Research On Multi-chip Modules Life Prediction Method In The Condition Of Small Sample

Multi-Chip Modules have have achieved high-efficiency operation of electronic systems and miniaturization of the whole machine,and have received extensive attention.Their complex structure and service environment make their life-span research of great significance.This article will start with the long life and complex structure of Multi-Chip Modules.Its life data has small sub-sample characteristics,study the processing methods of small sample data,draw on relevant research literature,and combine the design process and application environment of Multi-Chip Modules to accelerate the test.The life prediction method of a small amount of failure data and a small amount of time-series state data in the accelerated test conducted by it is studied.The main contents are as follows:Considering the characteristics of long life of Multi-Chip Modules,large packaging density,and slow heat dissipation,a brief description of the design,materials,processes,etc.of Multi-Chip Modules,analysis of the temperature is most likely to cause it failure,and failure of key components or weak links will directly lead to the failure of the entire component,from the perspective of applied stress,briefly describe the accelerated life test such as constant acceleration test and temperature cycle.The test record data is divided into two cases to estimate its life.The first one is a life prediction method based on a small amount of failure data.It requires all test pieces to be tested for all test failures,recording the failure time of each test piece,and failing the obtained small sample.The method named Bayes Bootstrap&k-means to process and analyze and predict its life is proposed;the other is the life prediction method without failure data but with a small amount of state information that changes with time.This method does not require testing until all specimens fail,only for the weak parts of the product,the performance data is measured and recorded during the test.Finally,it is proposed to use the Markov-tail residual gray model to predict its future changes.The data processing methods of the two cases are explained in combination with the case.Using matlab software programming simulation,the prediction accuracy after processing is verified,and the two methods are proved to be effective and applicable.