Policy Brief

Who is developing the chips of the future?

Authors

Julia Christina Hess

Anna Semenova

Programmes

Global Chip Dynamics

Published by

Interface

June 16, 2021

Introduction

Semiconductors are a foundational technology and indispensable for any modern society. Their application goes far beyond typical information and communication technologies such as smartphones, laptops and cloud infrastructure. Hospitals, car manufacturers and electricity companies, to name just a few, all rely on access to increasingly complex chips. The global semiconductor value chain that produces those chips has received a lot of attention from policy makers and the media. It is at the heart of the on-going US-China technology rivalry and several regions, including the European Union, try to strengthen their own semiconductor industry in an effort to become less reliant on foreign technology providers.

But an often overlooked aspect of the semiconductor industry is the amount of research & development (R&D) necessary to advance the cutting-edge. The chips industry has one of the highest R&D margins across all industries – semiconductor companies easily spend on average more than 18% of their revenue on R&D. Furthermore, the vast majority of R&D is done by just a handful of countries and they are at the center of this analysis.

Why focus on R&D? Recent policy initiatives in the US, Europe and South Korea pay a lot of attention to semiconductor manufacturing – how best to subsidize fabrication plants (fabs) and increase domestic wafer capacity. This is understandable, not least because of skyrocketing investment costs of modern fabs. But the semiconductor industry also faces a lot of technological challenges in terms of energy efficiency, sustainability, new materials and many more. Thus R&D at all process steps is quintessential. While it might matter, from the perspective of geopolitics and geoeconomics, where chips are manufactured, it matters at least as much who develops – defines and shapes – the chips of our future.

In a first attempt to assess the R&D power in semiconductors of different countries and regions, SNV’s Data Science Unit teamed up with the Technology and Geopolitics program to analyze paper contributions to three of the leading academic semiconductor conferences:

The International Electron Devices Meeting (IEDM) covers technological innovations from semiconductor and electronic device technology to design, manufacturing, physics and modeling. International research throughout the whole semiconductor value chain is presented. IEDM started in 1955.
The International Solid-State Circuits Conference (ISSCC) brings together leading experts in the field of solid-state circuits and systems-on-a-chip covering scientific achievements at the different stages of the design-process of semiconductors. ISSCC started in 1954.
The Symposia on VLSI Technology and Circuits (VLSI) connects two international multistakeholder-conferences on semiconductor technology and circuits with the aim to create synergies on topics of joint interest ranging from process technology to systems-on-chip. The VLSI symposia are focused on research about manufacturing and design of very-large-scale integration (VLSI) circuits. VLSI started in 1981.

Following are some key insights from this quantitative analysis, and we invite you to explore the data yourself in the interactive charts at the end of the report. To learn more about our method, its limitations and challenges please have a look at the Frequently Asked Questions at the bottom. To learn more about SNV’s Data Science Unit feel free to reach out to Pegah Maham and if you are interested in SNV’s work on semiconductors & geopolitics you can contact Jan-Peter Kleinhans.

Analysis

Insight #1: For the past 25 years US and Japan developed the chips of the future

Accumulated Paper Contributions from 1995 to 2020

What you see: The above chart shows the accumulated paper contributions per country over the past 25 years (1995-2020). For example, Taiwanese organizations or universities (co-)authored 1201 research papers in the past 25 years across all three conferences (IEDM, ISSCC, VLSI). Thus, for 1201 papers at least one author came from a Taiwanese organization or university.

What it means: First, it is hard to overestimate the importance of US and Japanese organizations and universities for semiconductor R&D. Together, both countries account for more paper contributions over the past 25 years than the rest of the world combined (10.338 against 8.187). While the research power of countries is shifting over time, especially in the case of Japan, their overall importance and contribution to semiconductor R&D is substantial.

Second, only a handful of regions develop the chips of the future. The US, Japan, Europe, South Korea, Taiwan and China are not just the most important regions for the semiconductor value chain but also by far the most important regions for semiconductor R&D. The five leading countries – US, JP, KR, TW, BE – contributed to around 75% of all papers over the past 25 years.

Insight #2: The US and EU have consistently high research power

What you see: The above chart shows the share of paper contributions per year for the US and EU28 (including UK). As an example, in 2005 the EU (co-)authored 21% of all conference papers and the US (co-)authored 40% of all conference papers.

What it means: First, The US is a semiconductor R&D powerhouse – on average more than 40% of the conference papers have been authored or co-authored by an US organization or institution. Furthermore, the US was able to sustain such a high level of R&D contribution over the past 25 years, partly due to the US’ large chip design industry. US semiconductor companies control around 50% of the global chip market (sales) since almost three decades. Since the semiconductor industry has one of the highest R&D margins – companies invest around 18% of their revenue in R&D – higher revenues directly translate into more R&D power.

Second, the EU was able to almost double their paper contributions over the past 25 years (1995: 13%, 2020: 25%) even though European companies only hold a miniscule market share of below 10% in global chip sales since three decades. Comparing a regions research power to their market share refers to the question of how well innovation is translated into invention. Having a significantly lower market share in contrast to research capabilities reveals potential hurdles and inefficiencies in the European ecosystem to get value from R&D.

Insight #3: Deep Dive Europe - member states with leading RTOs contribute the most

What you see: The above chart shows the relative share of paper contributions from EU member states by year. As an example, in 2020 Belgian organizations and universities (co-)authored 8.7% of all conference papers, while France contributed to 6.1% of all conference papers. The accumulated contributions by member state in this chart are higher than EU’s total contributions in the previous chart because the same paper could be co-authored by France, Belgium and Germany – giving each country +1 but the paper would only count as +1 for EU.

What it means: First, the vast majority of EU’s R&D power comes from a few member states – Belgium, France, Germany, Netherlands, Italy and the UK consistently accounted for more than 80% of EU’s paper contributions in the past 25 years. In 2020 Belgium and France alone made up more than half of EU’s total paper contributions.

Second, Belgium, France and Germany – the member states with the most paper contributions – also have important Research and Technology Organizations (RTO) for the semiconductor industry: imec (Belgium), CEA-Leti (France) and Fraunhofer (Germany). Belgium’s share of paper contributions increased substantial over the past 25 years since imec is at the forefront of node scaling and closely collaborates with TSMC, Samsung, Intel and others to manufacture ever smaller transistors.

Insight #4: China, South Korea and Taiwan significantly increased their R&D power

What you see: The above chart shows the relative share of paper contributions by China, South Korea and Taiwan per year. For example, China contributed to 5%; South Korea contributed to 8.8%; Taiwan contributed to 13% out of all conference papers submitted in 2014.

What it means: First, China, South Korea and Taiwan play an increasingly important role in semiconductor R&D – the three countries together contributed to more than 1/3rd of all conference papers in 2020. They are not just the manufacturing hubs of the semiconductor value chain anymore but deeply embedded in developing the chips of the future.

Second, especially China was able to significantly increase its R&D power over the past ten years. This is in line with the rise of globally competitive Chinese semiconductor companies, such as HiSilicon (Huawei), SMIC and Goodix. While China is highly dependent on foreign technology providers across the semiconductor value chain its raw research paper contributions are already higher than those of Belgium.

Third, Since South Korea and Taiwan are the most important countries for cutting-edge wafer fabrication, it makes sense that their companies (TSMC in Taiwan, Samsung and SK Hynix in South Korea) and research institutions (ITRI in Taiwan) are also heavily engaged in R&D.

What you see: The above chart shows the relative share of paper contributions from Chinese and Japanese organizations and universities per year across all three conferences. In 2015, Japan contributed to 18% of all conference papers while China contributed to 3.8%.

What it means: First, Japan’s research power fell significantly over the last 25 years – from almost 40% in 1995 to less than 10% in 2020. The same is true for Japan’s overall share in the semiconductor value chain: while in 1990 six out of the ten largest semiconductor companies (sales) were Japanese, in 2020 not a single Japanese company was among the top 10.

Second, against that, China’s development is particularly impressive. Only 10 years ago, they had no stake at all. Not until the last 5 years, China succeeded in approaching Japan and finally overtook Japan in 2020. As latecomer and region that has seen the fastest growth throughout the last 25 years, they more than doubled their share from 4 percent in 2015 to 10 percent in 2020.

Insight #6: EU’s most important research partner is the US, China overtook Japan as second in line

What you see: The above chart shows the number of collaborations between the EU and foreign countries. A hypothetical joint paper from imec (Belgium) together with Intel (US), Sony (Japan) and JCET (China) would count as +1 collaboration for each of these countries. Here, we selected the EU research collaborations with China, Japan and the United States. The lower chart shows the EU’s total count of paper contributions. In 2020, the EU published 11 papers in collaboration with Chinese organizations or universities (7% of all 164 EU papers), 4 in cooperation with Japan (2% of all EU papers) and every fifth paper (34 papers in total) is based on collaboration with the US (21%).

What it means: First, for Europe the most important R&D partner by far is the United States. In the last five years, at least 15% of EU’s contributed papers were based on R&D collaborations with the US.

Second, interestingly, next in line is China – since 2016 there are more co-authored papers with China than with Japan.

Insight #7: China teams up with prominent R&D partners

What you see: The above chart shows the number of collaborations between China and foreign countries. Here, we illustrate China’s paper collaborations with the EU, Taiwan and the US. In 2005, China contributed 3 (lower chart) conference papers – one of these papers was in collaboration with Taiwan and one in collaboration with the US.

What it means: First, as a fast-follower in the semiconductor industry, China focused on research collaborations from the beginning. The most important R&D partners for China are the United States and Europe: almost half of China’s conference papers are co-authored with Europe or the United States.

Second, the smaller chart at the bottom illustrates how quickly and substantially China’s R&D participation rose since 2010.

Conclusion

Who is developing the chips of our future? The answer to that question started to change over the past 25 years, even more so in the past 10. First, while the US is and continues to be a semiconductor research powerhouse, Asian countries – specifically China, South Korea and Taiwan – play an increasingly important role in semiconductor R&D. This should not come as a surprise in a value chain that heavily depends on R&D: Taiwan and South Korea are currently the only places in the world with the most advanced manufacturing capabilities at 7nm and below. To advance the cutting-edge and develop future manufacturing processes companies such as TSMC (Taiwan) or Samsung (South Korea) substantially invest in their R&D efforts – often in collaboration with European RTOs or US chip design companies. Thus, China, South Korea and Taiwan became more than just manufacturing hubs for chips but important research partners.

Second, research collaboration plays an increasingly important role. While the current policy debates about semiconductors in Europe, the US and China are dominated by narratives about “technological sovereignty”, “economic security” and “self-reliance”, international collaboration within the semiconductor research community substantially intensified: in 1995 only 11% of US’ paper contributions were co-authored with foreign research partners, while in 2020 around 36% of US’ paper contributions were based on international collaborations. Today, it takes 18 times more researchers to keep pace with Moore’s Law than in the 1970s – strengthening international research collaboration is the only way forward to overcome future technological challenges facing the semiconductor industry. Thus, policy makers would do well to incentivize international cooperation in semiconductor research.

Third, diminishing market shares seem to go hand in hand with decreasing R&D power: Japan’s global chips sales dropped from 49% in 1990 to a mere 6% in 2020 and in the same time its research contributions fell from 40% in 1995 to less than 10% in 2020. Whether these developments simply happened in parallel or if one lead to the other cannot be answered just through this data analysis. But there might be lessons (and warnings) for Europe, whose R&D power is substantially higher than its market shares partly thanks to very successful RTOs such as imec, CEA-Leti and Fraunhofer.

About

This is SNV’s first analysis of semiconductor R&D power to better understand who is developing the chips of the future, industry dynamics, power balances and cooperation within the global semiconductor value chain. Further analysis and additional insights will follow.

SNV’s Technology and Geopolitics project was made possible by the generous support of (in alphabetical order) the Dutch Ministry of Economic Affairs and Climate Policy, the Finnish National Emergency Supply Agency, the Finnish Ministry for Foreign Affairs, the German Federal Foreign Office and the Swedish Ministry for Foreign Affairs. The views expressed in this paper do not necessarily represent the official positions of these ministries.

SNV’s Data Science Unit is made possible by Stiftung Mercator.

Interactive Charts: Explore the data

In this section, you can explore our data on global paper contributions and paper collaborations throughout the last 25 years.

Paper Contributions (1995 - 2020)

You can toggle between a bar or a line chart and the share of contributions or the absolute values using the buttons above. Select the countries you want to compare on the right side next to the chart. The smaller chart at the bottom provides you with information about the total count of paper contributions per year.

Paper Collaborations (1995 - 2020)

First, you select one country from the drop-down list above the chart. Next, you choose which countries research collaborations with the selected county you want to display. You can then toggle between a bar or a line chart using the buttons underneath the drop-down list. There are two smaller charts at the bottom: The first one shows the selected countries total count of paper contributions. The second one refers to the total count of paper contributions per year.

The count for "No International Cooperation” includes all papers without any international collaboration. This means that even papers from a single author or institution are included.

FAQ

Metrics and Definitions

Our data basis are the paper sessions of the ISSCC, IEDM and VLSI (Technology and Circuits) conferences, for the years 1995-2020. In the introduction you can find the reasons why we have chosen these.

At the beginning of our count, in 1995, the lion’s share of semiconductor research was reviewed at the three conferences (IEDM; ISSCC; VLSI) our analysis is based on. Since then, the research landscape diversified and broadened. This goes hand in hand with the need for more and more application specific chips, e.g. in field of Artificial Intelligence or more Advanced Packaging Technologies. Additionally, these fora do not only cover semiconductor related research. The NeurIPS 2021 as an AI forum is one example for these developments, as research on ASICs is one major pillar of the conference. Thus, assessing R&D power tends to rely on multiple new conferences and our insights tend to be less representative if you focus on the last ten to five years.

Papers can be authored by multiple researchers or multiple organizations from the same country. We only count unique countries per publication. For example: a paper with three authors from Japan and two authors from Taiwan would add one point to Japan’s count and one point to Taiwan’s count.

Regarding the EU, a paper written by authors from multiple EU countries will increase the EU count only by one. For example, a paper written by three German and two Belgian authors will increase the EU count by one, as well as the German and Belgium count by one.

This is the reason why the sum of the contribution of EU members states can exceed the count of the EU: Papers authored by researchers from multiple EU countries contribute just one point to the EU count while adding a point to each country.

Collaborations are counted with respect to a country of interest (selectable in the drop-down menu in the chart above). With respect to this country, other countries get a count for every paper they have co-authored. We only count if any international cooperation took place, not to which extend: A USA, USA, Japan cooperation is treated like a USA, Japan cooperation. We do not distinguish between a paper by one author from a country and a paper with several different authors or institutions from the same country: Japan, Japan counts as the same as Japan and is displayed as “no international cooperation”.

This means that the sum of a countries’ collaborators can exceed the total amount of publications: Each paper can be co-authored with multiple countries. For example, if a country has only contributed one paper in a year, collaborating with two other countries, its total count for that year is one but the sum of collaborations is two.

A paper is often co-authored by more than one country: a paper co-authored by Japan, US, Belgium and Canada would give each country a +1 collaboration/contribution count for that year. Thus, the number of collaborations or contributions in year X can be higher than total conference papers in year X. Which is why the total % collaborations or contributions can be higher than 100%.

See “How do you count the paper contributions?”

See “How do you define and count collaborations between countries?”

We count all countries that are or have been at any time part of the EU28. This means that the set of countries of the EU count does not change over time. Even in years where a country was not yet a member state, they are included. The UK is included as well.

The China counts do not include Taiwanese or Hong Kong papers.

Data Collection and Accuracy

The data is obtained from the IEEE Xplore website, and was collected via the API for the ISSCC, IEDM and VLSI (Technology and Circuits) conferences, for the years 1995-2020. We have formulated our query based on each conference year Digital Object Identifier (DOI), which returns all available information on IEEE on publications including their authors and affiliated institutions.

Our conclusions rely on trend changes on a bigger scale than the scale of the error noise that we identified by various tests. Thus, we are confident that the accuracy is high enough for our conclusions.

We performed manual tests on a random sample of 30 country-year pairs where we compared our API based counts to manually obtained counts. In 21 cases there was no deviation between the counts. In the remaining cases the count only deviated by one. Reason for this are non-systematic errors in the API data, like missing or incorrect entries of the authors’ country affiliations. Another source of inaccuracy is the inclusion of 3-4 plenary session papers for the VLSI conferences, as they could not be excluded due to their data format.

Moreover, we collected 27 PDF booklets of ISSCC and IEDM conferences and compared our results against them. These comparisons confirmed our counts as mostly correct or with minor errors as described above.