Gate health state assessment of SiC MOSFET based on clustering algorithm and 2D-cloud modeling

Purpose The rapid deployment of silicon carbide SiC MOSFETs in power electronic systems has raised growing concerns regarding their safety and reliability, particularly in terms of gate reliability. This study aims to achieve an accurate and robust assessment of the health state of SiC MOSFETs during operation by focusing on the gate degradation process. Design/methodology/approach A health state evaluation method based on a clustering algorithm and a two-dimensional cloud model is proposed. The method first applies z-score normalization and nuisance attribute projection to preprocess the raw data, eliminating the influence of temperature on the characteristic parameters. Subsequently, a clustering algorithm is used to automatically partition health states. On this basis, an evaluation framework is established based on a two-dimensional cloud model, which fully accounts for the uncertainty in the evaluation process and the fuzziness of the boundaries between health states. Finally, accelerated aging experiments are conducted on fresh devices to validate the proposed approach. Findings The proposed method enables quantitative characterization of intra-level differences within each health state, thereby providing more detailed health information than conventional state assessment methods. Originality/value The proposed method provides a novel and effective framework for health evaluation, enabling quantitative characterization of intra-state differences within each health state. As a result, it offers more detailed health information than conventional state assessment approaches and provides a scientific reference for the operation, maintenance and health management of SiC MOSFETs and other power devices.