First principles modelling of nucleation and growth during atomic layer deposition onto III-V substrates
- Author
- Sylwia Klejna
- Year
- 2013
- Abstract & Cover
- Source of Information
- Author
- University
- University College Cork, Tyndall National Institute (Cork, Ireland)
- External Link
- Read Thesis
In this thesis we have presented theoretical investigation of nucleation during the incubation period of ALD of dielectric thin films on technologically important substrates: III-V semiconductors and silicon nitride. Atomic-scale simulations in the framework of DFT were conducted on the bulk structures, surface models and gas phase molecules. Ab initio thermodynamics provided a bridge between zero-temperature, zero-pressure DFT and real experiments. We have analysed and described the interaction of the precursors with the considered substrates: starting from initial adsorption of the gaseous molecule, through its transformation in contact with native oxides, until formation of surface intermediates that can be lost as byproducts. The structure and chemical state of surface intermediates were described in detailed. Some kinetic aspects of the incubation mechanisms were also revealed. Based on the trends arising from the initial computations, we have developed a comprehensive model that allows us to compare the operation of different classes of precursor chemicals and assess their properties for self-cleaning ALD. On the example of TMA we identified two separate factors governing the clean-up effect: formation of the metal oxide as the driving force and affinity of the precursor ligand to the III-V oxide substrate as the ancillary force. ‘Clean-up’ of an oxide film is shown to strongly depend on the electropositivity of the precursor metal, and thus always results in formation of dielectric film from native oxide. However, self-cleaning ALD does not necessary result in substrate-enhanced growth, as the clean-up effect can be spread over a few cycles. The choice of ligand determines bonding at the interface and the overall type of clean-up. The predominant pathway for a metalloid oxide such as arsenic oxide is reduction, producing volatile molecules or gettering oxygen from less reducible oxides. An alternative pathway is non-redox ligand exchange, which allows non-reducible oxides to be cleaned-up. TMA was shown to be a superior precursor for self-cleaning ALD, considering the good properties of Al2O3 film as an interfacial layer in the transistor gate stack. Thanks to the lability of the methyl ligand that has affinity to both Lewis basic and acidic sites, a number of mechanisms are exothermic and competitive. We proposed TMA clean-up byproducts, interface bonding and the most thermodynamically abundant surface intermediates. TDMAH was shown to have slightly different properties to TMA. However we proved that this precursor is also a clean-up reagent, with a number of exothermic reactions participating. The dimethylamide ligand was shown to be resistant to oxidation when intact, however very prone to decomposition and subsequent oxidation. Some of the decomposition intermediates can migrate from Lewis acidic to Lewis basic site as in the TMA case. What is characteristic in the operation of alkylamide precursors is that their self-cleaning behaviour is increased with temperature and results in accumulation of As suboxides at the interface. Our kinetics study has revealed the reason for source of this behaviour. In general, we predict that the investigated methyl precursors are the best reagents for deposition of dielectrics and performing clean-up. Unfortunately, most of them are very unstable compounds. Having identified Mg precursors as the most effective at clean-up, one possible avenue for future research would be to evaluate the performance of some specific Mg precursor molecules on III-V surfaces. For instance, Mg(C5H5)2 is a commonly-used precursor in ALD [147]. It would be interesting to look in particular at whether this precursor too can carry out ‘clean-up’ and improve the electrical properties of the interface. Our study showed also that the chloride family cannot act according to the same mechanisms as organometallic and metalorganic precursors, but according to different mechanisms resulting in self-cleaning ALD that could also be investigated further. Successful ALD relies on growth reactions being faster than side-reactions. For these reasons, calculating the kinetics of self-cleaning ALD reactions could contribute to the research in this subject. However, DFT is not suitable for cases where detailed treatment of electron correlation is necessary. This includes some types of transition state (TS) that we study within this thesis to evaluate the kinetics of the ‘clean-up’ effect. Therefore, it would be profitable in the future to benchmark the DFT barriers reported here against more accurate calculations. The eigenvalue-following Trust Radius Image Minimization algorithm provides more accurate geometries and energetics of transition states. These geometries, at the DFT level, can be used as an input for the computationally-heavy configuration interaction (CI) calculations that account for electron correlation more completely than DFT and yield high-quality energy. It is worth mentioning that the heterodeposition (deposition on the initial substrate) mechanisms investigated here could serve as an input to the multi-scale simulation, e.g. kinetic Monte Carlo [148], of the actual film growth corresponding to an experimental timescale. The major aim of this thesis was to assist and to complement experiment and as a result to obtain better understanding of the mechanisms occurring during initial ALD cycles of dielectric films. This goal was definitely achieved. Some of the suggested here mechanisms were subsequently confirmed, in the case of TMA clean-up by Tallarida and co-workers, while other mechanisms provided explanations for prior experimental findings, in the case of alkylamides by Gougousi and co-workers. What is more, the clean-up by products reported here may be detectable with, e.g. FTIR or mass spectrometry experiments of sufficient surface sensitivity. Our calculations assisted analysis of the growth details on silicon nitride substrates. Theoretical calculations accounted for the different starting substrates and provided possible explanations for the experimentally-revealed anomalous growth step.
