Predictive models for μBGA and QFN reliability estimation

The estimation of product reliability during the design is one a key question in microelectronics. Lifetime is a function of such parameters as geometry, material properties and loads. These parameters have the tendency to vary from the designed ones. A probabilistic FE-simulation combined with statistical analysis is used in this work to create predictive models for the evaluation of inelastic strain amplitudes. Soldered joints are subjected to thermal-cycling conditions as a function of the different design parameters of muBGA and QFN. The models include local and global models and can be used for the prediction of the reliability of muBGAs and QFNs. In order to obtain statistical information on the effect of input parameters the analysis was carried out with parametric FE-models by multiple repetitions according to a three-factor Box-Behnken-design. Material properties (coefficient of thermal expansion), geometry (component size, substrate and chip pad diameters) and also load parameters (temperature range in thermal cycling) were used as random input variables for the deterministic model. The regression coefficients for evaluation of the influence of the independent variables and the cross-terms for BGAs and QFN were calculated. The developed quadratic regression models allow the reliability prediction with a precision almost comparable with complex finite element models. Thus results can be made available efficiently and this will allow user to understand the influences on the thermal-mechanical reliability. The model can be also used for the evaluation of the relative improvement of component design rather than the prediction of the absolute value of the lifetime.