Adaptive Test Method Of Integrate Circuit Based On Mechine Learning

With the rapid development of integrated circuit technology,chip functions are becoming more and more complex,and feature sizes continue to decrease.Correspondingly,the requirements for integrated circuit testing technology are constantly increasing.In order to ensure the quality of testing,both of the cost difficulty of test are increasing.Traditional test methods cannot well solve some of the emerging problems.In order to solve many deficiencies in traditional IC test,adaptive test methods have emerged.This method refers to a type of method that changes test conditions,test flow and test content based on the analysis of chip manufacturing,test data and statistical data to reduce test costs and improve test quality.Along with the brilliant development of machine learning in various fields,the application of machine learning to adaptive test of chips has also received widespread attention.In order to reduce the test time of IC,an adaptive test method based on machine learning is proposed to predict the chip quality,which is based on the current mainstream classification algorithm and the parameter test in wafer test.Aiming at the problem of inefficient screening of effective test items and slow update in traditional test methods,a method of screening test items using the importance of features in the tree model is proposed.Retaining more important test items can ensure the prediction quality of the model reduce the test time.In addition,statistical analysis of the number of defect chips that can be tested for all test items is carried out in the subsequent process in an optimized sequence,so that defect chips can be tested as early as possible.Finally,a random forest was used to establish a model to predict the quality of the chip.The experimental results show that the accuracy of the model can reach 99.2% and the test time can be saved up to 28% under a low level of test escape.Compared with the other two representative adaptive test methods,the proposed method performs better in test time under the premise of guaranteeing test quality.After that,the extreme gradient boosting algorithm was used to further optimize the model effect and it was compared with several classic classification algorithms.Experimental results show that the number of test escape is only1 and the average accuracy can reach 99.8%,which show the effectiveness of the method.