In July this year, the Nikkei reported that Japan would cooperate with the United States to open a 2nm chip R & D center by the end of the year. The facility would be established by a new Japanese chip research institution that plans to make its debut this year. The Japanese government will provide financial support to this end, and assist Japanese enterprises to research semiconductor manufacturing technologies of 2nm and below. The goal is to start mass production of 2nm chips in China as early as 2025.
Although Japan obtained a cooperative relationship with TSMC last year, it obviously has no intention to do so. Tang Shanglong, a Japanese technical operation consultant and director of the precision processing research institute, interpreted the report on the mass production of 2nm in Japan.
He believes that Japan’s newly added amendment law to subsidize TSMC’s Kumamoto factory has also failed to increase the market share of Japanese semiconductors. If the self-centered chip policy is enforced as in the United States, semiconductor manufacturers in other countries may deviate from themselves. In addition, if Japanese professional officials who do not know anything make policies, they will have no effect at all. In addition, “Japan will start mass production of 2nm from 2025”, which is simply a policy that makes people laugh.
Author profile:
Mr. Tang Shanglong is the director of the Japan Institute of precision machining. He has been engaged in semiconductor research and development in the production line of Japan’s manufacturing industry for a long time. In 2000, he received a doctor’s degree in engineering from Kyoto University. Since then, he has been engaged in teaching, research, consulting and news workers related to the semiconductor industry. He has written the failure of Japan’s “semiconductors”, “lessons from the collapse of” motors and semiconductors ” The lost manufacturing industry: the failure of Japanese manufacturing industry.
The following is the full compilation:
“2nm mass production is so difficult and unrealistic, which is a false measure to strengthen Japan’s semiconductor manufacturing industry”
01
Semiconductor Enhancement Act of the United States
On August 9, 2022, US President Biden signed the chip bill to promote domestic manufacturing of semiconductors, which officially took effect. TSMC, Samsung and SK Hynix, which originally thought that semiconductor manufacturing in the United States would be strengthened, were caught in a dilemma as to whether they should receive subsidies.
The bill includes US $52.7 billion for semiconductor manufacturing and research and development in the United States, including US $39 billion for application subsidies for domestic semiconductor factories in the United States (including US $2 billion for automobiles and traditional chips for defense systems), US $13.2 billion for research, development and talent development, and US $500 million for international information and communication technology security and semiconductor supply chain activities.
The recipients of this subsidy are shown in Figure 1. Intel will invest $20 billion to $30 billion respectively in Arizona and Ohio to build processors and wafer foundries. TSMC, which was applied for by the US government, invested US $12 billion to build a 5nm wafer foundry in Arizona. Samsung announced that it would invest $17 billion to build a cutting-edge wafer foundry in Texas. SK group also plans to invest $22 billion in Semiconductor R & D centers and clean energy.
According to the data released by Intel and other companies, every $10 billion will receive $3 billion in subsidies. Therefore, these semiconductor manufacturers hope to get subsidies through the chip act anyway.
02
Disadvantages of chip act subsidies
If US $52.7 billion is invested in the semiconductor field, will us semiconductor manufacturing be strengthened? This is a question worth pondering. The chip act clearly mentions that enterprises receiving subsidies are restricted to expanding or building new production capacity of some advanced semiconductors in specific countries that pose a threat to the national security of the United States, such as China, for a period of 10 years, and violators may need to return the subsidies in full.
According to this restrictive measure directed at China, TSMC, which produces 40-16nm chips in Nanjing, Samsung, which produces 3D NAND in Xi’an, and SK Hynix, which produces DRAM and 3D NAND in Wuxi and Dalian, cannot invest in the above Chinese factories for 10 years if they receive subsidies.
Among them, TSMC’s factory in Nanjing accounts for a relatively small proportion of the company’s business, but the 3D NAND produced by Samsung in Xi’an accounts for about 40%, the 3D NAND produced by SK Hynix in Dalian accounts for 30%, and the DRAM produced in Wuxi accounts for about 50%.
Therefore, if Samsung and SK Hynix receive subsidies, they will not be able to invest in Chinese memory factories, not only to produce the most advanced memory chips, but also to expand production. This is fatal for storage manufacturers. Of course, some people will say, “in that case, it’s OK not to receive subsidies?”, However, it is also a great blow to establish semiconductor factories and R & D centers in the United States, where costs are expensive, and not to receive such subsidies.
In addition, the United States intends to further upgrade this restrictive measure.
03
Further US restrictions on China
At present, the United States has included SMIC in the list of entities and prohibited the export of goods required for the production of semiconductors with a wavelength of 10nm and below. Therefore, the Netherlands ASML’s most advanced EUV equipment is prohibited from being exported to it. This provision applies not only to SMIC, but also to sk Hynix’s Wuxi plant.
However, according to a Bloomberg report on July 6, the US government is exerting pressure on ASML and Nikon of Japan to ban the export of EUV lithography machines to China, as well as the export of earlier generation DUV lithography machines. In addition, Bloomberg reported on August 1 that the United States has issued a ban on Lam research and KLA, prohibiting the export of process equipment below 14nm to China. Sooner or later, the ban will also affect applied materials and Tokyo electronics.
The US government will impose stricter restrictions on the export of semiconductor manufacturing equipment. As a result, TSMC, Samsung and SK Hynix will not be able to invest and build factories in China, regardless of whether they receive subsidies or not, and will be in an extremely difficult situation.
In short, the United States has very thorough restrictions on semiconductors in Chinese Mainland. However, such strict regulations may lead to deviation of semiconductor enterprises in Taiwan, China and South Korea. In my opinion, the strict regulation of the United States is too self-centered.
04
Japan Amendment Act
Subsidies for semiconductor factories
The motivation of the Ministry of economy, industry and technology’s semiconductor industry policy is shown in Figure 2. The figure was published on page 7 of the “semiconductor strategy” document of the “semiconductor digital industry strategy” published by the Ministry of economy, industry and trade for the fourth time in June 2021.
As shown in the figure, the market share of Japan’s semiconductor industry was 50.3% in 1988, and dropped to 10% in 2019. If this continues, it will almost drop to 0% in 2030. The Ministry of economy, industry and technology, which has a sense of crisis about this, in order to stop the decline in market share and turn it into an increase, has filed an amendment law on investment subsidies for new and expanded semiconductor factories. On December 20, 2021, the amendment was adopted by the majority of the ruling party at the plenary session of the Senate.
According to the amendment law, TSMC Kumamoto factory received subsidies of up to 476 billion yen on June 17, 2022. In addition, on July 26, the factories of Kaixia and Western Digital in shirishi received subsidies of up to 92.9 billion yen (Figure 3). Meguiar’s Hiroshima factory has not yet decided on investment subsidies.
Then, if the Kumamoto factory of TSMC, the Kaixia factory and the West digital shirishi factory are subsidized, will the market share of Japanese semiconductors increase?
05
Will subsidies increase Japan’s market share?
TSMC Kumamoto factory is a foundry capable of producing 55000 28 / 22 to 16 / 12NM logic chips per month. If all Fabless IC design companies that outsource production to the plant and only specialize in the Fabless IC design involved are Japanese companies, Japan’s share in the semiconductor market will increase accordingly.
However, in fact, there are almost no IC design companies without Fabs in Japan: although semiconductor website semilinks shows that there are 70 such companies in Japan. However, if you look closely at the content of the website, you will find that the information is mainly product introduction and product information, without describing the company’s business. In fact, only about five companies in Japan are really engaged in semiconductor design business, including zain, megachips, Magna design, logic research and socionext. At the same time, the United States and Taiwan, China have hundreds of Fabless IC design companies, while China has nearly 3000.
Then, which companies will outsource the production of logic chips to the Kumamoto factory of TSMC? Sony and Denso are shareholders, so these two companies may outsource. However, these two companies alone cannot fill the factory’s monthly production capacity of 55000 wafers. Of the 55000 wafers it produces each month, only 10000 to 20000 may be shipped to Japan, and the rest will be purchased from other countries. Therefore, even if Japanese taxpayers subsidize the factory, there is no guarantee that it will give priority to Japanese production outsourcing.
What about the factories of kioxia and WD in sijishi? The two companies jointly funded the production of NAND, and then divided NAND in half, adopting a style of independent management. Kaixia is a Japanese company, while Western Digital is an American company. Therefore, even if subsidies are given to the factories in sijishi, they will only contribute half of Japan’s market share.
To sum up, the subsidies for TSMC Kumamoto factory, Kaixia factory and Western Digital’s shirishi factory will hardly increase Japan’s market share, and even if they do, they can be ignored.
06
Funny “2nm” mass production declaration
According to the article of Nikkei Shimbun entitled “Japan and the United States will hold the first economic version 2 + 2 and cooperate in semiconductor mass production”, on July 29, the two governments held the first meeting of “economic version 2 + 2” in Washington, D.C., within this framework, Japan and the United States will jointly develop 2nm chips. Japan will set up a new research institution before the end of this year, tentatively called the next generation semiconductor manufacturing technology development center, and will realize large-scale mass production in Japan in 2025.
Although the Ministry of economy, trade and industry (METI) has been formulating irrelevant and absurd policies (which is also the reason why the semiconductor industry in Japan is so chaotic), it is a very ridiculous and stunning policy that “Japan and the United States will jointly develop 2nm technology, and Japan will start mass production in 2025”. This is an impossible policy, and it is ridiculous to even discuss it seriously.
The Ministry of economy, industry and Technology (METI) seems to attach great importance to this ridiculous policy. At the “flash forward Japan semiconductor memory innovation Symposium” held at the Akasaka Inter City Conference on August 10, Kazumi Nishikawa, director of IT industry of the Ministry of economy, trade and industry, discussed the content under the title of “the technological status of Japanese semiconductors”.
07
What is the difficulty of 2nm?
The size of semiconductors is decreasing by 70% per generation. In Fig. 4, Japan failed to surpass 45nm, and will further advance to 32nm, 22nm, 16 / 14nm, 10nm, 7Nm, 5nm, 3nm and 2nm.
Judging from the 45nm that Japan can barely produce, 2nm is the miniaturization level ahead of the ninth generation. How difficult is the leap in the middle?
First of all, as long as the chip miniaturization is promoted to a new generation, problems will inevitably be found, and these problems must be solved through various repeated experiments. The shape of the transistor is planar before 28 / 22nm. The changes of FinFET from 16nm and gate all around (GAA) from 2nm will not achieve the expected performance unless a new structure is adopted. Besides transistors, there are many other problems discovered due to miniaturization.
For example, Intel has been a leader in miniaturization around 2015, but it failed to advance from 14nm to 10nm in 2016, and it still failed to launch 10nm in the next five years (Figure 5). In 2021, Patrick Gelsinger, who became the eighth CEO, changed the name of 10nm to “Intel 7 (i7)”, and changed the name of 7Nm adopting advanced lithography technology EUV to “Intel 4 (i4)”. However, I4 is currently being pushed forward with difficulty.
In 2019, Samsung and TSMC launched 7Nm at the same time, followed by 6nm, 5nm, 4nm and 3nm. However, it seems that only numbers are improving, and the yield after 5nm has not improved at all. When it comes to 3nm, which adopted GAA structure earlier than TSMC, the situation will be even worse, and it is rumored that Samsung has begun to practice 2nm technology.
Even TSMC, which is at the forefront of the industry, initially only planned to advance one generation within two years. However, 3nm did not start as planned. It has taken two and a half years. It finally reached the edge of mass production in the second half of 2022. At present, it is not surprising that the 2nm process using the new GAA transistor structure takes more than three years. Therefore, its mass production date will be 2025 at the earliest and may be postponed to 2026.
This means that the assumption that “Japan will start mass production of 2nm from 2025” may be earlier than TSMC. How unrealistic it is to skip nine generations with the current capacity of 45nm and mass produce 2nm earlier than TSMC.
08
not only
“The difference between junior high school students and college students”
In the deep insight released by Nikkei Shimbun on August 3, 2022, head office commentator tunzhi Nakayama expressed the difficulty of 2nm with the metaphor of “plane type is junior high school student, FinFET is high school student, GAA is college student”.
However, the two cannot be generalized. Because, although it is possible for a gifted teenager to jump directly from junior high school to a first-class university, it is impossible for semiconductors to jump directly from 45nm to 2nm and jump for nine generations at one go.
The difficulty of promoting the miniaturization of semiconductors by 70% per generation is exponential: more than 20 years ago, I experienced the development and large-scale production of 4m, 16m, 64M and 256M DRAMs. Every generation is walking on the steel wire of technological development. For example, in the development process from 64M to 256M, there were several times when it was so difficult that I doubted the possibility of 256M. This is the difficulty of promoting miniaturization.
In order to better answer “what does 2nm mean from the perspective of 45nm?”. Suppose a primary school baseball teenager dreams of becoming a major league player in the future, like Shohei Ohtani (a player of the Los Angeles Angels in the major league baseball, and a rare player in the modern professional baseball who is skilled in both shooting and playing. For a baseball boy, having a dream and playing baseball is certainly a good thing. However, if he said, “in three years, I will play second rate in the major league!” What should I do? Naturally, this is impossible, so he would say: “first of all, I will play in Jiaziyuan, and then I will be selected into the Japanese professional baseball team. After playing well there, I will consider transferring to the major league.”
The gap between 45nm and 2nm is just like that between baseball teenager and Shohei Otani.
09
What to do with EUV?
Why is it impossible for Japan to mass produce 2nm from 2025? Let’s talk about reality rather than sensibility. To achieve mass production of 2nm, ASML’s EUV lithography machine is indispensable.
It is estimated that by the end of 2022, TSMC will have 100 EUV lithographic machines, Samsung will have 35, and Intel will have 16 (Figure 6). However, these three enterprises hope to obtain more EUV lithographic machines. In addition, in addition to the above three enterprises, DRAM manufacturers such as SK Hynix, micron and South Asia technology also plan to import EUV.
ASML began to supply euvs in 2016 (Figure 7), and the number has steadily increased: 5 in 2016, 10 in 2017, 18 in 2018, 26 in 2019, 31 in 2020, 42 in 2021, and 55 in 2022 (note that this year, due to insufficient supply of parts, only 40 can be delivered at most).
However, by the middle of 2020, the backlog of orders (accepted but undeliverable) had reached 56 units, and it is expected to reach 100 units this year.
In short, even if you order EUV lithography machines, it is difficult to get them on time. If Japan aims at mass production of 2nm by 2025, it must have owned and skillfully used EUV lithography machines before that. Even if ordinary baseball youngsters can be changed into Ohtani Xiangping, if there is no EUV lithography machine, mass production of 2nm will be impossible
Recent Comments