Data-driven Modeling Research On Electrochemical Migration Failure Of High-density Printed Circuit Boards Under Multiple Factors

The miniaturization of electronic devices increases the possibility of insulation failure between the wires in high-density printed circuit boards due to electrochemical migration(ECM).The main factors affecting ECM failure of printed circuit boards are ambient temperature,the relative humidity and electric field intensity.In addition,the dust in the air pollution environment enters the electronic products and attaches to the surface of circuit boards,which will have a complex effect on the ECM of circuit boards,exacerbates the ECM failure and reduces the reliability of the system.The life model of printed circuit boards with multi-environmental factors under the influence of dust will give an indication to the design,processing and reliability evaluation of printed circuit boards under the environment of air pollution.In this paper,the ECM life model of printed circuit boards based on the time to failure(TTF)under the combination condition 1 of four environmental parameters(the temperature,the relative humidity,the bias voltage,the concentration of the soluble salt solution in dust)and the combination condition 2 of four environmental parameters(the temperature,the relative humidity,the bias voltage,the distribution density of insoluble particles in dust)were studied.By using the algorithm of multivariate non-linear statistical regression,support vector regression machine(SVR),gradient boosting regression tree(GBRT),random forest regression(RFR)the ECM life model of circuit boards under the combination condition 1 was established.Then by using the algorithm of multivariate non-linear regression,gradient boosting regression tree(GBRT)and random forest regression(RFR)in machine learning,the life model of circuit boards under the combination condition 2 was established on high and low dust distribution density respectively.In this paper,the failure data obtained based on the temperature humidity bias(THB)tests has the characteristics of small sample size,large discreteness and clear characteristics of univariate,so that the applicability of the modeling theory was analyzed and discussed firstly.The results show that the support vector regression(SVR)algorithm in machine learning is generally applicable to small sample data,the gradient boosting regression tree(GBRT)and the random forest regression(RFR)in ensemble learning have the advantages of making full use of data,strong learning ability and strong robustness of abnormal data.Based on the characteristics of failure data,the machine learning method,especially the ensemble learning method,is more suitable for the life model of printed circuit boards based on ECM in small sample size than the statistical data regression method based on failure physics.Through the comparison of the predictive performance of the different modeling methods on the test set,it is shown that,whatever under the combination condition 1 or the combination condition 2,the accuracy of the model based on machine learning method is higher than that of the statistical mathematical regression.For life modeling under the influence of dust soluble salt concentration,the model established by random forest regression(RFR)algorithm has the best prediction performance.For the life modeling under the influence of dust insoluble particles,the model established by gradient boosting regression tree(GBRT)algorithm has the best prediction performance in low-density zone,while the model established by random forest regression(RFR)algorithm has the best prediction performance in high-density zone.The model application effect combining with the algorithm theory analysis further proves the validity and the feasibility of the strategy of establishing the TTF modeling of ECM of circuit boards in complex environment by using machine learning with limited data.