Circuit for realizing activation function and processor comprising circuit

Abstract

The invention discloses an operation circuit which is suitable for performing operation on data. The operation circuit comprises an index operation unit, a division operation unit, a logistic regression function operation unit and a selection unit. The selection unit is suitable for selecting the operation output of the index operation unit, the division operation unit or the logistic regression function operation unit as an operation result according to an operation mode. The logistic regression function operation unit is coupled to the index operation unit and the division operation unit, and is suitable for performing an index operation in the logistic regression function operation by using the index operation unit and performing a division operation in the logistic regression function operation by using the division operation unit when the logistic regression function operation is performed on the data. The invention further discloses a processor comprising the operational circuit, a system on chip and intelligent equipment.

Description

technical field [0001] The invention relates to the field of processors, in particular to circuit design for neural network operations. Background technique [0002] In the field of artificial intelligence, various neural network models have been proposed to deal with problems in image, video, audio and natural language processing. The neural network model generally includes multiple network layers, and each layer of the network has a large number of nodes. The value of each layer of nodes can be passed to the next layer of nodes and activated by the value of the previous layer of nodes. Therefore, in the neural network model, a large number of activation function calculations are required. [0003] In the field of existing technologies, activation function calculations are usually performed in a software manner. Since this method only uses conventional processor instructions for processing, when performing a large number of activation function calculations, there is a prob...

AI Technical Summary

Problems solved by technology

[0003] In the field of existing technologies, activation function calculations are usually performed in a software manner. Since this method only uses conventional processor instructions for processing, when performing a large number of activation function calculations, there is a problem of poor computing performance, which significantly affects neuroscience. Network training and inference speed

[0004] In the field of existing technologies, there are also hardware acceleration solutions, but the existing hardware acceleration solutions only target one or several activation functions for hardware acceleration processing, and do not see the integration of operations of various activation functions, and Simultaneously provide a variety of hardware acceleration solutions for activation function operations

Images