PFN, IIJ and JAIST to Launch Joint Research Project on Ultra-High-Efficiency AI Computing Infrastructure

Japan

Preferred Networks, Inc. (PFN), Internet Initiative Japan Inc. (IIJ) and Japan Advanced Institute of Science and Technology (JAIST) will launch a joint research project on ultra-high-efficiency AI computing infrastructure. The project has been selected by Japan’s New Energy and Industrial Technology Development Organization (NEDO) to participate in its “Research and Development Project of the Enhanced Infrastructures for Post-5G Information and Communication Systems/Development of post-5G information and communication systems (Commission).”

The three participants will launch the joint research project in response to the exponential increase in demand for computing power that supports AI technologies. The joint project will focus on energy efficiency of AI accelerators and density of arithmetic devices in data centers, which are essential for sustainable development of AI technologies.

PFN, as the project’s lead, will develop hardware and software including compilers that can efficiently process practical workloads of AI systems including foundation models. PFN will lead the joint project based on its expertise including its MN-Core™ series of AI accelerators, MN-3 supercomputer which has topped the Green500 list of the world’s most energy-efficient supercomputers three times, and its deep insight in foundation model and AI workloads. PFN’s other roles will include:

• Research and development of ultra-high-efficiency AI accelerator system and its control technology
• Research and development of water-cooled high-density servers

IIJ will use its expertise in building and operating large-scale data centers and cloud services to conduct research on high-density data center infrastructure that will support large-scale commercial services. IIJ’s other roles will include:

• Development of data center reference model with economy, continuity and readiness for practical use
• Establishment of hybrid cooling system that combines cooling and air-conditioning technologies for high-heat IT devices
• Development of assessment method and metrics for energy efficiency for AI computing infrastructure

In addition, PFN, IIJ and JAIST will jointly conduct research on operation efficiency of commercial AI computing infrastructure in a multi-tenancy environment as well as streamlining of practical AI workloads. Other roles of the three participants will include:

• Development of technology for efficient AI workload execution
• Development of technologies that enhance efficiency of data center facilities
• Assessment metrics and method for effective efficiency on AI computing infrastructure testbed

Through the joint project, PFN, IIJ and JAIST plan to collaborate to build an environment where companies and organizations can build internationally competitive next-generation AI foundation models.