采用BP-GA算法的一种LSI神经网络的电路设计

将误差反传 (BP)算法和遗传算法 (GA)有机地结合在一起 ,提出了一种新的算法 BP- GA。采用 BP- GA算法 ,设计了一个两层前向 L SI神经网络。作为神经网络的关键部件 ,提出的新型神经元性能优越。它的激活函数与理想sigmoid函数拟合很好 ;可实现对阈值及增益因子的编程并且不同增益因子下饱和输出电压值相同。采用标准 1.2 μmCMOS工艺的模型参数 ,对该两层前向神经网络电路进行的HSPICE模拟证明了它有解决异或 (XOR)问题的能力

Circuit design of a LSI neural network using BP GA algorithm

A new algorithm is proposed to combine the Back-Propagation algorithm (BP) and the Genetic Algorithm (GA). The combined algorithm is used to design a Large Scale Integrated circuit (LSI) for a two-layer feedforward Artificial Neural Network (ANN). A novel neuron is proposed as the key element of the neural network. The neuron's activation function fit the sigmoid well and the bias weight and the gain factor of the neuron can be modulated. Further more, the saturation levels of the sigmoid remain constant for different gain values. HSPICE simulations were done using the neural network using transistor models for a standard 1.2μm CMOS process. Results using the exclusive or (XOR) benchmark demonstra te its effectiveness.