Physics-based AI models for optimization of a multistep, multistate RF pulsed contact hole etch process

Traditional trial-and-error approaches for optimizing advanced etch recipes are becoming impractical as the number of recipe variables vastly increases and process nonlinearities make it near impossible to rely solely on intuition. We address this challenge by demonstrating the use of a physics-informed model, coupled with Artificial Intelligence (AI), to optimize a contact hole etch that employs three-state RF pulsing across multiple processing steps. This approach successfully found optimal process settings for 21 recipe variables that satisfied 9 on-wafer dimensional targets for a three-dimensional structure. The model was then used to predict new optimal process conditions for a modified incoming stack geometry and additional processing steps, using a minimal number of experiments.