High-linearity neuromorphic calculation circuit realized by passive voltage stabilizing circuit

Abstract

The invention discloses a high-linearity neuromorphic calculation circuit realized by a passive voltage stabilizing circuit. A 1R1T binary storage unit array is connected with an input sequence control switch and a plurality of bit line capacitor voltage isolation transistors, and the input sequence control switch receives input data of a neural network. Each bit line capacitor voltage isolation transistor is connected with a biasing circuit through a respective passive voltage stabilizing circuit, the plurality of bit line capacitor voltage isolation transistors are jointly connected to an analog-to-digital conversion circuit, and a calculation result is output through the analog-to-digital conversion circuit ADC. According to the invention, the influence of the channel length modulation effect caused by the drop of the capacitor voltage in the integration process can be effectively eliminated, so that the integration current is more stable; and meanwhile, the stability of the source voltage of the T0 under different loads can be ensured, and the technical problem that the linearity of the weight value of the circuit is poor due to the fact that the load change is affected is solved.

Description

technical field [0001] The invention belongs to a neural network computing circuit in the field of neuromorphic computing, and relates to a multi-bit parallel binary neuromorphic computing circuit based on a passive voltage stabilizing circuit. Background technique [0002] In the past decade, as the Internet of Things, cloud computing, computer vision, and artificial intelligence have become more and more closely integrated, sensing devices will become a key interface for smart products to connect to the real world. However, due to the generation of a large amount of sensing data, data transmission and calculation will also be a huge challenge. Traditional intelligent systems have problems such as high construction cost, high energy consumption, low energy efficiency, and high delay due to the separation of data collection, processing, and analysis. In order to solve these problems, it is necessary to put part of the calculation of artificial intelligence on lightweight Io...

Images