The Design & Verification Of Monitoring Evolutionary Algorithm In Device Production Testing

It is unavoidable to bring in some errors and parameter drift during semiconductor chip fabrication and packaging process.At the same time,device testing environment also introduces some uncertain factors under automatic production process.These disturbances can affect test result,even test quality.These effects are particularly prominent in DC parameter testing section.In order to detect these disturbances and improve test quality,a real-time monitoring method that can be used in device mass production is needed.This paper,based on the actual production testing environment,will analyze the DC parameter testing from the system level,list and research various factors that affect the testing accuracy.Then,through the analysis of the fluctuation of test results,the theories of evolutionary strategy(ES)and quality control are imported as the means to implement the monitoring algorithm.Next,this paper combined with the characteristic of device production testing,optimized the process of standard ES,including the establishment,recombination and evaluation of the initial population,the fitness function is established that process capability index(CPK)as the parameter,the dynamic test lower and upper limits generating mechanism,and so on.In the end,the monitoring evolutionary algorithm in testing(MEAT),that can be running in automatic test equipment(ATE)to monitor production testing in real time,is realized in this paper.Through the software test and actual production verification,the stability and sensitivity of MEAT can be guaranteed.Compared with the traditional monitoring methods,MEAT innovatively introduces the framework of ES and sets CPK as constraint condition.That makes the establishment process of dynamic test limits is self-adaptive,and can be controlled by CPK.Through the simulation test of abnormal batch test data in actual production,the ratio of screening to abnormal data is 97.4%,then the defect parts per million products(DPPM)is reduced by at least one order of magnitude.