具有四值离散态的复合神经元网络及其电路实现

Complex Neural Network With Four Discrete Values and Its Hardware Implementation

介绍了由两种不同类型或功能的神经细胞耦合而成的具有四值离散态的复合神经元网络(CNN),并给出了时间连续、状态连续的CNN(complex neural network)动力学方程及稳定性证明;在此基础上,给出了具有半并行结构的电路实验方案,并实施了四对复合神经元的电路。实验证实理论结果及电路的稳定性,表明CNN有良好的联想记忆和容错能力,为进一步用VLSI实现CNN提供了一种可行的方案。

A complex neural network (CNN), which consists of two types of nerve cell with four discrete values, functionally differing from each other, is introduced. The dynamic equation of continuum CNN and the proof of its stability are given in the paper. Based on the discrete dynamic equation of CNN, a hardware implementation with semi-parallel and hybrid structure for 4 pairs of CNN is realized. It offers a pssibility to realize the CNN with VLSI techniques. The experiments conducted on the implementation show that the CNN has good associative memory and high tolerance.