Semiconductor industry has been experiencing the swell of so-called “emerging technologies” such as 5G (5th generation mobile communication), AI (artificial intelligence), and quantum computing for the past few years. On the other hand, in 2019, the industry was shaken by the Japan-Korea trade tensions. And now, the US-China tensions that involve Taiwan.

However, looking at Japan’s industrial structure, semiconductor device manufacturing has fallen considerably compared to the 1980s and 1990s. Although both Kioxia and Sony are doing their best, the global semiconductor industry is an oligopoly . Still, Japan’s semiconductor manufacturing equipment has a global market share of over 35%, and semiconductor materials have a market share of about 60%. The major issues are how to strengthen this area and how to strengthen Japan’s semiconductor and electronics supply chains, including the device industry.

The stage is about to change dramatically in the semiconductor industry. It should be recognized that the trends from the past 5 to 10 years have entered a turning point where the future should not be simply extrapolated.

There are three major things that have changed. The first is COVID-19 pandemic which has greatly accelerated digital innovation. DX (Digital Transformation) using IoT (Internet of Things), 5G, and AI development accelerated all over the world, and the fabs are full of orders.

The second thing is that energy problems are becoming more and more serious. DX cannot exist without overcoming energy constraints. No matter how convenient DX is, it is not a sustainable solution if it consumes a huge amount of energy. We are on the verge of being on the brink, and how to overcome the unique “energy crisis” of the data society is important.

The reason why this crisis is happening is because we did “Big Data x AI”. Data will increase significantly and AI processing will become more and more complicated, and we are multiplying them, which naturally requires a huge amount of power. On the other hand, dark silicon has been increasing since 2000. Although miniaturization progresses and the degree of integration increases, most of the transistors must not be turned on, and if all of them are used, tremendous heat will be released or energy will be consumed. Dark silicon is increasing due to energy constraints, and it is not possible to bring out the performance. This, combined with the acceleration of digital innovation in the post-corona times, is accelerating the crisis.

The third is economic security response, in a nutshell, the US-China high-tech war. In the 1980s, Japan also experienced “US-Japan semiconductor friction,” but at that time it was a bilateral trade problem. However, today it is not just a trade issue between the United States and China, but a serious issue that involves the world and puts pressure on the world’s semiconductor supply chain to divide it into two colors, red and blue. The whole country is discussing how Japan will respond to this geopolitical risk.

Starting with the GPU, there is the “calculation ability” of AI, the “store ability” relating to huge amounts of data using NAND flash memory, and the “communication ability” of 5G.  Pandemic unexpectedly accelerated the needs of DX when these three beats were met, but at the same time, are the energy needs satisfied? That is the argument that came out.

If the latest CPU is running only at a few percent, I think that it has more capacity, but in reality, if you move it further, problems such as great heat and battery consumption will occur, so there are limits and that is very interesting.

Then, what are the challenges for Japan?  Japan’s national goal is “Society5.0”, to lead new growth with DX, which is an image of a data-driven society, but semiconductors are indispensable for that scenario. Data-driven services in a data-driven society are computing, and if you don’t win there, you can’t win in DX.

However, in the last 10 to 20 years, Japan has been forced to retreat in the semiconductor business to component materials, and the management has been traumatized. Many of the managers’ said, “I stopped investing in hardware and turned to software. From now on, it’s DX. I’ll buy hardware from the outside.” It is worrying.

Alan Kay’s words, “People who really think about software, will make their own hardware” are a simple expression of that problem. This word is very esoteric. A good example would be GAFA (Google, Apple, Facebook, Amazon).

Most people think that GAFA is a software company, but they are currently the hardware investors. They bring together designers from around the world for the computing they want and invest tens of billions to make dedicated semiconductors at mega foundries such as TSMC and Samsung Electronics.

The von Neumann architecture and semiconductor integrated circuits were invented as a solution to the connection problem of large-scale systems, and after that, the semiconductor industry became a world where the one that controlled the processor and memory wins. This was a world of competition on how to prepare cheaper and better general purpose parts, and you have to win with capital while making risky investments. Here Japan has lost to South Korea and is now about to lose to China.

In addition, the division into this processor and memory has become an issue called “von Neumann bottleneck”. As a large amount of data moves back and forth between the processor and memory, it consumes a lot of energy, and at most one-tenth of its performance can be extracted.

AI was introduced at that time. The “neural network” used for AI processing is a “non-Von Neumann architecture” that is not divided into a processor and memory anywhere, and it is possible to improve energy efficiency, so dedicated chips are being developed all over the world. At the same time, there is a growing demand for 3D integration technology to reduce the distance between memory and processors and improve energy efficiency as much as possible.

In this way, the game in which the person who sells a lot of processors and memory wins as before is about to collapse.

Also, especially from an AI perspective, if it took two years to design, it would be an ancient project. Therefore, it is also important to improve development efficiency. From the cost performance in the simple capital competition as before, to the competition of know-how, the time performance of who made the chip first has become important. From now on, how to improve development efficiency and energy efficiency will compound into an index, and it will be very important to do it on a dedicated chip.

This can be said to be “a new stage for Japan to fight.” When it comes to winning the processor and memory markets with the power of capital, Japan can no longer compete with NVIDIA, Intel, and AMD, and it will be difficult to build a factory that can compete with Samsung and Micron Technology. However, there is no limit to the amount of AI know-how that can be gained. And when it comes to chip manufacturing, you just have to work well with Mega Foundries. 

Major changes in the semiconductor industry are expected to occur once every half a century, and we are about to enter a new stage that is completely different from the past 50 years. The role played by the universities is to “make a lot of academic assets into management resources suitable for a new stage,” and the University of Tokyo is also actively working.

One of them is d.lab and RaaS, which are industry-academia-government collaboration mechanisms. D.lab is positioned as an open collaboration, and RaaS is positioned as a closed collaboration.

D.lab started its operations only in October 2020. Companies in a wide range of fields, from chemicals to electronic components are participating, and each company openly discusses academic and social collaboration with a comprehensive perspective, finds issues and partners, and collaborates. Currently, there are more than 20 participating companies, and the aim is to exceed 50 this spring.

The d.lab sponsorship project is to gather on campus and work together openly, but if you want to actually go to industry, you need to manage information and work closely. RaaS is such a platform. At RaaS, the aim is to create two pillars. One is the demand for semiconductors, which is an industry-academia collaboration with Panasonic, Hitachi and Toppan Printing. The other is supply, and we would like to collaborate with industry, government, and academia in areas such as manufacturing, materials, and equipment.

Specific areas of focus are AI and 5G dedicated chips. We believe that AI and 5G are in great demand when moving from general-purpose businesses such as conventional processors and memory to dedicated businesses. In addition to improving energy efficiency through 3D integration, we will research and develop “agile design methods” for quick design with as little human intervention as possible to improve time performance.

Japan has some wonderful assets. One is human resources and the other is the expansion of the industrial ecosystem. The industrial ecosystem is a great asset, but it will die if left unchecked. When we decide to make our own chips, we will be in trouble without these two. From now on, the stage will surely turn and Japan will perform well, so it is key to maintain the ecosystem of human resources and industry without dying out. d.lab and RaaS are also working for that.

And with regard to geopolitical risk, university channels will save you a little more long-term network disruption. The important thing is that technology resides in people. Academia channels can preserve a network of people, researchers and engineers. This is high tech, not the world of ideology. The important thing is how to bring together the talents of the world and create new values, and to make use of an “academic” approach.

 In terms of human resources and ecosystem, we believe that it is important for people to gather, connect, and disseminate information correctly in order to bring about further innovation.

Innovation is something that is born where A and B are multiplied, and I am very interested in the dialogue between the leaders of the country, industry, and academia. 

Japanese materials and equipment are very strong in the semiconductor industry, so when industry enters the new stage Japan should utilize its strong voice to say, “The new stage should be like this.” It is important to provide the environment that will allow the equipment industry to win.

Especially since there is the old way of thinking of the parts business. Up until now, parts businesses had to work hard to provide the products that customers are looking for in terms of quality and cost, but this role might be reversed in the future. The “parts” from now on are data. Collect data with IoT, process it with AI, and make it a service. Semiconductors become even more valuable as they are used to finally deliver the service to consumers or to put it on networks across the country. 

RR