CIES Consortium

Industry-Academic Collaboration

AI Hardware Circuit Technology

R&D of brain-type processing circuit technologies

In this research and development, we have developed edge AI processors with low power consumption and high speed which are important for Society5.0. In the conventional edge AI processors, after learning has been performed offline and using GPU, learned network weighting has been used for reasoning of AI chip. Generally, since the data acquisition conditions at reasoning (on the spot) are different from those at learning, recognition accuracy is decreased.
In order to solve thi s problem, we have researched and developed the architecture / circuit / device technologies of the ul tra-low power consumption and ultra-high-speed edge AI processors which can learn and reason on the spot. MRAM attracts attention as nonvolatile memory, so we establish a base in the architectures / circuits to utilize MRAM thoroughly, and continue to research and develop fine-grain power gating of system level, high efficiency data transfer mechanism to dissolve data transfer bottleneck, and circuits / devices of MRAM suitable for uses.