As power consumption increases due to machine learning technology, the demand for AI devices with low power consumption and high computational performance is increasing. “Physical Reservoir” is an AI device that performs efficient brain-based information processing called reservoir computing, and is characterized by low calculation load and low power consumption. However, so far, the drawback has been low computational performance compared to software processing.
A research team from NIMS, Tokyo University of Science, and Kobe University has recently developed an ion-gel/graphene electric double layer (EDL) transistor-based ion gate reservoir (IGR) that achieves high computational performance comparable to deep learning while reducing computational load by an order of magnitude. By combining graphene, which has high electron mobility and ambipolar behavior, with ion gels, complex interactions generate various responses with different speeds (various ways in which ions and electrons move), making it possible for devices to respond to input signals with widely varying time constants (rates of change).
It has demonstrated the highest level of computational performance among conventional physical reservoirs, comparable to software-based deep learning, and has succeeded in reducing the computational load by approximately 100 times. This high-performance system based on ion gel/graphene is also highly compatible with flexible electronics, which are expected to become next-generation edge devices.
