基于PSO⁃BP神经网络的可缩放微波晶体管非线性电流建模

Scalable Nonlinear Current Modeling of Microwave Transis⁃tors Using PSO⁃BP Neural Network

提出了一种基于人工神经网络的InP异质结双极晶体管(Heterojunction bipolar transistor,HBT)可缩放非线性电流模型,准确地表征了不同尺寸、不同温度下的直流特性。该模型采用了具有良好泛化能力的反向传播(Back⁃propagation,BP)神经网络,并使用粒子群优化(Particle swarm algorithm,PSO)算法找到最佳的网络权重和偏置来规避其容易陷入局部最小值的问题。对不同尺寸、不同温度和不同偏置下的器件非线性直流特性测试数据,将其分为训练数据集和测试数据集,基于PSO⁃BP神经网络进行训练。最终,通过对比收敛迭代次数、不同温度下的可缩放非线性电流精度以及建模结果,验证了PSO⁃BP神经网络能够很好地表征InP HBT器件在不同温度下的可缩放非线性电流特性,证实了该模型具有较快的收敛速度、建模精度和泛化能力。

An artificial neural network⁃based scalable nonlinear current model for InP heterojunc⁃tion bipolar transistors(HBTs)was proposed to accurately describe the DC characteristics at different temperatures in this paper.The model employed a back⁃propagation(BP)neural network with good generalization capability and used a particle swarm algorithm(PSO)to find the optimal network weights and biases to prevent its tendency to fall into local minima.The measured data of nonlinear current of devices with different sizes,temperatures and biases were divided into training and test data⁃sets and trained based on the PSO⁃BP neural network.Finally,by comparing the number of conver⁃gence iterations,the accuracy of the scalable nonlinear current at different temperatures,and the mod⁃eling results,it is verified that the PSO⁃BP neural network is able to perform well representing the scalable nonlinear current characteristics of the InP HBT devices at different temperatures.Mean⁃while,it is confirmed that the model has a fast convergence speed,modeling accuracy,and generaliza⁃tion ability.