ALD/ALE 2026 Session ALE1-TuM: Wet ALE and ALE Modeling

We demonstrate a hybrid dry-wet quasi-atomic layer etching (Q-ALE) process using tungsten as a model case, combining a self-limiting O₂ plasma oxidation (modification) step with a wet-chemical oxide removal step in 1 M HCl. In the study of the modification step, O₂ plasma oxidation was selected for its self-limiting behavior and benchmarked against wet oxidation, which instead produces highly soluble oxides in aqueous solution. The removal step was quantified by ICP-MS, revealing an initially high etch rate that decreases and stabilizes to a background value. This behavior is attributed to the rapid dissolution of plasma-generated bulk WO₃, followed by progressively slower removal as the interface approaches a suboxide-rich layer, and finally a steady-state regime governed by continuous re-oxidation and metal dissolution. Although the background etch rate is very slow (0.03 Å·min⁻¹), in Q-ALE operation the removal step is intentionally stopped before reaching this background regime, enabling controlled and selective material removal. These findings are further supported by post operando XPS, TEM and conductive AFM analyses. The dry-wet Q-ALE sequence was demonstrated on both PVD blanket and ALD-patterned tungsten wafers using existing 300-mm fab tools, achieving ~8 Å per cycle. These results highlight lab-to-fab scalability. More broadly, the combined dry-wet ALE framework provides a versatile platform for layer-by-layer etching of technologically important materials, particularly where purely wet oxidation is not self-limiting (or is chemically challenging), while wet removal offers strong oxide-to-metal selectivity.

Molybdenum (Mo) has gained significant attention from semiconductor industries for its applications on logic BEOL, buried power rails, and 3 D NAND. Some of these applications require partial etch back of Mo where post-etch morphology is critical to device performance. Wet atomic layer etching offers material removal with Angstrom-level precision at or near room temperature and at ambient pressure by utilizing two sequential self-limiting steps. In the first step, the Mo surface is exposed to an oxidizing solution to form self-limiting surface passivation. In the second step, the modified Mo surface is selectively removed via dissolution in suitable chemistry. Etching of polycrystalline Mo is susceptible to surface roughness increase and uncontrolled etch rate due to solubilization of modified surface products in oxidizing solutions. Here, we present a novel method for controlled Mo etch with improved post-etch surface morphology via ligand assisted surface modification of metallic Mo in aqueous oxidizing solution. A ligand binds with the metal center to form a complex surface product that is insoluble in an aqueous oxidizing solution, preventing the continuous Mo background etch, and resulting in preserved post-etch surface smoothness.

Cyclic etch experiments were carried out by dipping Mo coupons in an oxidizing solution with varying ligand concentration followed by selective removal of modified layer in a low concentration dissolution chemistry. Mo etch rate decreases with increasing ligand concentration. The etch rate, however, can be enhanced via surface oxidation at an elevated temperature (Fig. 1). For the same ligand concentrations, Mo etch rate can be greatly improved by tuning the oxidizer concentration (Fig. 2). A significant enhancement in Mo etch rate [from ~0.12 nm/cycle with p% oxidizer +250 mM ligand to ~0.43 nm/cycle in 5p% oxidizer+250 mM ligand] with increased oxidizer concentrations suggests the formation of thicker surface oxide as surface passivation. For higher oxidizer concentration, Mo ER is independent of ligand concentrations under study (Fig. 2). The measured RMS roughness [~ (0.63±0.04) nm] of the post-etch Mo coupon with higher ligand concentration is the same as RMS roughness of [~ (0.62±0.03) nm] of the reference coupon (Fig. 3). We attribute the preserved morphology in post-etch Mo coupon with higher ligand concentration to the suppression of continuous Mo etch due to the formation of a stable metal complex that is insoluble in oxidizing solution. The ability to remove Mo at a substantial rate with atomic-level precision and preserved surface morphology in post-etch coupons using less harsh oxidizer and ligand in an aqueous solution near room temperature may provide a cost-effective alternative solution to recess Mo in the industrial scale.

References

[1] J. R. Vella, Q. Hao, M. A. I. Elgarhy, V. M. Donnelly, and D. B. Graves, “A Transient Site Balance Model for Atomic Layer Etching”, Plasma Sources Sci. Technol., 2024, 33, 075009.

It is well established that during the thermal and plasma-assisted atomic layer etching of silicon nitride – a process widely used in the semiconductor industry – the main solid by-product is ammonia fluorosilicate (AFS) salt, if the feed gas contains both hydrogen and fluorine. In fact, such chemistries initially convert silicon nitride into AFS salt, which subsequently decomposes into gaseous by-products. During high-aspect ratio etching, the formation of AFS significantly affects the overall process performance by influencing critical characteristics such as etch rate, selectivity, and feature shape distortion, which are vital for semiconductor manufacturing.

Despite this, the thermodynamic properties of AFS – such as its decomposition temperature and the dependence of this temperature on pressure – have not been reliably determined. To address this gap, we performed calculations of standard Gibbs free energies for various crystalline structures of AFS using density functional theory (DFT). The distinctive feature of AFS is that this ionic salt decomposes and sublimates through distortion of ionic bonds, bypassing the liquid state; therefore, DFT level of approach is required to accurately describe such complex chemical bond transformations. The calculations show that the AFS decomposition temperature varies in a wide range depending on its crystal structure and decreases significantly with increasing water content, in agreement with our previous publication [1].

In addition, our simulations show that similar ionic salts with the similar properties – such as NH₄F, NH₅F₂, (NH₄)₂SiF₆∙NH₄F, NH₄BCl₃F, (NH₄)₂TiF₆, and (NO)₂SiF₆ – can also form during etching. These compounds can be classified into reactive and unreactive salts. The simulations predict that some of these salts are stable only at temperatures below 0 ^◦C. Given that low-temperature etching is an increasingly important direction in the semiconductor industry, such salts may play a more significant role in future etching processes.

[1] T. Lill et al., “Low-temperature etching of silicon oxide and silicon nitride with hydrogen fluoride,” Journal of Vacuum Science & Technology A, vol. 42, no. 6, p. 063006, Dec. 2024, doi: 10.1116/6.0004019.

Atomic Layer Etching (ALE) is a key process for achieving conformal and selective material removal in advanced semiconductor manufacturing, particularly for high-aspect-ratio (HAR) structures in 3D memory and logic devices. Owing to its self-limiting nature, ALE enables uniform etching over extended process times. However, predictive modeling of feature-scale behavior remains challenging due to the interplay of surface modification, diffusion, and volatile etching mechanisms.

In this work, we present a feature-scale simulation framework for cyclic ALE of SiO₂ using HF/NH₃ co-dosing in a chemical-oxide-removal (COR) process. The model is implemented in the in-house process simulator ViennaPS [1] and couples geometry-agnostic Monte Carlo ray tracing for gas-phase transport with a level-set method for surface evolution. During reactant pulses, precursor transport is modeled to be ballistic with cosine angular distributions, appropriate for plasma-less processes. Local surface fluxes obtained from ray tracing are converted into surface concentrations that drive surface reactions. Formation of ammonium fluorosilicate (AFS) is described using a parallel-resistance approach, enabling simultaneous treatment of surface-limited reactions and precursor diffusion-limited reactions through the modified layer. In addition, direct volatile etching of SiO₂ by HF is incorporated via a thickness-dependent coverage function which activates under the conditions of incomplete AFS surface coverage.

Model parameters are calibrated against experimentally reported low-pressure-TEOS thickness data by Hagimoto et al. [2] for planar substrates over a range of precursor pulse times. The calibrated model is subsequently applied to HAR trench structures with aspect ratios up to 20 to investigate the influence of precursor partial pressures and sticking coefficients on etch profiles. The simulations reveal transitions from stoichiometry-limited surface modification to diffusion-limited growth regimes, as well as the emergence of inverse aspect-ratio-dependent etching under asymmetric HF/NH₃ sticking conditions, highlighting the sensitivity of ALE conformality to surface kinetics. To address the computational cost of simulating large numbers of ALE cycles, we introduce a cycle-skipping strategy in which surface concentrations are recalculated only after several cycles of constant-rate surface propagation. Comparison to gauge simulations, where surface concentrations are reevaluated for each ALE cycle, reveals that up to five cycles can be omitted while maintaining maximum profile deviations below 1% along the surface, corresponding to a significant reduction in computational cost.

Overall, the presented framework provides a physically motivated and geometry-agnostic approach for predictive ALE modeling in HAR structures capturing different reaction pathways and offers insight into the process windows governing conformality, selectivity, and throughput in HAR applications.