Semantic Autoencoder for Modeling BEOL and MOL Dielectric Lifetime Distributions

Abstract

This paper presents a physics-based machine learning framework for modeling a dielectric lifetime distribution in the presence of manufacturing process variations. It uses a semantic autoencoder that provides insight into the dielectric thickness distribution and parameters of the underlying percolation model. Experiments show that the model is applicable to various types of dielectric films and that including time-zero leakage current as an input improves the model performance. The autoencoder may be configured to model intrinsic break-down or to model breakdown resulting from competing failure mechanisms, e.g. intrinsic and extrinsic.