The reliability prediction of electronic packages – an expert systems approach

The exponential growth of the electronics packaging industry has fueled the availability of a variety of area array packages. The reliability of these packages, as characterized by their capacity to withstand the IPC- (formerly Institute of Interconnecting and Packaging Electronic Circuits) prescribed swings in temperature, differentiates one from the other. With design cycles shrinking and competition surging, the capability to make instant package selection decisions by leveraging prior empirical data could pose as a potential alternative for exhaustive experimentation. By employing expert systems techniques, this research developed suitable models that accurately depict field conditions in order to assist in delineating trends in package reliability data. The fatigue behavior of the solder joints subjected to accelerated thermal cycling is often used as an indicator of the reliability of electronic packages in field conditions. Design for reliability (DFR) could be pursued if the thermal fatigue behavior can be predicted in the design phase of a product. The finite element method (FEM) and accelerated testing such as air-to-air thermal cycling (AATC) have been used extensively to study second-level package reliability. Factors like incorrect assumptions or unknown material properties involved in the development of the FEM models are the cause of deviations between actual and predicted values The mathematical complexity and the time needed for model development further aggravate the situation. The focal point of this research was to develop a generic method that could be used to predict the second-level solder joint reliability of area array packages from the analysis of empirical data. While package characteristics play an important role in identifying the similarities between various subsets of packages, the role of assembly parameters is crucial in terms of their impact on reliability. Weights, in terms of the parameters’ impact on reliability, are computed by examining each individual experiment. Based upon identified trends and the separation of qualitative and quantitative impact of the contributing parameters, regression models may be developed to capture the second-level reliability behavior of the package. These models make it possible to predict reliability and potentially save time and resources for an end-user.