Tokyo, Japan: A team of researchers from Tokyo University of Science (TUS) has developed a groundbreaking solar-powered optoelectronic synaptic device designed to enhance the performance of low-power edge AI sensors.
Artificial intelligence (AI) is increasingly used to predict emergency events such as heart attacks, natural disasters, and infrastructure failures, requiring advanced technology for quick data processing.
Reservoir computing, particularly physical reservoir computing (PRC), is a promising method for handling time-series data with low power consumption. PRC with optoelectronic synapses, designed to mimic human synaptic functions, offers exceptional real-time processing capabilities.
However, current self-powered optoelectronic devices cannot efficiently handle complex time-series data across multiple timescales, which is necessary for monitoring health, infrastructure, and environmental conditions.
The researchers have produced a self-powered Optoelectronic Photo polymeric human synapse, using a dye-sensitized solar cell that adjusts its time constant based on light intensity. This advancement, published in ‘ACS Applied Materials & Interfaces’, is designed for processing time-series data across various scales.
The device integrates light input, AI computation, and power supply into a single unit, and exhibits synaptic plasticity, such as paired-pulse facilitation and depression. This enables efficient processing of time-series data, with high accuracy and minimal energy consumption, making it ideal for applications in surveillance, car cameras, and health monitoring systems.
The device demonstrated a significant breakthrough by achieving over 90% accuracy in classifying human movements while consuming just 1% of the power required by conventional systems. It is expected to revolutionize edge AI sensors, reducing both energy usage and costs in applications like vehicle cameras, smartwatches, and medical devices, contributing to more sustainable technology development.
This solar-powered, low-power device represents a major leap toward energy-efficient AI sensors with wide-reaching applications in both consumer and industrial sectors.