Skip to content
Theme
OnDevice Tools

GadgetsLaptops & PCs

Rapidus 2nm node reportedly misses key SRAM milestone

Sources cited by Bloomberg say the Japanese chipmaker has not yet shown working SRAM cells on its 2nm process, a step normally completed well before mass production begins.

Illustration of a semiconductor wafer in a cleanroom with a 2nm chip design shown alongside

Key takeaways

  • Bloomberg reporting, cited by Tom's Hardware, says Rapidus did not demonstrate working SRAM cells by summer 2026 as planned.
  • The report also mentions cleanroom contamination and does not specify whether single cells, larger arrays or yields are the problem.
  • Rapidus says mass production is still set for the second half of fiscal 2027, which ends on 31 March 2028.
  • TSMC's N2 process had passed 90% average SRAM yield on a 256Mb test chip about half a year before production began.
In this article
  1. A missed deadline in Hokkaido
  2. Why SRAM is the first real test
  3. The gap to TSMC's N2
  4. What has to happen before 2027
  5. Why the timeline matters

A missed deadline in Hokkaido

The Rapidus 2nm node has reportedly failed to reach one of its first real tests. According to Tom's Hardware, which cites reporting by Bloomberg, the Japanese company was expected to demonstrate working SRAM cells during summer 2026 and has not done so. The same report mentions contamination problems inside its cleanroom.

Rapidus has been running test structures built with 2nm-class gate-all-around (GAA) transistors at its IIM-1 plant in Hokkaido since July 2025. At the time, the company said the prototypes showed the electrical behaviour it expected, which suggested its equipment and basic process steps were working as intended.

A single working transistor is one thing. Wiring many of them into a circuit that stores data reliably is another matter entirely. The Bloomberg account does not say whether the difficulty lies with individual cells, with larger arrays, or with the electrical results and yields, so the exact scale of the problem remains unconfirmed.

Why SRAM is the first real test

Illustration: privacy, building, calendar, performance and the labels 2NM, Bloomberg, Gaa Transistors

SRAM is the fast memory used as cache inside processors, and a single chip can contain hundreds of millions of cells. Because the pattern repeats, arrays of these cells expose manufacturing faults and transistor variation that a lone device will not reveal. That makes them a standard tool for judging a new process.

A conventional six-transistor cell uses two cross-coupled inverters, four transistors in all, plus two access transistors. It also needs contacts, local wiring, wordlines, bitlines and power rails. A complete macro adds row decoders, sense amplifiers, write drivers and control logic, which turns even a 256 KB block into a complicated circuit.

Chipmakers therefore rely on SRAM arrays and macros to evaluate and qualify a technology at each stage. Low defect density in those arrays is also essential for good yields on chips that lean heavily on cache memory.

The gap to TSMC's N2

TSMC publishes SRAM yields as a progress marker for its upcoming nodes. In April 2023, its N2 process reached more than 80% of target device performance with over 50% yield on a 256Mb SRAM test chip. By April 2024, that had risen to more than 90% of target performance and over 80% SRAM yield.

In May 2025, roughly half a year before production began, performance was close to target and average SRAM yield was above 90%. Rapidus has published no comparable figures, and on Bloomberg's account its SRAM yield stands at zero.

Volume manufacturing of GAA chips is currently limited to three companies: Intel, TSMC and Samsung. Rapidus was founded only in 2022 and builds on process technology licensed from IBM, which makes reproducing these steps consistently in a new fab a steep task.

What has to happen before 2027

Rapidus says it remains on schedule for mass production in the second half of fiscal 2027, which ends on 31 March 2028, according to comments given to the Japan Times. Earlier statements named a year without saying whether they meant the calendar or the fiscal year.

Before high-volume manufacturing can start, the company must show working cells, scale them into larger arrays, reach acceptable yields, qualify the process and then run risk production about a year ahead of the ramp. Each of those steps depends on the one before it.

GAA transistors stack silicon nanosheets vertically and wrap the gate around them, which improves control of current but complicates every stage of fabrication. One known difficulty is removing the silicon-germanium layers between the nanosheets without damaging the silicon channels. Too little etching leaves material behind, while too much thins the channels or roughens their surfaces.

Why the timeline matters

Rapidus is a government-backed startup with plans to make processors for customers, and its schedule has been watched as a test of Japan's effort to rebuild a leading-edge chip industry. A delay to SRAM work narrows the window for the remaining development milestones.

There is no fixed rule that a working SRAM array must exist a set number of quarters before production. Even so, the reported setback leaves the company with little room before the end of fiscal 2027. Rapidus has not confirmed any delay to mass production, and the details of the SRAM problem come from unnamed sources rather than from the company itself.

Sources

  1. Rapidus reportedly struggles to produce working SRAM on its 2nm node, putting 2027 mass production at riskTom's Hardware

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.