Research On Reliability Test And Life Models Of GaN-on-Silicon LED

In recent years,semiconductor lighting has become more and more widely used in industrial and daily life,and has become a hot topic of academic research.The substrate materials for GaN-based LED are mainly sapphire,SiC and Si.GaN-on-Silicon LED has the advantages of lower cost,better conductivity and higher thermal conductivity than the other two types,but there is less research on the GaN-on-Silicon LED.In addition,the reliability of GaN-on-Silicon LED is related to the improvement of the product and industrialization,and LED's life is the ultimate manifestation of the reliability.Therefore,the research on the reliability test and the life model of GaN-on-Silicon LED is particularly important.In view of the above reasons,this thesis has conducted the following research work:1.This thesis set up an automatic test system of the LED's reliability to combine the LED's test of the photometric,colorimetric,electrical and thermal characteristics.Then,we developed PC software to achieve the automatic control of LED's test.In addition,in the aspect of controlling LED's temperature,a multiple isolated Modbus protocol converter was designed to allow multiple thermostats to be connected to Ethernet for off-site control.This converter improved the efficiency of the test.This test system provided data support for reliability test and life prediction of GaN-on-Silicon LED.2.This thesis was based on the LM-80-08 test data of GaN-on-Silicon LED,and presented a new life prediction model based on BP neural network optimized by genetic algorithm.The parameters were obtained by LM-80-08 reports served as the inputs,such as electric current,junction temperature,initial luminous flux and initial chromaticity coordinates,while the life under this stress served as the output.Then,the life of the LED was predicted under any current and junction temperature.The traditional method TM-21-11 can be used to predict the life of the LED under 2 or 3 stresses based on the test data of the LM-80-08 reports.And this test requires at least 6000h.The results show that this GA-BP model is more flexible than the TM-21-11.Compared with the traditional BP neural network,the prediction error of the GA-BP model is reduced by 65.5%,the average relative error is 1.47%,and the two parameters are better than the Adaboost model of 54%and 3.16%.The correlation coefficient of training samples reaches 99.4%.The GA-BP model has better accuracy and is more adaptable,which has practical significance in the area of GaN-on-Silicon LED's life prediction.In addition,we analyzed the weight of each input parameter.The results show that the electric current and junction temperature play a vital role on the GaN-on-Silicon LED's life and provide the guidance for the improvement of the GaN-on-Silicon LED.3.This thesis used self-built LED reliability test system to carry out high-current and high-temperature accelerated aging of the LED,and analyzed the effect of different accelerated stress on the LED's life.We built a model based on the LED's photometric,electrical and thermal characteristics.We presented a life prediction model based on BP neural network optimized by genetic algorithm and the simulated annealing algorithm.The parameters served as the inputs,such as electric current,the ambient temperature and the LED's working time,while the lumen maintenance rate under this stress served as the output.Then the lumen maintenance rate of the LED was predicted under any electric current,ambient temperature and the LED's working time.In addition,we used the simulated annealing algorithm to solve the problem,which was to calculate the LED's life when the input of the network was unknown.The results show that the relative error of the LED's working life calculated by the model is 4.47%.The relative errors of the predicted life of the Erying model for single stress and the inverse power law model are 29.3%and 20.7%respectively.It is verified that the error of the new model is smaller,and the new model contains both ambient temperature and the electric current.The application scope of the new model is more extensive.