A Neural Network Processor and Design Method Based on Efficient Multiplexing Data Stream

Abstract

The invention puts forward a neural network processor based on efficient multiplex data stream, and a design method, and relates to the technical field of the hardware acceleration of neural network model calculation. The processor compares at least one storage unit, at least one calculation unit and a control unit, wherein the at least one storage unit is used for storing an operation instruction and arithmetic data; the at least one calculation unit is used for executing neural network calculation; and the control unit is connected with the at least one storage unit and the at least one calculation unit for obtaining the operation instruction stored by the at least one storage unit via the at least one storage unit, and analyzing the operation instruction to control the at least one calculation unit, wherein the arithmetic data adopts a form of the efficient multiplex data stream. By use of the processor, the efficient multiplex data stream is adopted in a neural network processing process, a weight and data only need to be loaded into one row of calculation unit in a calculation unit array each time, the bandwidth of data on chip is lowered, a data sharing rate is improved, and energy efficiency is improved.

Description

technical field [0001] The invention relates to the technical field of hardware acceleration for neural network model calculation, in particular to a neural network processor and a design method based on high-efficiency multiplexing data streams. Background technique [0002] With the continuous development of machine learning technology, deep neural network has become the best solution for cognition and recognition tasks, and has attracted widespread attention in the fields of recognition detection and computer vision, especially in the field of image recognition, deep neural network has reached or even surpassed Human recognition accuracy. [0003] The deep network structure obtained by deep learning is an operation model, which contains a large number of data nodes, each data node is connected to other data nodes, and the connection relationship between each node is represented by weight. The mainstream neural network processing hardware includes general-purpose graphics ...

AI Technical Summary

Problems solved by technology

[0003] The deep network structure obtained by deep learning is an operation model, which contains a large number of data nodes, each data node is connected to other data nodes, and the connection relationship between each node is represented by weight. The mainstream neural network processing hardware includes general graphics processing However, with the continuous improvement of the complexity of the neural network, the problems of the neural network technology occupying a lot of resources, slow operation speed, and large energy consumption in the actual application process are becoming more and more prominent. , the applicability in mobile platforms or embedded platforms is not high, so there are serious energy efficiency problems and computing speed bottlenecks when this technology is applied in embedded devices or low-overhead data centers.

Images