Space Mapping Technique Using Decomposed Mappings for GaN HEMT Modeling

• Published in: IEEE transactions on microwave theory and techniques 2020-08, Vol.68 (8), p.3318-3341
• Main Authors: Zhao, Zhihao, Zhang, Lei, Feng, Feng, Zhang, Wei, Zhang, Qi-Jun
• Format: Article
• Language: English
• Subjects: Artificial neural network (ANN); Computational modeling; Data models; decomposed mappings; Equivalent circuits; Gallium nitride; Gallium nitrides; GaN high-electron-mobility transistor (HEMT) modeling; HEMTs; High electron mobility transistors; Integrated circuit modeling; Logic gates; Machine learning; Mapping; Mathematical model; Modelling; Modules; Neural networks; Semiconductor devices; space mapping (SM); Training; Trapping; trapping effects
• ISSN: 0018-9480; 1557-9670
• DOI: 10.1109/TMTT.2020.3004622

A novel space mapping (SM) modeling approach for gallium nitride (GaN) high-electron-mobility transistors (HEMTs) with trapping effects is presented in this article, advancing the SM technique for nonlinear device modeling. Existing SM modeling approach uses an external mapping to map an existing device model onto device data. When different branches inside the existing device model need to address very different behaviors, such as trapping effects and frequency dispersion in GaN HEMTs, it is hard for one external mapping to simultaneously map different behaviors. The proposed SM technique develops separate mappings for different branches, such that different behaviors can be mapped separately. Each mapping module is formulated to map a specific behavior in the overall model. Each mapping module is developed through machine learning to systematically overcome the gap between each internal branch and each set of target data, accelerating the process of model development. The proposed SM technique is a fast and systematic modeling approach, compared with the existing empirical function/equivalent circuit approach. Compared with the pure neural network modeling approach, the proposed SM technique employs less training data. Measurement data of a 2\times 350\,\,\mu \text{m} GaN HEMT device are employed for model training and verification. Good agreement can be achieved between the developed large-signal model and the measurement data, including dc, pulsed I-V (PIV) at seven quiescent biases, S -parameters, and power characteristics. Reasonably close predictions of load-pull figures of merit are achieved by the developed model. The model development illustrated in the example shows that the proposed SM technique is a fast modeling approach to develop an accurate large-signal model for GaN HEMTs.