TSMC is unique in the world, why?

Semiconductor Industry Observations

2024-05-15 12:08Posted on the official account of Anhui Semiconductor Industry Observation

TSMC is the absolute leader in chip manufacturing, with a stable market share of about 60%, compared to 11% of its biggest competitor, Samsung Electronics. As an oligopoly in the industry, TSMC's core competitiveness mainly includes business model, business philosophy, technology accumulation, capacity management, ecological construction, and enterprise structure, which is of great reference value to enterprises in the same industry. This article analyzes the core competitiveness of TSMC one by one.

Business model: Create an innovative business model of vertical division of labor in the chip field

Even in the era of artificial intelligence, Zhang Zhongmou always believes that the most valuable thing is still the innovation of business model. For example, eBay, Google, and Amazon in the United States, and Tencent and Alibaba in the mainland are all Internet-based business model innovations. Even in the seventies and eighties, when the Internet was not developed, companies such as McDonald's and Starbucks also showed the effectiveness of business model innovation.

At the beginning of the establishment of TSMC, the chip production market was the IDM model, that is, the business model of "one-stop" from chip design, wafer manufacturing to packaging and testing, and IDM manufacturers were Intel, AMD and other large manufacturers. The development of the semiconductor industry follows Moore's Law, which means that between one and a half to two years, the line width of the smallest size of a transistor is reduced by 0.7 times (about 50% of the area). Wafer manufacturing and packaging and testing in chip production involve heavy asset investment, and each generation of technology update involves large new capital expenditures such as new equipment and plants, so only IDM manufacturers have the financial strength to follow Moore's Law. The capital threshold restricts the main players in the industry to the leading enterprises, limiting the speed of technological development in the industry.

In view of the above-mentioned industry characteristics and pattern, the founder Zhang Zhongmou played the role of a professional wafer foundry at the beginning of TSMC's establishment, only manufacturing chips, not designing chips, playing the role of an "enabler" in the industry, allowing IC design customers to focus on design and innovation, and supporting a large number of potential IC design companies for professional production, prompting the entire semiconductor industry to open the "Fabless-Foundry" vertical division of labor operation model, which has gradually changed the global semiconductor industry landscape after 30 years.

Zhang's strategy is not only to work with well-established semiconductor companies, but also to find companies with growth potential as partners. He believes that only looking at ready-made large companies and not cultivating small chip design companies is not conducive to the long-term development of the wafer foundry industry. From around 2000 to the present, the IC design companies that have been supported by TSMC from small companies all the way to occupy a place in the industry include Qualcomm, Broadcom, NVIDIA, Xilinx, MediaTek, Marvell, Altera, LSI, Avago, Dialog, HiSilicon, etc.

Business principles: do not compete with customers, close service to customers

1. Do not compete with customers

Since its inception, TSMC has established the business principle of "never competing with customers" and continues to this day. TSMC's foundry customers are IC design companies, and the cooperation model is that the IC design companies deliver the internal design details of the chip to the foundry, and work deeply with the foundry to verify the performance and determine the manufacturing details. Therefore, in addition to its professional technology, TSMC's business principle of "not competing with customers" ensures that the interests of customers are synergistic. Although IDM manufacturers such as Samsung and Intel can carry out wafer foundry for IC design companies, design companies such as Apple, Qualcomm, and Nvidia are competitors of IDM's IC design team. In terms of customer trust, TSMC has incomparable advantages from IDM manufacturers such as Samsung and Intel, and it is also the secret of TSMC's success in the field of wafer foundry.

TSMC has become Apple's exclusive supplier by virtue of this advantage. For example, in 2015, in order to obtain sufficient stock, Apple adopted a "dual-supplier" model, placing orders for the A9 processor of the iPhone 6S to Samsung's 14nm factory and TSMC's 16nm factory. Apple's A-series chips are customized based on the ARM instruction set, which has a higher performance improvement than the A9 designed by the ARM standard architecture. In the first half of 2016, Samsung launched the Exynos 8890, a next-generation SoC that also uses 14nm, using the above-mentioned custom design architecture, and its single-core performance is similar to that of the A9. When it comes to the A10, Apple chooses TSMC as the exclusive supplier. Compared with Samsung, TSMC has no competition with Apple, and Apple does not have to worry about technology leakage when choosing TSMC, which is one of the reasons why TSMC has become the exclusive supplier of Apple A10.

In May 2017, Samsung spun off its foundry business after a business diagnosis and adopted the same business structure as TSMC to avoid direct competition with customers, demonstrating that it recognizes that customer trust is the most important foundation. After the division split, it will be beneficial for Samsung Semiconductor Division to further compete for foundry orders, so that it has the opportunity to close the gap with TSMC in the foundry business.

2. Serve customers closely

Maintain close liaison with the client's senior management. TSMC has a high concentration of customers, with the top 10 customers contributing 2/3 of the revenue and the top 20 customers contributing 80% of the revenue. The CEO of TSMC keeps in close contact with these 20~30 customers, so that customers can feel that they can communicate at any time if they have any problems.

At the executive level, TSMC has a dedicated team of several people for each customer to assist from design and development to production. For example, in order to allow Apple's IC design to be introduced into the TSMC production system faster, a team of teams stationed at Apple for one or two years at that time, successfully introduced Apple's design into the TSMC plant for production, significantly improving energy consumption, yield, and efficiency.

Technology: independent innovation, steady and steady, and maintain long-term advantages

TSMC has long occupied a leading position in technology, in addition to the annual R&D expenses maintained at more than 8% of the company's net income, the R&D focus is accurate, and the R&D model is reasonable, mainly due to the insistence on independent innovation, solid and stable technology accumulation, complete categories, and the adoption of correct competitive strategies.

1. Adhere to independent innovation and grasp the correct technical route

TSMC entered the market with 3.0 micron technology and has led technological innovation many times. TSMC grasps the correct technical direction at key nodes, adheres to independent technological innovation, and completes technological catch-up.

Before the 0.13 micron node, TSMC and UMC were known as the "two heroes in Taiwan", and the two wafer foundries were neck and neck, and UMC's operating income was once close to TSMC. The tipping point occurred at the 0.13 micron technology node. At that time, IBM was the first to release the new 0.13-micron technology and planned to sell it to TSMC and UMC. TSMC rejected IBM and developed its own copper process technology, and successfully developed it; UMC chose to purchase technology from IBM for joint development. IBM's technical strength is limited to the laboratory, and the yield rate in manufacturing is too low to reach the mass production level. In 2003, TSMC's 0.13 micron in-house process technology made a stunning debut, with customer orders of nearly NT$5.5 billion and UMC's NT$1.5 billion.

At the 28nm node, the process technology can be chosen as either gate-first or gate-last, with Zhang Zhongmou choosing the latter, while GF and Samsung choose the former. Since then, TSMC yields have increased rapidly, while Samsung and GF have been slow to make progress.

In the field of advanced packaging, as early as 2011, when the 28nm technology node was first launched, Zhang Zhongmou announced that TSMC had stepped into the field of packaging, which shocked the industry at the time and has now proved to be fruitful. While advanced manufacturing processes are approaching their physical limits and require astronomical investment for each step, advanced packaging is still at a cost-effective stage to improve chip performance in a more cost-effective way. TSMC's integrated fan-out (InFO) wafer-level packaging is first applied to the iPhone's A10 processor, further widening the gap with rival Samsung. And after more than ten years of deep cultivation, we have seen considerable revenue space in the AI era. AI and high-performance computing chips have an increasing demand for advanced packaging technologies, among which TSMC's CoWoS has become the main packaging technology used by AI server chip manufacturers.

2. Technology iteration is steady and steady

TSMC has a solid accumulation of technology, basically following the "Moore's Law" (between one and a half to two years, the line width of the smallest scale of the transistor is reduced by 0.7 times), and successively launched 90 nm, 65 nm, 45 nm, 28 nm, 20 nm, 14 nm, 10 nm, 7 nm, 5 nm, 3 nm, 2 nm and other technologies.

Based on the accumulation of the learning curve, whether it is in the process or the advantages of building a factory, it is a step-by-step experience inheritance, competitors can not achieve corner overtaking, TSMC can maintain its technical advantages for a long time.

3. All kinds of technologies are readily available

The characteristics of TSMC are that it carries out technical layout around terminal applications, and fills in the technologies required for terminal applications through advance layout, so as to ensure that it grasps and fully taps market opportunities in the first time and eats up the market dividends.

For example, in the field of smart phones, according to the iPhoneX teardown report, in addition to the main chip A11 processor is manufactured by TSMC, iPhoneX's wireless charging IC, short-range wireless communication chip, LTE transceiver chip, module, power management IC chip, wireless transceiver module and other functional modules are designed by Qualcomm, Broadcom, Texas Instruments, and Intel, but the aforementioned IC design companies have placed orders with TSMC. Therefore, the iPhone X sold well after it was launched, and TSMC was a big winner.

In the field of the Internet of Things, when the "Internet of Things" is frequently mentioned and shows its development momentum, TSMC set out to build a "super fleet" of the Internet of Things, copying the experience of smartphone technology layout to the field of the Internet of Things. Specifically, further planning was proposed for the future blueprint of the Internet of Things, which was personally led and supervised by Wei Zhejia, the co-CEO at the time, and selected elite teams from various departments such as process technology, R&D, strategic development, and business. Before the Internet of Things becomes a new market hotspot, the "Super Fleet" will step up the development of Internet of Things technology, especially the special process technology that is widely required by the Internet of Things, from 0.35/0.25/0.18 micron to 90/65/50/40 nm RF process, embedded flash memory process, etc., to complete the possible technologies of the Internet of Things as much as possible to ensure that any process technology and product specifications required by customers are readily available, so as to accurately grasp market opportunities.

4. The technical competition is surprisingly winning

The first is to "leapfrog" research and development on the basis of solid technology. In 2010, Zhang Zhongmou, then CEO, highlighted the strange trick of skipping 32nm and launched the 28nm technology generation with a 20% reduction in chip area, and the efficiency and cost were far superior to those of the opponent's 32nm products. Samsung, IBM, and GLOBALFOUNDRIES were caught off guard, and 28nm became the most profitable and longest-reigning technology node in TSMC's history.

The second is to speed up research and development with the "Nighthawk Plan" to catch up with competitors. In 2014, TSMC proposed the "Nighthawk Plan", that is, from more than 40,000 employees, a team of more than 300 R&D engineers was organized to work small night shifts or large night shifts to achieve 24-hour uninterrupted acceleration of R&D, relying on investment of 1.5 times and 2 times the manpower and time to catch up with technology. At the end of 2016, the Nighthawk program was initially successful, conquering the 10nm node. After 10 nanometers, there were 7 nanometers, and after 7 nanometers, there were 5 nanometers, and the Nighthawk unit was not disbanded on the spot, but was broken into pieces and became a permanent system.

The third is to accelerate the process of technology iteration through parallel research and development. In the face of Samsung's step-by-step approach to advanced process technology, TSMC changed the previous process of developing the next process after the R&D department completed one process R&D and handed over to the manufacturing department, and directly used two teams to develop 10nm and 7nm processes at the same time, instead of waiting for 10nm to complete R&D and then do 7nm. Thus, the period from 16 nm to 10 nm is almost two years, but the time from 10 nm to 7 nm is 5 quarters.

The fourth is to improve the iteration speed by improving the compatibility of seasonal products. The R&D team and the operation team work together to improve the speed of technology migration and reduce "conversion loss". When converting from 20nm to 16nm, the conversion capacity is used to replace the new capacity, and the commonality of the two technology nodes on the equipment is about 95%, which not only saves investment costs of $1 billion, but also saves equipment and material research and development time. When switching from 16nm to 10nm, the device reuse rate exceeds 95%, and only the remaining 5% is new devices that are converted to 10nm, developed by partners such as equipment and materials.

Capacity: Pay attention to layout, deployment and overall management

1. The production capacity is based on Hsinchu as the "base camp", and the backup production capacity is set

TSMC's manufacturing focus is concentrated in Taiwan Province, mainly located in Taichung, Hsinchu and Tainan. Among them, Hsinchu Science and Industrial Park is the "base camp", which is not only the earliest industrial park, but also the most comprehensive technology node, including the mature node from 450nm to the advanced node at 2nm; The technology of Tainan Science Industrial Park is positioned as a mature node of 16nm and above and an advanced node of 5nm and 3nm; The technology of Taichung Science Park is positioned as a key node (mature node at 28nm and advanced node at 7nm). From the perspective of the layout of technology nodes, the corresponding production capacity of technology nodes is set in double parts, Hsinchu, Tainan + Taichung cover the complete technology nodes respectively, and to a certain extent, Hsinchu, Tainan + Taichung are each other's backup production capacity.

2. The production capacity is distributed in "settlements", and the allocation within the region is flexible

Whether it is a wafer fab in Taiwan Province or an overseas fab, TSMC's fab construction strategy is to cluster Giga Fabs to form "settlements", that is, after selecting an industrial park, the site will be continuously expanded, so that expensive facilities can use the existing infrastructure and workers, thereby effectively reducing the average cost of factory construction for each fab and achieving economic scale when it is ready for full production. For example, the Nanke Fab18 factory in the province includes 4 5nm wafer fabs and 4 3nm wafer fabs; The Kumamoto plant in Japan, overseas, is also planning to expand its second plant to form an "industrial cluster".

TSMC's core factories are mainly located in Taichung, Hsinchu and Tainan, and the capacity allocation between the three places is efficient and flexible, and more than 300 engineers can be efficiently dispatched in one day.

3. Automatic overall management of production capacity in the cloud

TSMC's assets are increasing year by year, and it is necessary to improve management efficiency in order to maintain TSMC's core competitiveness, and even if the factory is expanded, the number of employees will still maintain more than 40,000 in the future, so as to maintain high salaries for employees. Cloud-based automation is the only way to improve overall efficiency. Specifically, the existing and new plants are all connected through the cloud system to achieve cross-plant cooperation and improve comprehensive capacity utilization. For example, factories with insufficient production capacity can share production materials with the cloud and use machines from other factories to achieve production, or use machines from other factories to verify the production procedures of the local plant through the cloud system.

4. Conform to the trend of anti-globalization and lay out international production capacity

In the development trend of geopolitical and anti-globalization industries, TSMC's customers pay more and more attention to supply chain security, and TSMC responds to customer needs and selects several key countries or regions around the world to deploy production capacity to enhance customer trust. In order to maintain its technological leadership, TSMC has formulated an N-1 development strategy, specifically, TSMC introduces overseas process technology that is one generation lower than the mass production process of local companies in Taiwan Province, and retains the most advanced technology in Taiwan Province.

(1) Build a factory in the United States

TSMC's move to Arizona to build a factory was a "trade-off" due to multiple factors such as geopolitics, customer and market pressures. For TSMC, the most important thing is the market and customers, with seven of the top 10 customers in 2022 headquartered in the United States, accounting for about 60% of total revenue that year, with Apple alone contributing 23%.

The advantage of building a factory in the United States is that the customer type is comprehensive, covering most of the application fields of smartphones and PCs, and it can easily digest the production capacity of Arizona. The disadvantages are high costs, labor shortages, and scarcity of water in raw materials. Compared with the cost of expansion in Taiwan Province, it is 4~5 times higher, and the difference is mainly due to labor costs, license costs, inflation costs, personnel and learning curve costs. It is expected to absorb the higher costs of overseas fabs through government subsidies and cost reductions.

(2) Build a factory in Japan

In November 2021, TSMC announced that it would build a joint venture with Sony Semiconductor Solutions (SSS) and Denso Co., Ltd. (DENSO) to build a "JASM wafer factory" in Kumamoto Prefecture, Japan, which is expected to be put into production by the end of 2024, with a monthly production capacity of 55,000 wafers.

The advantages of building factories in Japan mainly include government subsidies, market demand and material supply. First, about half of the construction funds come from the government. Second, before the mass production of the Kumamoto plant, the orders were full, including the large orders for automotive semiconductors from Japanese automobile Honda and the CIS orders from Sony. Based on the order situation, TSMC plans to build a second wafer fab in Kumamoto Prefecture, which will remain the same size as the first fab. Third, Japan's semiconductor materials are large and complete, accounting for half of TSMC's material suppliers, and after TSMC decided to open a factory in Kumamoto, Japanese manufacturers flocked to ensure that they had no worries about material supply.

(3) Build a factory in Germany

TSMC chose Dresden in Saxony, Germany, and plans to start construction of a wafer factory in the second half of 2024, with a target of putting into production by the end of 2027, with a monthly production capacity of 40,000 pieces of 12-inch wafers, which can provide 28/22nm and 16/12nm processes, with a total investment of more than 10 billion euros, TSMC will hold 70% of the shares of the joint venture and be responsible for the operation, with Bosch, Infineon, and NXP each accounting for 10%, of which NXP, Infineon, Automotive semiconductor companies such as STMicroelectronics are themselves important customers for TSMC.

The advantages of building a factory in Germany include government subsidies, the automotive chip market, and infrastructure. First, the government subsidy budget is at least 7 billion euros. Second, there is a huge gap in German automotive chips, broad market prospects, the automotive industry is a pillar industry in Germany, the shortage of core crisis affects German automobile production, and when Germany wants to revitalize the automobile industry, its own wafer factory is somewhat stretched, so TSMC's German factory has become a public expectation. As an old industrial city, Dresden is one of the largest semiconductor industry centers in Europe, known as the "European Silicon Valley", and has a complete semiconductor production chain and supply ecosystem, so TSMC is located here.

5. Adopt a preferential price strategy to implement off-peak production

The off-peak season for chip production is obvious, generally the first half of the year is the off-season, and the second half of the year is the peak season. Chip customers must grab TSMC production capacity in peak seasons, but they can't grab it every time, disrupting the production and sales plan of the upstream and downstream industry chain and the inventory management plan of the client.

TSMC may adopt a preferential price strategy according to the market conditions of the current year to achieve "off-peak production" of orders placed in the off-season, so as to improve the comprehensive capacity utilization rate of existing equipment throughout the year. For example, in the first half of 2019, due to the poor visibility of orders, TSMC took the initiative to adopt a price concession strategy to increase orders in the off-season in order to resist the sluggish market demand and the new market uncertainty created by the US-China trade war. In terms of specific strategies, due to the full capacity of 7nm and below, the price reduction strategy in 2019 mainly targets mature and special processes above 12nm, as well as 8-inch processes and other projects, and the degree of discount depends on different customers. First, the 2019 wafer capacity demand plan should be put forward as soon as possible, and it must be produced in the traditional off-season in advance.

From the perspective of the implementation effect, it not only improves the company's capacity utilization, but also produces a strong siphon effect, and IC design factories will generally withdraw the original orders from other wafer foundries and give priority to TSMC, resulting in the squeezing out of orders from competitors such as SMIC and UMC.

Ecosystem: Cooperate upstream and downstream to establish a "virtual IDM" with TSMC as the core

At the 65nm stage, TSMC launched the Open Innovation Platform (OIP) initiative. OIP gathers all important players involved in IC design, such as EDA and IP companies, and joins forces with TSMC wafer manufacturing to form a "virtual IDM". The purpose is to ensure that IC design enterprises receive sufficient technical support, use mature and complete EDA tools, effectively reduce various obstacles that may be encountered in design, further focus on IC design, and improve the probability of success in the first chip production, and enhance TSMC's customer service and order receiving capabilities. Currently, OIP has more than 3,000 TSMC employees, as well as 10,000 employees from more than 100 OIP partners. OIP now has 43,000 technical files and 2,800 PDKs.

Enterprise structure: Follow the example of international leading enterprises and create a two-head succession system

1. The shareholding structure is dispersed

The decentralized shareholding structure allows the management team to independently make the best business decisions for the company. Founder Zhang Zhongmou holds 0.48% of TSMC's shares, and the two successors hold even smaller shares, and TSMC's top 10 shareholders are all global heavyweight investment institutions, which is completely different from Silicon Valley companies, and Asian companies are mostly concentrated in the founding family.

2. The Board of Directors is composed of world-class professionals

TSMC directors are nominated and elected by shareholders at the general meeting of shareholders, so the board of directors must show performance to ensure the rights and interests of shareholders in order to gain the support of investors. The TSMC Board of Directors is made up of world-class professionals who have a wealth of experience in making appropriate recommendations. TSMC's independent director, Yan Gibbs, is a key figure in leading Texas Instruments' comeback, and another director, Spin Lint, is a former chairman of Applied Materials, the world's largest semiconductor equipment company. The composition of the board of directors is similar to that of Intel, and the number of seats on the board of directors of Intel is 80%, and in addition to the chairman and CEO, the other directors are composed of well-known lawyers, financial experts, and well-known business figures.

3. Succession system of double heads

Referring to the implementation of "chairman and general manager in parallel" in half of the Fortune 500 companies in the United States, the founder Zhang Zhongmou created the "dual leaders and parallel leadership" model, that is, TSMC represented by Liu Deyin as chairman for shareholders and the government, and TSMC represented by Wei Zhejia as CEO for customers and suppliers, and Wei Zhejia is also responsible for finance and legal affairs. The company reports to the CEO, who in turn reports to the Board of Directors.

Stock price and operating performance are the most important indicators to evaluate the new management team, which are specifically characterized as: if the stock price and operating performance continue to rise, it proves that there is no problem with the quality of decision-making; However, if the share price continues to fall, the board of directors will take steps to protect the rights of shareholders.