64-device memristor prototype detects edges at 31.25 fps, with 92.3% lower energy than logic circuits
Researchers built a volatile Ag/AlOx/n-Si memristor and a 32-channel edge detection system on it. The authors report its unit uses 92.3% less energy than traditional logic circuits.
Researchers reported in Science Advances a full-hardware edge detection system built on memristors. They fabricated a volatile Ag/AlOx/n-Si memristor and combined two such devices with a logic circuit. The result is a low-power edge detection unit based on dual random number hardware computation science.org
Science Advancesscience.orgOpen source ↗.
The work addresses a known limit. Conventional edge computing hardware at autonomous driving terminals is constrained by chip area and power efficiency when processing complex image data. Memristors have already been used for pattern recognition and edge detection, but a full-hardware system that combines anti-interference, low power and high real-time performance remains challenging, according to the authors.
High-resolution electron microscopy visualized the formation of Ag conductive filaments in the device. The authors report that the unit's energy consumption is 92.3% lower than that of traditional logic circuits, and that it shows strong anti-interference.
To test practical use, the team built a 32-channel parallel full-hardware system with 64 devices. It reached 31.25 frames per second with good edge detection results science.org
Science Advancesscience.orgOpen source ↗. The authors state that the system is scalable, stable and resistant to interference, and name autonomous driving, safety monitoring and surgical assistance as possible areas for edge vision hardware.
