Honey-Based EGOFETS Could Deliver Energy-Efficient Machine Vision Systems as Sweet as Can Be
Built using recyclable and biodegradable materials, EGOFETs are orders of magnitude more efficient than CMOS electronics, say researchers.
Researchers from the University of Glasgow, São Paulo State University, and Hong Kong Metropolitan University have come up with a machine vision system that's as sweet as can be — as the brain-inspired device is built, in part at least, of natural honey.
"In conventional computing, there's an inherent latency from having to fetch and transfer data in CMOS [Complementary Metal-Oxide Semiconductor]-based systems due to the physical separation between the processing and memory units," explains project lead Theodoros Serghiou. "“Our new memory-based device, however, performs these functions simultaneously in-memory, similar to how synapses in the human brain work, helping to overcome the bottlenecks of conventional systems. Our device’s ultra-low power consumption and sustainable materials could pave the way for eco-friendly, scalable artificial vision systems in the years to come."
The device at the heart of the team's system is dubbed the Electrolyte-Gated Organic Field-Effect Transistor (EGOFET), and the "sustainable materials" in question may raise an eyebrow: natural honey, which serves as an electrolyte in a sandwich made of glass, gold, and organic photosensitive perylene. Together, the materials act as a photodetector whose output "spikes" in response to different wavelengths and intensities of light — and produce a device that can be programmed into memory states, which are retained even when power is removed.
Using the EGOFET, the team was able to produce a highly-efficient machine vision system that is inspired by the operation of the human brain — drawing only 2.4 picojoules of energy for each spiking event, orders of magnitude less than conventional CMOS electronics. Better yet, the device is easily recycled at its end of life — and the organic materials are fully biodegradable. In testing, the single-EGOFET imaging system proved able to distinguish and remember different primary colors at differing power densities with an overall accuracy of 95.6%.
"Our device is able to emulate key synaptic behaviors such as short-term and long-term plasticity, spike-time dependent plasticity and paired-pulse facilitation with high fidelity," claims corresponding author Jeff Kettle. "The implications of this research, which brings together experts from around the globe, extend beyond artificial vision to broader applications in sustainable neuromorphic computing, image processing and energy-efficient bio-inspired electronics. We plan to scale this single-device prototype into arrays for enhanced image recognition capabilities, which could provide enhanced performance for artificial vision systems before being more sustainably disposed of at the end of their lifecycle."
The team's work has been published under open-access terms in the journal Advanced Functional Materials.
