Dynamic On-State Resistance and Threshold-Voltage Instability in SiC MOSFETs

. Dynamic on-state resistance has been experimentally observed in all commercially-available SiC MOSFETs studied on the time scale of normal device operation, and can be explained by the presence of dynamic threshold-voltage instability. The magnitude of this dynamic on-state resistance varies from vendor to vendor, but in every case this magnitude generally corresponds to the magnitude of that device’s threshold-voltage instability, as described by standard textbook equations—especially in the case of large threshold-voltage instabilities.


Introduction
This work presents very recent results investigating the direct relationship between the well-known dynamic instability of the threshold voltage (VT) [1,2], and the newly-revealed dynamic on-state resistance (RDS-ON) [2,3] in commercially-available SiC MOSFETs.Both planar-channel (Vendor M) and trench-geometry (Vendor T)-as defined in [2][3][4]-SiC power MOSFETs were characterized in terms of both dynamic RDS-ON and, independently, VT instability under similar gate-switching conditions comparable to standard power-switching operations, details of which are given elsewhere [2,3].The VT instability data was then applied to standard semiconductor device equations relating VT to the channel resistance, and the channel resistance in turn to the total resistance, resulting in a calculated expected change in RDS-ON due to a change in VT, which was then compared to the actual insitu measured change in RDS-ON.
The specific channel resistance is given as wherein this specific channel resistance is a function of the channel length, cell pitch, inversion channel mobility (μch), gate-oxide capacitance, and the difference between the applied gate-to-source voltage, VGS, and VT.The specific on-state resistance is the sum of all component resistances, normalized by the active area of the device.Fig. 1 shows the theoretical curve for RDS-ON versus threshold voltage for a generic SiC MOSFET, as a function of μch, for the same VGS that was applied when measuring RDS-ON directly.Lower channel mobilities result in larger channel resistances, thus affecting the total resistance to a greater degree.In addition, Fig. 1 shows that not only does total RDS-ON increase with decreasing channel mobility, but more importantly, the slopes of the curves increase as well, resulting in larger changes in RDS-ON for the same change in VT -and these slopes increase even more for larger values of VT.

Results and Discussion
VT hysteresis is a fundamental phenomenon occurring in all SiC MOSFETs.When subject to an alternating gate bias, near-interfacial oxide traps are alternately charged and neutralized, resulting in non-permanent drifts of VT [1].Although such dynamic VT does not represent device degradation, if large enough, it may affect device (and therefore circuit) performance.
The magnitude of this dynamic variation in VT depends on several factors, in particular the magnitude and polarity of the applied gate bias, and perhaps most importantly the duration in time that such bias is applied [1].Naturally, the longer the applied bias, the greater the effect, although even gate biases applied on the time scales of normal device operation can result in noticeable VT instabilities-but only if measured on a similar or faster time scale.Faster VT measurements reveal more of the VT instability that is present, but even the fastest measurements do not show the full extent of the VT drift experienced by the device-although the effect of such drift is likely captured more readily by in-situ measurements of changes in the RDS-ON [2,3].
Fig. 2 shows the measured VT instability of a previously unstressed DMOSFET from Vendor M, both at 25 and 125 °C.Both the low side (measured following the application of a negative gate bias for the indicated hysteresis stress interval time) and high side (measured following the application of a positive gate bias) of each VT hysteresis envelope is plotted.As is commonly observed, VT decreases with increasing temperature, and its VT hysteresis also decreases [1].Nonetheless, significant VT instabilities are observed on the time scale of normal device operation.For example, at room temperature, the Vendor M device shows a VT instability of 2.9 V for a hysteresis stress interval time of 50 μs, which corresponds to an operating frequency of 10 kHz with a 50% duty cycle.At 125 °C, the VT instability decreases to 1.7 V under the same bias-switching conditions.Dynamic RDS-ON was characterized (independently of VT instability) for both planar and trench SiC MOSFETs, up to 100 kHz and up to 125 °C, under conditions comparable to standard powerswitching operations.A standard gate driver with VGS of +15 and −5 V was used to supply the onstate and off-state voltages of the MOSFETs, respectively.The change in on-state resistance, ∆RDS-ON, is defined as the difference between RDS-ON measured at the beginning of each on-state interval and RDS-ON measured at the end of the on-state interval immediately before the turn-off transition.The VT instability data, such as that from Fig. 2, was then applied to (1) relating VT to the channel resistance,

118
Solid-State Power Devices: Operational Reliability and Parameters' Stability and the channel resistance in turn to the total resistance, resulting in a calculated expected change in RDS-ON due to a change in VT -which was then compared to the actual in-situ measured change in RDS-ON.This is seen in Figs. 3 and 4, which compare the calculated change in on-state resistance (∆RDS-ON) -for different assumed values of μch -based on the VT instabilities (∆VT) shown in Fig. 2, with the actual in-situ measured values of ∆RDS-ON (labeled "Experiment")-at 25 °C and 125 °C, respectively.The magnitude of the dynamic RDS-ON varies between 56% (at 10 kHz) and 42% (at 100 kHz) for the 25 °C case, whereas it is much smaller for the 125 °C case, varying between just 18% and 14%.This is unsurprising, given that the curves in Fig. 1 are much flatter for lower values of VT (and which naturally vary less with smaller changes in VT).Even so, a good fit to the data also requires that μch increase with increasing temperature (μch = 1 cm 2 /V⋅s at 25 °C; μch = 10 cm 2 /V⋅s at 125 °C), consistent with what is reported on its data sheet.Fig. 5 compares the VT instability of the same Vendor M device from Fig. 2 with a trench-geometry Vendor T device-both at 25 °C.The Vendor T device has a much, much smaller VT instability.Fig. 6 plots the dynamic RDS-ON of a Vendor T device as a function of gate-switching stress (GSS) cycles [3,4], again comparing actual in-situ results ("Experiment") with calculated values based on measured VT instability-all at a switching frequency of 10 kHZ (the GSS frequency was about 1 MHz).It is interesting to observe that before any increase in VT instability due to GSS-induced degradation (due to negative bias overstress effects [2,4]), there is a poor match between the measured ∆RDS-ON of about 22% and a calculated value well below 5%.As degradation occurs and ∆VT increases, the experimental and calculated values of ∆RDS-ON converge.For the very large VT instabilities measured when GSS cycles exceed 3×10 12 , the calculated ∆RDS-ON exceeds 160%.
In general, the standard textbook equations relating RDS-ON and VT were found to be consistent with our experimental results.The main difficulties appeared when measuring dynamic RDS-ON with small VT instabilities.In these cases, the measured values of dynamic RDS-ON were much larger than what might reasonably be expected based on the VT hysteresis measurements.One possible explanation is that even the fastest I-V measurements do not observe all the VT instability that actually occurs [1,3], yet is present and contributes to the dynamic RDS-ON that is measured in-situ.On the other hand, large VT instabilities, which would predict large variations in RDS-ON, were confirmed by experiment.Expected variations in RDS-ON with temperature were generally confirmed as well.Regardless of the cause of the large dynamic variability in VT, a large VT instability of more than a few volts can result in the presence of a significant dynamic variability of RDS-ON.If such dynamic RDS-ON effects are large enough, they may result in reduced device reliability due to an unanticipated increase in power dissipation [2,3].

Fig. 1 .
Fig. 1.Theoretical curve for RDS-ON versus VT for a SiC MOSFET, as a function of channel mobility.

Fig. 2 .
Fig. 2. VT hysteresis data of a Vendor M planar SiC DMOSFET, for temperatures of 25 and 125 °C.Similar line colors indicate the same bias polarity of the hysteresis stress, and similar symbol shapes and colors indicate the same stress-and-measurement temperature.

Fig. 3 .
Fig. 3. Results for Vendor M planar SiC DMOSFET at 25 °C.Comparison of experimentally measured change in RDS-ON, as a function of switching frequency, with calculated changes in RDS-ON -applying measured values of VT instability (for VT hysteresis measurements with hysteresis stress interval times corresponding to the switching frequency) to the theoretical curves from Fig. 1, as a function of assumed channel mobility.

Fig. 4 .
Fig. 4. Results for Vendor M planar SiC DMOSFET at 125 °C.Comparison of experimentally measured change in RDS-ON with calculated changes in RDS-ON -again applying the theoretical curves from Fig. 1 as a function of assumed channel mobility.

Fig. 5 .
Fig. 5. VT hysteresis data at 25 °C for a Vendor M planar SiC DMOSFET, and for a Vendor T trenchgeometry SiC MOSFET.Similar line colors indicate the same bias polarity of the hysteresis stress, and in this case similar symbol shapes and colors indicate the same device.

Fig. 6 .
Fig. 6. Results for Vendor T trench-geometry SiC MOSFET at 25 °C.Comparison of experimentally measured change in RDS-ON, as a function of total GSS cycles (with constant switching frequency of 10 kHz), with calculated changes in RDS-ON -applying measured values of VT instability (as a function of total GSS cycles with a fixed hysteresis interval time of 100 μs) to the theoretical curves from Fig. 1, as a function of assumed channel mobility.
Power Devices: Operational Reliability and Parameters' Stability