Run-to-Run control of Area-Selective Atomic Layer Deposition of SiO2 on Al2O3 and SiO2 surfaces with intermediate etching steps

Area-Selective Atomic Layer Deposition (ASALD) has emerged as a critical bottom-up patterning technique for sub-5 nm semiconductor manufacturing, offering a path to circumvent the alignment challenges of traditional top-down lithography, and prior work involved intermediate Atomic Layer Etching (ALE) to build ASALD-ALE pipeflow to retain selectivity under imperfect inhibition. However, the sensitivity of ASALD-ALE cycles to kinetic disturbances necessitates the development of advanced process control to sustain selectivity. In this work, a Run-to-Run (R2R) control framework based on the Exponentially Weighted Moving Average (EWMA) R2R control algorithm was developed for an integrated multiscale ASALD process. The framework was evaluated against a simulated 20% kinetic disturbance in adsorption rates using a multiscale model that couples microscopic surface kinetics with reactor-scale pressure dynamics. Furthermore, a severe 40% kinetic degradation was simulated to validate the robustness of non-rate-limiting steps (oxidation and fluorination), confirming they achieve full saturation within baseline times and justifying their exclusion from the active control loop. A primary innovation of this study is the implementation of a dynamic slope ( α ) updating scheme—utilizing both pure EWMA and a linear regression hybrid model—to re-characterize the process gain of the adsorption system as the system shifts into a slower kinetic regime. This adaptive approach ensures the controller accurately mirrors the exponential nature of surface reactions, significantly reducing estimation errors compared to static-slope models. Furthermore, a novel sequential protocol was established to pass refined model parameters ( α , β ) and updated reaction times between interdependent steps. Simulations demonstrate that the connected control architecture effectively mitigates the staircase-like profile caused by discrete time resolutions and eliminates the transient recalibration periods typically observed in isolated control loops. The results show that the integrated sequence requires only 10 cumulative batches to reach target coverages which is 100% saturation for adsorption and 0% for etching, while maintaining high process selectivity. This study provides a robust, high-efficiency digital twin and control methodology for optimizing commercial ASALD operations under realistic manufacturing variability.