Neural Network Assisted Speed Up of High Fidelity Warpage Simulations towards Design for Reliability in Ultra-Thin Packages

Thin electronic packages are usually achieved by thinning the die and swapping the standard core substrates to thin core and coreless variations. The decrease in thickness reduces the flexural rigidity of the package, resulting in a warpage challenge that must be addressed in the design phase for a high yield assembly process and good reliability. Finite element analysis (FEA) is a useful tool to predict warpage during the design process. However, computational time can be too long, especially when the models are large and the parametric studies are extensive. In this work, the FEA model is replaced with a neural network (NN) to speed up the analysis time. Results show that a single layer NN can replace the high fidelity FEA models for package warpage predictions with good accuracy and significant speed up of more than three orders of magnitude. In the future, the NN will be combined with Markov Chain Monte Carlo (MCMC) analysis to learn the material properties of the substrate given any observed warpage profile from digital image correlation studies.