ALD/ALE 2022 Session ALE1-TuA: In situ Studies, Mechanisms and Modeling of ALE

Transition metals are known as hard-to-etch materials for reactive ion etching (RIE). In recent years, atomic layer etching (ALE) based on the formation of highly volatile metal compounds from surface metals and organic gases has been proposed as atomically controlled and damage-less etching for transition metals [1]. In chemical vapor deposition (CVD) and atomic layer deposition (ALD) processes, various volatile metal-organic compounds such as β-diketonates (such as nickel(II) acetylacetonate: Ni(acac)₂ and Nickel(II) hexafluoroacetylacetonate: Ni(hfac)₂) and metallocenes (such as nickelocene) are known as metal precursors. The metal ALE process mentioned above, where volatile metal β-ketone enolates or metallocenes are formed on a metal surface, may be considered as a reverse process of such ALD. The development of new ALE processes for various transition metals requires a better understanding of surface reactions of metals with gas-phase organic molecules. Therefore, the goals of this study are to understand how organic compounds absorb on transition metal surfaces and react with transition metal surfaces. Moreover, we want to clarify the mechanisms of the formation of volatile metal compounds and their desorption processes. In this study, we focused on the surface reactions of Ni and Co with hexafluoroacetylacetone (hfacH) and acetylacetone (acacH) gases, which were analyzed experimentally with the Atomic Layer Process (ALP) Analysis System and theoretically with density functional theory (DFT) calculations [2,3]. The ALP Analysis System, which consists of an in situ high-resolution X-ray Photoelectron Spectroscopy (XPS) system and an ALP reaction chamber, allowed us to examine the chemical compositions and bond formation of the metal and metal oxide surfaces on which hfacH or acacH molecules adsorbed. ALE processes of Ni and Co films were also confirmed experimentally with oxidation and hfacH exposure at elevated temperatures. DFT calculationsalso clarified the mechanisms of self-limiting reactions in thermal ALE of Ni with hfacH and acacH exposure, which were found to be consistent with experimental observations.

[1] J. Zhao, et. al., Appl. Surf. Sci. 455, 438 (2018).

[2] A. H. Basher, et al., J. Vac. Sci. Technol. A 38, 052602 (2020).

[3] A. H. Basher, et al., Jpn. J. Appl. Phys. 59, 090905 (2020).

Ultra-sensitive techniques are required to observe the final cycles of atomic layer etching (ALE) as the etch process becomes a surface cleaning procedure at the interface of the film and the substrate/etch stop. With a sensitivity of ~1% of a monolayer, Auger electron spectroscopy (AES) is well suited for measuring the removal of a film with ALE. AES peak height ratios of the film and the underlying substrate provide a quantitative measurement of the film thickness during ALE. The peak height ratios were evaluated using a model developed by Holloway.

ALE of Al₂O₃ was performed using hydrofluoric acid (HF, 10% solution in pyridine) and trimethylaluminum (TMA). The sample stack was Al₂O₃/Au/Y₂O₃ (adhesion layer) /SiO₂ (native oxide)/Si. The gold underlayer provided a chemically inert, highly conductive interface ideal for AES studies. AES measurements were conducted periodically throughout the etch process and used to determine the thickness of the Al₂O₃ film. The instrument used for these measurements allowed for the transfer of the sample from an ALE process chamber (~1 Torr) to the ultra-high vacuum analysis chamber used for AES (10^-10 Torr) without exposure to atmosphere.

The reduction in the aluminum and oxygen AES peak intensities from the Al₂O₃ film and the increase in the gold substrate peak intensity was observed as a function of the number of ALE cycles at 250°C (Figure 1). As monitored by AES, the etch per cycle of Al₂O₃ with HF and TMA at 250 °C was determined to be 0.3 Å·cycle^-1. At the Al₂O₃ – Au interface, once the Al₂O₃ film had been nominally removed, Al and O atoms remained on the surface. Additional cycles of HF and TMA continued to remove this material, effectively cleaning the substrate surface (Figure 2). AES observations for the Al₂O₃ ALE process on Au substrates will be presented for different reaction conditions.

Atomic layer etching (ALE) is one of the most advanced etching processes today, which enables atomic precision in a cyclic etching, where the material is removed layer-by-layer. In an implementation of ALE, a monolayer of reactive gas is chemisorbed on the surface of the etched material providing surface activation. The activated surface is then removed by plasma - usually by low-energy ions extracted from the discharge. They provide sufficient energy to remove only the reaction products leaving inactivated material.

In practice the ions may contribute to the physical sputtering of the etched material, thus increasing the total etch rate above the values of pure ALE. The residual reactive gas in the reaction chamber can dissociate in the plasma and also increase the etch rate. This etching mode can be defined as a “quasi-ALE” with contributions from the sputtering and reactive ion etching (RIE). To realize a “real or pure” ALE regime with negligibly small unwanted effects, careful optimization of process conditions is necessary. In the current work, we apply optical emission spectroscopy (OES) in a quasi-ALE cyclic process to monitor the relative concentration of active Cl₂ species in the chamber and relate it to the observed etch rate of Si.

In our experiments, we used a commercially available Inductively Coupled Plasma RIE Takachi™ tool (Plasma-Therm LLC, USA) with Cl₂ and Ar as process gases. Silicon samples were patterned with Talbot displacement lithography to produce specimens with resist masks. The etch rates were measured after the samples were cleaved and inspected with a scanning electron microscope. The OES measurements were performed in the wavelength range 191 - 887 nm using an in-situ spectrometer. The etch cycle included the following steps: injection of molecular Cl₂, purging with Ar to remove the excess chlorine, Ar-plasma etching and purging again.

Future technological nodes (N3 and below) will see further shrinking of not only the pattern dimensions, but of photoresist film-thickness as well. The etching of Spin-On-Glass (SOG) using a Chemically-amplified-resist (CAR) mask following lithography is strongly affected by this trend as the selectivity of the conventional etch processes will soon no longer be sufficient. Furthermore, roughness reduction techniques as we know them will also be challenged as they typically consume some of the already-meagre CAR budget. Therefore, the reactive-ion-etching – based atomic layer etching (RIE-ALE) of SOG with a CAR mask is investigated as a softer and more selective alternative to conventional RIE. The RIE-ALE of SOG and the transfer into underlying hard mask have been demonstrated on coupons as well as on 300 mm full wafers, highlighting a strong increase of selectivity compared to RIE in coupons. However, the ALE process also yielded a higher roughness than RIE which doesn’t meet the requirements for EUV patterning.

An investigation of the mechanisms leading to this improvement of selectivity together with the deterioration of the roughness is provided, highlighting the lack of chemical contrast due to the reactivity of the fluorocarbons, the role of redeposition mechanisms and the importance of interfacial interactions. Indeed, the mechanisms of fluorination during the patterning process become essential to understand and optimize the processes, as the very thin layers become deeply modified within their bulk.

The selectivity observed on patterned wafers can be explained by redeposition mechanisms, as the fluorocarbon-mixed layers formed in ALE are sputtered and redeposit on the CAR mask. Redeposition mechanisms are not suitable for an optimal roughness control as their statistical distribution leads to an increase of the LER and LWR. Moreover, the surface properties of both the CAR photoresist and SOG underlayer being deeply altered during the fluorination, the formation of complex interfaces as the ALE process occurs are expected to occur which induce further stress – hence roughness – within the layers. Surface free energy (SFE) measurements highlight the evolution of the dispersive (i.e., London force) and polar (i.e., Coulomb force) components of the layers during the patterning. Those interactions driving the gas to solid interactions before the formation of the mixed layer and the interface with the bulk, they provide valuable information on the potential stress within the stack.

Copper interconnects are widely used in ULSI circuits due to their low resistivity and superior resistance to electromigration. Traditionally, the patterning of copper and barrier metals in the damascene process relies on chemical mechanical planarization (CMP), which may cause damages to fragile low-k interlayer dielectrics. Thermal atomic layer etching (ALE) is an emerging technique that uses sequential, self-limiting reactions to remove materials with high selectivity [1]. Thermal ALE is being considered an alternative to CMP as a final material removal step in the copper damascene process [2]. Recently, the thermal ALE of copper was demonstrated using O₂ or O₃ and H(hfac) (hexafluoroacetylacetonate) as the reactants [3]. Besides the oxidation mechanism, metal ALE can also be achieved through halogenation and ligand volatilization. For example, Co ALE has been demonstrated using Cl₂ and H(hfac) as the reactants [4]. In this work, theoretical calculations based on density functional theory (DFT) were performed to explore the possibility of Cu ALE through halogenation and ligand volatilization.

Using ab initio thermodynamic calculations, we have compared the Gibbs free energies of the reactions between H(hfac) and CuO, Cu₂O, and CuCl₂, as shown in Figure 1. It is found that the CuO + H(hfac) and Cu₂O + H(hfac) reactions are exothermic and have negative ΔG values between 300 K and 700 K. In contrast, the CuCl₂ + H(hfac) reactions are only favorable at temperatures higher than 425 K. Thermodynamic calculations suggest that a higher reaction temperature is required for Cu ALE using Cl₂ and H(hfac) as the reactants. In the next, we performed DFT calculations to investigate the elementary reaction steps of Cu ALE. During the Cl₂ exposure, the Cu surface is chlorinated to CuCl₂. During the H(hfac) exposure, the surface CuCl₂ is converted to gaseous HCl and Cu(hfac)₂. The surface reaction between CuCl₂ and H(hfac) consists of several elementary steps including H(acac) dissociation, hydrogen diffusion, Cu(hfac)₂ formation, and HCl and Cu(hfac)₂ desorption. The rate-limiting step is found to be the desorption of HCl and Cu(hfac)₂, with an activation energy of about 1.7 eV. Lastly, we focused on the computational evaluation of ligands for CuCl₂ volatilization. Various candidate ligands including acac, NacNac, dki, pyrim, dmap, and amd have been considered in this work.

[1] George, S.M. et al., ACS Nano, 2016, 10, 4889-4894. [2] Carver, C.T. et al., ECS J. Solid State Sci. Technol., 2015, 4, N5005-N5009. [3] Mohimi, E. et al., ECS J. Solid State Sci. Technol., 2018, 7, P491-P495. [4] Konh, M. et al., J. Vac. Sci. Technol. A, 2019, 37, 021004.

Molecular dynamics simulations were performed for a better understanding of the formation mechanisms and extent of damage in the underlying silicon (Si) material when a silicon nitride (Si₃N₄) layer above it was removed by atomic layer etching (ALE). ALE is an etching technique known to remove a surface layer of a material with atomic-level control and precision. On account of this, extensive research was focused on devising ways to obtain a uniform and stable etching performance with good selectivity to other materials, to cope with high demand from the semiconductor industry. However, though ALE is known to etch a very thin layer at a time, problems such as over-etching or damage on the underlying material layer might still occur during the fabrication [1]. Therefore, in this study, ALE of Si₃N₄ formed on a Si surface was examined, with the use of molecular dynamics simulation. As in experiments, the surface modification step was performed with the adsorption of CH₂F radicals, and the removal step was performed with the irradiation of the modified layer with Ar ions. To assist the removal of the remaining HFC species, O ions were introduced after the removal step. Recent experimental observations showed that the HFC species penetrated deeply into the underlying Si during the first ALE cycle [2]. Our simulation showed that the HFC and Si₃N₄ species penetrated the underlying Si material to a depth of around 4 nm, in the removal step with 300 eV of Ar ion incident energy. The higher the Ar ion incident energy was, the deeper penetration depths of the HFC were observed. Moreover, the bond distributions showed that those penetrated species formed chemical bonding with surrounding Si atoms. As a consequence, the extent of damage or the thickness of the amorphized Si layer was also found to be around 4 nm. The results also showed that physical sputtering tended to dominate the ALE process as the Ar ion incident energy increased, where a large number of monoatomic species left the surface during the removal step. The results suggest that the Ar ion incident energy has a significant contribution to the damage generation in the underlying Si material.

References

1.Tinacba, E., Isobe, M., and Hamaguchi, S., JVST A39, 042603 (2021).

2.Hirata, A., Fukasawa, M., Kugimiya, K., Karahashi, K., Hamaguchi, S., Hagimoto, S., and Iwamoto, H., DPS 2021 (2021).