Developing a framework for uncertainty quantification of a chemical kinetic parameter in MOCVDS

Open Access
- Author:
- Khosa, Kartik
- Area of Honors:
- Mechanical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Yuan Xuan, Thesis Supervisor
Dr. Jacqueline Antonia O'Connor, Thesis Honors Advisor - Keywords:
- Latin Hypercube Sampling
Monte Carlo Simulation
Uncertainty Quantification
CFD
Nanomaterials - Abstract:
- Transitional metal dichalcogenides (TMDs) have been an increasingly explored research area due to their various applications in solar cells and quantum computing hardware. However, the key challenge has been finding the optimum way to manufacture TMDs. The two most popular TMDs are WSe2 and MoSe2. Both of these are produced by chemical vapor deposition. It has been difficult to maximize the yield of TMDs because the cost of experimentation is high, and the experiment takes longer to set up and perform. Therefore, to overcome these challenges, simulations were run to model the conditions and understand the impact of various parameters on the yield of the TMDs. But the accuracy of the chemical model is still not fully understood. In this thesis, the impact of the uncertainty in one of the key chemical kinetic parameters, activation energy, is explored. This thesis focuses on the production of WSe2 specifically. The chemical model was split into five different categories according to the reaction types. It was found there was no uncertainty in one set of reactions and so that set was not perturbed during the sampling process. The other four sets were perturbed individually so the effects of activation energy of different reactions could be understood. However, due to time constraints, only three sets were fully sampled and simulated. From the CO release set, it was found the most sensitive reactions were the initial release reactions 1, 2, and 4 that decompose W(CO)6. For the H2 gain set, all the sensitive reactions had a positive correlation. For the H2 loss set, similar reactions were important but had negative sensitivity coefficients. MATLAB was used to create our own model to predict simulation results that ended up giving an underpredicted output. From an L2 norm analysis, it was found that slowing down even one reaction negatively affected the yield.