With regard to reliability of electronic systems, lifetime prediction on board and system level is key to understand occurring failure modes and, consequently, to optimize the system design. Simulation using nominal design values, however, represents the behaviour of real parts under reliability testing conditions only to a limited extend.Using the example of a Machine-to-Everything (M2X) communication module comprising a coupon for hot oil testing, life-time of copper $\mu -$vias under harsh thermal loading is predicted by finite element simulation based on a stochastic design variation approach. When setting up the model of the coupon, product and process variations like thickness tolerances of printed circuit board base materials or via shifts caused by drilling accuracy are taken into account; by doing so, hundredths to thousands of representative models can be acquired. The accumulated plastic strain during thermal cycling is then evaluated based on the Coffin-Manson approach for reliability assessment.Comparison of simulated to measured life-time of the coupons during hot oil testing showed a significant improvement of prediction accuracy when taking not only nominal values but design variations into account. Not only the average life-time of the coupons can be predicted more precisely, but especially the critical early failures caused by variations close to the respective specification limits can be replicated by simulation. These results clearly demonstrate a major advancement in reliability prediction when using a product and process tolerance-dependent simulation approach.