Development of a Surrogate ANN to Bypass Multiphysics FE Simulation

Prediction of deformation in silicon-based devices is a pivotal concern, given the widespread application of the material at various loading conditions and the non-linear temperature dependent behavior of the material. Existing ways to calculate silicon deformation include numerical and experimental methods. Experimental methods such as X-ray diffraction, optical interferometry, and Raman spectroscopy provide direct observations of the crystallographic structure under various conditions. Numerical approaches include finite element analysis and molecular dynamics simulations to give the predictions. These computational methods allow researchers and engineers to model complex deformation scenarios and predict corresponding behavior under various conditions. Although experimental approaches give accurate observations obtained directly from the laboratory, numerical methods help to replicate the conditions that may be difficult to analyze or perform in a laboratory setting. However, these methods can be computationally intensive and may not always capture the complex, non-linear behavior of silicon under various loading conditions. Recent advancements in machine learning have created avenues for predicting silicon deformation with greater efficiency and accuracy. This paper proposes a comprehensive approach to finding the optimized deformation of the silicon material when subjected to electrical and thermal stresses with the help of surrogate artificial neural network (ANN). The developed model outmanoeuvres the finite element analysis and predicts the corresponding output values.