Convolutional neural network chip acceleration method based on 5G

Abstract

The invention belongs to the field of computer hardware and fusion of chip design and a convolutional neural network and relates to a convolutional neural network chip acceleration method based on 5G. The convolutional neural network chip acceleration method based on 5G comprises the steps of cpu data processing, spiflash data reading and writing and convolutional neural network algorithm optimization. The acceleration method is based on an FPGA processor. According to the convolutional neural network chip acceleration method based on 5G, high-performance and high-degree parallel computing can be achieved, complex data operation can be supported, and therefore the method is suitable for a more efficient data processing mode; the FPGA can provide a better solution for the problem of poor effect during the processing of matrix operation by a cpu; and meanwhile, FPGA programming can change the network programming and the algorithm at any time, so that the method is more suitable for optimizing the algorithm, can be better suitable for parallel processing of data streams by the HLS toolbox, and improves operation speed.

Description

technical field [0001] The present invention relates to the field of computer hardware and the integration of chip design and convolutional neural network. More specifically, the present invention relates to a 5G-based convolutional neural network chip acceleration method. Background technique [0002] With the innovation of the new generation of communication technology, 5G communication has higher and higher requirements for chip processing performance and baseband cooperation performance. Chip optimization and chip performance reform have gradually become the focus of major manufacturers. How to improve The processing capability of the chip, designing a chip with low power consumption and no lag has also become a difficulty in research at this stage. The traditional cpu can no longer handle the rapidly growing amount of data. At the same time, simply increasing the number of cpu is not the best way to accelerate, and it cannot meet the needs of today's simple and portable...

AI Technical Summary

Problems solved by technology

The traditional cpu can no longer handle the rapidly growing amount of data. At the same time, simply increasing the number of cpu is not the best way to accelerate, and it cannot meet the needs of today's simple and portable mobile terminals. Deep learning and neural networks, with their powerful computing capabilities, seem to be It can better solve the problem of data processing, but the framework required for its runtime requires a higher storage space for the cpu

Images