You may have heard it before. There’s a silicon shortage, chip shortage, graphics card shortage, or whatever name it’s given. I’ve recently done extensive research into the issue for a supply chain management course at the London School of Economics. I’ll share my findings with you in two parts. Today we’ll focus on understanding the industry.
Subscribe to my newsletter so you don’t miss the second part. I write about finance, management, valuation, Excel, supply chains, and much more. Try it out, I won’t spam you and it’s easy to unsubscribe (no hard feelings).
Let’s dive right in!
Introduction
“It’s now clear to all: we are living in a semiconductor world” is the opening of the McKinsey report on the semiconductor industry (McKinsey, 2021). Whereas the chips powering our modern economy were largely invisible and taken for granted in the years prior, the COVID-19 crisis suddenly made them a hot topic on the world stage. The root cause: supply chain issues and upstream shortages.
The semiconductor industry is a consolidated, highly interdependent, and globally integrated network of specialized firms. Many of these firms hold virtual monopolies in their fields due to high research & development (R&D) costs (Hamilton, 2021). To unravel the effects of the COVID-19 crisis on its supply chains, we must first seek to understand the industry and its key players.
The supply chains of the semiconductor industry are global by nature and feature few highly specialized and interdependent players. The industry is typically divided into seven sectors: intellectual property (IP) cores, electronic design automation (EDA) tools, specialized materials and chemicals, wafer fab equipment (WFE), fabless chip companies, integrated design manufacturers (IDMs), and chip foundries (Blank, 2022).
The industry
The first part of chip design is IP cores. IP cores can be seen as modules designed by one party for use by others. These come in two variants: soft cores (which can be modified under the license) and hard cores (which cannot be changed). IP cores are created by many firms (Blank, 2022). At first sight, this may logically mean that there is little chance of supply chain issues created by IP cores. However, this disregards the nature of IP cores. IP cores are used very early in the design process, meaning that changing out an IP core can mean an entire overhaul of a chip’s design. Furthermore, hard cores are created with one specific foundry in mind, and a single chip can contain many IP cores. In the event of a foundry disruption, that means all the chip’s IP cores have to be recreated for the new foundry by their creators before they can be produced in the new foundry (Yeo et al., 2010). Even with EDA tools, it can take a large team of engineers multiple years to design a chip, meaning that switching costs are prohibitively high and reacting to short-term supply chain issues is nearly impossible.
The second part of chip design is EDA tools. EDA tools are used to integrate and extend the licensed IP cores. The market is an oligopoly, with Cadence, Mentor, and Synopsis (all US-based) together holding a 60% market share (Zhihao, 2020). EDA tools greatly automate parts of the chip design process (Blank, 2022).
Moving from chip design to chip production, an obvious place to start is the elements the chip is made with. Specialized materials and chemicals are the building blocks of the physical chip. A chip is primarily made of silicon delivered to the chip factory in the form of a silicon wafer. A silicon wafer is ready to be used in a lithography process (basically printing the chip design on the wafer using light). The lithography process uses a variety of materials and chemicals. According to Gasworld, over 100 gases are used in semiconductor production (Stockman, 2018). Lastly, a collection of chemicals is used intensively in the process. Their main purpose is to be used as photoresist and top coats, vital parts of the photolithography process.
Diving deeper into silicon wafer production, we again observe a highly consolidated market. The two largest players, Shin-Etsu and GlobalWafers, hold 29.4% and 26.7% of the market, respectively (Lapedus, 2021). Importantly, silicon wafers come in different sizes used by different fabs to produce different chips. Historically, wafer size has been increasing. Currently, the newest and largest size is 450mm, but production of this size is still ramping up. 300mm is the most common size. 200mm has been around since 1989 and saw its use decline initially after the introduction of 300mm. In recent years demand for 200mm production has rebounded, leading to shortages of 200mm fab capacity. For example, many of the chips used by the automotive industry are 200mm.
Wafer fab equipment (WFE) is the heart of any chip fabrication plant (‘fab’). They are the machines that print chip designs on silicon wafers using light. There are currently two main ‘lines’ of WFE: DUV (deep ultraviolet) and EUV (extreme ultraviolet). The difference between the two is that the wavelength of EUV light is smaller, allowing for smaller transistor ‘nodes’ (such as today’s 5nm). Components typically needed for their construction are lenses and mirrors, among others. The WFE industry is highly consolidated, with ASML, KLA, LAM, Applied Materials, and Tokyo Electron producing DUV-based photolithography machines (Davuluri, 2021). The newer EUV machines are only produced by ASML. Currently, 5nm chips can only be produced using EUV, meaning that ASML holds a monopoly on the current ‘node’(Tasdemir et al., 2018).
Fabless chip companies can be seen as chip designers that do not have a proprietary chip fabrication plant (‘fab’). Well-known examples are Apple, AMD, Nvidia, and Qualcomm. These firms combine IP cores and proprietary designs with external fab capacity to create proprietary silicon, such as Apple’s recently introduced M1 chip. These companies can either sell the produced chips to other manufacturers and systems integrators (AMD) or use them in their own production processes (Apple). Either way, they do not own WFE, nor do they arrange the supply of chemicals & materials. They are, however, completely reliant on fab capacity that is beyond their control.
Integrated design manufacturers are vertically integrated semiconductor companies that span the entire production process. A notable example of this is Intel for logic processors. Others include SK Hynix and Micron (memory) and Texas Instruments (analogue processors). These firms have their own IP cores, use EDA tools, own WFE, and have their own foundries (fabs). They historically only produce their own designs. Intel has, however, been making future capacity available to external parties.
Chip foundries, or ‘fabs’, are the factories that produce the chips. They integrate WFE and materials & chemicals in complex production processes. Crucially, they do not design chips. The foundry industry is highly concentrated, with leader TSMC holding 52.1% of the market in Q4 2021. The second-largest firm is Samsung, at 18.3% (Statista, 2021). Foundries are very capital intensive. With construction costs typically increasing with each new node, estimates for a 3nm fab are around USD 20Bn (Hamblen, 2019). The high consolidation of the industry has sparked geopolitical debate around the reliance on Taiwan for vital production of semiconductors, with the US and EU developing plans to incentivize self-reliance through subsidies on domestic foundry construction in a bid to harden supply chains (Hamilton, 2021). Figure 1 provides a schematic overview of the industry, including consolidation at each point along the supply chain.
What’s next?
Now that you have a broad overview of the industry I’m curious to hear your initial impressions and thoughts. Why do you think we’re seeing a semiconductor shortage? What’s the supply chain bottleneck? Next week I’ll publish my own take on it.
Subscribe to my newsletter so you don’t miss the second part. I write about finance, management, valuation, Excel, supply chains, and much more. Try it out, I won’t spam you and it’s easy to unsubscribe (no hard feelings).
References
Blank, S. (2022). The Semiconductor Ecosystem — Explained. https://steveblank.com/2022/01/25/the-semiconductor-ecosystem/
Davuluri, A. (2021, January 29). Chip Equipment’s Bright Prospects Are Etched in Stone and Silicon | Morningstar. Morningstar. https://www.morningstar.com/articles/1019474/chip-equipments-bright-prospects-are-etched-in-stone-and-silicon
Hamblen, M. (2019, November 4). TSMC starts building 3nm plant in Taiwan worth $20B | Fierce Electronics. Fierce Electronics. https://www.fierceelectronics.com/electronics/tsmc-starts-building-3nm-facility-taiwan-worth-20b
Hamilton, D. S. (2021). Enhancing Semiconductor Supply Chain Resilience and Competitiveness : Recommendations for U . S . -EU Action. Transatlantic Leadership Network.
Lapedus, M. (2021). More Silicon Wafer Consolidation. Semiconductor Engineering. https://semiengineering.com/more-silicon-wafer-consolidation/
McKinsey. (2021). McKinsey on Semiconductors. Mckinsey & Company, 1, 80.
Statista. (2021). Top semiconductor foundries market share 2021. https://www.statista.com/statistics/867223/worldwide-semiconductor-foundries-by-market-share/
Stockman, P. (2018). gasworld | 53.
Tasdemir, Z., Wang, X., Mochi, I., Lent-Protasova, L. van, Meeuwissen, M., Custers, R., Rispens, G., Hoefnagels, R., & Ekinci, Y. (2018). Evaluation of EUV resists for 5nm technology node and beyond. Https://Doi.Org/10.1117/12.2502688, 10809, 85–94. https://doi.org/10.1117/12.2502688
Yeo, Ng, Kong, & Dang. (2010). Intellectual Property for Integrated Circuits. In Paper Knowledge . Toward a Media History of Documents. J. Ross Publishing.
Zhihao, C. (2020, April 28). Analysis of the global EDA software industry market competition landscape in 2020 — Code World. Qianzhan. https://www.codetd.com/en/article/12224646