A memory impedance associative memory neural network circuit with a delay learning function

Abstract

The invention provides a memory impedance associative memory neural network circuit with a delay learning function, and the circuit comprises a voltage control module, a synaptic neuron module and a delay module. The input end of the voltage control module is connected with the input signal, and the voltage output end of the voltage control module is connected with the input end of the synaptic neuron module and the voltage input end of the delay module respectively. The input end of the delay module is connected with the input signal, the output end of the feedback signal of the delay moduleis connected with the input end of the feedback signal of the delay module and the voltage control module respectively, and the output end of the delay module is connected with the input end of the synaptic neuron module. The output end of the feedback signal of the synaptic neuron module is connected with the input end of the feedback signal of the voltage control module, and the input end of thecontrol signal of the synaptic neuron module is connected with the input signal. The invention has a learning function, and the learning rate is variable, has three forgetting processes and delay learning functions, and is more close to the actual situation.

Description

technical field [0001] The invention belongs to the technical field of analog-to-digital circuits, and in particular relates to a memristive associative memory neural network circuit with a delay learning function, which is realized based on the idea of ​​Pavlovian associative memory neural network. Background technique [0002] In 2008, Hewlett-Packard Labs in the United States first produced a physical object of memristor, and its results were published in the journal Nature. Memristor has natural non-volatile memory function and good switching characteristics. field has great application potential. [0003] The research work of artificial neural network has experienced ups and downs since the beginning of the 20th century. In the 1980s, the backpropagation algorithm was used to train multilayer perceptrons, bringing neural networks back to life. The introduction of the concept of deep learning in 2006 pushed the research of neural networks to a new climax. [0004] As...

AI Technical Summary

Problems solved by technology

But the previous neural network circuits either only have a constant rate learning process, or only have a simple forgetting process, and all circuits require that the food signal and the bell signal appear at the same time to generate the learning process

Images