Hideharu Amano

Development of an integrated multi-node system for multi-access edge computing

Research Director

Hideharu Amano

Collaborator

Masahiro Iida	Professor Research and Education Institute for Semiconductors and Informatics Kumamoto University
Midori Sugaya（Shimazaki）	Professor Faculty of Engineering Shibaura Institute of Technology
Hiroaki Nishi	Professor Faculty of Science and Technology Keio University
Kazutosi Wakabayashi	Senior Fellow System Design Research Center The University of Tokyo

Outline

Crust-core HUB is an integrated switch for ensuring the bandwidth and latency which combines a reconfigurable accelerator and a general-purpose processor. Connecting FPGAs and other domain-specific computers with Crust-core HUB, a powerful yet energy-efficient computing system is built for the future multi-access edge computing system. System software and design CAD for such a system are also developed, and the prototype system will be applied to the real smart city.

Isao Inoue

Platform for real-time learning at the edge with spiking neural networks

Research Director

Isao Inoue

Collaborator

Tetsuya Iizuka	Associate Professor Graduate School of Engineering The University of Tokyo
Kantaro Fujiwara	Project Associate Professor International Research Center for Neurointelligence The University of Tokyo
Yasushi Hotta	Asssoiate Professor Graduate School of Engineering University of Hyogo
Takeaki Yajima	Associate Professor Graduate School and Faculty of Information Science and Electrical Engineering Kyushu University

Outline

Back-propagation (BP) of machine learning consumes immense energy for its enormous logic operations. On the other hand, the so-called “attractor formation”, in which firings of neurons of the spiking neural network (SNN) spontaneously converge to some specific loops, has the same function as BP does in the machine learning. With this intriguing feature, we are sure that we will be able to perform on-site ultra-low-power real-time learning/inference. Thus, in this project, we aim to corroborate this scenario by building the attractor SNN circuits consisting of our uniquely developed neuromorphic devices. The SNN circuits are implemented to a personal-authentication device. Finally, the device should learn and infer the unconscious behaviours of a human without accessing the internet nor consuming considerable energy.

Shigeo Sato

Fundamentals of spintronics-based edge computing hardware

Research Director

Shigeo Sato

Collaborator

Takahiro Hanyu	Professor Research Institute of Electrical Communication Tohoku University
Shunsuke Fukami	Professor Research Institute of Electrical Communication Tohoku University

Outline

Towards the realization of Society 5.0, functionalities such as low-power consumption and real-time processing are required in edge computing systems. We therefore develop spintronic devices having non-volatile analog-storage functions and rich dynamics to meet these requirements and research analog-digital-mixed CMOS integrated circuit technologies to maximally utilize the aforementioned device characteristics. We also investigate the optimal hardware architecture for spintronics-based edge computing and finally identify the best way for the practical implementation of our hardware through experimental verifications.

Nozomu Togawa

Research on Ising Computation for Real-World Geographic Information Systems

Research Director

Nozomu Togawa

Collaborator

Toshinori Takayama

Expert Engineer Technical Division ZENRIN DataCom CO., LTD.

Outline

In this project, we focus on geographic information applications, which are necessary for Society 5.0, and newly propose Ising programming for them. Firstly, we pick up variety of geographic information problems, including cyclic-route search problems as well as load packing problems with many types of constraints and represent them onto Ising models. After that, we embed these Ising models into practical Ising computers and finally solve real-sized geographic information problems with real types of constraints.

Naofumi Honma

Secure Information Processing Technology Based on Post-Quantum Encrypted Computing

Research Director

Naofumi Honma

Collaborator

Takashi Sato	Professor Graduate School of Informatics Kyoto University
Masanori Hashimoto	Professor Graduate School of Informatics Kyoto University
Yuichi Hayashi	Professor Graduate School of Science and Technology Nara Institute of Science and Technology

Outline

This research project aims at studying the theory of secure information processing technologies that can execute secret computation based on post-quantum homomorphic encryption in a tamper-resistant manner with extremely high speed and high efficiency, and developing a secure secret computation platform that integrates the technologies. In addition, as new applications pioneered by the platform, we will develop those of secret statistical analysis and secret inference (that conceals input/output and learned parameters in machine learning) based on high-speed/high-efficiency secret computation.