2026 54 38+No. 100 1 was published in China Newsweek.
In mid-April, the Taiwan Province factory of TSMC, the world's largest OEM, suffered a power outage. Some research institutions predict that the loss of scrapped wafers will exceed $20 million only after a half-day power outage, and a large number of customers will be affected. At the beginning of June, Jingyuan Electronics of Taiwan Province sealing and measuring factory was infected by foreign employees. More than 200 people were diagnosed, and more than 2,000 people stopped working and were isolated at home. It is estimated that the production will be reduced by 30% ~ 35% in June. The epidemic in Taiwan Province province spread to the semiconductor industry, which made the global chip production worse.
Earlier, people were worried that Taiwan Province Province would suffer the worst drought in half a century. Because of the need to clean the workshop and silicon wafers, the wafer foundry consumes a lot of water. In order to ensure its water use, about one fifth of the irrigated farmland in Taiwan Province Province stopped irrigation.
This is enough to explain the status of Taiwan Province wafer foundry industry. The American Semiconductor Industry Association even put forward an extreme assumption that if the semiconductor foundry in Taiwan Province Province is shut down for one year, the global electronics industry will face a loss of $490 billion. Zhang Zhongmou, the founder of TSMC, regards the establishment of 1987 TSMC as a creation to change the semiconductor industry. The rise of TSMC and foundry has indeed profoundly changed the semiconductor industry, making it one of the most globalized industries.
However, this round of "lack of core tide" has caused security concerns in the supply chain. Europe and the United States are concerned about the excessive concentration of semiconductor industry in Japan, South Korea and Taiwan Province Province of China, and put forward a series of stimulus plans to attract the chip industry back. The localization process of China's chip industry chain is also accelerating. The "arms race" of global chip manufacturing has begun, and the existing pattern of semiconductor industry may be reshaped.
Mode dispute
From the changes in the proportion of semiconductor production capacity in various countries and regions, it can be found that the semiconductor production capacity in Taiwan Province Province was almost zero in 1990, and then increased to 22% in 2020, which is the result of the rise of wafer foundries such as TSMC.
There are two modes in the semiconductor industry. One is IDM mode, that is, chip design and manufacturing are integrated in one company, such as Intel and Infineon. The other is OEM mode, which was initiated by TSMC and rose in the 1990s. The core is that chip design companies do not need to be involved in manufacturing, testing and sealing, and accordingly a number of fabless manufacturers like Qualcomm, NVIDIA and MediaTek have been born. Because there is no need to undertake high R&D investment in design and heavy asset investment in manufacturing at the same time, the revenue growth of fabless manufacturers is often faster than that of IDM manufacturers.
Although IDM manufacturers still have nearly 70% of the world's semiconductor production capacity in 20 19, OEM mode occupies nearly 80% of the production capacity in the field of logic chips belonging to mobile phone SoC (system on chip) and is considered as the mainstream of the industry.
"One disadvantage of the IDM model is that enterprises tend to pursue products with high profit margins. If the design and manufacturing links are separated, the foundry can earn OEM fees regardless of whether it produces products with high added value or low added value, which is conducive to a more balanced industrial ecology. " Dr Xu Hongtao, a professor at the School of Microelectronics of Fudan University and co-founder of Silicon Micro, told China Newsweek that this is also the reason for the rapid development of the semiconductor industry under the OEM mode.
However, in this round of "core shortage tide", IDM enterprises show the advantage of more stable supply chain. Zhang Rujing, the founder of SMIC, once said that at this stage, it is very important for the downstream manufacturing link to support the upstream design link. However, under the OEM mode, wafer foundries are often cautious when expanding their production capacity, which is also an important reason why the global semiconductor production capacity is always in a tight balance.
In fact, since last year, the foundry has been expanding its production capacity frequently. At the end of March, TSMC announced that it would invest US$ 654.38+000 billion in the next three years to increase production capacity, and support high-end technology research and development, changing the previous style of "steadily expanding production" and expanding production capacity at a rate of five times. SMIC also announced twice in a year to expand the mature process capacity of 28nm and above.
Nevertheless, many people in the industry told China Newsweek that the foundry is still cautious in terms of the scale of capacity expansion. This is related to the characteristics of OEM business. In the total capital expenditure of the chip industry chain, manufacturing links account for 64%, but the added value only accounts for 24%. "The gross profit margin of wafer foundry is not high. Once the capacity utilization rate is not high, the wafer factory may fall into a loss. " Yao, chairman of Cushing Microelectronics, told China Newsweek.
A foundry person explained to China Newsweek, "Before the expansion, the foundry needs the capacity utilization rate of the existing factory to reach a considerable level. If the capacity utilization rate of the existing production line is only 70% to 80%, expanding production often means losses, and the depreciation pressure of equipment and other large capital investment is not small. "
Compared with the caution of the foundry, Xu Hongtao noticed that some fabless manufacturers began to participate in the construction, "because they are more aware of the risks and needs".
In the second half of 2020, MediaTek spent NT$ 654.38+62 million to purchase semiconductors and lease them to power generation. However, the most typical example is UMC. UMC's capital expenditure this year is only $654.38 billion+500 million euros. Nevertheless, compared with last year, it has increased by 50%. It expands its production capacity by cooperating with Samsung, a chip design company, that is, the chip design company invests to buy equipment, provides it to UMC, and then lets the latter OEM chips. At the end of April, UMC announced cooperation with several chip design companies to expand the production capacity of Tainan 12 inch wafer fab. The chip design company pays the deposit in advance at the agreed price to ensure the long-term guarantee of future production capacity.
As a result, two long-standing models of semiconductor industry seem to be blurred with the spread of "lack of core". In order to ensure the production capacity, fabless manufacturers began to move closer to IDM mode, and some IDM manufacturers, such as Intel, announced the start of OEM business this year.
In February this year, Pat Gelsinger became the CEO of Intel. In a speech in late March, in addition to the news that Intel will invest 20 billion US dollars to build two new fabs, and it is expected to mass-produce 7nm or more advanced processes in 2024, it is also announced that Intel will return to the wafer foundry business.
This news directly impacted TSMC's share price that day, but in a speech nearly a month later, TSMC founder Zhang Zhongmou responded, "It's ironic that Intel wants to do wafer foundry business. TSMC was founded in 1986. During the fund-raising period of 1985, it sought Intel Capital, but it was rejected by Intel. Although the economic situation in that year was not very good, it was still a bit despicable. "
"Intel has tried to enter the OEM business many times before. I have been involved in similar projects. I was working as an OEM for Altera. We said half-jokingly that Intel finally acquired Altera because the OEM products could not be made. Intel simply acquired this company. " An engineer who has participated in several process research and development at Intel told China Newsweek that an important obstacle for Intel to do OEM work is the lack of service awareness. "It's hard to imagine that Intel is willing to grow up with small customers in a low profile. Some small customers, such as TSMC and Qualcomm, have now become giants. However, Intel will choose customers. When Apple asked Intel to do OEM work, it was rejected due to limited orders. Now this is considered by Intel executives to be an extremely stupid decision. "
At the same time, technical problems are also considered by him as one of the obstacles for Intel to engage in wafer foundry. "The tools in the Intel factory are relatively closed, and more high value-added chips such as CPU are produced. However, for example, the same 28nm process chip, different types of chips often need different processes, so it is doubtful whether Intel's production line can serve mobile phone chips and car code chips well. "
But obviously, Intel's return to the wafer foundry business is not just a company that has taken a fancy to the chip manufacturing market. The meaning behind it may be attributed to Huguel Singh's words, "American companies should put one-third of their semiconductor production in the United States." At present, this ratio is only 12%.
Vigorously stimulate semiconductor reflow
Gail Singh said that the status quo stems from the rise of the OEM model. According to the data of Boston Consulting Group, the market share of American semiconductor manufacturing industry has dropped from 37% in 1990 to 12% now. If the current trend continues, it may be reduced to 6%, but in contrast, American semiconductor companies account for 47% of global chip sales.
Employment America, a think tank that studies the labor market and economic policies, published a document reviewing the history of the semiconductor industry in the United States, saying that since the 1980s, in order to compete with semiconductor companies in Japan, South Korea and other countries, the semiconductor policy in the United States has gradually turned to encouraging the reduction of operating costs and improving company profits, ignoring the construction of the semiconductor supply chain, making the semiconductor industry form a fragile supply chain around giants. "Under the business philosophy of' going to the factory and ignoring the capitalization', although the balance sheets of various semiconductor companies look more stable, the advantages of American chip manufacturing have been transferred to Chinese mainland, Taiwan Province Province, South Korea and other regions.
At present, 75% of the global chip manufacturing capacity is in East Asia. The United States hopes that the chip manufacturing industry will return, but it faces talent and cost bottlenecks.
Zhang Zhongmou mentioned that the wafer manufacturing conditions in the United States are definitely better than those in Taiwan Province Province, including water, electricity and other resources, but the talent dedication in the United States is not as good as that in Taiwan Province Province. There are a large number of excellent engineers, technicians and operators in Taiwan Province Province who are willing to invest in manufacturing.
The aforementioned Intel engineers also explained to China Newsweek that one of the reasons for the continuous loss of chip manufacturing in the United States is that the American cultural system is not easy to generate service awareness. The chip belongs to the precision processing manufacturing industry, which is more in line with the East Asian cultural circle and requires workers to have strong discipline and obedience. Therefore, the most important foundries in the world today are concentrated in East Asia. Even an American company like Intel, its factory management system is different from other departments, and it implements militarized management.
The cost disadvantage of the United States is even more obvious. 10 the total cost of building a new chip factory in the United States is about 25% ~ 50% higher than that in Asia. If we want to meet the semiconductor self-sufficiency, the United States needs to invest 350 billion to 420 billion dollars in the early stage, which is also much higher than Chinese mainland's175 billion to 250 billion dollars.
(The staff observed the pre-drying of photoresist under the working environment of yellow light source. Photoresist, also known as photoresist, is the core material of semiconductor chip manufacturing industry. Figure/Xinhua)
The American Semiconductor Industry Association sent a letter to US President Biden in February this year, mentioning that competing countries have invested heavily in attracting semiconductor manufacturing and research. The absence of the United States leads to the loss of its competitiveness and the reduction of its share in global semiconductor manufacturing. The United States needs to encourage the construction and upgrading of semiconductor manufacturing facilities and invest in research.
On June 8, local time, the US Senate passed the American Innovation and Competition Act, approving the allocation of 52 billion US dollars to vigorously promote the production and research of semiconductor chips in the United States in the next five years. Schumer, the leader of the Democratic Party in the Senate, once called it a "historic $52 billion investment to ensure that the United States maintains the leading position in chip production" and bluntly said that "this bill will ensure that the United States will no longer rely on foreign chip processors." According to Reuters, it includes $39 billion in production and R&D incentives, $654.38+005 billion in implementation plans, including the National Semiconductor Technology Center, the National Advanced Packaging Manufacturing Plan and other R&D plans, and $654.38+05 billion in emergency funds.
Gina Raimondeau, US Secretary of Commerce, said at the event of Micron Technology, a chip manufacturer, that the $52 billion fund will generate more than $654.38+05 billion investment for chip production and research, including the contributions of state and federal governments and private enterprises, that is, more private capital will be released through federal funds. "By the time it is completed, there may be seven, eight, nine or ten new factories in the United States." She predicted that states will compete for federal funds for chip facilities, and the Ministry of Commerce will have a transparent fund allocation procedure.
Since last year, members of both parties in the U.S. Congress have been proposing bills to encourage the development of the U.S. chip industry, such as "American Wafer Foundry Act of 2020" (AFA) and "Creating Useful Incentives for Chip Production Act" (CHIPS). The chip bill was also included in Biden's $2.3 trillion infrastructure plan promulgated earlier this year, and approved incentives and research plans for semiconductor manufacturing, but did not provide funds. Biden also called for $50 billion to promote semiconductor production and research.
Previously, technology giants such as Apple, Amazon, Google, and Microsoft joined hands with chip industry chain enterprises including Intel, Qualcomm, and TSMC to form a lobbying group, the Semiconductor Alliance of the United States (SIAC), with the goal of putting pressure on the US government and asking the US Congress to provide $50 billion for the chip bill.
TSMC and Samsung have planned to expand their chip factories in the United States, and a war for government subsidies has begun. As early as last May, TSMC announced that it would invest 1200 million US dollars to build a new12 inch factory in Arizona, USA. It is expected to be completed and put into production in 2024, with an initial monthly production of 20,000 5nm chips. The investment scale and production capacity of this plan have been exposed many times this year and are still expanding.
In the document submitted to the Texas government of the United States, Samsung also disclosed the specific details of its plan to build a factory in the United States: the planned cost is 654.38+07 billion US dollars, and 654.38+0800 local jobs will be created in 654.38+00. It is located in Austin and covers an area of 7 million square feet. Samsung also warned that the project is "fiercely competitive", with Phoenix, Arizona, Genesee County in upstate New York and South Korea as potential competitors. If settled in Austin, it will break ground in the second quarter of this year and is expected to be put into production in the third and fourth quarters of 2023. It is said that it will be used to produce advanced 3nm technology. Samsung explicitly requires that within 20 years, the tax relief for Samsung chip factories in Travis County, Texas and Austin will reach about $65.438+48 billion, higher than the aforementioned $805.5 million.
Talking about the active revival of the semiconductor manufacturing industry in the United States, Zhang Zhongmou believes that the United States will always do things together with carrots and sticks, and the subsidy is only a few years, which cannot make up for the long-term competitive disadvantage. After so many years of subsidy policy, it still depends on strength.
Supply chain security is compromised.
The continuous growth of strong demand is prolonging the prosperity cycle of semiconductors. Not only the United States, but also South Korea, Japan, Europe and other countries or regions are attracting the semiconductor manufacturing industry to return. The Japanese government has promised to expand the existing fund scale of about 200 billion yen to support the domestic chip manufacturing industry. The South Korean government announced that it would provide a long-term loan of 1 trillion won to the local chip industry, expand the production capacity of an 8-inch wafer fab, and increase investment in materials and packaging. One of the goals of "Digital Guide 2030" proposed by the European Union is that by 2030, European semiconductor production will account for at least 20% of the global output value.
With the rise of division of labor, the semiconductor industry was once considered as one of the most globalized industries. According to the data of 20 19, the contribution of six countries and regions (the United States, South Korea, Japan, Chinese mainland, China, Taiwan Province Province and Europe) to the added value of the whole industry is above 8%. However, after this round of "lack of core", the semiconductor industry is shrinking to the mainland for the sake of maintaining supply chain security, and China is no exception.
Boston Consulting Group has made a prediction, assuming that the establishment of a completely self-sufficient local supply chain in each region will require an incremental upfront investment of 900 billion ~ 1.225 trillion US dollars, which will lead to an overall increase in semiconductor prices by 35% ~ 65%, and eventually lead to an increase in the cost of consumer electronic equipment.
"The trend is already obvious, which has been fully reflected in the products of Japanese manufacturers. When you disassemble Japanese electronic products, you will find that the chips used are basically Japanese, and all localities will follow this trend in the future. Simply put, where to design, where to produce. " Dong Liu (pseudonym), CEO of a chip design company in Shanghai, told China Newsweek.
Ye, vice chairman of China Semiconductor Industry Association and chairman of Integrated Circuit Branch of China Semiconductor Industry Association, believes that the lack of core in the United States and the European Union is certainly a factor to strengthen the local industrial chain, but the fundamental reason is a concern about supply chain security. The regional division of the global development of the chip industry chain leads to the hollowing out of industries in some areas. Now all localities hope to establish an industrial system that can at least maintain the minimum feasible manufacturing capacity of the local area.
However, in his view, it is very difficult for Europe and America to build a local wafer manufacturing industry, which is a problem of cost, supply chain system and industrial ecology. First of all, supply chain enterprises should follow the past and rebuild the supply chain, and the economic cost and subsequent operation and maintenance cost will be high; Secondly, the development of manufacturing industry needs the support of human resources, and whether the talents of European and American universities can support the reconstruction of manufacturing industry is also worth considering. "Continuous innovation" may be a path for Europe and the United States to develop manufacturing industry, but it is difficult to operate industrial returns in the usual sense.
As for the future of China's chip industry, Ye pointed out in an interview with China Electronic News that domestic integrated circuits are "stuck in the neck" and are passive in some fields, but they are completely different from the state of near "shock" before 10. However, he is worried that after the immediate problems are alleviated, he lacks a sense of urgency for the subsequent layout. After two years of delay, he found that the problem of "stuck neck" always exists.
His suggestion is that for China, the first thing is to ensure the security of the supply chain. The supply chain above 28nm should be absolutely safe, and the technical shortcomings of 14nm and 7nm should be filled as soon as possible. In addition, to forge a long board, we must master enough countermeasures to really get rid of "being controlled by people." We should grasp the "degree" of global division of labor and independent control of supply chain industry chain.
How to accelerate domestic substitution
Dong Liu pointed out that since last year, domestic chip manufacturers, including some manufacturers of end products, have gradually shifted their supply chains from overseas to Chinese mainland. At that time, it was mainly impacted by Huawei sanctions. With the outbreak of this round of "core shortage", this trend will become more obvious.
Shendi Semiconductor supplied six-axis IMU inertial sensor chips to a domestic first-line mobile phone manufacturer this year. Huang Du, head of public relations for Shendi Semiconductor, told China Newsweek that it took two years to test from the first sample delivery in 20 19 to the final negotiation. "The cost of replacing a chip by a mobile phone manufacturer is not small, so once you get used to purchasing chips from foreign manufacturers, there will be no incentive to take risks."
Many domestic mobile phone factory suppliers confirmed to China Newsweek that in the field of consumer electronics, domestic terminal manufacturers are transferring their industrial chains this year. "If you can use domestic substitutes, try to use domestic substitutes, even if domestic suppliers can't be the main suppliers, they will be auxiliary suppliers."
Xu Hongtao even thinks that one of the reasons for this "lack of core" is that domestic substitution leads to the increase in the scale of new product introduction of foundry enterprises. "For example, in the original capacity allocation of a foundry, the ratio of mass production to new product introduction may be 8:2, but with the increase in demand for new product introduction, the proportion of mass production has been squeezed out. Part of the reason is that domestic substitution and tight production capacity have intensified the demand for new suppliers, and a new product has to go through a long new product introduction period before mass production. "
Not only do terminal manufacturers use more chips from domestic chip manufacturers, but some domestic fabless manufacturers are also transferring their production capacity to Chinese mainland.
For companies in Dong Liu, they cooperate with foundries all over Chinese mainland and Taiwan Province Province, South Korea and the United States. "It's just that each company puts into production in a different way. On the one hand, it is to find a more suitable process for a specific type of chip, on the other hand, it is also to avoid risks. However, once the production capacity is in short supply on a global scale, even if the production capacity is redistributed, the risk is hard to avoid. " Dong Liu asked, "In this round of" lack of core ",we can already see that the color of local protection has increased. For example, if a Korean foundry faces the needs of a Korean customer and a China customer, how will he choose? "
A person in charge of a chip design company invested by Samsung told China Newsweek that it was precisely because of Samsung's endorsement that its production capacity was hardly affected, but it was expanded to compete for orders from competitors with insufficient production capacity.
In Dong Liu's words, "everyone has become less moral". To some extent, this also stems from government intervention. The newly established SIAC said in an open letter that "when the industry tries to correct the imbalance between supply and demand caused by shortage, the government should avoid intervention." The outside world believes that this implies that the US government has previously put pressure on foundries including TSMC to ensure the production capacity of automotive chips.
Someone from SMIC told China Newsweek that SMIC will consider terminal applications when allocating production capacity, that is, if one chip is missing, will it affect people's lives and even the national economy? "The capacity demand of enterprises will be carefully screened according to the real needs of enterprises and will not give enterprises more capacity."
Many people in the industry feel that in this round of "lack of core tide", the significance of SMIC can be better seen. "If the production capacity is in China, at least in extreme cases like the epidemic, we can meet and communicate, but if the foundry in South Korea and Taiwan Province Province of China is put into production, there is no possibility of meeting and coordinating." A person in charge of a chip manufacturer revealed that since last year, the company has gradually transferred its production capacity from overseas to Chinese mainland, and now it is nearly half, and even directly invited a domestic foundry company to become a shareholder, "in order to transfer production capacity more smoothly in the future".
This kind of capacity transfer doesn't just happen in one company. Yao also admitted that the company will take SMIC's 14 nm and 12 nm processes as the main platforms. "14nm below the advanced technology OEM space is not big, in addition to Samsung, TSMC, SMIC. At present, the capacity utilization rate of SMIC advanced technology is still not high, so this year's capacity expansion is concentrated on mature technologies such as 28nm. Because its advanced technology has just been developed, it will take time for domestic design companies to make targeted designs. It is estimated that SMIC's advanced process capacity will become full in the middle of next year. "
What China lacks is not only the production capacity of advanced technology, in fact, the production capacity of process nodes above 20nm accounts for 82% in the market at present, and more chip products depend on the production capacity of mature technology.
"In addition to SMIC and Hua Hong, China Resources Shang Hua has mass production capacity in China, but the monthly production capacity of 8-inch wafers may be only 20,000 to 30,000, which is too small to support larger customers. The production capacity of the new generation factory can't keep up at all, especially at present, some products need specific processes, such as BCD high pressure. In fact, only SMIC, Hua Hong and China Resources Shang Hua have the technical preparation, and there is no other choice. This is the current status quo. " The aforementioned SMIC people told China Newsweek.
Huang Du mentioned, "Some time ago, the government solicited opinions and suggestions on industrial support policies. We propose that in addition to encouraging advanced standard processes mainly based on line width, MEMS characteristic processes independent of line width should also be supported. In the era of artificial intelligence Internet of Things, MEMS inertial sensor chips will be more and more widely used. "
MEMS technology is a special chip manufacturing technology, which is widely used in the manufacture of inertial sensor chips. Inertial sensor chip has entered every smart phone, and the conversion between horizontal screen and vertical screen of mobile phone depends on this chip.
"The situation of unfinished fabs in various places worries the government, but overall, the production capacity in the Mainland is still in short supply." A person in charge of a wafer manufacturer told China Newsweek, "A few years ago, the company originally planned to invest in the construction of a wafer factory, with a planned investment of 5 billion yuan and an annual production capacity of 6.5438+0.8 million 8-inch wafers, but later the project died because the local government no longer supported it." "At that time, the project encountered the state tightening semiconductor investment, and the local government had no confidence in the project's financial support from the state, requiring the local government to bear most of the financial pressure." The person in charge said that it took three years for the factory to "put into production", and according to the calculation at that time, it took eight years to return to the capital, which was too long for the local government.
At present, the government's support for fabs is undoubtedly very important. Zhang Rujing, the founder of SMIC, summed up the conditions for the success of the project, and said that if there is government support, the central government usually gives support in policies and taxes, and local governments usually give support in land and project incentives. In order to introduce some new projects, local governments at all levels need to formulate some guidance on access.
He suggested to China Newsweek that under the consideration of avoiding the loss of state property and funds caused by wrong investment to a certain extent, the semiconductor companies needed by such countries should be actively promoted through the window guidance of the National Development and Reform Commission. However, if the control is too strict, it may also inhibit the development of this industry and slow down the development of the national integrated circuit and semiconductor industry. "If the investment amount is less than 654.38 billion yuan, it shall be filed in the existing way; For government investment below a certain amount, such as below 5 billion yuan, the relevant provincial and municipal development and reform commissions will guide it; If the amount is large, it will be controlled by the higher level NDRC. As for private enterprises or foreign capital, the guidance window can be moderately relaxed because of the small risks borne by the government. "
In Ye's view, the "unfinished" project of chip manufacturing broke out in many places before because individual projects were not prepared for market positioning, technology research and development, and team configuration. , and rushed to launch. For well-prepared projects, whether to start or not.
He pointed out that according to incomplete statistics, the domestic logic ic has a capacity gap of 400,000 chips 12 inch per month, and the memory has a monthly capacity of at least 200,000 to 300,000 chips. It is conservatively estimated that the monthly capacity gap is between 600,000 and 700,000 pieces. In the case that the overall capacity gap is so large, we should expand production on a larger scale and more efficiently. "China seems to lack the latest technology and products, but more than 80% of the products in the world do not adopt the most cutting-edge technology. The technology above 14 nm can cover most of the demand-although the market share may be only 60% to 70%, there is much to be done."