太赫兹 InP HEMTs 的神经网络建模方法

包含新技术、新材料的非传统器件的不断涌现使现有的模型已不能完全表征THz 器件的特性。而采用神经网络 建模的方法，可极大地提高建模的效率和精确度，解决一系列传统模型所无法解决的问题，是一种新型的CAD 建模方法。 本文采用神经网络空间映射的方法，在传统的粗模型的基础上对输入信号进行有效地修正，从而得到适合太赫兹器件的 精确模型，器件的截止频率Ft 和最高振荡频率Fmax 分别为220GHz 和310GHz。模型在直流IV 和1-110GHz 范围内的S 参数与测试结果吻合较好，比传统粗模型的精度有了较大的提高。

Modeling of Terahertz InP HEMTs with Artificial Neural Networks

Due to the appearance of novel devices with new materials and new process, it is too hard for traditional models to characterize those devices operating in terahertz frequency. As a non-traditional modeling method, Artificial Neural Networks (ANN) can improve both the efficiency and accuracy of model of terahertz devices, which is difficult for the existing models. In this paper, we used the spacing-mapping neuro-modeling technique to map the input signals of coarse model, and then a more precise model of Terahertz InP HEMTs was obtained. The cutting frequency Ft and maximum oscillation frequency Fmax of our InP HEMTs are 220 GHz and 310 GHz, respectively. Good agreement between the model and measurement was achieved both in DC and S parameters from 1GHz to 110 GHz frequency