By the end of their 138th Zoom meeting, the group of Japan’s leading experts in the world’s most critical technology finally had their plan: a blueprint for the country’s biggest industrial comeback in more than half a century.
The secret project, presented to the country’s prime minister in 2020, was designed to create, out of nowhere, a world-leading semiconductor manufacturer. Japan was once the leader in this $600bn industry but surrendered that position to rivals in the US, South Korea and Taiwan. Now it wants its crown back.
“I explained to [then prime minister Shinzo] Abe-san that this was the most important project for Japan since the Meiji period,” says Atsuyoshi Koike, a veteran chip industry executive who led the group, referring to the transformational epoch in the 19th century that vaulted the country into the modern world.
That blueprint has become a company, Rapidus, which has raised billions of dollars from both the government and Japan’s leading companies and banks.
With Koike as chief executive, it is one of the world’s most capital intensive start-ups — and among the riskiest technological bets ever placed by the Japanese government.
At the heart of the Rapidus project is an attempt to prove that bespoke chips can be efficiently and profitably produced in small quantities rather than large batches, an idea that overturns the received wisdom in advanced semiconductor manufacturing.
About 90 per cent of the world’s advanced chips are produced by Taiwan Semiconductor Manufacturing Company, whose model involves operating at massive scale — with the huge capital costs that involves. If Rapidus succeeds, it would challenge both the economics and geography of the industry.
“Ever since TSMC was founded by Morris Chang in 1987, it has been all about scale and Rapidus is really something nobody has ever attempted before. If they succeed, I think it’s going to be a game-changer,” says Bernstein’s Hong Kong-based semiconductor analyst David Dai.
In the three years since that final Zoom call, Rapidus has leapt from concept to reality. A massively expensive plant is rising from the forests of Hokkaido, 900km north of Rapidus’s headquarters in Tokyo.
In December, it will take delivery of an extreme ultraviolet lithography (EUV) machine from Dutch equipment maker ASML, vital for manufacturing the two-nanometre chips the company is targeting for trial production from April.
If all goes to plan, mass production is set to begin in 2027.
But there is deep scepticism among analysts, rivals and industry executives about the chances of Rapidus actually making its untested technology a success.
Konrad Young, a professor who retired from TSMC in 2018 after more than 20 years, calls Rapidus’s attempt to go straight to making the most advanced 2nm chips a “ridiculous idea”.
Given the likely cost of high-end chips made in such small volumes, he says it is not clear who the customers might be. “I don’t think Nvidia will even bother. The cost will be much higher. Will it be Japanese companies? Nobody in Japan could use that 2nm [chip], it’s way too expensive,” he says. “So who else? It’s for the beauty of scientific research? Sometimes Japan will think that way.”
The Japanese government is also exposed if the project fails, given the state support for the project. The government has already pledged ¥920bn for Rapidus and in November unveiled a package of ¥10tn ($65bn) for the artificial intelligence and semiconductor industries over the next seven years that could include funds that would double state backing for the company. It is also exploring potential loan guarantees to attract more private investment.
Yoshihiro Seki, a senior politician in the ruling Liberal Democratic party, says that he cannot remember a time when the government had so singularly poured its focus into a single technology.
“When it comes to cutting-edge industrial technology, giving several tens of billions of yen to a single company is putting all of our might behind it,” he says. “There is no choice but to create a system to somehow make Rapidus a success.”
Japan is taking such a gamble because of the potential pay-off. Alongside huge returns and investment for Japan, Rapidus could open up an entirely new vista for cutting-edge chip production, allowing new companies and countries to enter the industry.
It could also ease one of the key geopolitical issues of the day: the concentration of manufacturing expertise in Taiwan.
By many standards, Japan is perfectly placed to make the attempt. At the end of the 1980s, “Japan dominated the world semiconductor manufacturing with more than 50 per cent market share,” says Dai. “Today there’s only 15 per cent.”
The industry is now dominated by South Korea, the US and Taiwan. Koike and others blame an arrogance among Japan’s policymakers and an inability of companies to adapt and collaborate. But Japan was also outflanked by America, which was keen to build its own industry, and South Korea, which was poaching engineers.
But Japan is still home to a deep reservoir of expertise and niche dominance of semiconductor tools and equipment businesses. Geopolitical forces are also now aligned. In an ironic reversal, the US and its allies are keen for Japan to build a counterweight to TSMC, whose dominant position in the chip supply chain has put it front and centre of geopolitical tensions between the US and China.
Japan’s powerful Ministry of Economy, Trade and Industry, or Meti, has also has softened decades of policy that sought an all-Japan solution to industrial challenges in an effort to create a full semiconductor ecosystem in the country, with Rapidus as the linchpin.
“We don’t just want to create strong Japanese companies, we want the industrial capacity to be in Japan, no matter who owns it,” says one senior Meti official. “That is perhaps something that would have been harder for us to say 20 years ago.”
It is not Koike’s first attempt to revive Japan’s chip prowess. In 2000, he launched a joint venture between Japan’s Hitachi and United Microelectronics, the Taiwanese group that pioneered the independent foundry model of manufacturing chips under contract for other companies.
Despite initial success the company failed, says Koike, due to the same dilemma that faces South Korea’s Samsung and America’s Intel today: their core businesses of designing and developing semiconductors put them in competition with customers for their foundries.
However, his ambitions were reignited in the summer of 2020 after Tetsuro Higashi — a longtime friend and former Tokyo Electron executive — got a call from IBM.
Higashi was told that the US computing giant was trying to reduce its reliance on its existing partners — including Samsung — and find a way to produce its newly designed 2nm chips in Japan. Higashi, now chair of Rapidus, took the proposal to Koike and before the end of 2020 the first Zoom calls began.
Today, Rapidus and Japan’s hope of success rests on two highly contested propositions. The first is that the surging AI market means that there will be sufficient demand from smaller customers for customisable special-use chips — bespoke designs that prioritise efficiency and can outperform more generic chips, such as those produced by Nvidia, in specific tasks.
Rapidus believes that such customers will pay a premium for speed of production and because they cannot get the required capacity from TSMC, which has its hands full with bigger orders. Rapidus thinks it can win 10 per cent of what it estimates to be the $90bn foundry market.
The second, more controversial, bet is that it can reject the core industry premise of large-scale batch manufacturing — printing hundreds of wafers at the same time — in favour of a much quicker single-wafer process.
Koike claims that producing individual silicon wafers, one after another and at high speed, generates data that can improve efficiency in real time. This increases quality and consistency and raises the “yield rate” — the percentage of chips produced deemed shippable to customers.
“In the future, the single wafer concept will be key. That is why we can change everything,” he claims. The company will also use so-called advanced packaging to improve performance by integrating multiple chips more closely together to increase speed and efficiency.
Rapidus will, says Koike, boast the “world’s shortest total cycle-time”, meaning the total amount of time it takes to process a wafer in a fabrication plant. And he thinks he can hit yield rates of up to 90 per cent within a year.
“Usually it takes up to one year to reach 30 per cent yield and to start the production. But our speed is so fast we can move easily to reach 50 per cent yield at the start of production,” he predicts. “Within one year it might be possible to hit 80 to 90 per cent. The key is how to generate feedback quickly.”
Stacked against Rapidus are a few stark realities. Seki, the Japanese politician, believes that the “biggest problem for Rapidus is people”. Even though engineers who drifted away from the country over the past decades are being poached back, there remains a serious shortage of skilled workers in a labour force already under pressure from its ageing population.
“You can get the money, laws and structures in place but whether Rapidus can secure the geniuses and specialists to do the development — here lies our biggest challenge,” says Seki.
There is no existing ecosystem of suppliers in Hokkaido and the distance from Tokyo means attracting talent might be difficult. Rapidus says it is partnering with local academia such as Hokkaido University to begin training new workers.
The most fundamental issue, according to industry executives and analysts, is that Rapidus is attempting to jump directly to 2nm, the very cutting edge of the technology, something that only TSMC is currently set to bring into mass production.
“I think it doesn’t work. At the 2nm level, only TSMC can do that,” says one senior executive at a Japanese semiconductor company. “Intel can’t do it and Samsung can’t yet do it successfully — so how can Rapidus expect to catch up in such a short period of time?”
The most advanced chip produced right now in Japan is a 40nm variant made by TSMC in plants subsidised by Japan, and the recent history of semiconductor manufacturing is littered with TSMC rivals dropping out of the race to develop and commercialise cutting-edge process technology.
In the early 2000s, the field narrowed from hundreds to a little more than 20 manufacturers. Since then, a few more have been giving up with every new process technology generation, leaving just TSMC, Samsung and Intel.
Rapidus says its partnership with IBM, which developed the key technology for 2nm chips, gives it faith it can succeed where others failed. “I think of Rapidus as being a very expensive start-up,” says Macquarie analyst Damian Thong. “We’re not quite sure if the market they want to address exists in the size they need, or if their model for addressing it actually works.”
A TSMC executive says Rapidus is not a competitor. “They are not really about the business or the margin — it feels more like an incubator.”
The other big challenge facing Rapidus is the immense funding needed for developing new production technology and for building factories.
The cost of developing each new generation of chip has ballooned from about $1bn-$1.5bn at 28nm to $6bn at 3nm, according to consultants Bain & Company. So far, Rapidus has secured only 20 per cent of the ¥5tn it says it needs to get to mass production and with every passing month of development, the challenge of recouping the investment grows.
“People pay a premium when you are first on a node . . . You get a little bit behind and you’re not catching up,” says Peter Hanbury, a partner at Bain focused on technology manufacturing and semiconductors.
That means that Rapidus’s plan to reach mass production in 2027 is itself a risk, given the possibility the technology will have ceased to be cutting edge by the time it launches. If that happens, it could prove to be a costly embarrassment to the government.
“There’s a question how far Japan should go in providing financial support,” admits Seki.
While there are plenty of questions about Rapidus and its ambitions, some in the industry believe it could work.
“At first I was really sceptical. But as I’ve seen more things, I’m kind of like, ‘Yeah, OK . . . they’re getting the latest and greatest from everybody and there are benefits to having a clean technology slate’,” says Douglas Lefever, chief executive of Japan’s Advantest, the world’s largest supplier of chip testing machines.
The senior Meti official believes that “so far the tech development is going well, and better than expected in some respects”.
Another senior executive of a Japanese company in the semiconductor industry says he thinks Rapidus will be able to produce the 2nm chips — but is not sure if it can drive down the unit costs far enough to make the maths work for customers.
Although he would not be drawn on exact targets, Koike remains confident of achieving costs that are “competitive” with TSMC, in part because Rapidus will not have to carry as much inventory. The first demonstrations of Rapidus’s success or failure will come in July, adds Koike, a few months after his trials begin.
If he can prove that his theories work in practice, Rapidus can move on to expanding the facility in Hokkaido. That could involve more expensive EUV machines, say people familiar with the matter, and a possible initial public offering of the company. Koike says he wants to try to pull in more investors before considering an IPO and will not be drawn on a date.
But although he refuses to countenance failure, others in the market are asking what happens if the company does not live up to its promises. Will Japan still get the semiconductor ecosystem it craves without Rapidus?
One certainty is that even if Rapidus fails, there will still be an extremely valuable fabrication plant up for grabs in Hokkaido — a prime target for a host of companies.
And even with all its advantages, there is no guarantee that Japan can break into the first ranks of chip production. “What the world has to ask”, says one industry expert based in Tokyo, “is if Japan cannot do this now . . . who can?”
Additional reporting by Christian Davies in Seoul
Read the full article here