Breakthrough in Low-Power AI Sensors for Time-Series Data Processing
Breakthrough in Low-Power AI Sensors for Time-Series Data Processing
Artificial intelligence (AI) is becoming increasingly effective in predicting emergency events like heart attacks, natural disasters, and infrastructure failures, thanks to advanced technologies capable of quickly processing data. One promising technology is reservoir computing, which is designed for processing time-series data with low power consumption. Among the various frameworks for reservoir computing, physical reservoir computing (PRC), which uses optoelectronic artificial synapses that replicate human neural functions, stands out for its exceptional recognition and real-time processing capabilities, similar to the human visual system.
However, current PRC devices based on self-powered optoelectronic synaptic elements struggle with processing time-series data across multiple timescales, which is crucial for monitoring infrastructure, natural environments, and health conditions.
A recent breakthrough by a research team at the Department of Applied Electronics, Graduate School of Advanced Engineering, Tokyo University of Science (TUS), led by Associate Professor Takashi Ikuno, has addressed this issue. The team, including researchers Hiroaki Komatsu and Norika Hosoda, has developed a self-powered, dye-sensitized solar cell-based optoelectronic photopolymeric human synapse with a time constant that can be adjusted by input light intensity. Their study was published in ACS Applied Materials & Interfaces on October 28, 2024.
Dr. Ikuno explains the motivation behind their work: "To process time-series optical data with diverse timescales, it's crucial to create devices tailored to the desired timescale. Inspired by the eye's afterimage effect, we developed a novel optoelectronic human synaptic device that can serve as a computational framework for power-saving edge AI optical sensors."
The solar cell-based device uses squarylium derivative-based dyes and incorporates optical input, AI computation, analog output, and power supply functions at the material level. It demonstrates synaptic plasticity in response to light intensity, exhibiting characteristics like paired-pulse facilitation and paired-pulse depression. By adjusting the light intensity, the device achieves high computational performance for time-series data tasks, regardless of the pulse width. When used as the reservoir layer in PRC, the device successfully classified human movements—such as bending, jumping, running, and walking—with more than 90% accuracy. Remarkably, its power consumption is just 1% of what is required by conventional systems, significantly reducing associated carbon emissions. Dr. Ikuno emphasizes, "This is the first demonstration of a device that operates with extremely low power consumption while accurately identifying human motion." The device represents a major step toward the realization of edge AI sensors for various time scales, with applications in surveillance cameras, vehicle cameras, and health monitoring systems. Dr. Ikuno notes, "This invention can be used as an edge AI optical sensor on any object or person, offering a potential impact on reducing power consumption costs in applications like car-mounted cameras and computers." He further adds, "The device could be used in stand-alone smartwatches and medical devices, significantly lowering costs to be on par with or even cheaper than current medical devices." In conclusion, this innovative solar cell-based device holds the potential to accelerate the development of energy-efficient edge AI sensors for a wide range of applications, from health monitoring to smart technologies.
