Surface Smoothing by Atomic Layer Deposition and Etching for the Fabrication of Nanodevices

Abstract

In many nano(opto)electronic devices, the roughness at surfaces and interfaces is of increasing importance, with roughness often contributing toward losses and defects, which can lead to device failure. Consequently, approaches that either limit roughness or smoothen surfaces are required to minimize surface roughness during fabrication. The atomic-scale processing techniques atomic layer deposition (ALD) and atomic layer etching (ALE) have experimentally been shown to smoothen surfaces, with the added benefit of offering uniform and conformal processing and precise thickness control. However, the mechanisms which drive smoothing during ALD and ALE have not been investigated in detail. In this work, smoothing of surfaces by ALD and ALE is studied using finite difference simulations that describe deposition/etching as a front propagating uniformly and perpendicular to the surface at every point. This uniform front propagation model was validated by performing ALD of amorphous Al₂O₃ using the TMA/O₂ plasma. ALE from the TMA/SF₆ plasma was also studied and resulted in faster smoothing than predicted by purely considering uniform front propagation. Correspondingly, it was found that for such an ALE process, a second mechanism contributes to the smoothing, hypothesized to be related to curvature-dependent surface fluorination. Individually, the atomic-scale processing techniques enable smoothing; however, ALD and ALE will need to be combined to achieve thin and smooth films, as is demonstrated and discussed in this work for multiple applications.

Figure 1

Illustration of the UFP model in 1D where the white arrow indicates the direction of propagation for growth and etch for nine steps of the model. The starting surface is shown in gray. Each step of the model indicates one ALD or ALE cycle. (a,b) Conformal growth on sine-shaped surfaces with different spatial frequencies. (c) ALD on a randomly rough surface. (d) ALE on a randomly rough surface. In both (c,d), it can be seen that the roughness is reduced, with small-scale features being removed from the film.

Figure 2

(a) AFM heightmap of the 98 nm ZnO sample, which was used as an input for the model. (b–d) AFM heightmaps measured after Al₂O₃ ALD of different thicknesses. (e–g) Heightmaps calculated using the UFP model for films of the same thickness as in (b–d).

Figure 3

(a) Comparison of the RMS roughness as a function of deposited Al₂O₃ thickness on the 47 nm ZnO sample from experimental and model data. (b) PSD from the AFM data compared to PSD from the model for different deposition thicknesses. The measured AFM data deviates in the high-wavenumber range due to contribution of the inherent roughness (see Section S.A). The PSD is corrected for the contribution of the noise using a constant value for the noise (see Section S.C).

Figure 4

(a) RMS roughness after ALE using TMA and SF₆ plasma on 41 nm Al₂O₃ on 47 nm of ZnO. The curvature-dependent propagation (CDP) model is fitted using ε = 1.5 × 10^–9 m. (b) The PSD of 41 nm Al₂O₃ ALD on 47 nm ZnO, the PSD after 35 nm of ALE using TMA and SF₆ plasma, and model data of 35 nm of ALE using ε = 1.5 × 10^–9 m for the CDP model and ε = 0 m for the UFP model. The deviation between the model and experimental data for high wavenumbers is caused by the measurement noise.

Figure 5

(a) Fluorination of a flat surface. Fluorine radicals adsorb on the surface and diffuse through the fluorinated layer, reaching the underlying material where it reacts to fluorinate the material. (b) On a convex surface, there is more surface area for radicals to absorb, relative to the surface area of the nonfluorinated material interface. This results in relatively faster fluorination of the material compared to a flat surface. The convex surface is also less constrained by the surrounding material and is thus able to more easily undergo the surface expansion required during fluorination. (c) On a concave suface, the opposite effects are true, leading to relatively slower fluorination.

Figure 6

(a) Description of the ZnO/Al₂O₃ stack. (b) TEM image and EDX map of the stack. The interface between layers 2 and 3 is visibly smoother than between layers 4 and 5. (c) Roughness values measured by AFM and TEM analysis. Both techniques follow a similar trend in roughness and show that layer 2 is the smoothest.

Figure 7

AFM height maps of samples prepared by (a) 52 nm ZnO, (b) 52 nm ZnO and 41 nm of Al₂O₃ deposited using ALD, and (c) 52 nm ZnO and 41 nm of Al₂O₃ deposited using ALD followed by 34 nm of Al₂O₃ etching using ALE. (d) The PSDs of samples (a–c), where the vertical dashed line indicates 50 nm. Rough features below 50 nm (wavenumbers above 2 × 10^–1 nm^–1) are smoothed significantly more than larger features. Film (e) roughness and (f) thickness as a function of ALD + ALE cycles plotted alongside model predictions over multiple ALD + ALE supercycles.

Figure 8

(a) Possible method to create thin and closed films by first depositing a thick film, to ensure film closure, and then to use ALE to reduce the film thickness. This figure was produced using the UFP and CDP models, with 20 randomly placed hemispherical nucleation sites of 0.4 nm initial radius. (b) Surface coverage, (c) film thickness, and (d) RMS roughness as a function of the number of cycles. The roughness peaks during island coalescence and decreases afterward for ALD, after which ALE reduces the RMS roughness even further.