Virtual lifetime estimation of microelectronic devices has become essential, due to
the tightening requirements and the trend of electronification. Thermal cycling load
cause fatigue crack in solder joints and therefore functional failures. The application
of the fatigue models viable in the literature is circumstantial for industrial scale
problems. The exploitation of the already existing simulation data in data-driven
models is a promising solution for speeding up the virtual lifetime estimation process.
This paper benchmarks data-driven methods to predict the lifetime of solder joints
with different geometric properties using a dataset originated from finite element
simulation results. This case study shows that the nonlinear function fitting and
the neural network are applicable.
