ALD2019 Session ALE2-TuA: Modeling & Instrumentation II

Atomic layer etching (ALE) is a fundamental part of semiconductor processing as device critical dimensions must be controlled to the order of nanometers [1]. One known issue in ALE, as in other etching processes, is aspect ratio dependent etching (ARDE) [2], which is the reduction of etch rates as the aspect ratio of a feature increases. One of the mechanisms linked to ARDE is the depletion of neutral species towards the bottom of a feature. This phenomenon has been investigated using a three-dimensional Monte Carlo method [3]. However, this method requires a complex setup and it is computationally expensive. For deposition processes, Knudsen diffusion [4] models provide analytical results and are actively developed. These models have been used for estimating surface parameters in some atomic layer deposition processes [5]. The Knudsen diffusion approach arises from physical considerations to the mass balance at each volume element. Alternatively, given isotropic reflections and particle source, the fluxes can be calculated exactly over the whole domain via the radiosity equation [6]. The radiosity approach requires the assembly and inversion of a matrix describing the exchanges, being notably unsuitable for low sticking regimes.

We propose a model extending the standard deposition Knudsen diffusion approach by including the direct flux from a particle source and a geometric factor to enable a more rigorous picture of ARDE in ALE. The inclusion of the direct flux is motivated by the radiosity equation, while avoiding the costly matrix inversion step. The geometric factor enables a more accurate description of the geometry by integrating over the whole feature at each volume element. We compare our extended Knudsen diffusion model against a reference radiosity model [6], achieving good agreement. Our results highlight one shortcoming of the standard Knudsen diffusion model: The flux near the bottom of a high aspect ratio feature is underestimated. We also show that the geometric factor describes the particle transport more accurately near the extremities of finite cylinders.

The financial support by the Austrian Federal Ministry for Digital and Economic Affairs and the National Foundation for Research, Technology and Development is gratefully acknowledged.

[1] K. Ishikawa et al., Jpn. J. Appl. Phys. 56, 06HA02 (2017).

[2] C.G.N. Lee et al., J. Phys. D: Appl. Phys. 47, 273001 (2014).

[3] C. Huard et al., J. Vac. Sci. Technol. A 35, 05C301 (2017).

[4] A. Yanguas-Gil, Growth and Transport in Nanostructured Materials (Springer, 2017).

[5] M. Ylilammi et al., J. Appl. Phys. 123, 205301 (2018).

[6] P. Manstetten et al., Solid-State Electron. 128, 141 (2017).

View Supplemental Document (pdf)

One of the main challenges in designing novel atomic layer deposition (ALD) and atomic layer etch (ALEt) processes is to ensure the self-limiting (SL) nature of reactions during the individual precursor pulses at the target temperature of interest. It is important to establish a SL temperature window to ensure that the precursors will not produce chemical vapour deposition (CVD) or chemical vapour etching (CVE). To that end, we have developed a first principles based computational methodology to study the competition between continuous and SL reactions when a material surface is exposed to precursor gases.

ALEt processes for oxides have been reported using sequential fluorination by HF and ligand exchange reactions at elevated temperatures.¹ Herein, we investigate the nature of the HF pulse when treating a range of materials including B₂O₃, TiO₂, HfO₂, ZrO₂, Al₂O₃ and TiN by comparing the free energy profiles of the potential CVE and SL reactions computed with density functional theory (DFT). In general, elevated temperatures favour CVE of these materials, whereas fluorination of the surface is self-limiting at lower temperatures. HF was computed to continuously etch B₂O₃ and TiO₂ even at temperatures below 100°C by forming volatile fluorides. This is in good agreement with experimental FTIR, where the onset of continuous etching of TiO₂ by HF was found at 80°C – 90°C and where etching of B₂O₃ proceeded even more rapidly than that of TiO₂. However, in another example, HF is predicted to preferably etch HfO₂ above 160°C, whereas experiment¹ shows the process to be still self-limiting at 200°C. Therefore, this simple thermodynamic analysis, which does not include reaction kinetics, is only able to provide a lower threshold temperature at which a CVE reaction may become favourable, subject to overcoming any kinetic barrier. We also computed that Al₂O₃ resists continuous etching up to 340°C while TiN can not be etched until temperatures exceed 1300°C.

This methodology is also used to understand the competing CVD and SL reactions in the ALD of Ru using the RuO₄ precursor and H₂.² The DFT calculations show that RuO₄ undergoes an SL decomposition into RuO₂ on electron-rich surfaces and that this is in competition with CVD.

These examples illustrate that this relatively quick computational approach can be effectively used to screen candidate precursor molecules to be selected for material processing.

1. Y. Lee et al., Chem. Mater.28, 7657-7665 (2016).

2. M. M. Minjauw et al., J. Mater. Chem. C3, 132-137 (2015).

Alternating application of reactive species from a plasma with no bias energy and Ar plasma with low bias energy to a Si-based film (such as Si, SiO2, and SiN) can cause atomic layer etching (ALE) of its surface. In this process, the initial application of reactive species to the surface will leave a sufficient number of reactive species (e.g., Cl atoms) bonded with the surface atoms (e.g., Si, forming Si-Cl bonds) and, in the second step where low-energy Ar+ ions irradiate the surface, a thin mixed layer of the reactive species and surface atoms is formed and volatile molecules (e.g., SiCl4) desorb from the surface. Since the incident energy of Ar+ is too low to physically sputter surface atoms, etching or desorption of volatile species ends as soon as reactive species are exhausted from the surface - even if Ar+ ions continue to bombard the surface with low incident energy. This self-limited etched depth of a single cycle (i.e., application of reactive species and application of low-energy ions) allows “nearly” layer-by-layer etching of the material surface. At the end of each cycle, where self-limited etching takes place, some reactive species tend to remain on the surface or in the subsurface region, causing surface roughness. Depending on how deep such reactive species diffuse into the subsurface region or how thick the formed mixed layer becomes, the etched depth per cycle changes. In this study, etched depth per cycle and remaining reactive species are discussed based on the results of molecular dynamics (MD) simulation of single cycles of various ALE processes. Example considered here include Si ALE by fluorine reactions, SiO2 ALE by fluorocarbon reactions,

And SiN etching by hydrofluorocarbon or hydrogen reactions.