Newshot

Rapidus reportedly struggles to produce working SRAM on its 2nm node, putting 2027 mass production at risk

Summary

Rapidus is reportedly facing challenges in developing functioning SRAM cells using its 2nm-class node, which relies on EUV lithography. According to a new report, Rapidus has missed a critical milestone to demonstrate working SRAM cells this summer, despite the Japanese company's efforts in producing semiconductor test structures at its fab in Japan in mid-2025. In addition, the company is allegedly fighting contamination in its cleanroom. This puts the plans of the government-backed startup to begin mass production of chips in 2027 at risk. Rapidus misses SRAM milestone Bloomberg's report spells bad news for the Japanese firm . Rapidus began to produce test structures with 2nm-class Gate-All-Around (GAA) transistors at its IIM-1 facility in Hokkaido in July 2025, nearly 16 months ago. At the time, the company announced that its prototype transistors exhibited the expected electrical characteristics, which suggested that its equipment and fundamental process steps worked as intended. "…

Original text(via Tom's Hardware)
Rapidus Logo
Rapidus Logo

Rapidus is reportedly facing challenges in developing functioning SRAM cells using its 2nm-class node, which relies on EUV lithography. According to a new report, Rapidus has missed a critical milestone to demonstrate working SRAM cells this summer, despite the Japanese company's efforts in producing semiconductor test structures at its fab in Japan in mid-2025. In addition, the company is allegedly fighting contamination in its cleanroom. This puts the plans of the government-backed startup to begin mass production of chips in 2027 at risk.

Rapidus misses SRAM milestone

Bloomberg's report spells bad news for the Japanese firm. Rapidus began to produce test structures with 2nm-class Gate-All-Around (GAA) transistors at its IIM-1 facility in Hokkaido in July 2025, nearly 16 months ago. At the time, the company announced that its prototype transistors exhibited the expected electrical characteristics, which suggested that its equipment and fundamental process steps worked as intended.

"Then we had a testing and July 10th we confirmed a major milestone: we successfully obtained electrical characteristics of 2-nanometer node GAA transistors," said Atsuyoshi Koikem, chief executive of Rapidus, at Hot Chips 2025. "We can see the transistor actually works."

Rapidus
Rapidus

However, production of individual transistors is considerably easier than fabrication of functional integrated circuits (ICs) with multiple interconnected devices. According to sources at Bloomberg, Rapidus was supposed to demonstrate functional static random-access memory (SRAM) cells by summer 2026, but has yet to achieve this milestone. It is noteworthy that the report does not reveal whether Rapidus is struggling with individual SRAM cells, larger memory arrays, or achieving acceptable electrical characteristics and yields, so we will have to speculate.

SRAM is fundamental to modern processors because it serves as cache memory and forms other on-chip memory structures, which can contain hundreds of millions of bitcells. Furthermore, SRAM arrays are important vehicles for early process evaluation, as their repetitive structures expose manufacturing defects, transistor variability, and other problems that may not become apparent during tests of individual transistors.

A conventional 6T SRAM bitcell consists of six transistors: two cross-coupled inverters (four transistors) and two access transistors. However, these transistors also require contacts and local interconnects, as well as connections to wordlines, bitlines, and power rails (VDD and VSS). Meanwhile, a complete SRAM macro also requires peripheral circuits, such as row decoders, sense amplifiers, write drivers, and control logic, which makes a 256 KB SRAM macro quite a sophisticated IC. As a result, chipmakers usually use both SRAM cell arrays and SRAM macros as primary means to evaluate, monitor, and qualify process technologies at various stages. Furthermore, low defect density of SRAM macros is crucial for good yields of SRAM-intensive chip designs.

The importance of SRAM makes Rapidus' reported difficulties particularly concerning given its intention to begin commercial production in 2027. While the company does not confirm delays in mass production using its 2nm fabrication technology, in an interview with the Japan Times, representatives at Rapidus already said that the company was on track to mass-produce chips in the second half of fiscal 2027, which ends on March 31, 2028. Previously, Rapidus had specified its production targets without specifying calendar years or fiscal years, which by default implied the calendar year.

Time is running out

Before it can mass-produce processors for customers, the company must demonstrate functional SRAM cells, integrate them into larger arrays, achieve acceptable yields, qualify its process technology, and then initiate risk production around a year before the start of high-volume manufacturing (HVM).

Rapidus
Rapidus

Chipmakers do not generally disclose detailed schedules for SRAM development and milestones as far as new manufacturing technologies are concerned. However, TSMC tends to share progress on its upcoming nodes using the performance and yield of a 256 Mb SRAM array or '256 Mb SRAM yield on an internal product-like vehicle.' For example, TSMC's N2 demonstrated more than 80% of its target device performance and over 50% yield on a 256Mb SRAM test chip in April 2023. By April 2024, the process had advanced to more than 90% of its target device performance and over 80% SRAM yield. About half a year before mass production started, in May 2025, N2 transistor performance was close to 100%, whereas 256 Mb SRAM average yield was over 90%. By contrast, Rapidus has not announced any SRAM progress so far, and Bloomberg reports that the company has failed to produce functioning static random access memory cells, meaning that its SRAM yield is 0%.

While there is no rule of thumb that a company must have a functional SRAM array or macro N quarters before mass production of chips with that array or macro, it is evident that the reported setback leaves Rapidus with limited time to complete its crucial development milestones before kicking off mass production by the end of fiscal 2027.

Exceptional complications

Production of 2nm-class chips with GAA transistors and extreme ultraviolet lithography is exceptionally complicated, even for established chipmakers. So far, only Intel, TSMC, and Samsung have begun mass production of chips using nodes that rely on GAA devices. Unlike FinFETs, GAA transistors feature multiple vertically stacked silicon nanosheets surrounded by gate materials, which improves electrostatic control but introduces additional fabrication challenges. Furthermore, extremely small feature dimensions require exceptional precision in deposition, etching, lithography, and process control. For Rapidus, which was founded only in 2022 and relies on process technology IP developed by IBM, the task of reproducing these processes consistently in a new fab represents an incredibly tall task.

Rapidus
Rapidus

One potential problem involves selective removal of silicon-germanium (SiGe) layers that separate silicon nanosheets during fabrication. These sacrificial layers must be removed without damage to the silicon channels or significant changes to their dimensions. Insufficient etching can leave unwanted material between nanosheets, whereas excessive etching can reduce channel thickness or increase surface roughness. A research paper published by imec identifies nanosheet release as one of the critical challenges of GAA manufacturing, particularly because variations in channel geometry can affect transistor performance and consistency.

In fact, EUV lithography itself introduces additional complications. At extremely small dimensions, stochastic effects associated with photon absorption, resist chemistry, and subsequent pattern transfer can produce variations in feature dimensions, edge roughness, and occasional missing or malformed structures. Such defects can affect pretty much everything, including transistor gates, contacts, and interconnects. In fact, there is a company called Fractilia that specializes solely in stochastic controls. Although individual test transistors may function correctly, circuits with millions of interconnected devices are much more vulnerable to failures caused by stochastic defects.

Another potential challenge concerns inner spacer formation. These extremely small dielectric structures electrically isolate the gate from the source and drain regions. Their fabrication requires precise lateral etching of SiGe followed by dielectric deposition. Variations in spacer geometry can affect effective gate length, parasitic capacitance, leakage current, and transistor performance. A paper published in Nanomaterials in January 2026 demonstrates that even subtle differences in etching conditions can affect spacer geometry and nanosheet integrity, which potentially introduces electrical variability that becomes particularly problematic in SRAM circuits.

Gate stack integration presents another possible source of difficulties. GAA fabrication requires conformal deposition of gate dielectrics and metal materials around multiple nanosheets, including narrow spaces between them. Variations in material thickness, interfaces, and effective metal work function can alter transistor threshold voltages, which is not a big problem for individual transistors or tiny test patterns. However, for SRAM structures, excessive differences between nominally identical transistors can reduce read, write, and retention margins, which could prevent cells from operating reliably at their intended voltages and greatly narrow the manufacturing process window.

Finally, contact resistance, interconnect defects, and particle contamination can further complicate circuit integration. Bloomberg specifically reports that Rapidus is working to improve dust and particle control in its cleanroom. Normally, there might be no connection between contamination and the SRAM woes the company is experiencing. However, one of the distinct features of IIM-1 is that it uses an unconventional grid-based wafer transportation system to enable its single-batch processing as well as increase the efficiency of expedited processing. Usage of this system raises questions about whether the new infrastructure has undergone sufficient contamination and reliability qualification, though, again, a transportation system alone could not be responsible for zero (or near-zero) SRAM yield.

On-track for 2027?

Rapidus reportedly stresses that it remains committed to mass production in the second half of fiscal 2027. However, until the company demonstrates functional SRAM and subsequently larger integrated circuits, questions about its ability to manufacture commercially viable 2nm chips will persist.

Anton Shilov
Anton Shilov

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by Tom's Hardware
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗