Run-to-Run Control of an Atomic Layer Etching Process with a Machine Learning-Based Endpoint-Detection Control System

This study introduces a novel control approach for an Al<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>O<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf> Atomic Layer Etching (ALE) process, combining an ex-situ Run-to-Run (R2R) controller and a real-time Endpoint (EP) controller. The method dynamically adjusts process times for both ALE half-cycles to address disturbances and optimize outcomes. The EP controller employs a machine learning-based soft sensor, using a transformer architecture to analyze variable-length pressure profiles from five wafer inspection points and make binary classification decisions. To evaluate the integrated controller performance, the controller is applied to a multiscale process model which integrates Computational Fluid Dynamics (CFD) with kinetic Monte Carlo (kMC) simulations, offering accuracy and low computational cost. Various R2REP combinations, including traditional Exponential Weighted Moving Average and a novel Standard Case Corrector, were tested, showing significant improvements in reducing mispro-cessing rates compared to traditional controllers.