Fujitsu is building Japan’s next supercomputer: FugakuNEXT

Fujitsu has been awarded the contract by the Japanese research institute RIKEN to develop a new supercomputer, provisionally named FugakuNEXT. The project, with its design phase scheduled until February 27, 2026, includes the entire system, the compute nodes, and the CPU components. The development is driven by the rapid growth of technologies such as generative AI, which bring an increased demand for computing resources for research and development. FugakuNEXT is intended to strengthen Japan’s leadership role in science, technology, and innovation and to advance social and industrial development by supporting the so-called “AI for Science” initiatives, which combine AI with simulation technologies, real-time data, and automated experiments.

Performance and Technological Innovation

FugakuNEXT will utilize advanced technologies from the currently developing Fujitsu MONAKA CPU. The designated successor for FugakuNEXT, Fujitsu MONAKA-X, is intended not only to continue and accelerate existing applications from the predecessor supercomputer Fugaku, but also to integrate state-of-the-art capabilities for accelerating AI processes to meet the growing demands in this area. The Fujitsu MONAKA CPU is based on state-of-the-art 2-nanometer technology and uses, among other things, a microarchitecture optimized for advanced 3D packaging and particularly low-voltage operation. The goal is to deliver both high performance and energy efficiency for a wide range of applications, from edge computing to data centers. Another important feature is the ability for seamless integration with graphics processing units (GPUs) and other accelerators. Fujitsu also intends to incorporate the expertise gained from this project into the development of future neural processing units (NPUs) and other advanced AI processors.

The Characteristics of 2-Nanometer Chip Technology

The term “2 nanometer” for chip processes is primarily a marketing term in the semiconductor industry and does not refer to an actual measurable physical property of transistors, such as gate length or metal pitch. Intel, for example, used the term “20 Angström” for this. The 2-nm technology represents a new, improved generation of chips characterized by higher transistor density (a greater degree of miniaturization), increased speed, and reduced power consumption compared to the previous 3-nm node generation. A significant technological advance in this generation is the transition from FinFET transistors to Gate-All-Around FETs (GAAFETs). Companies like IBM have already produced chips with 2-nm GAAFET-transistors that use three silicon layers. Intel’s version of this technology was called ‘RibbonFET’, while Samsung has developed its GAAFETs as MBCFET (Multi-Bridge Channel FET).