Why is Japan anxious to revive semiconductors?

Chip manufacturing jumped directly from 40nm to 2nm. What does Japan think?

Moore's Law is coming to an end. Where will the development of chip technology go?

Why is time cost becoming more and more important?

What role will AI play in semiconductor development?

Why not a chip war?

When it comes to Japanese semiconductors, the same keywords always come to mind: the lost thirty years, the Plaza Accord, the king of upstream materials...

Japanese semiconductors, which have been almost silent for the past 30 years, have been extremely active in the past two years. The most eye-catching one is Rapidus, a semiconductor manufacturer jointly established by eight leading Japanese companies. It is directly targeting 2nm and intends to compete with TSMC and Samsung.

Japan is betting everything on its semiconductor renaissance.

This year, Chris Miller's "Chip Wars" has become a hot book in the semiconductor circle. Across the sea in Japan, there is a book that is placed in the most eye-catching position on the bookshelf together with "Chip Wars", and that is Tadahiro Kuroda. "Semiconductor Superevolution".

The author Tadahiro Kuroda is a professor at the Department of Electrical Engineering and Systems Science at the Graduate School of the University of Tokyo. He has worked at Toshiba, Keio University and the University of California, Berkeley. He is now a professor at the University of Tokyo. He is a key figure in Japan's semiconductor renaissance plan. Some media call it The first person in 3D stacking technology.

The Chinese version of "Semiconductor Super Evolution" "The Future of Chips: Technology to Check and Balance the World" will be released soon. In this book, issues such as how Japan can revitalize its semiconductor industry, how chip technology will develop in the future, and the direction of the chip industry will all be discussed. get answered.

01 Japan Renaissance Semiconductor:

Directly to 2nm

The Japanese semiconductor industry has been dormant for almost 30 years. It lags behind TSMC and Samsung in chip manufacturing technology by about 10 years and is currently at the 40nm process node.

Under the sense of urgency that advanced processes are lagging behind, Japan's eight giant companies formed Rapidus in August last year to cooperate with IBM to develop 2nm technology in order to build one or more wafer fabs in Japan for production.

I can understand Japan's goal of developing advanced processes, but it will mass produce 2nm. Who gives him the confidence?

Perhaps it is difficult for Japan itself to say what its confidence is. It is more about the urgency of what it has to do. This is their "last stand."

The global semiconductor market is expected to grow rapidly at an annual rate of 8%. By 2030, the market value may reach twice the current size, and the total market value will exceed 100 trillion yen. Japanese semiconductor companies' global market share peaked at 50% in 1988, but has now fallen to just 10%. It has been almost in a "dormant state" for more than 20 years. Under this situation, Japan decided to bet on the fate of the country and bet on the revival of the semiconductor industry.

What is needed for Japan's semiconductor renaissance?

It is difficult to recover the lost 30 years with inherent strategies alone, so anticipating the competitive arena and investing in advance becomes key. To this end, the author puts forward three changes in the current industry in Chapter 2 "Comeback".

First, the replacement of industry protagonists. The main battlefield of logic chips is shifting from general-purpose chips to special-purpose chips. There are three reasons behind this: With the rapid growth of data and the complexity of AI processing, the energy crisis is intensifying; the emergence of AI will increase the efficiency of electricity consumption by more than 10 times; the semiconductor industry structure is divided into labor, and companies can, according to their own business models, Independently develop special chips.

Secondly, market fluctuations. Every quarter of a century, major fluctuations will hit the semiconductor market, such as home appliances, PCs and smartphones before. Japan only caught the first wave. The upcoming fourth wave is to use sensors, AI and motors to highly integrate cyberspace and physical space, that is, to use "digital twin" technology to create a human-centered society, also known as "Society 5.0".

Third, there is a paradigm shift in technology. One is to shift from von Neumann architecture to neural networks, and the other is to shift from miniaturization to 3D integration. The miniaturization of chips is approaching the limit, and 3D integration can greatly reduce the energy consumption of data transmission.

Moreover, the above three changes are essentially energy issues. The sharp increase in energy consumption is caused by semiconductors. Therefore, the key to solving this problem is also in semiconductors, which is to improve semiconductor energy efficiency.

The author puts forward two points as the key to improving energy efficiency: miniaturization technology and 3D integration.

The representative of action in micronization is Rapidus. Compared with TSMC's "everything" production line, Rapidus only produces the world's most advanced products in a short period of time and in small batches. Providing foundry starting from 2nm process, using only the most advanced 3rd generation process technology.

Even in the Japanese industry, many people question Rapidus's strategy. Is this too aggressive?

The author expressed optimism about Rapidus's strategy. First of all, the most advanced technology is profitable. In fact, TSMC’s most profitable part comes from its most advanced technology.

In the past, even the most advanced technologies relied on low-price competition to make profits. Now, the number of first-to-market players is decreasing year by year, turning it into an oligopoly market. However, the demand for the most advanced chips will always exist and expand, and players who enter the market first will have the confidence to negotiate prices.

Secondly, the 5nm market has been depreciated by TSMC, Samsung and Intel, and it is easy for them to prevent new players from entering. 2nm uses GAA technology instead of 5nm and 3nm FinFET technology. Instead of expecting "old technology" to gain experience, it is better to start anew directly from the starting point of "new technology".

In addition, instead of competing with large OEM manufacturers, it is better to target small-volume orders that they cannot cover. As long as they can quickly enter the market, they can coexist with large manufacturers and form complementarities. And, there are customers who expect this kind of service.

3D integration is another key to improving energy efficiency in addition to miniaturization.

Judging from the current situation in Japan, in terms of micronization, Japan has fallen far behind the most advanced technology in the world and needs to learn from overseas; in terms of 3D integration, Japan has basic technologies in materials and manufacturing equipment used in 3D integration. It still has many advantages.

The development cycle of special-purpose chips is long and expensive. If a silicon compiler can be created that can programmable and automatically design chips, it may be possible to develop hardware quickly. The performance of automatic design may only be 80 points, but it can increase development efficiency by 5 times, which is acceptable.

From 2019 to 2020, the University of Tokyo has successively established a cooperation center d.lab and a technology research organization RaaS (Research for Advanced Systems), intending to complement Rapidus and jointly increase energy efficiency and development efficiency by 10 times.

In terms of energy efficiency, Rapidus pursues miniaturization, while the University of Tokyo pursues 3D integration; in terms of development efficiency, Rapidus chooses to shorten the production cycle, while the University of Tokyo studies how to shorten the design cycle.

02 The future of chip technology

How will it develop?

"Moore's Law is dead."

In recent years, countless people have questioned this "golden iron rule" of chip development. Moore's Law has indeed reached its limit. Some people continue to explore miniaturization, calling it "Deep Moore", while others are beginning to focus on new values in other aspects, calling it "Beyond Moore."

The author's choice is the latter. As Moore's Law is coming to an end, he believes that instead of choosing an extension of traditional technology routes, choosing disruptive technologies and studying how to make them practical is a huge opportunity that is about to come.

The current challenge in chip development is the limitation of performance improvement. When the power, that is, the heat generation, reaches its limit, no matter how far the circuit is integrated, it cannot further improve the performance. Processing performance per unit of electricity, known as electrical efficiency, will determine the fate of Moore's Law.

One side effect of miniaturization is an increase in the power consumption of semiconductor components. There are three strategies to reduce power consumption: reduce voltage, reduce capacitance, and reduce the number of switches. In response to the above three aspects, people have carried out reforms in materials, processes and structures.

1) From 2D to 3D

When the rapid growth of data hits the von Neumann bottleneck, communication between chips becomes an important factor in reducing energy efficiency, because data transmission consumes more energy than calculation.

The computing performance of chips increases by more than 70% every year. If you want to make full use of the improved chip performance, the signal transmission speed between chips must increase by 44% every year. But in fact, the communication speed between chips can only increase by 28% every year.

How to fill this gap?

Shorten the connection distance between memory and processor, and increase the number of connections to transmit signals at reasonable speeds. That is, making short-distance connections by stacking chips and using the entire surface to communicate at appropriate speeds is why chips have evolved from 2D to 3D.

Through silicon via TSV and magnetic coupling communication TCI technology are new technologies born under this trend.

2) From landscape to portrait

3D integration starts with memory, first stacking two DRAM HBMs, then 4, 8, and 12 DRAMs. They are all stacked flat and are called pancakes.

But theoretically there is a possibility of stacking. That is, stacked upright and called sliced bread type.

In fact, the sliced bread type has more advantages than the pancake type. First, it is easier to dissipate heat; second, it is easier to communicate.

3) Synchronous design and asynchronous design

To improve the performance of the chip, careful timing design is required. There is a need to calculate the effects of component manufacturing variations, supply voltage and temperature changes on signal propagation delays in logic circuits, as well as calculations of fluctuations in clock generation and time differences in distribution.

The cost and waste of the synchronous designs being used today is highlighted, so people are rethinking the asynchronous designs that were not available in the past.

Although asynchronous design uses more transistors and wiring than synchronous design, it should still be profitable compared to the waste of synchronous design.

The author originally thought that 7nm would realize asynchronous design, but FinFET technology allowed the performance of transistors to far exceed expectations, so it was not realized. Because structural reform of transistors continues, it may be some time before asynchronous designs are adopted.

In 1959, physicist Richard Feynman proposed that "there is still a lot of room at the bottom," and then microelectronics was born. Today, the author faces Moore's Law, which is about to reach its limit. He said, "There is still a lot of room for development at the top." In other words, it is more important to let more people participate in developing chips.

Currently, only large companies can develop specialized chips, because the industrial system has always been built for large-scale mass production, and cost performance is the most valued indicator in the semiconductor industry. But now, “time performance” is also important.

First of all, society is transforming from capital-intensive to knowledge-intensive. Wisdom creates value and promotes digital innovation. Semiconductors have transformed from low-cost components to important tools that help build society; secondly, semiconductors have transformed from industrial infrastructure to social infrastructure. This means that more semiconductors have entered robots, communications and other products, and these products will not be easily replaced by companies.

In addition, the author also mentioned a variety of solutions such as agile development, silicon compilers, and chip automation design platforms, which will bring new possibilities to chip development, especially the development of specialized chips that are time-consuming and costly.

03 “No chip war”

The voice of a Japanese semiconductor researcher

"Survival of the fittest."

It has been more than 160 years since Darwin proposed the theory of evolution, and cutting-edge science is revealing a more secret evolutionary mechanism: support and cooperation.

The author once saw the idea of "super evolution" on a TV show.

Five billion years ago, the earth was a desolate land. About 400 million years ago, plants covered the earth. In the Cretaceous, the emergence of one thing caused a dramatic increase in the number of terrestrial biological species: flowers.

Flowers use pollen to attract insects, then give pollen to the insects, and spread the pollen through the insects, that is, they establish a "symbiotic" relationship. Before this, plants were only preyed upon by insects.

Then, flowers and insects began to influence each other, each evolved because of the other, and everything began to flourish.

This symbiosis hidden under competition can better maintain the continuation of life, and this is the key to the development of the semiconductor industry.

A large semiconductor factory is like a tree. More important than the tree is the soil that cultivates life underneath and the forest formed by countless large trees.

A set of semiconductor manufacturing equipment is composed of more than 100,000 parts, compared with about 30,000 parts for a car. Moreover, the list of more than 100,000 parts for each device is different, and the suppliers are also diverse.

Therefore, around semiconductors, a huge network with multiple levels and spanning multiple industries has formed. In addition, the semiconductor manufacturing process is complex and requires suppliers, manufacturers and users to cooperate, support and adjust each other at any time. It can be said to be like a forest ecosystem.

The public often only sees big trees, and the media only focuses on the rise and fall of large companies. What really supports large companies is the rich soil, that is, the power of the industrial ecosystem network. This is where Japan's strength in semiconductor manufacturing equipment and materials lies.

This is also the reason why TSMC chose to build a factory in Kumamoto, Japan. Kumamoto is an old industrial land and has such soil.

The Japanese semiconductor industry has been committed to cultivating a "forest" that helps each other and is closely connected to each other, and now this effort is attracting the world.

Along with TSMC's new Kumamoto plant came various ancillary investments and the support of Japanese material and equipment suppliers. The emergence of this huge tree has revitalized the underground network.

Therefore, the author proposes in Chapter 6 "Super Evolution" that the key words of the new era of semiconductors are: international cooperation, international talent flow, network, symbiosis and evolution.

As early as 20 years ago in 2003, the author started working on this. The author Tadahiro Kuroda, who was a professor at Keio University in Japan at the time, Liu Huijun, a professor at KAIST University in South Korea, and Wang Zhihua, a professor at Tsinghua University, reached a consensus: the center of gravity of the integrated circuit industry in the future will shift to Asia, and young people in Asia will integrate in the world in the future. Plays an important role in the circuit industry.

To this end, the three of them co-founded a small closed-door academic seminar involving graduate students from three universities in China, Japan and South Korea - KKT Workshop. Keio University in Japan, KAIST University in South Korea and Tsinghua University hold it in turn every year. It has been going on for more than 20 years now. As the author transferred to the University of Tokyo, the current Keio University student became a student of the University of Tokyo.

"We should not incite chip wars, but build chip networks." The author put forward his expectations for the future of the semiconductor industry.

It is still unknown whether Japanese Semiconductor can be "resurrected" with Rapidus. But "it's finally starting."

View more at EASELINK