Japan’s New Semiconductor Offensive

Japan’s New Model of Advanced Chip Manufacture: An Interview with Rapidus President Koike Atsuyoshi (Part 1)


Japan is banking heavily on Rapidus, an advanced logic chip startup, to propel the nation’s comeback as a semiconductor superpower. In the first part of this two-part interview, Rapidus President Koike Atsuyoshi sheds light on the company’s inception, its partnership with IBM, and his own involvement in the ambitious venture.

Koike Atsuyoshi

President, Rapidus Corp. Born in 1952 in Chiba Prefecture. Received his master’s degree in materials science and engineering from Waseda University and his PhD in electronic engineering from Tōhoku University. Held several senior management positions at Hitachi Ltd. prior to his appointment as president of Trecenti Technologies in 2002. Served as president of SanDisk Ltd. and Western Digital Japan before assuming his current post in August 2022. Author of Jinkō chinō ga ningen o koeru shingyurariti no shōgeki (The Shock of the Singularity, When Artificial Intelligence Surpasses Human Beings).

TAKENAKA HARUKATA  Rapidus Corporation was launched in August 2022, and the following November it announced a partnership with IBM that will give it access to the latter’s breakthrough 2-nanometer node semiconductor technology. It has also entered into a cooperative arrangement with Imec, a leading European hub for semiconductor research and development. The goal is to begin volume production of cutting-edge logic chips in 2027. Can you give us the background to the launch of Rapidus and its partnership with IBM?

KOIKE ATSUYOSHI  The connection with IBM goes back more than twenty years. Shortly after the terrorist attacks of September 2001, IBM began building a major research hub for the development of advanced logic chips in Albany, New York. It invited equipment and materials suppliers from around the world, including Japan, to form a consortium. One of the most active participants was Tokyo Electron under the leadership of Higashi Tetsuro, who is now chief executive officer of Rapidus.

There were quite a few companies involved at the beginning, including TSMC and Samsung, as well as Sony and Toshiba. But as time went on, the number of participants tapered off. That said, there are also some companies, like Intel, that have only recently joined the consortium.

The focus of this massive R&D effort was the development of a next-generation chip architecture known as gate-all-around, or GAA. Why was IBM pursuing that technology so aggressively? Because GAA logic chips are needed for the supercomputers that are the lifeblood of IBM’s own business. Having developed a prototype, the next step was to perfect a process for the commercial manufacture of 2-nanometer GAA chips, and for this purpose, IBM must have felt the need to expand the scope of collaboration. It looked all over the world to find a suitable manufacturing partner, and apparently came to the conclusion that it had to be Japan.

Lessons from a Failed Venture

TAKENAKA  So, when did IBM contact Mr Higashi with this business proposition?

KOIKE  That was in the summer of 2020. Higashi then approached me, and we conferred with people from the Ministry of Economy, Trade, and Industry.

At that time, I was president of Western Digital Japan(*1). It was a rough time for the company, right in the middle of the COVID 19 pandemic.

I’d been working with semiconductors ever since I’d started out as an engineer at Hitachi. In 2000, I launched Trecenti Technologies, a joint venture of Hitachi and United Microelectronics Corp. of Taiwan. In those days, UMC was more or less neck-and-neck with TSMC. Trecenti was a logic-chip foundry that was trying to build a business manufacturing 300-millimeter wafers at a time when 200-millimeter wafers were the standard. This was before Japan fell so far behind in the semiconductor industry.

The venture was basically a success. We demonstrated that we could manufacture our product with half the standard lead time using a new single-wafer processing system. A major US firm told us it wanted us to be its sole supplier, but only on the condition that Trecenti would spin off from Hitachi. The client’s core business was designing and developing semiconductors, and as such it was in competition with Hitachi. So, it was perfectly natural for it to impose that condition. Foundries have to be able to preserve neutrality in their dealings with customers.

At the outset, Hitachi had a 60 percent share, and UMC’s stake was 40 percent. I had assumed that Trecenti would spin off and become independent once the business was on a firm footing, but Hitachi said no. At that point, I said to myself, “My job here is done,” and I left the company.

That was twenty years ago, but deep inside, I think I’ve always felt some sense of responsibility, as if Trecenti’s failure to succeed as a business were one reason Japan fell behind in logic semiconductors.

After that, I focused on memory and data storage for fifteen or sixteen years. By teaming up with Kioxia, SanDisk was finally competing head-to-head with Samsung. That was quite an accomplishment. I was already thinking that my next mission was to do something to help out in the logic-chip field. That’s why I couldn’t just dismiss Higashi’s request.

(From left) Chairman Higashi Tetsurō and President Koike Atsuyoshi of Rapidus Corp. pose with Senior Vice President Dario Gil of IBM and General Manager Yamaguchi Akio and Chief Technology Officer Morimoto Norishige of IBM Japan at a press conference announcing a semiconductor partnership, Tokyo, December 13, 2022. (© Jiji)
(From left) Chairman Higashi Tetsurō and President Koike Atsuyoshi of Rapidus Corp. pose with Senior Vice President Dario Gil of IBM and General Manager Yamaguchi Akio and Chief Technology Officer Morimoto Norishige of IBM Japan at a press conference announcing a semiconductor partnership, Tokyo, December 13, 2022. (© Jiji)

Optimizing Design for Manufacturing

TAKENAKA  Rapidus isn’t aiming for mass-production of cutting-edge logic chips, as I understand it. It’s looking to produce specialty semiconductors at lower volumes, tailored to individual clients’ needs. What does that mean in concrete terms?

KOIKE  What everyone’s talking about now—artificial intelligence chips. The key point is that these AI chips will be taking the place of general-purpose graphics processing units, or GPUs. They’re going to be needed for everything from mobile phones to automobiles, creating a huge market. But customers don’t want one company to monopolize production. Rapidus won’t be devoting an entire fab to chips for one client, the way TSMC does for Apple. We plan to supply customers with medium-batch needs.

TAKENAKA  Let me ask you about chip design. Companies like Google have begun designing their own chips in-house. As I understand it, more and more tech firms have decided that they can achieve better software speeds and so forth by designing chips optimized for their own purposes. Apple, which partnered with Intel for so long, now uses its own processors. So, is Rapidus targeting smaller firms that also want to design their own chips? And in that case, to what extent will Rapidus be involved in the design process?

KOIKE  Rapidus won’t be at the core of the design process, but we hope to provide design support oriented to reducing costs and shortening development time. This is a very important aspect of design. In my view, the semiconductor fabrication process today suffers from the fact that the engineers in charge of chip design don’t know that much about semiconductor manufacturing. The design and manufacturing departments are basically cut off from one another. The designers should be getting regular feedback from engineers with expertise in the manufacturing side of things.

In the semiconductor industry, people talk a lot about DTCO, or design-technology co-optimization, but I prefer to stress DMCO, design-manufacturing co-optimization. And this happens to be one of Japan’s strengths. At Rapidus, we intend to streamline production by teaming up with designers to make the most of Japan’s strengths in manufacturing technology.

TAKENAKA  Some experts have come out and said that it’s simply not possible for Japan, which currently has only 40-nanometer chipmaking technology, to suddenly start manufacturing 2-nanometer node chips. At Rapidus, you’re saying that it’s not a problem. Can you explain the basis for your optimism?

KOIKE  Generally speaking, today’s CMOS logic chips use FinFET [fin field-effect transistor] architecture. Japan got left by the roadside during the shift to FinFET technology. But 2-nanometer node semiconductors will use GAA technology, which is completely different. If FinFET were a necessary steppingstone on the way to GAA, then the project would be prohibitively expensive and time consuming. But since they’re two completely different technologies, we can just leapfrog FinFET. GAA is the technology developed by IBM through the same R&D consortium I was talking about, and IBM was looking for a new partner to manufacture its 2-nanometer GAA chips in Japan. That’s how Rapidus came into being.

Our competitors in the manufacture of cutting-edge 2-nanometer chips will most likely be Intel and Samsung. Of the three, only Rapidus is a foundry business dedicated to manufacturing semiconductors for other companies. What that means is customers can count on Rapidus to conduct its business from a position of neutrality.

(To be continued in part 2.)

(Originally written in Japanese by Ishii Masato of the Nippon.com Editorial Department based on a July 25, 2023, interview. Banner photo: Rapidus President Koike Atsuyoshi in Kōjimachi, Tokyo, on July 25, 2023. © Hanai Tomoko.)

(*1) ^ Western Digital Japan (previously known as Sandisk Ltd.) is a subsidiary of the US-based computer drive manufacturer and data storage company Western Digital Corp. It produces data storage devices in a partnership with Japanese memory-chip maker Kioxia Holdings.—Ed.

economic security semiconductors Rapidus