and, in most cases, Korean inventors.This trend flips starkly after 2010, when most Korea-filed patents were for inventions by only Japanese engineers.Thus, particularly during the early period, Korean engineers repeatedly were inventing and interacting with Japanese engineers, suggesting a dominant direction of knowledge flow: from Japan to Korea.
In Step 3, we aim to understand from where 'imported' knowledge was acquired.To this end, the 10 most central Japanese inventors in the network are identified and highlighted in figure 4(A) (see SI section 1.4 and table SI 9 for more information on node centrality and rankings).Figure 5 illustrates these 10 top inventors' academic and professional careers.The figure indicates that all 10 key inventors were trained in Japan, either at a Japanese university (light red) or a Japanese company (dark red), before (co-)inventing for Korean companies (blue).Moreover, most of the Japanese inventors were highly experienced, through academic and/or professional means.Interestingly, Korean companies like Samsung and LG did not necessarily require Japanese engineers to relocate to Korea: patent metadata and the top inventors' social media profiles suggest that at least five of the 10 top Japanese inventors worked at Korean-owned research facilities based in Japan (hatched blue).The results also suggest a certain mobility level within the global battery ecosystem, where two inventors worked for American and Chinese companies and two inventors returned to Japanese institutions.
At the beginning of the early 2000s, Korean companies held a negligible market share, but only five years later (2005), they already possessed almost 20% (figure 1).During this period, Samsung already had hired four out of the 10 top Japanese inventors from figure 5.In particular, these engineers brought a wealth of experience, with three of them having each accumulated over 15 years of professional experience in Japan.Furthermore, Japanese engineers' typical employment duration at Korean companies notably was relatively short, typically shorter than 10 years.
While the previous results largely target types of co-inventions (Steps 2 and 3), the remainder of the results is dedicated to types of transferred knowledge (Step 4) through co-invention.As LIB are complex in terms of product and process design, successfully catching up requires capabilities in both fields.Figure 6 presents the development of the product vs. process contributions from influential KR-JP coinventions with five or more forward citations.Most of the highly cited patents were product or hybrid inventions, i.e. patents that include product and process claims (cf figure SI 6).This suggests that Korean companies used co-inventions with Japanese engineers mostly to create intellectual property related to the product design of LIB, as triangulated through expert interviews (see SI section 1.6).
More intriguingly, we also find that process and hybrid shares are highest during the first half of the observation period (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006), occasionally surpassing mere product patents.This contrasts with the technology life-cycle literature, which predicts a phase of product innovation preceding process innovation for complex, mass-produced products (Utterback and Abernathy 1975, Murmann and Frenken 2006, Malhotra et al 2021).One potential reason for the high early proportion of process and hybrid patents is the data sampling of this study (coinventions as a subset of global LIB patents).The analysed patents were filed mainly by Korean companies seeking to catch up in an existing market, potentially missing the phase of product-focussed innovation during early market formation (1990s) 10 .
Besides understanding the share of product vs. process patents, understanding co-inventions' dynamics in terms of product architecture also is essential.Three key observations emerge from our results (figure 7).First, it becomes evident that almost all co-inventions entail core inventions (green).However, peripheral inventions (purple) played a minor role in the very early years (1996)(1997)(1998)(1999)(2000) and during the late 2000s before declining further in significance once again.Second, the focus of co-inventions evolved across core components over time.Initially, anodes were prominent, later shifting to electrolytes, then to separators around 2010, and finally back to electrodes, mirroring global trends (Malhotra et al 2021).Cell subsystem inventions consistently held moderate levels, ranking second in total patents.This high pertinence may point towards the ongoing challenge of integrating multiple core components and addressing product complexity.Third, upon combining product share (figure 6) and product architecture (figure 7) information as presented in figure SI 8, we note that no pure process inventions were found for peripheral patents.

Implications for catching up
This research examines the surprisingly fast catchup of the Korean LIB industry from Japan, focussing on knowledge transfer through international co-invention.Our findings suggest that KR-JP coinventions are significantly more influential than other cross-country co-inventions, as well as patents without cross-country collaboration.Furthermore, we observe two types of co-inventions and show that predominantly, by means of inventor-level coinventions, experienced Japanese engineers played a marked role in patenting activities of Korean companies, particularly in the early 2000s.Moreover, we also find that while most key patents targeted product inventions, the process and hybrid shares-contrary to theoretical expectations-are highest during the first half of the observation period.
LIBs were used predominantly for consumer electronics in the early 2000s, but have emerged as a crucial technology for a low-carbon future, garnering substantial attention from policymakers worldwide.In this study, we investigate a catch-up process viewed as unlikely within innovation research and demonstrate that, despite its low likelihood, rapid catchup in complex technologies is possible.As innovation theory (Lema and Lema 2012) has suggested, unconventional knowledge transfer mechanisms, e.g.overseas R&D and inventor collaborations, arguably played a significant role.
Of course, the long-standing knowledge exchange between Korea and Japan should be acknowledged (Kim 1999, Hu 2012, Park et al 2012).Additionally, the distinctive structure of Korea's industrial setup, which prominently features public-private partnerships and major industrial conglomerates known as 'Chaebols' , plays a crucial role in its economic landscape (Lee and Lim 2001, Kim 2010, Lim 2012).Beyond these specific factors, broader research underscores the importance of local capacity building (Lee 2009) and the utilisation of existing local knowledge and experience for effective industry localisation (Cohen and Levinthal 1990, Lema and Lema 2013, Binz and Truffer 2017, Scheifele et al 2022).Therefore, for companies pursuing advancement in battery technology or other complex technologies, it may be advisable to strategically align their existing knowledge base with the skills required for developing technology-specific capabilities.This approach is exemplified in figure 4(A), which, as suggested by an expert interview (see SI section 1.6), shows LG leveraging its strong background in chemistry to focus on knowledge transfer regarding separators, whereas Samsung, with roots in electronics, engaged in international co-inventions to enhance its expertise across various core components of LIBs.A more recent example is German BASF joint-venturing with Japanese Toda Kogyo for cathode material production and R&D in Japan to transfer knowledge for site construction and operation in Germany (BASF 2020(BASF , 2021)).
Contrary to traditional industrial policy and classical business strategy, which focus on industry localisation and strictly local capacity building, Korean companies successfully have established knowledge hub research centres in Japan.This example highlights how next to industry localisation through nationally tied subsidies, policy may also want to consider supporting overseas collaboration during periods of technological catch-up.Attracting experienced inventors and encouraging collaborations can be one way to accelerate globalisation of complex technologies.
From a research perspective, investigating incentive structures that drive experienced engineers to engage in knowledge transfers of this nature appears to be rewarding.A suitable research design for examining these incentives might entail conducting expert interviews with both engineers involved in such knowledge transfers and hiring managers.By gathering insights from these stakeholders, researchers could enhance the current understanding of the influence that corporate and public policies, along with their interplay, have had in promoting these forms of collaboration.This becomes increasingly important given that Korea is a prime example of successful catching-up through industrial policy that relies heavily on strong public-private relations (Lee andLim 2001, Lim 2012).Consequently, this information could help inform future policy work and shape the direction of research in this field.
Applying this research design to other cases, including comparative analyses and varying scopes on technologies and countries, could enhance our understanding of historic knowledge flows and accelerate global knowledge and technology transfer.To this end, the methodology developed in this study, supported by comprehensive documentation in the supplementary information and open-source code, could serve as a foundation for future analyses.To facilitate large-scale, meaningful regression analysis, enhanced data sets and corresponding methods should be made accessible, either centrally by organisations like the EPO or by individual research groups (Rassenfosse and Seliger, 2021).Automated language classifications, such as GPT API, can assist with data cleaning and patent coding, including abstracts or claims, thereby catalysing the potential to generate large-scale insights through the content of patents beyond metadata.

Figure 1 .
Figure 1.Aggregated historic Li-ion battery (LIB) market shares of Korean (blue), Japanese (red), Chinese (light grey) and other (grey) companies.Historic market shares are derived based on (NEDO 2010) and (KBIA 2014) and 2022 market shares for EV and stationary storage demand are based on (SNE Research 2023).

Figure 2 .
Figure 2.Estimated marginal effects on forward citations for specifications with different fixed effects (FEs) While categorical variables are denoted by 'I(…)' , for continuous variables, the plot displays the estimated impact of a +1 standard deviation to improve comparability.For the full regression table, see SI section 2).

Figure 3 .
Figure3.Japanese (red),Korean (blue) and Chinese (yellow) patent activity, as well as the share of KR-JP co-inventions (dashed) over time.The data presented uses triadic patent counts, i.e. patents granted in the US, Europe and Japan, and are based on a three-year rolling average, with a minimum rolling window size of two years.
) set up LIB research centres in Japan (LG Japan Lab 2023, Samsung SRJ 2023).The academic research output from the Samsung R&D Institute Japan, shown in SI section 2.2, mirrors the pattern seen in the dark blue area of figure 4(B).This pattern may point towards increased Samsung research

Figure 4 .
Figure 4. (A) Two-mode network diagram ofKorean and Japanese co-inventions (1995-2020).Nodes represent legal persons as either inventors (natural person: circle shape) or applicants (organisation: star shape).Edges connect two nodes if the nodes filed at least one patent collaboratively.The diagram's layout is optimised using a force-directed Fruchterman-Reingold algorithm (see section SI 1.3).(B) Types of co-inventions over time.Time-aggregated data can be found in table SI 8. 'Appln' refers to applicant(s) and 'inv' to inventor(s).The data presented are based on a three-year rolling average with a minimum rolling window size of two years.

Figure 5 .
Figure 5. Academic and professional careers of the 10 most central Japanese inventors.Academic affiliations predominantly pertain to research activities, although in a few ambiguous instances, they also refer to educational backgrounds.

Figure 6 .
Figure 6.Product vs. process share of KR-JP patents with ⩾ 5 forward citations.The numbers on top of the bars indicate post-rolling sample size.The data presented are based on a three-year rolling average with a minimum rolling window size of two years.The initial and final years of the observation period were aggregated to ensure a minimum of five observations after application of the rolling average.Raw data can be found in figure SI 7.

Figure 7 .
Figure 7. Product architecture of KR-JP patents with ⩾ 5 forward citations.Purple (green) colouring indicates peripheral (core) inventions.The numbers on top of the bars indicate the post-rolling sample size.The data presented are based on a three-year rolling average, with a minimum rolling window size of two years.The observation period's initial and final years were aggregated to ensure a minimum of five observations after application of the rolling average.Raw data can be found in figure SI 7.

Table 1 .
Overview of methodological steps.please refer to the supplementary information (SI) for a detailed description of the steps.
Lema R and Lema A 2012 Technology transfer?The rise of China and India in green technology sectors Innov.Dev. 2 23-44 LG Japan Lab 2023 LG Japan lab (available at: www.lgjlab.com/company/company.html)(Accessed 19 December 2023) Lim W 2012 Chaebol and industrial policy in Korea Asian Econ.Policy Rev. 7 69-86 Liu W, Tao Y and Bi K 2022 Capturing information on global knowledge flows from patent transfers: an empirical study using USPTO patents Res.Policy 51 104509 Malhotra A and Schmidt T S 2020 Accelerating low-carbon innovation Joule 4 2259-67 Malhotra A, Schmidt T S and Huenteler J 2019 The role of inter-sectoral learning in knowledge development and diffusion: Case studies on three clean energy technologies Technol.Forecast Soc.146 464-87 Malhotra A, Zhang H, Beuse M and Schmidt T 2021 How do new use environments influence a technology's knowledge trajectory?A patent citation network analysis of lithium-ion battery technology Res.Policy 50 104318 Matsuo T, Malhotra A and Schmidt T S 2022 Catching-up in green industries: the role of product architecture Innov.Dev.1-30 McKinsey&Company 2021 Why the automotive future is electric Meissner D, Burton N, Galvin P, Sarpong D and Bach N 2021 Understanding cross border innovation activities: the linkages between innovation modes, product architecture and firm boundaries J. Bus.Res.128 762-9 Meyer M 2002 Tracing knowledge flows in innovation systems Scientometrics 54 193-212 Miguélez E, Moreno R and Suriñach J 2010 Inventors on the move: tracing inventors' mobility and its spatial distribution Pap Reg.Sci.89 251-74 Mowery D C, Nelson R R and Martin B R 2010 Technology policy and global warming: why new policy models are needed (or why putting new wine in old bottles won't work) Res.Policy 39 1011-23 Murmann J P and Frenken K 2006 Toward a systematic framework for research on dominant designs, technological innovations, and industrial change Res.Policy 35 925-52 Nagaoka S and Naotoshi T 2014 Assessing the Effects of International Research Collaboration on the Invention Process: Some Evidence from Triadic Patent Data (Intellectual Property for Economic Development.Edward Elgar Publishing) ch 7, pp 159-79 NEDO 2010 蓄電技術開発室 2010-2011 (Energy storage technology development room 2010-2011) Park Y W, Amano T, Moon G and Hong P 2012 Benchmarking open and cluster innovation: case of Korea Benchmark.Int.J. 19 517-31 Peiseler L, Bauer C, Beuse M, Wood V and Schmidt T S 2022 Toward a European carbon footprint rule for batteries Science 377 1386-8 Peiseler L, Jun Y L, Schmid N, Waidelich P, Malhotra A and Schmidt T 2024 Repository for KR/JP co-invention battery patent innovation study Zenodo https://doi.org/10.5281/zenodo.10636806Popp D and Newell R 2012 Where does energy R&D come from?Examining crowding out from energy R&D Energy Econ. 34 980-91 Samsung SRJ 2023 Samsung R&D Institute Japan | company profile (available at: www.samsung-srj.co.jp/en/profile/) (Accessed 18 December 2023) Scheifele F, Bräuning M and Probst B 2022 The impact of local content requirements on the development of export competitiveness in solar and wind technologies Renew.Sustain.Energy Rev. 168 112831 Schmidt T S and Huenteler J 2016 Anticipating industry localization effects of clean technology deployment policies in developing countries Glob.Environ.Change 38 8-20 Seawright J and Gerring J 2008 Case selection techniques in case study research Polit.Res.Q. 61 294-308 Simon H A 1962 The architecture of complexity Proc.Am.Phil.