Series connection of low-voltage SiC MOSFETs is a practical and cost-effective approach for realizing high-voltage SiC switches. However, achieving dynamic voltage-balancing in series-connected SiC MOSFETs is a critical challenge, which is strongly influenced by variations in intrinsic device parameters (e.g., Vg threshold, transconductance, and packaging parasitics). Higher switching frequency and voltage/current levels further exacerbate voltage balancing. Moreover, since the intrinsic parameters are highly coupled, it is difficult to analyze their individual effect on dynamic voltage-balancing. In this paper, the influence of intrinsic parameter variations is systematically investigated and statistically quantified through an “Impact Factor” using a machine-learning based approach. Experimental data used for training the surrogate model was obtained using two series-connected 3.3 kV SiC MOSFETs, tested at bus voltages up to 4.2 kV and 40 A turn-off current. The results provide new insights into the mechanisms governing dynamic voltage-balancing and offer practical guidance for designing balancing circuits to ensure improved performance in high-voltage SiC modules.
