When IBM PCs set the standard for personal computing and Madonna topped the charts, Japan led the semiconductor industry. But that 1980s dominance faded as the fabless design and foundry model evolved.

Now for the comeback: Japan is using demand for sovereign chip production to re-invent its chip industry.

Leading the charge is Rapidus. A relative newcomer to the foundry game, the company tells El Reg it's on track to begin mass production at its 2nm wafer fab in Hokkaido in the second half of 2027, just five years after the company’s founding.

The fledgling foundry has already made significant progress toward this goal. Last year, the IIM-1 fab began pilot production of 300mm wafers based on 2nm gate-all-around transistor tech.

However, Rapidus isn't just a wafer fab. Modern semiconductors are now assembled from multiple dies, often fabricated at different foundries that employ process technologies, necessitating the use of advanced packaging technologies to stitch them together.

"We have to solve these challenges of power and space and heat," Stephen DiFranco, who leads Rapidus' partner ecosystem and marketing team, told The Register. "It's not going to be solved by one thing, but one of those things is much more intelligent, multi-dimensional packaging."

Alongside its fab, Rapidus is also developing its own advanced packaging capability, which should enable it to compete with giants Samsung, TSMC, and Intel.

Over the weekend, Rapidus revealed its chiplet solutions unit would now start full-scale operations following the success of its 600mm square redistribution layer interposer last April.

Alongside the IIM-1 Plant, Rapidus has also opened a new facility to conduct physical, environmental, and chemical analysis in collaboration with its partners.

The speed at which the Japanese startup has managed to stand up its foundry biz is notable . Unlike Intel, which began its foundry journey just a year earlier, Rapidus is starting from scratch.

It's hard to develop new processes for everything, though DiFranco notes that it also means that Rapidus hasn't needed to balance legacy technologies the way Intel has.

"We had the advantage of a senior executive team with a lot of decades of experience, getting an opportunity to start with a fresh piece of paper," DiFranco said.

A three-pronged approach made it possible for Rapidus to move quickly. A wafer fab is among the most complicated facilities in the world, with intricate supply chains requiring tight collaboration and planning for everyone involved.

So, in addition to its manufacturing unit, Rapidus also operates two other divisions responsible for laying the groundwork for the foundry business and working with the company's design partners on process technology, so that the technology, customers, and supply chains are ready to roll when the fab opens next year.

A Big Blue shortcut

One of the few areas where Rapidus isn't starting from scratch is process technology. That's because instead of developing its own, it partnered with IBM to commercialize its 2nm process tech.

IBM isn't a foundry. You can't knock on Arvind Krishna's door and order up some wafers. But it nonetheless has a long tradition of semiconductor research and process design, and continues to work on those technologies.

In early 2021, IBM showed off an experimental 2nm process node developed at its research lab in Albany, New York. At the time, TSMC's 5nm process tech had just hit the market and Intel was struggling to ramp production of its 10nm process node.

IBM's tech marked the first step away from the FinFet transistors Intel popularized a decade earlier. In their place, Big Blue offered a new type of transistor called gate-all-around, which used a gate that completely encircled the sources. The result was significantly reduced power leakage and improved transistor density, resulting in a much more efficient chip.

If any of that sounds familiar, that's because Intel, TSMC, and Samsung have all adopted this approach for their latest designs.

It's IBM's 2nm process tech that now underpins Rapidus' foundry push. After starting pilot production last year, the fab released a preliminary PDK, or process design kit, to early adopters.

On Saturday, Rapidus announced it would release the full PDK later this year to give customers ample time to adapt their designs before the fab opens its doors.

N-1 with no minimum order required

With mass production slated for the end of 2027, Rapidus will arrive a little late to the 2nm party.

As we understand it, all of the major fabs, including Samsung, began volume production of 2nm silicon last year, though only Intel and Samsung have actually shipped chips built using a 2nm process.

Rapidus' apparent tardiness may not be as big a deal as it might sound. That's because only a select few want to use cutting-edge silicon.

It's not uncommon for major chip designers like Nvidia or AMD to deploy on “n-1 nodes”. That is to say, they wait for a new node (n) to come out before starting production of the old one (n-1).

The reason for this comes down to yields and batch sizes. Buyers who book capacity with major fabs must commit to a certain volume of their capacity. New nodes often have growing pains which lead to low yields. This makes smaller, more loss-tolerant chips, like those used in high-end smartphones, more economical than reticle-sized datacenter chips.

Rapidus hopes to capitalize on this by making its 2nm manufacturing capacity available in much lower volumes and compressing the fabrication time considerably.

"What we're seeing is a lot more companies that need more modest volumes per project," DiFranco said.

The idea is to enable customers to iterate designs much faster and with less risk than if they'd had to commit to larger orders.

"In a single wafer fab, you actually get a lot more data, because each wafer gives you a unique piece of data," he explained. "We believe what makes Rapidus unique is that we are going to be able to deliver more data back into both the tools and IP vendors — Synopsys, Cadence and Siemens — but also to the designer groups themselves."

A focus on domestic capacity

None of this is cheap. Foundries can easily cost tens of billions of dollars to bring on line. However, Rapidus has the Japanese government on its side.

Over the weekend, Japan's Industry Ministry approved an additional ¥631.5 billion in funding ($3.96 billion USD) to advance development of Rapidus’ fab.

"Perhaps it was the pandemic, but I think even before that, it was becoming quite obvious that semiconductors were becoming the brain of everything, that those who had more production of it were probably in a better position geopolitically," DiFranco said.

We've seen similar efforts to bolster domestic semiconductor across the US, Europe, and China following in the wake of the pandemic-era chip shortage.

Not alone

Rapidus isn't the only company building fabs in Japan. In February, TSMC announced plans to mass produce chips based on its 3nm process tech at a fab in Kumamoto.

TSMC opened its first Japanese fab in 2024, where it produces chips on a variety of older or mature nodes. The decision underscores an effort to move cutting-edge chip production offshore.

Over the past year, the foundry giant has also announced several new chip fabs in the US, with its second slated to begin producing chips based on its 3nm process tech next year. ®