An ANN-Based GaN HEMT Large-Signal Model With High Near-Threshold Accuracy and Its Application in Class-AB MMIC PA Design

Artificial neural network-based gallium nitride high-electron-mobility transistor (ANN-based GaN HEMT) models have garnered significant attention due to their high accuracy, low development costs, and immunity from process fluctuations. They have demonstrated huge potential in class-A and class-C power amplifier (PA) and low-noise amplifier (LNA) designs. However, there is limited reporting on the application of ANN-based GaN HEMT models in class-AB GaN monolithic microwave integrated circuit (MMIC) PAs. One key challenge lies in the limited accuracy and sensitivity of the drain current and gate capacitances in these models in the near-threshold region, which in turn leads to various issues in class-AB PA design, such as inaccurate biasing, stability, gain, and matching network design. To address this challenge, this article proposes an ANN-based GaN HEMT modeling method with high near-threshold accuracy by applying a nonlinear transformation, which utilizes initial fitting functions to transform the targets and amplify their characteristics in the near-threshold region, thereby improving the ANN’s fitting accuracy in this region. The model verification results show that compared to the traditional ANN-based models that do not transform targets, the proposed method achieves significantly higher accuracy in the near-threshold region while maintaining global accuracy. The proposed model is subsequently applied to the design of a class-AB GaN MMIC PA, which is measured to have a center frequency of 9.5 GHz, a 3-dB fractional bandwidth greater than 32%, a saturated output power of approximately 7 W within the bandwidth, and a maximum power-added efficiency (PAE) higher than 50%. Simulations accurately predict these measured values. These verifications demonstrate that the proposed model offers high prediction accuracy in the near-threshold region and performs well in class-AB GaN MMIC PA design.