Neural networks based on doublet generator synapses and its \\applications in image processing}{Neural networks based on doublet generator synapses and its applications in image processing

Abstract

Memristor is a nanoscale device with memory and similar synaptic properties. The memristor bridge synaptic circuit, which has the advantages of a simple structure and precise control, can be applied to artificial neural networks, ultra-large-scale integrated circuits, image processing, pattern recognition, etc. However, traditional memristor bridge synaptic circuitry will cause some errors in the synaptic simulation process due to the accumulation of flux caused by unipolar pulses. In this paper, we propose a new synaptic bridge circuit, which effectively overcomes the memristance drifting problem of traditional bridge circuits. Then we apply the new synaptic bridge circuit to a synaptic neural network; the application in image processing has advantages that are more obvious. The feasibility of the structure was verified through simulation experiments, confirming the efficient biomimetic properties of neural networks based on doublet generator memristor bridge synapses. In addition, its higher degree of integration and the ability to replace templates more easily make it more effective in solving real-time complex intelligent problems.