基于神经网络嵌入式系统体系结构的研究

The Architectures of Embedded System Based on Neural Network

介绍了目前国内外硬件实现神经网络的方案,给出了3种基于共享存储器、总线结构、可编程器件的嵌入式体系结构,提出神经网络嵌入式体系的构想。最后采用神经网络的BP算法,结合数字信号处理器,在共享存储器的嵌入式体系结构中实现神经网络算法。

This paper presents a brief description of current research on the hardware realization of a neural network and describes three architectures of embedded system, respectively, based on shared memory, bus architecture and programmable device. Meanwhile, A conception of neural network embedded system(NES) is also proposed. As an example, a shared memory embedded architecture with DSP, is designed to realize a BP algorithm. The results show that both functions are matched with their ideal functions.