Research On Key Technologies Of Reliability Screening And Life Prediction Of Semiconductor Power Device

In the aerospace system,due to the particularity of the using occasion and environment,the key devices are subjected to harsh environmental conditions such as high-strength mechanical stress,thermal stress,radiation stress,and electrical stress.Key devices will be continuously worn and aged until they fail or break down,which will eventually lead to the failure of the system.Scientific design methods and the selection of high-reliability devices can improve the inherent reliability of the system;the status monitoring and life prediction of key devices to formulate maintenance and replacement strategies can improve the application reliability of the system.In order to improve the long-term reliability of various electronic systems,the dissertation studies the key technologies of reliability screening and lifetime prediction for semiconductor power devices.The main research contents are as follows:1.Aiming at the chip parameters with locational information,this dissertation proposes a screening method that uses the locational correlation analysis of chip’s parameters to reduce test escape chips.And it establishes a quantitative model of locational correlation analysis through key algorithms that solve the problems of the deviation quantification of chip parameter,the normalization of local area parameter fluctuation and the setting of screening limit.The SiC JBS test data is used to verify the model,and the results show that the method can screen out the test escape chips;the comparison between the method and the vehicle-level AEC Q101 parameter consistency evaluation method shows that the two algorithms can screen part of the same and part of different test escape chips.When the locational information of the chip on the wafer is known,the screening effect of this method is significantly better than the method specified by AEC Q101.It can be used as a supplement to the existing measurement technology for reducing the cost of packaging,testing,and aging tests.2.Aiming at the multi-parameter of the chip,this dissertation proposes a screening method that uses the structural correlation analysis of chip’s multi-parameter to reduce the test escape chips.It establishes a quantitative model of structural correlation analysis through key algorithms that solve the problems of the characterization of outliers by multi-dimensional covariance distance,the robust extraction of covariance,and the setting of screening limit.Two-dimensional normal distribution simulation data and SiC JBS test data are uesd to verify the model,the results show that the method can further identify the remaining test escape chips in single-parameter screening.Si MOSFET test data are used to verify the model,the results show that the locational correlation and structural correlation can be combined to screen the test escape chips.Without additional physical measurement,the use of the structural correlation of the chip’s multi-parameters can further identify the remaining test escape chips in single-parameter screening.3.Aiming at the problem of characterizing the health status of multi-parameter devices,this dissertation proposes a method of effective information extraction and fusion.It establishes a multi-parameter state characterization model through key algorithms that solve the problems of the robust extraction of fitting coefficients,the extraction of effective information by dimensionality reduction,the fusion of information,and the conversion of data.The online measurement data of the IGBT power cycle experiment and simulation data are used to verify the model,the results show that the correlation between the comprehensive index and the cycle period is greater than the single parameter;the correlation between the comprehensive index and each single parameter is greater than the correlation between them;the degradation trend of the comprehensive index is better monotonic and less random disturbance.The result of life prediction shows that the prediction error of the comprehensive index is smaller than that of a single parameter,and the prediction result is more accurate and stable.The comprehensive index obtained by using effective information extraction and fusion can reduce the interference of random disturbance,and more accurately and stably characterize the comprehensive health status of the device.4.Aiming at the interference of measurement error and dynamic environment in life prediction,this dissertation proposes a fusion method of least squares support vector machine and particle filter with adaptive adjustment capability based on time weighting.It establishes a life prediction model through key algorithms that solve the problems of the extraction of linear and non-linear trends,the adaptive adjustment capability based on time weighting,the correction of prediction results,and the provision of confidence limits.The IGBT test data are used to verify the model,the results show that the state prediction error and life prediction error of the fusion model are significantly reduced and the confidence limit can characterize the credibility of the prediction result.The MOSFET test data are used to verify the model,the results show that the time-weighted fusion model has adaptive adjustment capabilities and can solve the system deviation caused by the system dynamic environment.The fusion model proposed in this dissertation can not only reduce the prediction error and provide confidence limit to characterize the credibility of the predicted results,but also has the adaptive adjustment ability to reduce the system deviation.