Abstract
Computer modelling provides a bridge between theory and experiment, using the remarkable computational power available today to perform virtual experiments and examine theories in complex systems like the atomic layer deposition (ALD) process. The complete ALD growth mechanism is highly complicated, as it consists of acid/base reactions, structural relaxation and self-limiting surface chemistry, all of which are strongly influenced by factors such as steric hindrance. Moreover, timescales in ALD are even more complex than length scales as the reactions could be relatively fast (pico to nanoseconds) or slow (micro to macro seconds). This range of timescales influences the film growth rate and gases that are being pulsed and purged. This thesis provides a profound understanding of the ALD of oxides such as Al2O3, showing how the chemistry affects the properties of the deposited film. Using multiscale modelling of ALD, the kinetics of reactions at the growing surface is connected to experimental data found in the literature. Experiments were not conducted, yet the computational investigation provided a novel approach and process by reducing the development time and cost associated with the laboratory. The geometric technique, cellular automata and multiscale simulation are among the modelling methods based on the features of atomic layer deposition. Each model and simulation method's principles, together with their benefits and drawbacks, were outlined.
The study focused on numerical simulation of the multiscale atomic layer deposition (ALD). A particular focus was on the sticking coefficient in these varied scales of ALD modelling. One of the most significant industrial techniques is thin film deposition. Atomic layer deposition consists of a thin film deposition with high uniform deposition ability. Several ordinary thin films and atomic layer deposition models and numerical simulation approaches are discussed in this work. The study's emphasis was on the sticking coefficient with interest in controlling it when modelling the pulsing of the Trimethylaluminum and ozone precursors to deposit alumina Al2O3. The study presented computational fluid dynamics and adopted a large-scale approach to investigate the benefit of using short but multi-precursor pulses in an ALD process. As a derivative of the super-cycles found, it is mainly utilised in the doping strategies of an ALD process.
The study’s objective was to demonstrate the potency of short multi-pulses of the ALD super-cycles over the conventional precursor pulsing binary method. Three major findings were
6
revealed. The first major finding was related to the understating of the fact that the sticking coefficient of metal precursors during an ALD process was from the high aspect ratio of the substrate. This study shows that surface oxidation plays a vital role in maintaining the sorption of a molecule on a surface. This approach is undoubtedly appropriate for substrates with zero and ultra-high aspect ratios. It is a precise tool to control the sticking coefficient of a TMA precursor. The second major finding was to show an improvement in depositing thin films of alumina at low pressure and low-temperature reactors on a substrate. More film deposits were achieved quickly compared to the ABx strategy. There were more by-products removed too, meaning that more precursor diffusion took place and improved film growth. The growth per cycle (GPC) of literature ranges from 0.8–1.1 Å/cycle for GPC, but this study determined a 0.29–1.13 Å/cycle for oxides in low pressure and low temperature was attained. The advantage is that more surface area is covered at super-cycles. The co-reactant responds better because more by-products are removed, allowing molecules to react on the surface whilst most unreacted molecules are pumped out of the reactor. The exposure or settling techniques (closing at pulsing and purging) at the inlet is explored using the AnBnBnBx super-cycle (stop-flow inclusive), and more film deposits took place while conserving precursors. The third major finding regarding the sticking coefficient was from the microscale. An ab initio molecular dynamic (MD) of the ALD as an all-atom approach was modelled. A good, agreeable GPC with was found to be 4.257x10-3.
Overall, an initial sticking coefficient of Trimethylaluminum precursor at multiscale level was determined, with the critical takeaway that the S0, the sticking coefficient, remains identical to its relevant scale, not varying within the reactor and feature scale. Only its nature and attributes are notable in the geometrical feature of the substrate, temperature and exposure time of precursors. Conclusively, the study revealed that the role of the sticking coefficient to molecules dissociating on the chemisorbed on the surfaces at any ALD timescale is unchanged.