Memristor-based low-power-consumption pulse convolutional neural network hardware architecture

Abstract

The invention discloses a memristor-based low-power-consumption pulse convolutional neural network hardware architecture. In the prior art, an existing neural network acceleration hardware architecture adopts a digital circuit multiplier and a digital-to-analog and analog-to-digital converter so as to cause huge hardware power consumption and great area consumption; and a traditional neural network acceleration hardware architecture is difficult to adapt to the requirements that the network scale and the calculation complexity are continuously increased. Compared with the prior art, the technical scheme of the invention adopts a low-power-consumption memristor array to complete simulation multiply-accumulate operation so as to achieve the advantages of being high in operation efficiency and low in power consumption, and adopts an array and a sensitive amplifier to simulate the IF neurons of a pulse convolutional neural network so as to eliminate the digital-to-analog and analog-to-digital converters and save a large amount of hardware area consumption. According to the invention, compared with a traditional convolutional neural network with the defects of large operand and over-high energy consumption, the pulse convolutional neural network has the characteristic of high energy efficiency calculation, so that the memristor-based pulse convolutional neural network hardware architecture provided by the invention has the advantage that the performance is greatly improved.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a hardware architecture of a memristor-based low-power pulse convolutional neural network. Background technique [0002] In recent years, artificial neural networks have become a hot research direction in the field of artificial intelligence. Thanks to the development of big data and high-performance computing hardware, convolutional neural networks have achieved remarkable achievements in the fields of image recognition, object detection and pattern recognition. However, today's computing processors (CPUs) and graphics processing units (GPUs) are based on the traditional von Neumann computing system, which consumes a lot of money in the transfer of data from the memory to the processor. As the amount of data and calculations processed by the neural network is increasing, the traditional von Neumann computing system will consume a very large amount of hardware energy con...

AI Technical Summary

Problems solved by technology

[0004] Due to the in-depth understanding of the specific learning methods adopted by biological neural networks, the pulsed convolutional neural network currently lacks efficient training algorithms and lacks certain biological basis. Compared with the network, there is still a certain gap

Moreover, the hardware design of the traditional Von Neumann architecture consumes a lot of power and resources

Images