Research On Prognostics And Health Management Of Semiconductor Device Under Temperature Environmental Stress

With the vigorous development of information technology in China,electronic equipment is widely used in military,communications,automation,aviation and other fields.As the basic unit of electronic equipment and an important component of equipment system,the stability,reliability and health status of semiconductor devices directly affect the performance of electronic equipment.It is increasingly urgent to evaluate and ensure the health status of semiconductor devices.At present,the early failure rate and accidental failure rate of semiconductor devices caused by the design and manufacturing process are close to zero.Various environmental stresses such as temperature,electromagnetic,environmental radiation,mechanical vibration and other environmental stresses that the devices are subjected to during storage,transportation and use have become the main external cause of semiconductor device failure and health.Temperature environmental stress is the environmental stress that semiconductor devices are most frequently impacted by long-term continuous changes in the use process,and it is also one of the main factors leading to device performance degradation.The traditional evaluation and identification of semiconductor devices based on the reliability manual is based on the binary judgment of“normal” and “failure”.It only considers the functional failure and structural failure of the device,and does not consider the performance degradation caused by the parameter deviation or fluctuation.A single binary criterion can no longer meet the current equipment development requirements for systematicness,the stability and reliability of semiconductor devices.Therefore,it is increasingly urgent to carry out the research on fault prediction analysis,lifecycle management and health management of semiconductor devices in real time,and establish the relationship between multiple environmental factors and health status in the product life cycle.To solve the above problems,this paper takes typical semiconductor devices metal oxide semiconductor field-effect transistor and insulated gate bipolar transistor as the research object,and based on the performance degradation data,studies the state prediction,health management and remaining useful life of semiconductor devices.By analyzing the influence of temperature environmental stress on the electrical parameters of semiconductor devices,the logical relationship between temperature environmental stress,electrical parameters of semiconductor devices and the state of semiconductor devices is constructed,the coupling relationship between temperature rise and electrical characteristics of semiconductor devices is established,and the health state criterion is established to predict the working state and health state of semiconductor devices.The main research contents of this paper are as follows:(1)According to the reliability manual,the state of electronic components is evaluated,only the functional failure and structural failure of the components are considered,and the performance degradation caused by parameter out of tolerance is not considered.To solve this problem,the research method of fault prediction and condition monitoring of semiconductor devices based on device performance degradation data parameters is proposed by taking the temperature environment stress and the cumulative action time of stress together as the monitoring variables to characterize the device degradation state.The characteristics of devices changing with time series are revealed.A fault prediction model of semiconductor devices based on long-short term memory algorithm and a cumulative monitoring model based on kernel principal component analysis gated cycle unit state are designed.The state feature information contained in the degradation parameter data is effectively mined,and the internal relationship of these information in the cumulative action time is established to represent the health state of devices.The effectiveness of this method is verified by an example of a power device sensitive to temperature and environmental stress.(2)To solve the problem that the degradation state of semiconductor devices cannot be directly observed,a method to construct a health recognition classifier is proposed.The autoregressive model is used to extract fault eigenvectors.According to the change trend of fault eigenvectors during the degradation process of semiconductor devices,the device performance state is quantized into three levels: healthy normal,unsteady state and obstacle state.The mapping relationship between fault eigenvectors and degradation levels is established using the strong state description characteristics and self-learning ability of discrete hidden Markov model,according to the recognition results and combined with the fault threshold,the device health status recognition,false alarm elimination and fault time prediction are completed.The experimental results show that the recognition rate of health recognition classifier is 80%,and the prediction error of fault time is 2.3%.It shows that this method can effectively identify the health status of experimental devices.(3)Aiming at the problem of lack of fault data of semiconductor devices and limited application scenarios of existing models,a method for predicting the remaining useful life of semiconductor devices based on migration learning and multiple regression is proposed,which expands the use scenarios of the models built.Build intermediate state variables and multidimensional time domain array,and carry out data analysis and judgment.The maximum mean difference method is used to judge the suitability of the source domain and target domain,and the method combining parameter migration and fine tuning is used to complete the migration of the model from the source domain to the target domain,so as to realize the state monitoring of the sample target domain.On this basis,combined with multiple regression analysis,the dummy variables are introduced to predict the remaining service life of semiconductor devices.The experimental results show that this method can improve the prediction accuracy of the model,improve the training efficiency of the model,and more accurately predict the remaining life of the device.This paper provides new ideas and solutions for fault prediction,status monitoring and remaining useful life prediction of semiconductor devices.It is applicable to the health status analysis,assessment,prediction and health management of semiconductor devices with performance degradation caused by temperature environmental stress,and has significant value and engineering promotion and application.