Modeling Research On Electrochemical Migration Failure Of PCB With The High Density Caused By The Soluble Salts Of Dust Contamination

With the integration of electronic products and progress in electronic industrial technology,the assembly density of PCB is greatly improved,the spacing between wires is becoming smaller.At the same time,the working environment of electronic products is becoming more and more diversified.High temperature,high humidity,and dust contamination brings huge risk for the electrochemical migration of PCB with high density of circuits.The soluble salts in the dust dissolve in the water film adsorbed on the surface of PCB,which changes the ionic concentration of the water film,and increases the complexity of the electrochemical migration failure of PCB.Therefore,based on the electrochemical migration failure mechanism of PCB with high density of circuits under the dust pollution of soluble salts,it is very necessary for the research to build up the relation model of the environmental factors of temperature,humidity,voltage,ion concentration in dust contamintation and electrochemical migration failure time.This topic mainly studies the modeling of the relationship between the electrochemical migration failure time and the multi-environmental factors,which is divided into two parts.The first part is to establish the model of the relationship between the failure time of electrochemical migration and the temperature,humidity and voltage.Firstly,the data of failure time obtained by electrochemical migration accelerated experiment.Secondly,using the failure of physical modeling,factors combination modeling,machine learning methods modeling to establish the relationship between the failure time and influence factors.Support vector machine regression,random forest regression and gradient boosting regression tree algorithm in machine learning were used.Finally,models acquired by different methods were evaluated on the test data set,models established by machine learning have the optimal prediction performance,the model established by the factors combination is the second,the model established by physics of failure is the worst.The random forest regression model has the the optimal performance in machine learning modeling.In the second part,the relationship between the time of electrochemical migration and temperature,humidity,voltage and the ion concentration of soluble salt in the dust contamination is established.Firstly,the data of failure time obtained by electrochemical migration accelerated experiment.Secondly,models were built by support vector regression,random forest regression and regression tree gradient boosting in machine learning algorithm.In addition,this paper used the data augmentation to expand the training data set,and to establish the failure model using machine learning algorithm baesd on the extended train data set.Finally,models established by different methods were evaluated in the test set,random forest regression algorithm model has the optimal predict performance,and support vector machine regression model is the second,and gradient boosting tree regression is the worst.Support vector machine regression model has the remarkable prediction performance improvement after using data argumentation.The performance improvement of random forest regression and gradient boosting tree regression is not obvious.This paper obtains the following conclusions after comparing different modeling methods on the test data set.Firstly,the prediction performance of models gained by traditional modeling methods are lower than the model gained by machine learning method.Secondly,the prediction performance of models have some improvement after using data argumentation.Through this topic,the conclusion can be drawn that the machine learning modeling method can be applied in the research field of failure modeling,data argumentation of failure data can improve the prediction performance of model based on machine learning modeling.