Temperature Dependent Small-Signal Neural Modeling of High-Periphery GaN HEMTs

This paper analyzes the thermal dependence of high-periphery GaN-on-SiC HEMT performance. The proposed approach is based on artificial neural networks (ANNs) that are used to model the scattering parameters versus temperature and frequency under a high dissipated power condition for a GaN HEMT with a gate width of 1.5 mm. The modeling results agree very well with measurements up to 65 GHz in the whole considered temperature range going from 35°C to 200°C, confirming the high accuracy and the good generalization capability of the proposed ANN approach.