Rapidus, Japan’s effort to create a domestic chip-making alternative to TSMC and Samsung Foundry, is becoming a broad-based national industrial project of a kind Americans might dream about but that only China, Taiwan and South Korea could possibly match.
The plan is to build a state-of-the-art logic integrated circuit foundry and start commercial production at the 2-nanometer node in 2027. The facility is currently being built in Chitose near Sapporo on the northern island of Hokkaido. According to an unconfirmed but widely cited report in the Japanese press, Rapidus is targeting 1.4nm production by the end of the decade.
Established in 2022, Rapidus was originally backed by eight leading Japanese companies and the Japanese government. The private backers include Sony, Toyota and its group semiconductor arm Denso, NAND flash memory maker Kioxia, formerly part of Toshiba, national telecom carrier NTT, telecom equipment maker and industrial software developer NEC, investment group Softbank and Mitsubishi UFJ, Japan’s largest bank.
The project’s promoter and lead investor is NEDO, Japan’s New Energy and Industrial Technology Development Organization, an independent national research and development agency supported by the Ministry of Economy, Trade and Industry (METI).
Shortly after its founding, Rapidus and America’s IBM announced a “joint development partnership” to commercialize IBM’s 2nm semiconductor process technology in Japan.
On December 13, the Japanese press reported that 22 additional companies were considering investments in Rapidus. These include:
- Manufacturers including Honda, Canon, Seiko Epson, Fujifilm and Fujitsu, which is developing enterprise AI agents;
- Specialized technology companies including light source maker Ushio, components supplier Kyocera, JX Advanced Metals, photomask designer Dai Nippon Printing, fire prevention specialist Nohmi Bosai, design software developer Argo Graphics, and specialty chemicals maker Nagase Sangyo;
- Hokkaido Electric Power and logistics company Nippon Express;
- Seven additional commercial banks and the Development Bank of Japan.
Separately, Organo is building a water purification facility at the site that it will own and operate on behalf of Rapidus.
On December 17, Rapidus announced a new suite of design tools to support its Rapidus AI-Assisted Design Solution, or Raads, a core elements of its Rapid and Unified Manufacturing Service for advanced semiconductor manufacturing. Multiple tools will be released under the name Rapidus AI-Agentic Design Solution next year, starting with:
- Raads Generator: An electronic design automation (EDA) tool based on large language models (LLMs). When designers input semiconductor specifications, Raads Generator outputs design data optimized for the company’s 2nm manufacturing process.
- Raads Predictor: A tool for Register Transfer Level (RTL) debugging and optimization for physical design, placement and routing. It can rapidly estimate power, performance and area (PPA), the three variables used to determine optimal semiconductor designs.
- By utilizing Raads alongside existing EDA tools, Rapidus estimates developers should be able to cut design time by an estimated 50% and design costs by 30%.
Designers can use Raads Generator to convert design concepts and specifications into RTL source code, then feed the data into Raads Predictor along with Synopsys Design Constraints (SDC) to estimate the performance of silicon manufactured by Rapidus.
Synopsis is one of the world’s top suppliers of EDA technology for the semiconductor industy. It recently partnered with Nvidia on design, simulation, verification and accelerated computing using digital twins.
As explained by Synopsis, RTL design is an essential step in the design process of digital circuits. It defines and optimizes the logical functionality of a digital design before specifying the circuit’s physical layout.
Engineers convert the high-level desired behavior of their design to software code using a hardware description language. A hardware element that can store a set amount of data is called a register.
Following the launch of Raads Generator and Raads Predictor, Rapidus intends to release additional tools throughout 2026, including:
- Raads Navigator/Raads Indicator: Utilizes LLMs to provide quality assurance and assistance to designers to derive solutions to design problems.
- Raads Manager: A layout design tool that uses machine learning/AI to create a hierarchical configuration that minimizes design time.
- Raads Optimizer: Applies machine learning/AI to search for and derive parameters to optimize PPA.
Rapidus’ Innovative Integration for Manufacturing foundry is introducing single-wafer processing across all manufacturing processes. While slower than batch processing, the approach is more precise and better suited for complicated devices such as AI processors with extremely small features sizes.
The machines used in single-wafer processing can also collect more detailed data using sensors, facilitating AI-based control and feedback to improve control and raise yields.
Last June, Rapidus connected more than 200 advanced pieces of equipment to create a new automated materials handling system to support 2nm Gate-All-Around (GAA) prototyping. GAA, as explained by Applied Materials, is an advanced transistor architecture that supports scaling with lower variability, higher performance and reduced power consumption.
Like the rest of Japan’s semiconductor industry, Rapidus is integrated into the global ecosystem that emerged from Silicon Valley. What sets it apart is its ambition to return Japan to the leading edge of logic IC design and production with technology capable of fabricating the most advanced AI processors.
Japanese companies including NEC, Hitachi, Mitsubishi Electric dominated global semiconductor markets in the 1990s before losing their edge to the South Korean rivals and a new wave of innovation in the US.
The Japanese maintained an essential position in semiconductor production equipment and materials, but given the conflict with China, this is no longer enough. Japan cannot safely rely on chips made in Taiwan by TSMC, nor is total dependence on the US a viable option
TSMC began high-volume 2nm production in the fourth quarter of 2025 and is developing 1.6nm and 1.4nm process technology for introduction by the end of the decade. Samsung Foundry has also started production at 2nm, but will find it difficult to match TSMC’s production capacity.
TSMC controls about 70% of the global foundry market versus 7% for Samsung Foundry, according to market research firm TrendForce. Intel Foundry cancelled its 2nm node to focus on 1.8nm, which may be ready for commercial production in 2027, according to industry sources.
Rapidus is playing catch-up with TSMC and Samsung but appears to be operating on a timeline similar to Intel’s. Beyond IBM, Rapidus is working with California-based Tenstorrent, a designer of RISC-V CPUs and AI processors, and may attract other clients interested in competing with Nvidia and AMD while avoiding dependence on Taiwan.
In November, the Japanese government designated Rapidus an official business operator under a METI program aimed at ensuring stable semiconductor production. The company is now the official spearhead of Japan’s campaign to regain top-tier status in chip manufacturing.
Follow this writer on X: @ScottFo83517667
