Neural network unit circuit based on memristor bridge synapses

Abstract

The invention discloses a neural network unit circuit based on a memristor bridge synapse, and the neural network unit circuit comprises a memristor weight circuit which is used for weighting a voltage signal transmitted by an upper-level neuron and transmitting the voltage signal downwards; a differential amplification circuit which is used for converting the voltage signal transmitted by the synapse into a current signal; and a current mirror circuit which is used for adding and summarizing the current signals transmitted by the synapses and transmitting the current signals to the next stageof neurons. The invention aims to solve the problems that a transistor-based hardware neural network is high in power consumption and large in size, and a neural network circuit based on a memristorarray is low in fault tolerance is solved. The memristor bridge composed of five memristors is used for simulating synapses in neurons, the differential amplification circuit is used for simulating axons of the neurons, the current mirror circuit is used for simulating dendrites of the neurons, the three parts play a role in signal weighting, voltage-current conversion, current accumulation and cascade voltage output respectively, and a neural network unit circuit is completely achieved.

Description

technical field [0001] The invention relates to a neural network unit circuit based on memristive bridge synapse, which can be used in the technical field of hardware neural network. Background technique [0002] Memristor is a two-port nonlinear passive electronic device based on the resistance transition effect, which can memorize the amount of charge flowing through it. Oxide semiconductor) is compatible with the composition of hybrid (hybrid) units, etc., and has strong scalability and 3D stacking capabilities. Memristors also have characteristics very similar to synapses. Neural network has a great application prospect. Non-volatile memory has the characteristics of fast erasing and writing speed, low power consumption and multi-valued storage, and can use cross-array structure to realize high-density storage. [0003] Due to the unique properties of memristors such as plasticity, analog behavior, non-volatility, nanoscale size, and low power consumption, memristors h...

AI Technical Summary

Problems solved by technology

[0005] At present, transistor-based hardware neural networks such as CPUs, GPUs, FPGAs, and ASICs face the problems of large size and high power consumption, and the fault tolerance of hardware neural networks based on existing memristor arrays is very low.

Images