Device Requirements and Technology-Driven Architecture Optimization for Analog Neurocomputing

Switching dynamics of single and coupled VO2-based oscillators as elements of neural networks

“…The use of artificial neural networks (ANNs) 1 in information processing has been proposed to overcome critical bottlenecks of traditional computing schemes for applications in such fields as image and speech processing, pattern recognition, and associative memory. 2 Intensive study of this problem began in the 80s of the last century with the work by Hopfield . 3 Since then, various models of ANNs have been discussed in the literature 1 for simulation of the associative memory phenomenon with learning based on the Hebb's rule.…”

“…) of the corresponding generators designated by the numbers(1,2,3,4). The value of the Ri resistance is small, so the voltage across it is proportional to the current flowing through the VO2 switch.…”

Autoassociative Memory and Pattern Recognition in Micromechanical Oscillator Network

“…Following the initial proposals by Hoppensteadt and Izhikevich 2–4 , recent work has focused on schemes to implement the basic principle across a variety of different platforms. Spin-torque oscillators, in which the magnetization of a thin ferromagnetic layer is induced into sustained oscillation through the application of bias current or external magnetic field, have been shown to be capable of frequency locking via a number of methods, and show promise as a platform for neurocomputing 5–9 . Similarly, vanadium oxide relaxation oscillators, which rely on a precise switching between metallic and insulating states 10 , have been successfully synchronized using capacitive coupling with aims towards application to associative memory operation 11 .…”