On the Neural Network approach to the Kinetic Model for the Semiconductor Thin Film Deposition

다국어 초록 (Multilingual Abstract)

This Master’s thesis addresses the simulation of the kinetic model, which is used to describe the semiconductor thin film deposition. In this work, we consider a ki- netic model, with absorbing, specular reflection, and inflow boundaries. We adopt a thermal Atomic Layer Deposition (ALD) method that does not consider chemical re- actions. Our focus is on the precursor flow during various processes of thermal ALD. Using deep learning algorithms, we derive a Deep Neural Network (DNN) solution for the kinetic model. Through this approach, we observe the behavior of particles and investigate the associated macroscopic physical quantities.

목차 (Table of Contents)

• I. Introduction 1
• 1.1 Motivation 1
• 1.2 Problem 3
• 1.3 Our kinetic model 3
• 1.4 Boundary conditions 5
• 1.4.1 Specular reflection boundary condition 6
• 1.4.2 Absorbing boundary condition 7
• 1.4.3 Inflow boundary condition 7
• II. Methodology : The Deep Learning approach 8
• 2.1 Deep neural network framework for the kinetic model approximation . 8
• 2.2 Grid points 11
• 2.3 Loss functions for our kinetic model 13
• III. Neural Networks Simulations 17
• 3.1 DNN solutions by varying initial conditions for the kinetic model 17
• 3.2 DNN solutions by varying boundary conditions for the kinetic model . 21
• 3.3 Macroscopic Physical quantities for the kinetic model 25
• IV. Conclusion 28
• Summary (in Korean) 29
• References 30
• – II –