Maximum likelihood Weibull analysis of repetitive short-circuit stress in asymmetrical trench SiC MOSFETs

The repetitive short-circuit (R-SC) endurance of asymmetrical trench SiC MOSFETs is characterized through a maximum likelihood Weibull statistical framework with 90% Fisher-information confidence intervals, applied to a test matrix spanning two DC-link voltages (400 V and 600 V), six pulse durations per voltage, and four case temperatures (25, 75, 100, and 125 °C) at two selected operating points. The Weibull shape parameter β as a function of pulse duration is identified as a quantitative statistical signature of the underlying degradation behavior, with a strictly monotonically decreasing β ( t p ) profile at 400 V (verified by the Mann–Kendall trend test at the 1% significance level) and a non-monotonic profile at 600 V featuring a local minimum within the degradational regime followed by a sharp rise within the catastrophic transition window. The temperature-dependent analysis reveals a sign reversal of the pairwise apparent activation energy of the scale parameter at both operating conditions, with global linear Arrhenius regression coefficients of R 2 = 0.164 and R 2 = 0.528 that rule out a single-Arrhenius lifetime model over the tested case-temperature range. Post-stress transfer characteristics provide an independent electrical signature: the change in subthreshold swing and the threshold-voltage shift follow qualitatively different pulse-duration dependences, which a single trapped-charge population could not produce. The convergence of these complementary signatures — voltage-dependent β ( t p ) profile, sign reversal of the pairwise activation energy, quantile-dependent power-law exponent, and the divergent evolution of the two static-parameter metrics — is consistent with two competing degradation mechanisms whose relative weight depends on pulse duration, applied voltage, and case temperature. The engineering implication is that the population spread quantified by the B 99 / B 1 ratio reaches a factor of 30.3 at 400 V and 14 µs, requiring derating prediction based on the lower-tail percentile rather than on the median in regimes of strong β ( t p ) decline.