| 引用本文: | 徐坤，张金灿，王金婵，刘敏，李娜. 基于神经网络的GaAs HBT器件模型研究[J]. 雷达科学与技术, 2022, 20(2): 165-172.[点击复制] |
| XU Kun, ZHANG Jincan, WANG Jinchan, LIU Min, LI Na. Research on GaAs HBT Device Model Based on Neural Network[J]. Radar Science and Technology, 2022, 20(2): 165-172.[点击复制] |
|
| 摘要: |
| 建立精确的模型是使用砷化镓异质结双极晶体管器件（GaAs HBT）设计集成电路的必要基础，传统经验模型建立过程复杂，在输出功率、增益、功率附加效率等功率特性方面的模拟精度不太高，给电路设计带来了一定的难度。本文利用径向基函数(RBF)神经网络算法和反向传播(BP)神经网络算法分别建立GaAs异质结双极晶体管器件的大信号模型。这些模型的训练和测试数据分别来自于测试的双端口散射参数，以及测试的直流特性和功率特性数据。然后将模型数据与实测结果进行对比，结果发现，基于神经网络的器件模型能够精确地模拟器件特性，而且RBF神经网络模型相比BP神经网络模型，误差更小，预测更精确。 |
| 关键词: 砷化镓异质结双极晶体管器件 径向基函数神经网络 反向传播神经网络 器件模型 |
| DOI：DOI:10.3969/j.issn.1672-2337.2022.02.007 |
| 分类号:TN322+.8;TN304.2+3 |
| 基金项目:国家自然科学基金(No.61804046); 河南省科技攻关项目(No.202102210322); 河南省高等学校重点科研项目(No.21A510002，20B510006) |
|
| Research on GaAs HBT Device Model Based on Neural Network |
|
XU Kun, ZHANG Jincan, WANG Jinchan, LIU Min, LI Na
|
|
Electrical Engineering School, Henan University of Science and Technology, Luoyang 471023, China
|
| Abstract: |
| An accurate device model is the necessary basis of integrated circuit design for GaAs heterojunction bipolar transistor (GaAs HBT). The process of building conventional empirical models is complicated, and these empirical models are not very precise on modeling power characteristics, such as output power, gain, and power additional efficiency, which brings difficulty to circuit design. In this paper,the radial basis function (RBF) and back-propagation (BP) neural network algorithms are used to build GaAs HBT large-signal device models. The training and testing data for these models are obtained from the two-port scattering parameters, direct-current and power characteristics measurements. Comparison results of measured and modeled data show that the device models based on neural network can accurately model device characteristics. Compared with the BP neural network model, the RBF neural network model has less errors and more accurate prediction results. |
| Key words: GaAs heterojunction bipolar transistor radial basis function neural network back-propagation neural network device model |
