Rapidus, the Japanese foundry the government has made its bet on leading-edge chipmaking, has not yet produced a working SRAM cell on its 2-nanometer process, according to a Bloomberg report published November 10 and relayed by BigGo Finance and Startup Fortune. People familiar with the matter told Bloomberg that the milestone was due by summer 2026 and that development is at least several months behind. That puts at risk the company's goal of mass production in the second half of Japan's fiscal 2027, which ends in March 2028.
Why it matters: SRAM is the on-chip cache memory every modern processor relies on. A process that can't yet make reliable SRAM can't yet carry a customer's real design. And Rapidus is not a normal startup. BigGo puts Japan's cumulative R&D support at ¥2.354 trillion (about $14.9 billion), including ¥631.5 billion that the Ministry of Economy, Trade and Industry (METI) added in April 2026.
What's reported, and what Rapidus says
| Claim | Who says it |
|---|---|
| No working SRAM cell yet; the summer 2026 milestone was missed | Bloomberg, citing people familiar with the matter |
| Development is at least several months behind | Bloomberg |
| Particle and dust control in the cleanroom is another unsolved problem | Bloomberg, via BigGo and Startup Fortune |
| SRAM yield validation with IBM slipped past the summer deadline | TechSoda (paywalled preview) |
| Still aiming for mass production in the second half of FY2027 | Rapidus spokesperson |
| A government-commissioned outside expert panel said in September that development is progressing | Rapidus spokesperson, via BigGo |
| The plan is roughly on schedule | Chairman Tetsuro Higashi, in October |
Rapidus declined to discuss technical details, citing confidentiality. Nothing here is an official delay: the company has not moved its date. The report says the date is getting harder to hit.
Why SRAM is the tell
Per BigGo, which cites EE Times Japan, SRAM and high-current circuits are typically where problems show up early in gate-all-around (GAA) development. GAA is the transistor design used at 2nm. A high-end processor can hold hundreds of millions of SRAM cells, and each has to work.
Rapidus said in July 2025 that its Hokkaido pilot line had made working transistors. BigGo notes that since then its public progress has mostly been a series of process design kit releases. On October 5 it also launched a "Rapidus CORE" ecosystem with 17 design-service partners, including Synopsys and Infosys. Startup Fortune notes that no customer committed to volume production has been named.
There's also disagreement inside the company about what "mass production" means. In a September interview, independent director Yoshimitsu Kobayashi set the bar at a yield of at least 80%. BigGo reports that CEO Atsuyoshi Koike and representative director Atsuo Shimizu indicated not all executives share that definition. That matters, because a date can be "met" at very different levels of output.
How far behind the rest of the field
Startup Fortune notes that TSMC moved N2 into volume production in late 2025 and that Samsung began first-generation SF2 production the same year. Even on schedule, Rapidus would arrive roughly two years after both.
Japan's other leading-edge project is TSMC's own. On October 16, METI certified a revised plan for TSMC's second Kumamoto fab (run by its subsidiary JASM), switching it to 3nm. According to BigGo:
| Rapidus (Chitose, Hokkaido) | TSMC/JASM Kumamoto fab 2 | |
|---|---|---|
| Process | 2nm (GAA) | 3nm |
| Production target | Second half of FY2027 (by March 2028) | Start by December 2028 |
| State support | ¥2.354T cumulative R&D support | Subsidy cap cut to ¥645.15B from ¥732B |
| Initial capacity | Not disclosed | 6,000 300mm wafers/month |
The money question
The government has said it will keep backing Rapidus. BigGo reports that Minister Ryosei Akazawa said last month the state is prepared to support the company until it is financially self-reliant. Startup Fortune says Rapidus still needs to raise about ¥4 trillion from banks and private investors, and BigGo says it aims to go public by 2032. Visible technical slippage makes both harder. Koike has also talked up expansion: three more plants and an R&D site near Chitose, according to remarks at a Sapporo forum this month reported by BigGo.
What to watch
- A public SRAM result. The clearest sign of recovery would be Rapidus saying it has working SRAM, ideally with yield data. As of Bloomberg's report it had done neither.
- A named anchor customer. PDKs and partner programs don't fill a fab. A committed volume customer would.
- Budget signals. Watch METI's next tranche and any private financing round. Both will show whether Tokyo and lenders still trust the March 2028 date.
- TSMC Kumamoto. If Rapidus slips, TSMC's 3nm fab could become Japan's first domestic advanced-logic production line, built by a Taiwanese company. Startup Fortune notes Tokyo's case for Rapidus was insurance against relying on Taiwan.
For chip buyers outside Japan, nothing changes near term: leading-edge supply stays with TSMC and Samsung. For anyone tracking Japan's industrial policy, though, this is a credible report that the flagship project is behind schedule, and the company hasn't offered evidence against it.
Sources
- BigGo Finance — Japan's 2nm national team hits a wall: Rapidus struggles with key memory finance.biggo.com
- Startup Fortune — Rapidus is likely to miss its 2nm mass-production deadline, Bloomberg reports startupfortune.com
- TechSoda — Global Market Watch: Rapidus pushes back 2nm chip mass production schedule (preview) techsoda.substack.com