CNN coprocessor

Abstract

The invention discloses a CNN coprocessor, which belongs to the field of artificial intelligence algorithm circuit implementation and comprises a storage module, a convolution calculation module and a data cache control module, wherein the storage module is used for storing data required by forward reasoning; the convolution calculation module is used for realizing convolution calculation in a forward reasoning process; the data caching control module is used for reading data required by current forward reasoning from the storage module and caching the data, and when the CNN reasoning mode is continuous forward reasoning; and after the data required by the current forward reasoning is read and before the current forward reasoning is completed, the data required by the next forward reasoning is read from the storage module and cached. According to the invention, forward reasoning of the current round and data reading of the next round are executed in parallel, the time occupied by data transmission is greatly shortened, and the calculation speed is high; in addition, the invention further introduces a multiplication and addition circuit based on sign bit interpretation, redundancy calculation is effectively reduced, and system power consumption is reduced.

Description

technical field [0001] The invention belongs to the field of artificial intelligence algorithm circuit realization, and more specifically relates to a CNN coprocessor. Background technique [0002] With the continuous development of System on a Chip (SoC) and artificial intelligence algorithms, artificial intelligence algorithms are due to their high accuracy, strong robustness and self-learning ability, among which convolutional neural network (CNN) is due to its local connections and The feature of weight sharing can effectively prevent overfitting and greatly reduce calculation parameters. Compared with other network models, it has higher prediction accuracy and computational efficiency, and is widely used in mobile devices. Due to the large amount of data calculation and information access in the implementation of convolutional neural network algorithm, the calculation efficiency, energy consumption and data storage are issues that cannot be ignored. [0003] In terms o...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a CNN coprocessor to solve the technical problem of slow calculation speed caused by the long time of data transmission existing in the prior art

Images