Ab initio investigation of reaction mechanism in the initial phase of deposition by atomic layer deposition of oxides with middle and high permittivity on silicon

Leonard Jeloaica
Language of the thesis
Thesis name in original language
Etude Ab initio des mécanismes réactionnels dans la phase initiale du dépôt par couches atomiques des oxydes à moyenne et forte permittivité sur silicium
Abstract & Cover

The deposition of the dielectric layer to replace SiO2 remains a major challenge for the electronics industry in the coming years. After the invention of the technique of atomic layer deposition ALD, in parallel with its continuous improvement, many experimental studies have been carried out in this direction. In addition, in the last decade, numerical simulations based on quantum calculation, ab initio or semi-empirical, have begun to provide the additional information necessary for the development of this new technology. Yet, in the current stage of know-how, the deep understanding at the atomic level of the reaction mechanisms that will generate, layer by layer, new gate oxide will be essential for the pursuit and completion of this challenge. In this manuscript we have, firstly, reviewed the main problems and challenges related to the miniaturization of integrated circuits. We discussed the alternative of replacing SiO2, as gate oxide in CMOS or DRAM applications, by oxides with higher permittivities, as well as the main selection criteria for future dielectrics. Al2O3, HfO2 and ZrO2 have demonstrated remarkable thermodynamic compatibility with the Si substrate and are considered today among the most interesting candidates to replace SiO2. The ALD technique, to which we have devoted a section detailing the principles, characteristics and advantages over conventional techniques, was often used to test the three oxides.
Al2O3 remains the reference oxide for ALD, as it lends itself well to the basic models of this technique. Yet Al2O3 remains an alternative in the medium term given its slightly higher permittivity compared to SiO2. The other two, more interesting from this point of view, are unfortunately more difficult to handle at the technological level for the conformity of the deposit, in particular in the initial phase. The main difficulty remains the formation of oxide nanocrystals from the initial phase of the deposition, which induces leakage currents through the oxide along the grain boundaries between the nanocrystals. This rather complex problem has been little studied and the proposed mechanisms have not been studied explicitly by ab initio calculation. However, the reactions of ALD precursors with water constitute a very good start for the understanding of the formation of these oxides. In this thesis we approached this complex subject in a less conventional way, by trying to answer several problems and specific growth conditions (temperature in the reaction chamber, concentration of vapors of gaseous precursors, temperature and preparation of the surface of the substrate, gas phase reactions and surface reactions).
In the methodological part, we detailed the principles of the ab initio and DFT methods. In this context, we have focused our discussion on the treatment of electronic correlation and its correct application according to the systems to be studied. We have also devoted a section to methods and numerical algorithms for the optimization of molecular structures, where we have inserted some tips accumulated with experience, in particular for the search for transition states in chemical reactions. Our own work began with an extensive methodological study, using as molecular systems the molecules used as precursors for the formation of the three candidate oxides: Al2O3, ZrCl2 and HfO2. We tested the accuracy of the B3LYP hybrid density functional, by comparison, both with more accurate post Hartree-Fock type methods (CCSD(T) or QCISD(T)) and with different function bases, and with existing experimental data. The calculations were carried out on the static properties, in the fundamental states, as well as on the dynamic properties, of the vibrational spectra of these states. This led us to validate the very reliable B3LYP/TZVP calculation model for the prediction of all the properties studied.
In particular, we are explicitly interested in potential surfaces in the space of strongly anharmonic internal motions of the methyl groups of the TMA, which constitutes a first for the systems in question. Also, we have proposed two methods to calculate the scale factors of the normal modes for different thermodynamic functions and for different temperature regimes. Our results, and in particular the observations that we have made concerning the necessity of the treatment of strongly anharmonic motions for TMA, will serve as very precise and complete data for the study of the kinetics or/and the precise prediction of the thermodynamic properties of the systems. Then we presented our study of the reactions of H2O with the three molecular systems in the gas phase. Three-step hydrolysis mechanisms for TMA, and four-step hydrolysis for ZrCl4 and HfCl4 have been proposed and the results discussed in their most detailed aspects. These mechanisms correspond to conditions of very low water concentration, i.e. residual water in the containment. In particular, we noticed and discussed the strongly anharmonic movements in the formed complexes. We concluded that the reactions of TMA with H2O were exothermic, whereas the corresponding reactions for the two tetrachlorides, ZrCl4 and HfCl4, were found to be endothermic. We also discussed the possible consequences of the exothermicity of TMA reactions with H2O on the surface reactivity. This opens perspectives to study the possible modulation of the growth rate of the film in its initial phase. Subsequently, we presented the study of similar mechanisms, this time between the precursors formed at the surface (SiO2/Si(001)-2x1) in the initial phase of deposition by chemisorption/recombination. First, we made some important remarks from a methodological point of view on the choice and construction of surface aggregate models.
Concerning the results part, we focused our discussion on the main differences and similarities with the results of hydrolysis in the gas phase. We have again concluded to the strong reactivity of H2O with hydroxymethyl aluminum complexes. With regard to Zirconium and Hafnium hydroxychloride complexes, apart from the fact that their surface chemistry is very similar, the endothermic nature of the reactions studied was also confirmed. We have discussed these results with respect to very specific deposition conditions, i.e. low concentrations of OH active sites, and low concentration of water vapour. In the last part of our thesis, we presented and discussed our preliminary results on more complex mechanisms of the reaction of water with surface ALD complexes. These mechanisms, as well as the resulting complexes, are more appropriate for the usual deposition conditions, where the concentrations of the OH sites are high - which induces considerable interactions between the neighboring surface complexes - or high vapor pressure pulses of water. We have modeled such conditions by explicitly treating the effects of cooperative interactions of water molecules in the vicinity of the complexes.
We have presented here the cases of Aluminum and Hafnium. The results revealed mechanisms not yet considered, to our knowledge, and demonstrated the need to question classical models of reactions in ALD, at least for HfO2 (and implicitly for ZrO2, given their very similar chemistry). Even if our study is in an initial phase, and there are certainly other aspects to be discovered, we believe that we have captured one of the essential aspects of reactions with water, and in particular of the hydrolysis cycle. Finally, we proposed for study a reaction mechanism responsible for Cl contamination in ZrO2 and HfO2 films. Our results constitute a very important step in the understanding of ALD growth of these films. However, the conjunction of non-equilibrium thermodynamic constraints, of the non-linear structure of the evolution equations of the macroscopic physicochemical processes and of the nature of the initial state (very flexible), makes the complexity of the evolutionary dynamics of the systems envisaged inconceivably treatable by modeling studies only at the atomic scale. Thus, the modeling of the deposition by ALD of the three oxides must be part of a multi-scale strategy, capable of structuring and coherently interconnecting the results of numerical simulations at different scales. In this context, firstly, the ab initio results obtained in this work will be used to parameterize a Kinetic Monte Carlo simulation software, a first version of which, incorporating mechanisms appropriate to ALD boundary conditions, is already available at LAAS-CNRS. Work is currently underway to improve this software.

Université Paul Sabatier - Toulouse III
(Toulouse, France)
External Link
Read Thesis
linkedin invite