by Cutting-edge technology is transforming the way we process real-time data, and edge computing combined with reservoir technology is leading the way. Traditional cloud computing methods face challenges such as leaks, communication delays, slow speeds, and high power consumption. Edge computing offers an innovative alternative by distributing computations closer to users, reducing the load and speeding up data processing. Edge AI, which involves AI processing at the edge, is set to revolutionize various industries, including self-driving cars and anomaly prediction in factories.
To make edge computing efficient and cost-effective, reservoir computing technology holds great promise. Reservoir computing is a computational method designed to process signals recorded over time, transforming them into complex patterns using reservoirs that respond nonlinearly to the signals. Physical reservoirs, which utilize dynamics of physical systems, are particularly effective and efficient, but real-time signal processing is limited by the natural relaxation time of the physical system.
Researchers at the Tokyo University of Science (TUS) have recently made breakthroughs in this area. Professor Kentaro Kinoshita and Mr. Yutaro Yamazaki from TUS’s Faculty of Advanced Engineering and Department of Applied Physics have developed an optical device that supports physical reservoir computing and enables real-time signal processing across a wide range of timescales within a single device. Their findings were published in Advanced Science on November 20, 2023.
The motivation behind this research was to develop devices capable of processing time-series signals with various timescales that are generated in our living environment. The researchers aim to create an AI device that can be used in edge computing. They created a special device using Sn-doped In2O3 and Nb-doped SrTiO3, which respond to both electrical and optical signals. The team confirmed that the device functions as a memristor, a memory device that can change its electrical resistance. They also investigated the influence of ultraviolet light on the device and found that the relaxation time of the photo-induced current can be modified according to the voltage, making it a potential candidate for a physical reservoir.
To test the effectiveness of the device as a physical reservoir, the researchers used it to classify handwritten digit images in the MNIST dataset. The device achieved an impressive classification accuracy of up to 90.2%. In comparison, experiments without the physical reservoir resulted in a relatively lower classification accuracy of 85.1%. These results demonstrate that the ITO/Nb:STO junction device improves classification accuracy while keeping computational costs lower, making it a valuable physical reservoir.
Prof. Kinoshita highlighted the research group’s focus on developing materials applicable to physical reservoir computing. The aim was to fabricate devices that can control the relaxation time of the photo-induced current through voltage variation. The novel memristor device presented in this study showcases enhanced learning capabilities and promises to be a valuable AI device for edge computing applications.
In conclusion, this research introduces a groundbreaking memristor device that can adjust its response timescale through voltage variation. This device offers enhanced learning capabilities and shows great potential for use in edge computing as an AI device. It opens up the possibilities for single devices capable of effectively handling signals of varied durations found in real-world environments.
