New optical device enables real-time edge AI with over 90% accuracy
-
Researchers developed a new optical device for edge AI capable of real-time data processing across various timescales. It demonstrates high accuracy in classifying handwritten digits.
-
The device combines memristor properties with controllable relaxation time of photo-induced current using an ITO/NbSTO junction.
-
It acts as an efficient and low-cost physical reservoir, overcoming limitations in conventional cloud computing.
-
Tests showed 90.2% accuracy in classifying MNIST dataset, outperforming a model without the physical reservoir.
-
This technology enables rapid edge AI analysis for applications like self-driving cars and machine anomaly detection in factories.