Hidethugu Irie

Development of an on-demand approximate computing framework that dynamically fits user requirements

Researcher

Hidethugu Irie

Outline

This project realizes a flexible approximate computation infrastructure that estimates the user’s accuracy requirement at runtime and adjusts the accuracy in multiple steps, thus going beyond the limits of conventional fixed approximate computation. To achieve this goal, we are trying to address issues that have not yet been addressed, such as methods for estimating the required accuracy, mechanisms to achieve multi-step accuracy, and architectures to improve the balance between the overhead and gain. The instruction set, compiler, and chip reference design of the development infrastructure will be released as open-source.

Kiyoshi Kanazawa

Reduction theory of multi-agent stochastic processes for fast social simulators

Researcher

Kiyoshi Kanazawa

Outline

Numerical simulation of agent-based models (ABMs) is a useful computational method for understanding of social dynamics. While ABMs are described as stochastic decision-making process of individuals, their numerical simulation requires high-computational resource. This project aims to reduce numerical cost of large-scale multi-agent simulations, by developing reduction theory of stochastic processes. Reduction theory has been historically developed within the context of statistical physics to approximately reduce high-dimensional physical dynamics. The investigator develops and apply this methodology to reduce computational cost of general social simulators.

Jun Shiomi

Opening a New Frontier in Secure Computing by Photonic Integrated Circuits

Researcher

Jun Shiomi

Outline

Secure computing based on cryptography is a key technology for realizing the super-smart society. This research aims to develop secure computing platforms exploiting photonic integration technologies for realizing a safe and secure super-smart society. Tampaer-resistant optical computing methods are developed by fully exploiting the optical wave characteristics, which enables to push past the limits of the tampaer resistance in CMOS-based integrated circuits. Design methods for tamper-resistant optical cryptographic circuits are also developed.

Yutaka Shikano

Hybrid Secure Quantum Computation based on Quantum Random Numbers

Researcher

Yutaka Shikano

Outline

Quantum random numbers in principle generates the “true” uniform random numbers because this measurement process is probabilistic. In quantum foundations to testify a validity of quantum mechanics in the Nature, we will define the secure quantum random numbers. Furthermore, we propose the hybrid secure computatation system based on the secure quantum random numbers and advantages of classical and quantum computations.

Sumito Thunegi

Development of nano-oscillator neural networks

Researcher

Sumito Thunegi

Outline

The aim of this project is to develop information processing system named “Nano Oscillator Neural Networks” that partially mimic biological neural networks by using nanosized spin torque oscillators. Utilizing several features of the oscillator such as collective behavior similar to neural networks and low power consumption, the system can be contributed to the realization of a super-smart society.

Thi Hong Tran

Ultra Low Power Consumption Blockchain Accelerator for Society 5.0

Researcher

Thi Hong Tran

Outline

Decentralized blockchain (BC) is an important technology to secure the shared data in smart systems of Society 5.0. Motivated by the Global Sustainable Development Goals (SDGs) No.7 (clean energy), we target to develop a sustainable BC accelerator (BCA) consuming power less than 1mW for edge devices. In this PRESTO project, we will develop and evaluate a low power consumption (1W) high processing rate BCA architecture. We will prove that our tiny BCA wins the existing hardware platforms (GPU Tesla V100, Intel CPU, etc.) on Bitcoin mining task.

Song Bian

Efficient Multi-Party Secure Computation Platform for Safe and Remote Medical Diagnosis

Researcher

Song Bian

Outline

We investigate the design of secure computing platforms by dividing the design of such plaforms into three layers: the basic operator, the protocol, and the application layers. Novel design methods are proposed in each of the layers, while inter-layer optimization techniques are also developed to achieve a domain-specific framework for medical diagnosis. Besides medial applications, the visions and methodologies proposed in this research are expected to become the foundation for secure and efficient computing in many other real-world tasks.

Takeshi Fukaya

Development of numerical algorithms based on the low precision and reliability arithmetics

Researcher

Takeshi Fukaya

Outline

In exchange for performance improvement, the precision and reliability degradation of floating-point operations are expected on hardware in the Post-Moore Era. In this research, I aim for developing new matrix algorithms in which the low precision and reliability arithmetics are exploited, while comparable computational results are also assured. In application layer, such algorithms can fill the gap between hardware in the Moore Era and that in the Post-Moore Era, and realizes the seamless use of new hardware in the Post-Moore Era.

Yutaka Masuda

Testing of Approximate Computing Design with Fuzzing

Researcher

Yutaka Masuda

Outline

This project develops verification technology for approximate computing (AC) circuits using fuzzing. Firstly, this project tackles fundamental problems regarding fuzzing, i.e., preparing initial seed inputs, selecting a mutation strategy dynamically, and evaluating the coverage for the target hardware. Based on these essential components, a prototype of a fuzzing tool is constructed. The project then extends the prototype to establish functional verification technology and the timing verification technology for AC circuits. Lastly, the project verifies the developed technology in a case study of large-scale integrated circuits.

Mohamed Wahib

Parallel Programming Beyond Moore’s Law

Researcher

Mohamed Wahib

Outline

This project attempts the next leap by squarely addressing heterogeneous programming challenges with its heterogeneity-centric framework. We aim to realize a breakthrough in the area of extreme heterogeneity by proposing a completely different model of execution that has the potential to replace the mainstream offloading approach of heterogeneous parallel programming. We introduce an autonomous execution model where accelerators have the autonomy to launch and cooperatively execute a program with other accelerators.