Humidity Robustness of 3.3kV SiC-MOSFETs for Traction Applications - Compared to Standard Silicon IGBTs in Identical Packages

Silicon carbide (SiC) MOSFETs are gaining more and more market share in typical silicon (Si) IGBT applications such as traction or renewable energies. Especially in reliability sensitive traction applications, medium voltage IGBT-modules (3.3 kV-6.5 kV) are widely used and introducing SiC-MOSFETs to such industries is the next self-evident step already on the way. While their superior electrical performance has been generally accepted already (e.g. [1]), SiC-modules have not yet established a track record of high reliability in this voltage class. For this study, 3.3kV SiC-MOSFET-switches were compared to standard Si-IGBTs regarding their humidity robustness under high voltage bias. Both chip types had been assembled in the same traction rated packages to exclude this influence. The Si-IGBTs resembled the well-known industry standard performance, while the SiC-MOSFETs show no degradation within the reported test time of 2000 h. Given the fact [2], that the latest Si-IGBT generation offers a much better humidity performance as well, the standardised HV-H³TRB is no longer sufficient to provoke failures within a reasonable testing time. On the one hand, this suggests that humidity driven failures will not be an issue under field conditions anymore. On the other hand, even harsher tests are required to investigate differences in humidity performance.

• IEC60749-5 [3], general standard • ECPE guideline PSRRA01 [4], traction applications • ECPE guideline AQG 324 [5], automotive applications All of the said documents refer to the well-established HV-H³TRB (High Voltage, High Humidity, High Temperature, Reverse Bias) test. While the climate is constant at 85 °C and 85 % relative humidity, the electrical stress differs between the standards. For example, [4] defines the electrical stress based on the DC-link voltage of traction applications, while in [5] 80 % of the device's nominal voltage (Vnom) is recommended. In each case of the mentioned standards, the bias voltage is limited by the power loss and the resulting temperature increase at the chip interface. As described in [6], the bias voltage is an important acceleration factor, resulting in a faster degradation of silicon devices. However, well designed, latest Si-IGBT-modules proved to be highly robust against humidity and show little to no degradation in accelerated humidity testing anymore [2]. While the failure mechanisms and the corresponding degradation is well documented for standard Si-IGBTs, the situation for SiC-MOSFETs is completely different. SiC enables higher voltages even for the discrete market, leading to a variety of EMC-packaged SiC-MOSFETs with blocking capabilities up to 1.7 kV. Therefore, humidity robustness assessments are often carried out on discrete devices [7,8,9] showing an overall high robustness. However, the performance under high humidity is strongly influenced by the packaging materials and especially for high power modules, silicone gel insulated housings need to be considered. In contrast to Si-IGBT-modules, SiC-MOSFET-modules tend to be much more reliable when being exposed to constant high humid climatic conditions. Studies [10,11] show an extended lifetime under 80 % Vnom for 1.2 kV SiC-MOSFETs in a constant climate of 85 °C and 85 % rel.h. of more than 3,000 h. Referring to [10], a lifetime of up to 10,000 h is possible for SiC-devices, while conventional Si-IGBTs in the same package type failed before 2,500 h of accumulated test time. However, no benchmark of the humidity induced degradation of 3.3 kV SiC-MOSFETs has been published and the direct comparison to state-of-the-art Si-IGBTs with their known field performance shown in this work is the best method to determine the usability of 3.3 kV SiC-MOSFETs for traction applications.

Devices under Test (DUT)
Power modules in the voltage class of 3.3 kV are particularly interesting for traction applications and need to fulfil high reliability specifications due to the long service life of 20-40 years. At the moment, the low availability and the high cost prohibit fully populated SiC-modules as DUTs for reliability test and therefore, an engineering, single-switch/-chip-package was used to compare the two different device types. Fig. 1 (A) shows the module, developed and manufactured by Mitsubishi Electric Corp., which is specially designed for reliability testing of different chip concepts. Furthermore, this package type is manufactured with traction rated materials and produced on the same line like the standard modules. With the same packaging design and the same materials used, a comparison between Si-IGBT-and SiC-MOSFET-chips is more meaningful. The IGBT-chip is always operating with an external free-wheeling Si-diode, leading to a second chip per module, while the SiC-MOSFET can be operated individually due to its body diode. These configurations imply, that the actual comparison will include the minimum usable number of chips per switch of each technology.

Functional Materials and Materials Reliability
Since silicon IGBTs exhibit a higher leakage level compared to SiC-MOSFETs, different current resolutions have to be utilised for the respective device types. Therefore, a highly sensitive, 32 channel, leakage logging system is used to monitor the current under high voltage during the stress test. The same system ensures a fast turn-off in case of a failure, protecting the devices from excessive damage. If an over-current event is triggered, the voltage source is disconnected by an external halfbridge and ramped up again after opening a high voltage relay at the corresponding channel. After each test cycle, all devices are measured individually with a high voltage curve tracer up to ~110 % Vnom, although, the degradation can be identified long before the blocking capability reaches the nominal voltage of the device. The sensitive parameter for this type of measurement is the bendoff-voltage (premature avalanche), which is the first indication of the avalanche inception. Stopping the measurement at this voltage level avoids damaging the devices and improves the degradation monitoring. Fig. 1 (B) shows the blocking curves of the complete test group with differences in the leakage level as well as the shape of the curves. The scattering between the DUTs of each test group is in an acceptable range and just one SiC-MOSFET shows an elevated leakage current above 2 kV. The complete test group contains 18 devices, 10 SiC-MOSFET-and 8 Si-IGBT-modules. Two devices each are used as references to distinguish between voltage and climate related degradation (chamber reference, placed in the chamber without bias voltage), as well as calibrating the intermediate blocking curve measurements (shelf reference, stored at room climate also without bias).
The test system can provide two different voltage levels at the same time. A first test voltage is chosen with respect to the application relevance [4] at a level of 65 % Vnom, which offers a realistic electrical stress for the modules. To achieve degradation in an acceptable time frame, a higher acceleration factor is mandatory and therefore, a higher voltage level with 90 % Vnom is used as well (tab. 1). For Si-IGBTs, the acceleration factor between 65 % and 90 % of the nominal voltage is expected to be 2.1 [6] and this might enable an early degradation of the SiC-MOSFET-modules, too. However, due to the engineering status of the SiC-MOSFET-modules, a higher bias voltage might be an over-stress, resulting in early failures.   . 3). Within the complete test time, all IGBT-modules were turned off one by one due to a high leakage current during testing and due to a low blocking capability in the intermediate measurements (limit at 10 µA). The first failure occurred after 551 h and the last IGBT-module was removed after an accumulated test time of 1,661 h due to over current events at the voltage ramp up. All silicon devices showed the typical behaviour, already reported in [2,6] with an oscillation in the leakage current monitoring combined with a reduction of the bend-off-voltage in the intermediate measurements. Every Si-IGBTmodule was shut down during the test cycle when reaching the current limit of 450 µA. The split, connected to 65 % Vnom showed a significantly longer lifetime compared to the 90 % split. An example of the blocking curve evolution is shown in fig. 2 with the corresponding leakage monitoring in fig. 3. In contrast, the SiC-MOSFET-modules show no sign of degradation at the blocking curve measurements and no differences can be seen with respect to the different voltage levels ( fig. 3)

Functional Materials and Materials Reliability
While none of the SiC-MOSFETs showed any degradation, the bend-off-voltage of the IGBTs is significantly decreased after stressing these devices. All IGBT-modules contain a diode-and an IGBT-chip and therefore, the deterioration of both chips might be different. In this case, the diodes of two IGBT-modules were disconnected to verify the remaining blocking capability of the IGBTchip. In fig. 4, both devices were disconnected from high voltage at the same time and have a comparable state of degradation. The leakage current of the IGBT-chips is more or less following the trend of the full module curve and is therefore, the main cause for the shifted bend-off-voltage. With these highly degraded chips, the comparison to the SiC-MOSFETs is proved meaningful due to the negligible influence of the anti-parallel diode.  Overall, the static testing under high humidity and voltage is not sufficient to trigger any degradation of the SiC-MOSFETs, used in this experiment, at least not in a reasonable testing time of 2,000 h. This is in line with the previously published data sets and therefore, standard testing needs to be improved for SiC-devices. Some manufactures have already published extended testing parameters [13], but as presented in this study, a test time of 2,000 h under constant conditions will not lead to significant degradation. But, there are some approaches for application relevant testing such as dynamic climate testing (e.g. AC-HTC, [12]) or thermo-mechanical-preconditioning [14] to provoke delamination and degradation of package materials. Within the rest of the RECET4Rail project, the constant climate and voltage testing will be continued and extended to more application relevant test conditions.