Modular Hardware Implementation of SOM Neural Network Based on FPGA

Abstract

With the development of artificial intelligence technology, artificial neural networks have been widely used in many fields. And traditional artificial neural network implementation methods generally have the disadvantages of poor flexibility and low real-time performance. To solve these problems, this paper presents a modular hardware implementation method of SOM neural network based on FPGA. Firstly, based on the analysis of the model structure, operation process and learning algorithm of SOM neural network, combined with the conditions and limitations of FPGA hardware implementation, the network model is divided into five modules with relatively independent structure and function. Secondly, VHDL is used to describe the RTL of each module to build a common module library. Finally, in practical applications, the target hardware of the SOM neural network can be constructed by combining the modules according to requirements. When building a hardware network, flexible configuration of the arithmetic structure is achieved by setting and passing parameters. The performance test results show that compared with the software simulation method and the traditional hardware implementation method, this method has higher running speed and more flexible operation structure, and can meet the application requirements of high-speed and miniaturized intelligent information processing system.