Surrogate Model Optimizer for Purge Flow Uniformity

The patent applies parameter changes by dynamically adjusting pressure and flow rate parameters of non-concentric purge regions based on machine learning model predictions. The system modifies purge flow rates and pressures to achieve both effective station isolation and improved azimuthal/thickness uniformity, resolving the contradiction between separation reliability and manufacturing precision.

The patent implements feedback control through machine learning models that predict the impact of purge parameters on deposition uniformity. The system uses simulated and experimental data to create surrogate models that provide real-time feedback on optimal purge settings, enabling continuous adjustment to maintain both separation effectiveness and uniformity requirements.

The patent applies universality by creating a unified control system that manages both concentric and non-concentric purge regions through a single machine learning-based platform. The surrogate models handle multiple purge region types simultaneously, enabling coordinated control of both deposition process parameters and thermal uniformity across the chamber.

The patent introduces machine learning surrogate models as intermediaries between conventional process recipes and purge region control. These models translate process requirements into optimized purge parameters for both concentric and non-concentric regions, mediating the control relationship to achieve improved thermal uniformity while maintaining deposition process precision.

The patent applies preliminary action by pre-computing simulation data across a range of operating conditions before actual processing. Design of experiments is performed in advance to generate training datasets, and machine learning surrogate models are built beforehand to replace time-consuming detailed physics simulations during real-time operation, significantly reducing computational time while maintaining prediction accuracy.

The patent creates simplified copies of detailed physics models through machine learning surrogate models. These surrogate models replicate the behavior of complex physics-based simulations but execute much faster, enabling real-time optimization and control decisions without the computational burden of full physics model execution.