Statistical Methods For Discriminating Between Tin-Lead And Lead-Free Solder Interconnect Failure Data In High Cycle Fatigue Applied To Multi-Factor Test Vehicle Testing
Authors: Joseph M. Juarez, Ph.D., Craig Hillman, Ph.D., and Nathan Blattau, Ph.D. Company: Honeywell International and DfR Solutions, LLC Date Published: 5/14/2013
ICSR (Soldering and Reliability)
Abstract: Accelerated testing of Test Vehicles subjected to harmonic vibration specifically addresses methods for finding the most significant factors that lead to predictive failure modeling of tin-lead and lead-free solder interconnects. Here testing produced data using factors involving both tin-lead as a baseline and 96.5%Sn 3.0%Ag 0.5%Cu (SAC305); thermal preconditioning; and various vibration levels. Several factors are fixed like room temperature testing; only one full grid array type 1156 I/O BGA is used; and only a single geometry PBA, but the PBA is fully populated with 20 BGAs so location matters on the board for vibration and the BGA is large and susceptible to early failures compared to longer life for smaller components. Data for sixteen (16) Test vehicle runs are analyzed to demonstrate how predictive modeling might use likelihood-ratio (LR) to discriminate between failure times for vibration induced bending strain for tinlead and lead-free solder interconnects given two other factors: preconditioning or aging before vibration and surface chemical finishes. Both these factors have high processing cost implications. Aging can take 120 cycles with temperature excursions between -40C and 85C. The LR test is not enough and therefore the data is subjected to multivariable regression analysis to show that preconditioning may not be important. Two-way ANOVA provides more insight into differences between tin-lead and lead-free for increasing vibration levels, but these levels are discretized into four vibration levels. Further resolution on the data can be accomplished by using S-N curves to take discrete vibration levels and convert the vibration level to a more quantitative variable.