Japan's eight giants set up new companies to jointly build cores. Can the goal of 2nm be achieved?

According to Japanese media reports, Nishimura Nishimura, the Minister of Economy and Industry of Japan, announced a new strategy for large-scale production of the next generation of semiconductors for supercomputers and artificial intelligence (AI) in Japan last week. Its pillar is that the Japanese government will provide a subsidy of 70 billion yen to the new company "Rapidus" set up by eight companies, including Toyota, NTT, Sony, NEC, Electric Equipment, KIOXIA, Softbank and Mitsubishi UFJ Bank, for development and production.

"Rapidus" in Latin means "fast". According to external sources, the name of the new enterprise is related to the fierce competition of major economies in the fields of artificial intelligence and quantum computing. The new name implies the expectation of rapid growth.

The company set up this time was led by the former president of Tokyo Electric (the world's largest semiconductor equipment manufacturer), such as Toshiro, and attracted the joint investment of 8 Japanese companies, including Toyota, Sony, Armor, NEC, Softbank, and Electric Apparel, each contributing 1 billion yen.

In the future, the company also hopes to further expand investment and attract more Japanese enterprises to join. This company hopes to develop a new generation of logic semiconductor manufacturing technology. The process goal is to exceed 2nm, which is prepared for nodes below 2nm. At the same time, it will also cultivate talents supporting the semiconductor industry.

According to CCTV News, the goal of this "joint team" is to mass produce 2nm or smaller chips that have not been put into practical use in the world by 2027, and realize the localization of high-end chips.

In order to achieve this goal, in terms of technology research and development, the new company will establish a research base in conjunction with the University of Tokyo and other well-known Japanese universities within this year to promote the combination of research results of scientific research institutions and industry, and transform them into production capacity as soon as possible. In addition, the research and development of new products cannot be separated from high-end technical talents. The company also plans to attract Japanese technicians working overseas to return to China in many ways.

2nm is so difficult, can Japan make it?

According to Tang Shangzhilong's analysis, on July 29, 2022, Nihon Keizai Shimbun reported that Japan and the United States are cooperating in the mass production of semiconductors. The governments of Japan and the United States held the first "Economic2plus2" conference in Washington, D.C., on July 29, 2022. It was mentioned in the conference that Japan and the United States jointly develop 2nm semiconductors. Japan will establish a "new generation semiconductor production technology research and development center (temporary name)" before the end of the year, It is planned to mass produce 2nm semiconductor in Japan in 2025.

So far, the Ministry of Economy and Industry of Japan has formulated a number of incomprehensible policies (so Japan's semiconductor industry is in chaos), "Japan and the United States have cooperated in the research and development of 2nm semiconductors, and Japan has started to mass produce 2nm semiconductors since 2025" - which is really surprising, shocking and incredible! It is a stupid strategy. The relevant personnel in the Japanese semiconductor industry can't help asking, "Can this policy really be implemented?"?

Where is 2nm difficult?

The so-called "miniaturization" of semiconductors refers to the reduction of about 70% per generation. As shown in Figure 4 below, Japan has made no progress since 45nm. After 45nm, it is 32nm, 22nm, 16/14nm, 10nm, 7nm, 5nm, 3nm and 2nm.

From 45nm (which can be produced in Japan) to 2nm, there have been nine generations of miniaturization. How difficult is it?

First of all, with each generation of miniaturization, there will certainly be many problems. To solve these problems, many trial and error experiments need to be carried out. Before 28nm/22nm, the shape of the transistor was "Planner", after 16nm, it was FinFET, and after 2nm, it was GAA (Gate-All-Around). The reason for such a big change was that the expected performance could not be achieved without changing the transistor structure. In addition, with the development of miniaturization, there are various problems.

For example, before 2015, Intel, which was regarded as the "Top Runner" in the United States, could not smoothly transition from 14nm to 10nm in 2016. In the following five years, the launch of 10nm has not been smooth. Last year (2021), Patrick Gelsinger became the eighth generation CEO of Intel, and changed the name of 10nm to "Intel 7 (i7)" and 7nm (using the most advanced EUV exposure equipment) to "Intel 4 (i4)". Now Intel is working hard to achieve i4.

TSMC and Samsung started 7nm at the same time (2019). Later, although we can only see a series of digital advances of 6nm, 5nm, 4nm, 3nm, etc., it is difficult to improve the yield after 5nm, and the 3nm of TSMC adopts GAA structure, so TSMC fell into despair in the 3nm generation and resolutely entered 2nm.

Even TSMC, which has exclusive access to the most cutting-edge technology, has replaced one generation every two years, but 3nm has not been started as planned, which took two and a half years. In the second half of 2022, it finally ushered in the critical period of mass production. The 2nm inter-generation transistor adopts the new GAA, which may take two and a half or even three years to enter mass production. Therefore, the fastest time for mass production of 2nm is 2025, or even 2026.

If Japan "starts mass production of 2nm from 2025", it is likely to be earlier than TSMC. From the current 45nm, spanning nine generations, the mass production of 2nm earlier than that of TSMC is a fantastic plan!

In the Nihon Keizai Shimbun "Deep Insight" on August 3, 2022, Mr. Nakayama Junshi of Common Data made the following analogy for the difficulty of 2nm: "Planner is a middle school student, FinFET is a high school student, and GAA is a college student".

Japan's chip industry once occupied half of the global market in the 1980s, but its share continued to decline due to the influence of the United States and other factors. Last year, it accounted for less than 10% of the global market. Although Japan has tried many times to revive the chip industry in this process, it has not succeeded.