Power Electronics Europe February/March 2022

INDUSTRY NEWS 25 www.power-mag.com Issue 1 2022 Power Electronics Europe GaN transistors simultaneously for over 1000 hours of continuous operation. The purpose of this test is to show that the charge trapping mechanism responsible for a long-term increase of R DS(on) follows a log(time) trend. If this trend is maintained over the long-term, then data from the first few hours can be used to project the expected R DS(on) after 10 or 15 years. Figure 4 shows the normalized RD S(on) over time of all the samples under test, and Figure 5 shows the difference between the line fits using either the first five hours of data, or the full 1150 hours. The main source of error in the five-hour line fits are small temperature changes in the ambient temperature. These (random) temperature fluctuations tend to cancel out as the length of the test increases. Nevertheless, the short duration and long duration tests agree to within 10 % on the projected RD S(on) after 15 years. This lends credence to the idea that short-term data collection (over a few hours) can be used to accurately project long-term dynamic R DS(on) behavior. This log (time) extrapolation is valid when the changes in R DS(on) are relatively small. When the changes are larger, the case where a sizeable percentage of the available 2DEG electrons are trapped, there needs to be a more refined extrapolation. Inductive versus resistive hard switching Designers have raised concerns that resistive hard switching is not truly representative of the kind of hot-carrier stress that occurs during inductive hard switching. These concerns have also been voiced in the academic literature, at conference proceedings, and by other GaN manufacturers. The argument centers on the loci the part traverses in current-voltage space during an on- transition. For an inductive transition, the FET experiences higher current during the critical interval of time when both voltage and current are high, precisely the conditions that lead to hot-carrier effects. Though plausible, these arguments are mostly hand-waving, and are never supported by hard data or solid theory. To address this question, both inductive and resistive hard switching conditions were measured. The measurement system was able to alternate from inductive to resistive modes (and back) on the same device under test. For inductive mode, the test circuit is a boost converter operating in Continuous Conduction Mode (CCM). In both modes, the part is switching continuously at 200 kHz, and oscilloscope traces are captured periodically, allowing monitoring of both short term and long term dynamic R DS(on) . Figure 6 shows data for an EPC2204 GaN transistor switching at 80 V. For the first four hours, the part was operated in inductive mode. After that, it was operated in resistive mode for the ensuing four hours. To guarantee a fair comparison, the off-state voltage across the device, frequency, duty cycle, and current at turn-on were kept the same for the resistive and inductive cases. As can be seen in the figure, there is no discernable difference in the slope or intercept of the log(t) growth characteristic: resistive and inductive hard- switching are essentially indistinguishable in terms of dynamic R DS(on) . The same Figure 5: Comparison of log(time) fits to the RDS(on) data, where the dashed line represents the fit over the first 5 hours, while the solid line represent the fit over the full 1150 hours. Data for two samples of EPC2218 are shown. Note that the short-term fit has a similar projection to the long-term fit, with small random differences of ± 10% on the 15 year projection Figure 6: Comparison of inductive versus resistive hard switching on an EPC2204 FET switching at 80 V, 200 kHz. The same part was tested under inductive mode for the first four hours, followed by resistive mode for the next four hours. Both modes are essentially indistinguishable in terms of dynamic RDS(on) is true of short-term effects within the first microsecond of the transition; for neither mode displayed any “fast” recovery effects. This result implies that the mechanism responsible for R DS(on) shifts in GaN transistors is either independent, or weakly dependent on the detailed loci of current-voltage traversed during a transition. In both switching cases, there is simultaneous voltage and current during turn-on. While in resistive switching, the voltage across the transistor decreases as the current rises; whereas, in a purely inductive turn-on, the current rises before the voltage collapses. The fact that dynamic R DS(on) is so similar between the modes suggests that the electron current has a weak influence on hot-carrier trapping. Physics-based dynamic on-resistance models The physics-based model of dynamic R DS(on) explains the results. This model predicts that switch current (or the switching loci) has no impact on slope of the log(t) growth line, as observed. Furthermore, the model predicts that switch current does affect the intercept of the line, but only weakly. In fact, the