A 1t1r-based convolutional neural network circuit and its operation method

Abstract

The invention discloses a 1T1R-based convolutional neural network and its operation method, including an input module, a convolution calculation module, a pooling calculation module, and a fully connected calculation module, wherein the convolution calculation module, the pooling calculation module, the full The connection calculation modules are all composed of 1T1R arrays, which can store synaptic weight information in situ, realize the integration of calculation and storage, save the consumption of data interaction, greatly shorten the calculation time, and reduce energy consumption. In addition, the present invention realizes bidirectional gradual adjustment of the conductance value of the memristor by regulating the gate voltage of the transistor in the 1T1R device, so that in the process of reducing the conductance, it is not necessary to re-operate the 1T1R device to the lowest conductance state, and directly adjust the The gate voltage of the transistor in the 1T1R device can reduce the conductance to achieve the target conductance, and the power consumption is low.

Description

technical field [0001] The invention belongs to the technical field of artificial neural networks, and more specifically relates to a 1T1R-based convolutional neural network and an operation method thereof. Background technique [0002] With the advent of the era of big data, the demand for computing speed and power consumption of traditional computing systems is increasing, and the von Neumann problem has become an important bottleneck restricting the further development of current computer systems. Since the memristor was proposed, it has become one of the most promising development targets for the next generation of semiconductor memory due to its characteristics of non-volatility, high integration, and low power consumption. The fusion of computing and storage features of memristors has also become a powerful foundation for building new computing architectures. Since the memristor was proposed to simulate the synaptic function of the human brain, the research on synapti...

AI Technical Summary

Problems solved by technology

[0007] Aiming at the defects of the prior art, the purpose of the present invention is to provide a convolutional neural network based on 1T1R, which aims to solve the limitation of the prior art due to its conductance gradient, and the device must be re-operated to the minimum when the conductance is reduced. The problem of high power consumption caused by the conductance state to achieve the target conductance in the conductance-increasing operation mode

Images