Gas Transport Optimization for Atomic Layer Deposition Processes (ALD)

Abstract

[ES] El proyecto se centra en la optimización del flujo en procesos ALD. Para ello se pretenden mejorar tanto el tiempo de un ciclo completo como la uniformidad de la capa depositada. El método de optimización se desarrollará haciendo uso de un benchmark validable y, posteriormente, será aplicado al reactor ALD. Mediante el software CFD Ansys Fluent se modelará el comportamiento del flujo dentro del reactor, obteniendo así una aproximación de alta fidelidad del proceso. A partir de esta, se calculará un modelo de orden reducido (ROM) por medio del método de superficie de respuesta (RSM). Dicho modelo reducido será optimizado con programación cuadrática secuencial (SQP).

[EN] This Master’s Thesis proposes an optimization method for Atomic Layer Deposition (ALD) reactors. ALD is a manufacturing process that consists of the creation of thin layers of material over a surface (substrate) by deposition due to chemical reactions between the surface compounds and the precursor gas being injected into the working chamber. Numerous factors affect the resulting layer, the multi-disciplinary nature of the method makes it especially interesting for optimization processes. The objective of the project is to establish a procedure to reduce the purge time needed to evacuate the residual gas after the deposition as well as to maximize the uniformity of the precursor’s concentration over the substrate. In order to develop the method, a well-studied ventilation benchmark was employed. This case was modeled to mimic the phenomena expected to take place inside an ALD reactor. After the benchmark, the method has been successfully applied to a real reactor. The geometry used to represent it is a true replica of the Picosun R-200. This last one is a hot-wall thermal reactor and the simulated process is based on Trimethylaluminum precursor and N2 carrier gas. The formulated algorithm consists of using Computational Fluid Dynamics simulations to create a high-fidelity approximation of the system. These are later used in a Design of Experiments to obtain a relevant amount of samples to build a Response Surface Approximation surrogate model. This model constitutes a low-fidelity approximation which can be easily optimized with a Sequential Quadratic Programming algorithm. The calculation of optimum points for different permits to obtain the characteristic Pareto-Optimal Front of the reactor. The results from the application of the scheme to both response functions show the opposite needs in terms of mass flow to optimize the purge and uniformity. Temperature and pressure, on the other hand, have smaller effects on the results, but they are also discussed.