Methods for resolving the challenges of degradation diagnosis for SiC power MOSFET

Abstract

Modern society is in quest of more electric power in various sectors. The rising power demand has driven the development trend of power electronic converters toward higher power density. Power electronic converters, however, are subject to continuous electrical and thermal strains that threaten their reliability. Any unexpected failure can incur unnecessary economic loss or even worse, safety hazards. Semiconductor power devices are the backbones in a power electronic converter, and yet, they are recognized as one of the most failure-prone components. Therefore, early detection of a power device’s incipient abnormality due to either aging, electrical stress or defect is crucial. Degradation diagnosis is a process of identifying an evolving damage before it develops into destructive failure. A variety of degradation diagnosis techniques have been proposed in earlier works. Among them, the time-domain on-state characteristics and thermal path integrity are the most frequently chosen properties as indications of device wear-out. As the measurable electrical variations associated with degradation are typically subtle, the measurements are susceptible to system’s noise and loading conditions. Besides, as power applications are pursuing higher switching frequency to gain power density, the measurement system is demanding for higher bandwidth, which could reach its data rate bottleneck. For reliable and accurate prediction of developing faults, these challenges must be addressed. In this thesis, silicon carbide (SiC) MOSFET is chosen as the device for degradation diagnosis study, as it has shown great promise for high power density converter (HPDC) design owing to the excellent material property for high-frequency and high-temperature operation. The practical reliability issues responsible for power device degradation and challenges in degradation diagnosis are investigated. Special solutions are proposed to counter these challenges. Firstly, a voltage monitoring circuit with built-in isolation is proposed and experimentally verified up to 100 kHz. The embedded isolation eliminates EMI influence on the measured data, allows direct processing of the measured voltage information, and relaxes the data rate bottleneck of the data transmission interface of a measurement system. Secondly, a non-intrusive measurement method that measures the on-state impedance of a power device in frequency domain for offline degradation diagnosis is proposed. The method overcomes the shortcomings of the conventional time-domain methods, which are easily affected by system’s noise and may have difficulties on detecting subtle electrical changes associated with reliability issues. Finally, a temperature sensor-less method is proposed to estimate power MOSFET’s junction temperature. It has experimentally demonstrated that by means of device saturation current (Id,sat) measurement, junction temperature can be estimated either by Id,sat, √Id,sat, or the extracted threshold voltage.