Preferred Networks and Tenstorrent to Develop Photonics-Integrated AI Processors in NEDO-Supported Project

Image: Tenstorrent · Source

Preferred Networks and Tenstorrent have been selected by Japan's NEDO to jointly develop high-performance, energy-efficient AI processors integrating photonics and electronics, targeting advanced AI workloads over a five-year research program.

Preferred Networks, Inc. (PFN) and Tenstorrent USA, Inc. (Tenstorrent) announced their joint selection by Japan’s New Energy and Industrial Technology Development Organization (NEDO) for a project titled "xPU Design and Development Incorporating Photonics-Electronics Integration Technology." This research initiative falls under NEDO’s program focused on enhancing infrastructure for post-5G information and communication systems and developing advanced semiconductor manufacturing technologies.

The collaboration aims to address the burgeoning computational demands posed by expanding AI models and workloads, which are increasingly deployed across diverse industries. By developing a system-level integration of AI processors that utilize photonics-electronics integration technology and advanced packaging techniques, the partnership hopes to create AI hardware that delivers both high performance and efficiency for training and inference tasks.

Planned as a five-year research and development project, the initiative involves a clear division of responsibilities between the two companies. Both PFN and Tenstorrent will jointly develop the system integration technologies for AI processors leveraging the photonics-electronics approach. PFN will focus on creating AI chiplets featuring ultra-high memory bandwidth enabled by 3D memory stacking technologies using cutting-edge semiconductor processes. Concurrently, Tenstorrent will develop CPU chiplets based on an open architecture to support confidential computing.

In addition to the core companies, the project includes collaboration with academic and research institutions such as The University of Tokyo, Kobe University, Toyo University, the Japan Agency for Marine-Earth Science and Technology, and the Japan Atomic Energy Agency to advance domain-specific applications, including scientific AI and physical AI. Tenstorrent Japan will assist in the design of CPU chiplets and the development of software facilitating confidential computing.

Preferred Networks, founded in 2014 and based in Tokyo, is recognized for its vertically integrated AI solutions encompassing AI processors, cloud platforms, and generative AI models. Tenstorrent, led by renowned architect Jim Keller, develops RISC-V-based AI processors and systems with a global footprint and strong industry partnerships.

This project signifies an important step toward next-generation AI hardware optimized for scaling future AI workloads and new AI applications, backed by government support to strengthen Japan’s semiconductor and AI technology capabilities.