Research On Data-driven Electronic Device Fault Prognostic And Health Management

With the rapid development of new electronic devices and integrated circuit technology,the reliability requirements of new electronic devices in the system are increasing day by day.Fault prediction and health management（PHM）plays an important role in ensuring the reliability and maintainability of electronic devices and systems.For new electronic devices and integrated circuit systems,there is an urgent need for appropriate maintenance schemes and reasonable health management design to estimate the risk of potential fault states in electronic systems and predict the related trend of device performance degradation,so as to ensure the reliability of system use and avoid the occurrence of potential hazards.However,with the development of microelectronics technology and integrated circuit development of microelectronic technology,the failure characteristics of new electronic devices and the system design structure have changed,which has caused great challenges to the development and application of PHM design.On one hand,the changes in monitoring signal characteristics and the increased complexity of functional signals in novel electronic devices pose new demands on degradation prediction models.The microscopic variations in the degradation mechanisms of electronic devices make it difficult to analyze using existing mathematical models.On the other hand,the diversity of electronic circuits in novel devices is growing,and traditional techniques are unable to adapt to the diagnostic needs in different application environments and scenarios.In the view of those problems,taking new electronic devices and related systems as the research object,some key issues on degradation trend prediction and health management evaluation of third-generation power semiconductor devices and new analog circuits are discussed.Firstly,based on the existing device degradation mechanism and degradation feature analysis,design the degradation feature based on physical knowledge,construct dynamic model to describe the degradation process and study effective degradation prediction methods.Then,a closed-loop health management system for testability design and maintenance design is constructed to solve the system failure caused by device performance attenuation,and a test optimization method combining device performance evaluation is realized.Finally,the correlation between each module of the system and the test is analyzed,and the test optimization method based on the topological structure of the electronic system is studied to improve the utilization rate of test information and reduce the cost of system fault location.The main research achievements and innovations of the thesis is as follows.1.Aiming at the problem of voltage instability of new semiconductor devices under repeated short-circuit stress,the fault mechanism of the thermal-electric coupling degradation process caused by carrier migration and the influence of thermodynamic field-electric field is analyzed,and the voltage spectrum characterization quantity reflecting the carrier trap effect and the temperature characterization quantity reflecting the internal temperature change of the device are constructed.The intrinsic relationship between degradation mechanism and degradation process was constructed by temperature and voltage spectrum.The degradation trend prediction method based on physical information is studied by using deep neural network model.The short circuit process of new electronic devices is described and the V_th change is predicted.The experimental results show that the model based on physical information neural network has good prediction ability,more stable prediction performance and higher prediction accuracy than the existing data-driven model,which has practical significance for the study of the degradation process of wide band gap semiconductor devices.2.In view of the transition state characteristics of the threshold voltage degradation process of wide-band gap semiconductor devices,a double-phase adaptive neural network is constructed,and the extreme learning machine is combined with classical activation function and periodic activation function to predict the trend of threshold voltage degradation of devices.The experimental results show that the double-phase adaptive neural network has good prediction accuracy and stability in predicting the degradation trend,and the double-phase adaptive neural network can effectively evaluate the performance remaining life of the device.Therefore,the double-phase adaptive neural network is of practical significance to provide an efficient new method for predicting device degradation trend.3.In view of the potential impact of device performance attenuation on the reliability and safety of the electronic system,a closed-loop PHM design was constructed in combination with the application requirements of testability design and reliability design,and the generation method of fault diagnosis tree of soft sensor nodes was studied to improve the flexibility of diagnosis positioning and test efficiency.The proposed method has lower false alarm rate and higher accuracy than the existing methods.4.Aiming at the problem of correlation among test points and test information of electronic system,a sequential matrix was designed to describe the correlation of test nodes in combination with the topology of electronic system,a heuristic evaluation method for potential test schemes was designed,and a heuristic programming optimization method was constructed to generate a hybrid test optimization strategy.Simulation data and actual circuits are used to verify that the test analysis theory and algorithm model proposed in this thesis can effectively improve the efficiency of circuit system fault identification.The work in this thesis provides a theoretical and technical basis for the degradation trend prediction and health management of new electronic devices and systems,and has a certain reference value for promoting the application of PHM technology in new electronic devices and systems.