Novel Dynamic Characterization Method Applied to the Modeling and Understanding of Power GaN HEMTs to Enhance Static Converter Reliability

Ludovic Roche PhD defense

Soutenance

13.04.26 - 13.04.26

Wide Band Gap Transistors are a key asset for the ongoing energy transition. They allow for reduced energy losses and better density of power electronics converters compared to their silicon counterparts. These improvements stem from their faster switching transients and lower on-state resistance. With these criteria, Gallium Nitride High Electron Mobility Transistors are currently the best commercially available devices. However, significant fluctuations in the electrical characteristics of GaN HEMTs have been reported. In particular, quasi-static gate and drain biases shift their threshold voltage and capacitances. Yet, few studies directly correlate quasi-static and dynamic behaviors, likely due to the inherent difficulties of dynamic characterization. Indeed, GaN HEMTs fast transients cause increased voltage and current ringing in dynamic operation. To address this issue, conventional circuit design methods trade-off between 1) the swiftness of the switching transients 2) the amplitude of the ringing and 3) the tracks length. While this trade-off can be satisfactorily balanced in actual applications, characterization requires increased signal path length for modularity and measurement purposes. This constraint reduces the compromise to a choice between high ringing or slowed down transients. Both solutions result in unrealistic signals or measurement errors that hinder precise dynamic characterization. Hence, the dynamic behavior of GaN HEMTs is not fully understood and more in-depth studies of dynamic operation are needed. However, the high ringing vs. slowed down transients impasse is inherent to conventional design methods. New methods are thus needed to overcome this dilemma. In this work, propagation models and design methods from the radiofrequency and electrostatic discharge domains are adapted to design an original switching measurement setup and a custom gate driving generator. With this setup, the dynamic behavior of normally-off GaN HEMTs is then studied simultaneously with 1) parametrized gate transients 2) a realistic on-time and 3) in hard-switching condition for the first time in literature. We find that, contrarily to state-of-the-art power switching circuits, the signal integrity of the proposed setup is independent from its density and the shortness of its signal tracks. This results in current loops of up to several meters, with no ringing and while preserving GaN HEMTs very fast transients. The signal generator also allows for custom gate waveforms with software-controllable transients, or spikes. Dynamic operation of p-GaN HEMTs is found as reproducible as pre-conditionned static measurements, unlike raw static measurements. We evidence various and drastic impacts of gate transient spikes on the devices Ron, especially a four to six times increase in one manufacturer SOA. We explain the observed behaviors by a shift in the gate p-GaN potential. All observed behaviors fit a proposed two-diode model of the gate stack. This work demonstrates the benefits of a transmission-line based approach to overcome the conventional transient’s swiftness/ringing/signal path’s length compromise in power electronics. The proposed characterization method provides new insight in GaN HEMTs’ dynamic behavior. Especially, such insight in the semi floating pGaN layer behavior could be leveraged for better gate stack designs and better SOA definition. Moreover, the observed behaviors and models can be transposed to predict the switching behavior in actual converters, to enhance their reliability. Especially, gate ringing should be avoided in power converters to prevent the Ron increase phenomenon inherent to at least one tested technology. Such Ron increase could lead to higher conduction losses, overheating and reduced lifetime. Beyond the proven benefits for dynamic characterization, the broadening of the proposed design approach for actual static converter applications remains to be studied.

published on 01.04.26