ALD/ALE 2025 Session AF1-TuM: Mechanism and Theory I

Carbon-doping of silicon nitride or silicon oxide results in a lowering of the dielectric constant and etch rate. The C-doped derivatives are therefore being used to replace the parent films in many applications in FEOL semiconductor manufacturing, such as for spacers and gap fill. Traditionally these C-doped films are deposited by CVD methods, but recently ALD methods have been formulated using silicon precursors with Si-CHx bonds. Maintaining the Si-CHx bond has been challenging due to its propensity to cleave under plasma conditions or at high temperatures. For these reasons, precursors that can be used in a low temperature thermal ALD process are highly desired.In this work we investigate whether ammonia is viable as a thermal ALD co-reagent for this application and what mixture of chloro and alkyl ligands are most favorable in the silicon precursor.

We use density functional theory (DFT) to compute the atomic-scale structure of surface intermediates and the mechanism of potential ALD reactions and complement this with detailed characterization during substrate exposure experiments.

Looking first at the co-reagent, we investigate the kinetics of proton transfer from ammonia to a model silyl fragment on an aminated surface. Four proton transfer pathways are obtained for HCl elimination, all with high barriers (>1.2 eV) that are indicative of low Bronsted acidity and suggest that side-reactions will compete with the ALD process. Three pathways are obtained for CH4 elimination with even higher barriers (>2.5 eV). Therefore, up to moderate temperatures, terminal-CH3 is likely to survive ammonia treatment and be incorporated into the film, and this could be a route to C-doping. Residual gas analysis (RGA) and quartz crystal microbalance is used to monitor whether HCl or CH4 is in fact evolved during each half reaction.

The second part of the study looks at the effect of various combinations of chloro and methyl ligands in candidate Si precursors, both monomers (SiCl4, SiHCl3, SiCl3Me, SiH2Cl2, SiHCl2Me, SiCl2Me2) and dimers (Si2Cl6, Si2Cl4Me2). Structural models are generated efficiently of the >100 surface intermediates that can potentially occur through physisorption, chemisorption via ligand elimination and etching by exchange with surface amines. For each precursor, DFT-based thermodynamics reveal that the most favorable intermediate is surf-SiCl3 up to about 250C, surf-Cl2 up to 450C and surf-Cl at higher temperatures, with associated predictions of ALD growth rate. Film composition and growth rate is then validated experimentally by ellipsometry (SE) and infrared red spectroscopy (ATR-FTIR), ex-situ but with an inert atmosphere glove box.

The importance of self-aligned film deposition processes is obvious when considering the complexity of the devices being currently manufactured. Every process that can directly deposit only on selected areas may decrease the number of processing steps needed. When the growth is dictated by surface chemistry also issues related to pattern misalignment can be avoided. Finding new self-aligned thin film deposition processes becomes yet more important as the device dimensions are shrinking. Understanding the mechanism behind the selectivity helps in finding new process and surface combinations. In this work, the reason behind a selective growth of iridium on a set of substrates that are chemically alike is studied. The surface reactions and the selectivity mechanism are studied with ALD-UHV cluster tool by using in vacuo XPS, LEIS and TPD together with in situ QCM.

Excellent area selectivity in ALD of metallic iridium on a wide selection of substrates was initially found by Zhang et al. (Chem. Mater. 2022, 34, 8379-8388). The depositions were done with iridium(III) acetylacetonate, Ir(acac)₃, and O₂ as precursors. The growth is believed to start with a ligand exchange reaction between surface -OH group and the Ir(acac)₃ precursor. Initial hypothesis for the selective growth is that the weakly acidic byproduct formed in the ligand exchange reaction, Hacac, adsorbs preferably on more basic surfaces passivating them. In the current work, two surfaces on which iridium was deposited successfully, native SiO₂ terminated Si and in situ deposited Al₂O₃, are compared to two surfaces on which the deposition did not succeed, Al₂O₃ and HfO₂. The latter mentioned oxides as well as the native SiO₂ terminated Si were exposed to air and airborne hydrocarbons prior to the iridium deposition.

One of the main differences between Chemical Vapor Deposition (CVD) and atomic layer deposition (ALD) is gas-phase reactivity: we anticipate precursors to react together or independently in the gas phase in CVD but generally expect ALD precursors to arrive at the substrate intact. However, thermal activation of a precursor in the gas phase can benefit ALD saturation by producing a more active chemical species.

Our group often uses an in-situ solution-phase method to discover thermal behaviour and track the kinetics of decomposition. Samples are flame-sealed in a glass tube with an appropriate solvent and monitored by nuclear magnetic resonance (NMR) spectroscopy. This characterization method is common in synthetic chemistry and can be used to analyse a variety of nuclei independently, and this technique can commonly and easily assess protons. Using this technique in conjunction with thermogravimetric analysis (TGA) and differential scanning calorimetry (DSC), we can build structure-function relationships to help design better precursors.

The framework Mo(VI)(=N^tBu)₂Cl₂.(L-L) (where L-L is a chelating ligand) allows us to easily investigate the role of chelate ligands in precursor design due to the thermal stability of the imide/chloride core. This talk will center on measuring the thermodynamics of precursor thermolysis through Eyring analysis and detail the implications of β-H versus γ-H elimination, as well as differentiating inter-molecular vs. intra-molecular reactivity. The presentation will show detailed reaction coordinates and different mechanistic thermolysis pathways for a growing family of Mo(VI) compounds.

A promising strategy for the development of advanced atomic layer deposition (ALD) precursors is the design of heteroleptic metal-organic compounds in which multiple ligand types are attached to a metal center. These heteroleptic precursors offer advantages over homoleptic counterparts, such as improved thermal stability while maintaining high reactivity. A notable example is tris(dimethylamino)cyclopentadienyl zirconium (CpZr(NMe₂)₃), which has successfully replaced the homoleptic precursor tetrakis(ethylmethylamino)zirconium (Zr(NEtMe)₄) in the fabrication of zirconium oxide (ZrO₂) films for dynamic random-access memory (DRAM) devices. Compared to Zr(NEtMe)4, CpZr(NMe2)3 exhibits better thermal stability, enabling better conformality at high process temperatures of 300°C with a growth per cycle of ~0.9 Å [1]. However, the competition between the NMe₂ and Cp ligands during surface reactions complicates the identification of the final chemisorbed species, which is important for designing the precursor for high-temperature ALD. In this study, we investigated the reaction mechanism of ZrO₂ ALD using a combined approach of ab initio calculations and in situ characterization. Density functional theory (DFT) calculations were performed to model the stepwise surface reactions of CpZr(NMe₂)₃ on a ZrO₂ surface, while in situ quartz crystal microbalance (QCM) analysis was used to study the ALD process with alternating precursor and ozone (O₃) exposures. DFT results indicate that the first two NMe₂ ligands are readily released during chemisorption of the Zr precursor with low activation energies of 0.22 eV and 0.16 eV. In contrast, the release of the Cp ligand or the third NMe₂ ligand is thermodynamically unfavorable due to its endothermic nature and high activation energy. As a result, the final chemisorbed surface species is predicted to be O₂ZrCp(NMe₂)*. QCM analysis confirms the release of two NMe₂ ligands during the first ALD half-cycle, which agrees with the DFT results. This study demonstrates the effectiveness of integrating DFT calculations with QCM analysis to gain mechanistic insight into ALD reactions, especially when heteroleptic precursors are involved.

References [1] J. Niinisto et al., J. Mater. Chem. 18, 5243 (2008).

Atomic layer deposition (ALD)-grown ultra-thin In₂O₃ offers promising solutions with amorphous films at sub-3 nm thicknesses, allowing effective tuning of mobility and V_th based on thickness [1].

Materials and electrical properties of ultra-thin ALD films are significantly dependent on surface chemistry. For example, HfCl₄, with simple chlorine ligands, achieves a higher growth per cycle (~1.3 Å/cycle), lower interface trap density (1.86×10¹¹ eV⁻¹ cm⁻²), and reduced leakage current density (3.94 × 10⁻⁷ A/cm² at V_FB = −1V). This is due to efficient oxidation and minimal steric hindrance, resulting in stoichiometric films with fewer oxygen vacancies. In contrast, Hf(N(CH₃)₂)₄, with bulky dimethylamino ligands, exhibits slightly lower GPC (~1.2 Å/cycle) and higher oxygen vacancy concentrations, leading to increased D_it (8.96 × 10¹¹ eV⁻¹ cm⁻²) and leakage current (2.25 × 10⁻⁶ A/cm²). Steric hindrance limits oxidant access, resulting in sub-stoichiometric HfO₂₋ₓ phases [2]. This highlights the critical role of ligand structure in determining film stoichiometry, defect density, and electrical performance, emphasizing the need for a comparative study for ultra-thin film formation using ALD.

In this study, we investigate the reaction mechanisms and surface chemistry of two indium precursors, [3-(dimethylamino)propyl]dimethyl indium (DADI) and trimethyl indium (TMI), for ALD-grown ultra-thin In₂O₃ films using O₃ as the reactant. DADI, containing dimethylamino and methyl groups, introduces greater steric hindrance compared to TMI, which has only methyl groups. This difference affects adsorption density, oxidizer accessibility, and oxygen deficiency during film growth, leading to variations in film composition, density, and electrical properties. To evaluate these differences, quadrupole mass spectrometry (QMS) will analyze ALD byproducts, X-ray photoelectron spectroscopy (XPS) will assess impurity levels and oxygen vacancies, and transmission electron microscopy (TEM) will confirm film uniformity. Physical properties such as morphology and density will be analyzed using atomic force microscopy (AFM) and X-ray reflectometry (XRR). Finally, a bottom-gate TFT will be fabricated to correlate these properties with electrical performance, providing a comparative evaluation of the precursors for BEOL integration into monolithic 3D devices.

Atomic layer deposition (ALD) has been a key enabling technology for atomically fine-tuned manufacturing of cutting-edge semiconductor devices. Depending on the underlying chemical mechanism of the ALD process, growth rates can vary significantly from process to process. In semiconductor gate stacks, Al₂O₃ is considered as interface dipole inducer and SiO₂ has been used as gate oxide. It has been shown that plasma enhanced ALD (PEALD) growth rate of SiO₂ is orders of magnitude larger than that of Al₂O₃ [1]. In these processes, O₂ plasma is used as the source of oxygen. O₂ plasma includes reactive species such as O radicals which can, in principle, recombine with the previously deposited O species on the growth surface and form O₂ gas. It has been proposed that the growth rate difference between SiO₂ and Al₂O₃ can be attributed to the differences in the oxygen recombination rates at the respective growth surfaces [1].

In this talk, we present evidence from atomic scale modeling in terms of thermochemical, kinetic and dynamic analysis of reactions taking place at the SiO₂ and Al₂O₃ growth surfaces using industry-grade simulation frameworks within the QuantumATK software [2,3] developed by Synopsys. Dynamic simulations with Synopsys QuantumATK have been previously utilized to get insights and determine mechanisms of ALD of HfO₂ [4] and ALE of 2D TMD materials [5].

In this study, we employed Density Functional Theory to calculate the thermochemical and kinetic properties of oxygen recombination reactions at both SiO₂ and Al₂O₃ growth surfaces. Using thermochemical modelling, we computed the free energy change associated with oxygen recombination. Using kinetic modelling, we extracted the activation barrier for the recombination reaction and derived the corresponding reaction rates. MACE universal machine learned force field is chosen for dynamic simulations as ab initio methods are very time-consuming for such studies. Using dynamic modelling, we explicitly simulated the plasma species-surface interaction and compared the probability of recombination at both growth surfaces. From the results generated at different levels of atomic-scale modelling, we provide a comprehensive understanding of the oxygen recombination leading to growth rate differences observed in the PEALD of the two materials.

[1] K. Arts et al., J. Phys. Chem. C, 123, 27030−27035 (2019).

[2] QuantumATK V-2024.09, Synopsys. (https://www.synopsys.com/quantumatk )

[3] S. Smidstrup et al., J. Phys.: Conden. Matter 32, 015901 (2020).

[4] J. Schneider et al., ALD/ALE 2022, Ghent, Belgium.

[5] S. K. Natarajan et al., ALD/ALE 2024, Helsinki, Finland.