Quantified Uniformity and Selectivity of TiO2 Films in 45‐nm Half Pitch Patterns Using Area‐Selective Deposition Supercycles

Area‐selective deposition (ASD) shows great promise for sub‐10 nm manufacturing in nanoelectronics, but significant challenges remain in scaling to ultrasmall dimensions and understanding feature‐dependent nonuniformity and selectivity loss. This work addresses these problems by simultaneously quantifying uniformity and selectivity for passivation/deposition/etch supercycles in 45 nm half‐pitch TiN/SiO2 line/space patterns. This work employs three selective processes that are uniquely suited for supercycle processing: dimethylamino‐trimethylsilane (DMA‐TMS) inhibition, TiO2 atomic layer deposition (ALD), and HBr/BCl3 plasma etch. The DMA‐TMS inhibition selectively passivates the SiO2 nongrowth surface without affecting deposition on the TiN and TiO2 growth surfaces. The plasma etch removes TiO2 defect particles at a faster rate than the conformal TiO2 film or SiO2 lines. Using three supercycles of this process, this work demonstrates 8 nm of TiO2 with 88% uniformity and ≈100% selectivity according to transmission electron microscopy (TEM), a 2×improvement in film thickness from previous reports in similar nanoscale patterns. Integrated consideration of uniformity and selectivity at specific feature scales will facilitate the effective design of selective deposition processes for nanoscale electronic devices.


Introduction
Area-selective deposition (ASD) techniques have recently grown significant research interest to enable sub-10 nm resolution in semiconductor manufacturing. [1][2][3] ASD relies on chemical differences on different regions of a substrate to deposit material at a different rate in each region, thus resulting in bottom-up nanopattern growth. [1,4] Deposition on the desired "growth" region occurs more quickly than on the desired "nongrowth" region and is typically achieved using techniques such as chemical vapor deposition or atomic or molecular layer deposition (ALD/MLD). [1,5] ALD and MLD utilize chemically selflimiting vapor-solid surface reactions of alternating precursors to deposit conformal, pinhole-free films with atomic-scale thickness control and high surface sensitivity. [6][7][8] These processes have successfully demonstrated ASD of many metals, [9,10] dielectrics, [4,[11][12][13][14] and organic materials, [5,15,16] resulting in a wide range of applications. For example, TiO 2 thin films have a high refractive index, visible and near-IR frequency transmittance, and high chemical stability for use as antireflection coatings, sensors, photocatalysts, and etch resistant layers in lithography stacks, [17][18][19][20] and is the focus of our work.
Comparisons between different ASD processes are enabled by a consistent definition of the selectivity fraction, shown in Equation (1), based on the relative surface coverage of material deposited on each surface. [21] Here S represents selectivity and and t represent surface coverage and film thickness, respectively, on the growth (G) and nongrowth (NG) surfaces.
passivation occurs when the nongrowth surface is chemically modified to make it less reactive toward the subsequent ALD process. Alkylsilyl groups have attracted recent attention as passivating groups in ASD for their increased thermal stability, smaller size, and easier process integration (vapor phase, shorter dose time) compared to traditional self-assembled monolayer passivation. [25][26][27][28][29] For example, dimethylamino-trimethylsilane (DMA-TMS) has demonstrated successful passivation of SiO 2 against TiO 2 , TiN, Ru, [30,31] and ZrO 2 . [32] DMA-TMS reacts efficiently with Si-OH surface groups, [33,34] achieving a TMS site density of ≈2 TMS sites nm −2 , [30] and thus increasing the hydrophobicity of the surface to inhibit ALD. However, even with inhibitors such as DMA-TMS, surface selectivity is eventually lost. Further improvement in selectivity has been obtained for some processes using defect mitigation strategies (e.g., etching) or cyclical processes that embed defect mitigation and/or passivation steps into the ALD sequence. [6,[35][36][37] These cyclical processes are especially suited toward improving selectivity on surfaces where undesired nuclei are periodically generated, such as has been recently demonstrated for TiO 2 ASD on DMA-TMS passivated SiO 2 . [38] DMA-TMS is uniquely suited for such a cyclic ASD process, as furthermore, it is relatively unreactive on many deposited materials (TiO 2 , TiN, and Ru), [31] and therefore may be useful to enable ASD on these surfaces versus SiO 2 . The low reactivity of DMA-TMS toward TiO 2 surfaces is in line with the lower surface acidity of TiO 2 . [34] Additionally, it has been attributed to more hydrogen bonding occurring on TiO 2 surfaces compared to SiO 2 , thus lowering the reactivity of vicinal Ti-OH sites with the inhibitor. [31,34,38,39] However, improving TiO 2 selectivity using DMA-TMS in a cyclic dep/etch cycle has not previously been explored. Therefore, investigating the potential for DMA-TMS to improve selectivity in cyclical ASD is highly desirable.
An additional outstanding challenge in ASD is successfully demonstrating selectivity on feature scales commensurate with current electronic devices. Mechanisms that enable ASD on large (>100 nm) patterns [1] will likely not directly scale to extremely small (<20 nm) features in current commercial devices. Moreover, as feature scales continue to shrink, controlling film uniformity in the vertical direction is essential for complex 3D structures, e.g. fabrication of complimentary field-effect transistors (CFET). [40] Thus, more detailed understanding on pattern-dependent selectivity and uniformity is critical. [41] Herein we take a significant step forward in sub-50 nm processing using a cyclical TiO 2 ASD process with DMA-TMS as an inhibitor that doubles the best previously achieved TiO 2 film thickness at this scale. Moreover, we specifically investigate selectivity and uniformity in the nanopatterns with a critical dimension (CD) of 45 nm, emphasizing the importance of focusing research efforts on feature-scale processing. In addition, we analyze the pattern dependence of the nonuniformity in the industrially relevant CD range between 2 and 60 nm. The insight obtained in this study provides a new benchmark in the field of nanoscale ASD and demonstrates how understanding fundamental mechanisms is crucial to improving process performance on industrially relevant scales.

TiO 2 Selective Etch with HBr/BCl 3 Plasma
In this work, we develop and describe a cyclical passivation + deposition + etch strategy that enables high-selectivity TiO 2 ASD in industrially relevant sub-50 nm patterned features, and utilize the dep/etch sequence to reveal mechanisms leading to featureand scale-dependent selectivity. The cyclical process is shown in Figure 1a, with TiN as the growth surface and SiO 2 as the nongrowth surface. The patterned surface is first exposed to DMA-TMS at 250°C for 300 s (DMA-TMS partial pressure ≈3 Torr, as described in Experimental Section), which reacts minimally on the exposed TiN, but reacts with hydroxyl sites on the SiO 2 to produce nonreactive Si-O-Si(CH 3 ) 3 sites that passivate the surface and impede TiO 2 nucleation. [31] The sample is then exposed to TiO 2 ALD using TiCl 4 /H 2 O at 150°C. Under these conditions on receptive (i.e., unpassivated) SiO 2 , the TiO 2 grows at 0.037 nm cycle −1 according to Rutherford backscattering spectrometry (RBS) results, as seen in Figure 1b. [38] The Ti aerial density determined from RBS is used to calculate equivalent TiO 2 film thickness assuming a film density of 3.72 g cm −3 . On the unpassivated surface, TiO 2 grows at ≈0.037 nm cycle −1 . After ALD, the sample is treated for a short time with a low-RF power BCl 3 /HBr/Ar plasma at 50°C, without applied substrate bias, that works to etch TiO 2 from both the desired growth and nongrowth regions. Controlling the etch time removes undesired nuclei on the nongrowth surface (i.e., DMA-TMS passivated SiO 2 ) without completely removing the TiO 2 film from the growth surface (i.e., TiN) or damaging the underlying substrate (SiO 2 ).
For the etch process, ALD TiO 2 films were exposed to the etch conditions for various times, and the thickness of film etched was measured by ellipsometry. As shown in Figure 1c, the etch depth increases approximately linearly with time, corresponding to an etch rate of 0.028 nm s −1 . We also tested the etch process on the passivated SiO 2 . For this test, to mimic the SiO 2 surface after the dep/etch sequence, a 50 nm PEALD SiO 2 layer on silicon was pretreated with DMA-TMS. The sample was then exposed to BCl 3 /HBr/Ar plasma etching, and the extent of SiO 2 removed was monitored versus etch time. The results in Figure 1c show that on this nongrowth surface, the rate of etching is ≈0.0017 nm s −1 , more than 10× smaller than on TiO 2 . The difference in etch rates provides an opportunity for a selective etch to aid in removal of undesired TiO 2 without significantly damaging the substrate.

SiO 2 Surface After Passivation, TiO 2 ALD, and Etching
To understand reactions on the nongrowth (SiO 2 ) surface, wafer pieces with thermal SiO 2 were treated with DMA-TMS followed by TiO 2 ALD (50-100 cycles) and BCl 3 /HBr/Ar plasma etch (15-90 s), and analyzed by X-ray photoelectron spectroscopy (XPS), water contact angle (WCA), scanning electron microscopy (SEM), and RBS. As shown in Figure 2a, the starting SiO 2 + DMA-TMS surface is hydrophobic (WCA = 102°), and the WCA is only slightly decreased (≈89°) after 75 cycles of TiO 2 ALD, consistent with a small amount of TiO 2 growth. Upon etching, the WCA decreases to 21°-25°, equal to that measured on the starting SiO 2 , indicating the etch step removes any deposited TiO 2 as well as the underlying passivation layer.   Indeed, SEM images in Figure 2b-d show that after DMA-TMS treatment, 75 cycles of TiO 2 ALD produce visible nuclei, and nuclei are not present after 15 or 45 s of etching. XPS results in Figure S1 and Table S1 (Supporting Information) also show removal of TiO 2 . For some samples, the etched surfaces were exposed again to 300 s DMA-TMS. As shown in Figure 2a, after 15 s etch, the DMA-TMS exposure reproduced a hydrophobic surface (WCA = 95°). The same trend is observed after 45 s etching, with WCA = 88°. This confirms that DMA-TMS can successfully react with the etched SiO 2 surface and restore the TMS surface termination. We note that further investigation is required to fully elucidate the effects of surface etching (e.g., etch time) on surface repassivation.

Cyclical Passivation + Deposition + Etch Process on Blanket Wafers
Using the HBr/BCl 3 plasma etch process described above, we evaluated several cycles of the fully integrated passivation + deposition + etch process. For this experiment we analyze an SiO 2 sample after 75 cycles of ALD only and compare it to samples treated as shown in Figure 2, (i.e., 300 s DMA-TMS passivation, 75 cy TiO 2 ALD at 150°C, and 15 or 45 s etch time). Results from RBS and WCA analysis are given in Table 1 and Figure S2 (Supporting Information), respectively. The sample exposed only to ALD shows substantial Ti deposition, consistent with facile nucleation and growth, whereas very little Ti is present after the cyclic passivation + deposition + etch, indicating impeded nucleation and removal of unwanted growth. The WCA results after the multistep sequence are also consistent with the single supercycle results in Figure 2. Based on these results, to explore feature dependence in ASD, the passivation + deposition + etch conditions were fixed at 300 s DMA-TMS + 75 cycles TiO 2 ALD + 45 s BCl 3 /HBr plasma etch, unless otherwise noted.

Demonstrating Selectivity on Nanopatterns
To analyze ASD in nanoscale features, 45 nm half-pitch SiO 2 /TiN line/space patterns were prepared using immersion lithography on full wafers and cut into ≈2 × 2 cm 2 coupons before transfer to the reactor for DMA-TMS passivation + ALD + etching using conditions developed above on blanket coupons. For this pattern, the TiN is expected to react minimally with the DMA-TMS passivation, and therefore be receptive to TiO 2 growth, whereas the DMA-TMS will readily passivate the SiO 2 regions and impede nucleation. [31,38] All samples receive the same ALD treatment (225 cycles) and etch time (135s), but some are processed using one supercycle (consisting of 300 s DMA-TMS passivation + 225 ALD cycles + 135 s etching) while others are processed using three supercycles (i.e., [300 s DMA-TMS passivation + 75 ALD cycles + 45 s etching] × 3). We note that TiO 2 is not inherently selective to SiO 2 , in accordance with previous results on blankets, [31,38] supporting the need for the DMA-TMS inhibitor to obtain good selectivity. To analyze the deposition and etching, some samples are removed from the reactor after deposition (without etching). Samples are analyzed using transmission electron microscopy (TEM) with energy-dispersive X-ray spectroscopy (EDS), SEM, and cross-sectional SEM images. Figure 3a,b shows TEM images of patterned samples after 225 ALD cycles and 225 ALD + 135 s etching, respectively, with the thicknesses of relevant layers reported in Table 2 and Table  S2 (Supporting Information). After ALD only (Figure 3a), the TiO 2 layer on TiN is ≈11.2 nm thick. This is consistent with ≈3 nm of native TiO 2 on the TiN, plus ≈8 nm of TiO 2 ALD formed at the expected 0.037 nm cycle −1 measured on blanket substrates (Figure 1b). The thickness is uniform and consistent for 11 different spaces ( Figure S3, Supporting Information). Less TiO 2 (≈6.5 nm on average) is also present on the exposed SiO 2 top surface and trench sidewalls, consistent with nucleation inhibition. After 135 s etching (Figure 3b), the average TiO 2 thickness on the top and side of the SiO 2 lines is ≈1.8 and ≈4.9 nm, respectively, indicating that TiO 2 islands on the top surface etch faster than on the sidewalls. On the TiN region, the etching reduces the TiO 2 thickness to ≈8.1 nm in the center of the feature, which is reasonably close to the 7.4 nm expected with an etch rate of 0.028 nm cycle −1 (from Figure 1c). Note that after etching, the TiO 2 film thickness in the TiN region is larger near the SiO 2 trench wall (whereas it was uniform before etch), consistent with shadowing during the plasma etch process.
Similar TEM images from samples prepared with three cycles of passivation + deposition + etch are shown in Figure 3c,d, where the sample in Figure 3c was removed before the final etch step. Before the final etch ( Figure 3c) the TiO 2 layer thickness on the TiN growth surface is ≈8.1 nm, which is consistent with the ≈8.8 nm expected using the deposition [38] and etching rates on blankets (Figure 1), including a 3 nm native TiO x layer. In this case, the nonuniformity of the TiO 2 thickness on TiN is somewhat more pronounced than after the single supercycle results  in Figure 3b (see also Figure S5, Supporting Information). The average height of the TiO 2 islands on the SiO 2 top surface and sidewalls is ≈1.7 and 2.0 nm, respectively, which is much thinner than the single supercycle sample (Figure 3b). After three full passivation + deposition + etch cycles, (Figure 3d), the average thickness of the TiO 2 layer on the TiN is ≈7.8 nm, which is reasonably close to 6.8 nm TiO 2 expected after 45 s etching of the 8.1 nm film. On the SiO 2 region, few isolated particles (≈1-3 nm in height) are observed on the sidewalls with TEM, and no defects are observed on the SiO 2 top surface. Also, the width of the SiO 2 feature is not affected by the etching (Table S2, Supporting Information). Figure 3e and Figure S3 (Supporting Information) show an extended broad view of the patterned sample after three cycles of (passivation + deposition + etch), demonstrating the consis-tency of the TiO 2 film thickness and selectivity across a large area of the pattern. For the results in Figure 3d, the three supercycle process enables ≈7.8 nm of selective TiO 2 growth on TiN, which is ≈2x larger than previously reported for TiO 2 ASD on similar patterns. [31] Figure 4a,b shows EDS scans collected from samples in Figure 3c and d, respectively. A Ti-containing (TiO 2 ) layer is apparent on the TiN. In Figure 4a before the final etch, some Ti signal is also present on the SiO 2 sidewalls, with no measurable Ti on the SiO 2 top surface. Within the TiO 2 layer on the TiN in each sample, small horizontal contrasting layers are observed, which may be related to changes in C content (as observed in Figure 4) or film crystallinity, although low temperature (i.e., 150°C) TiO 2 deposition is typically expected to produce amorphous films. The C observed within the TiO 2 film is attributed to some residual  TMS groups on the TiO 2 surface during the repeated passivation steps. [31,42] After the final etch step (4b), the thickness of the Ti layer on the growth surface has decreased (while maintaining the "U"-shaped profile), and no Ti is visible on the SiO 2 top surfaces or sidewalls. There is a small native SiO x layer formed on top of the TiO 2 film on the TiN growth surface, which may be attributed to residue from the slowly etched Si lines, residual inhibitor species, or Sicontaining residues from intermediate etch processes. Some Cl is observed within the TiO 2 film for both samples, attributed to residual TiCl 4 precursor incorporated into the deposited film. No Br ( Figure S4, Supporting Information) is detected, consistent with XPS results in Table S1 (Supporting Information). Additionally, we note that there is no increase in defectivity near the corners of the SiO 2 lines using this supercycle process with DMA-TMS, in contrast to previous TiO 2 selectivity studies in similar patterns. [23] Figure 5 shows top-view and cross-section SEM images of the samples presented in Figure 3. Significant TiO 2 particle defectiv-ity is observed for the single supercycle process on both the SiO 2 top surface ( Figure 5a) and sidewalls (Figure 5c). The line edge roughness (LER) appears quite high before etching for these samples, which is an indicator of island-type nucleation on the sidewalls. While the extent of observed particles and excess SiO 2 line width are significantly reduced after etching (Figure 5b,d), significant LER still remains, which can be problematic for future processing steps. On the contrary, for the three supercycle process, minimal TiO 2 particles are observed on the SiO 2 lines after 225 cycles TiO 2 deposition even before the last etch step (Figure  5e,g). In this case, the pattern edges are significantly smoother than in the standard process, consistent with less sidewall defectivity. After the last etch from the third supercycle, no TiO 2 particles are observed on the SiO 2 top surfaces and little to no defects are observed on the sidewalls (Figure 5f,h). Corresponding line widths are reported in Table S3 (Supporting Information). A more detailed, quantitative analysis of LER during ASD in patterns merits further investigation outside the scope of this work.
www.advancedsciencenews.com www.advmatinterfaces.de Figure 6. Film uniformity for the cyclical TiO 2 process developed here in patterns as a function of a) half-distance between pattern lines (x 1/2 ) and b) characteristic uniformity length (x d ). Vertical dashed lines correspond to the space half-width and characteristic uniformity length used in our experiments in Figure 3d. c) Thickness profiles for deposited TiO 2 film from images in Figure 3, with associated modeling parameters and uniformity provided in Table S4 and Figure S5 (Supporting Information).

Quantifying Uniformity in Nanopatterns
Using cyclical passivation-deposition-etch as a reliable ASD process in nanopatterns, we next investigate feature dependence of uniformity. To begin to quantify the uniformity of the resulting ASD film, we note that based on the experimental results in Figures 3 and 4, the film thickness is largest near a feature edge, and the thickness decays exponentially with the distance away from the feature. This trend is represented by Equation (2): where t(x) is the film thickness (nm) at any distance (x, nm) away from the feature edge (in this case the SiO 2 line). The values t 0 and t ∞ are the film thicknesses at the SiO 2 line (e.g., x = 0) and at a point far away from any edge or feature, respectively. The parameter x d is a "characteristic uniformity length" representing the extent of thickness change with distance away from the feature, where higher x d indicates a more rapid change in thickness (from t o to t ∞ ) when moving away from the feature edge. Parameters for each image from Figure 3 are provided in Table  S4 (Supporting Information), with overlaid thickness profile fits and TEM images shown in Figure S5 (Supporting Information). We assume that the thickness profile is symmetric from the center of the space in this analysis.
Furthermore, the net uniformity, U, of a nonplanar film can be defined as: where is the standard deviation of film thicknesses (nm), and t avg is the average film thickness (nm). Following this definition, if the thickness of a film is independent of location, then approaches zero and U = 1, i.e., uniformity equals 100%. By assuming a value for one-half of the feature separation distance, x 1/2 , (i.e., one-half the distance between SiO 2 lines), Equation (2) can be used to determine and t avg . Therefore, we can predict how U will depend on t o , x d , and x 1/2 for the TiCl 4 /H 2 O process.
Results are shown in Figure 6. In this figure, the space half-width and characteristic uniformity length corresponding to our experimental patterns (Figure 3d, t 0 = 12.4 nm, t ∞ = 7.8 nm, and x d = 6 nm) are indicated by the vertical dashed lines. In these plots, U is calculated from Equations (2) and (3), where t avg is determined from Equation (2) using the experimentally determined thickness values, t 0 and t ∞ (listed in Table S4, Supporting Information). Figure 6a is a plot of uniformity versus feature spacing dimension for various values of x d , and Figure 6b plots uniformity versus x d for various feature spacings. The uniformity for etched samples is similar (≈88%) after the final etch for both one and three supercycles (Table S4, Supporting Information). From Figure 6a,b, for larger pattern dimensions, uniformity is higher when x d is small, consistent with the nonuniformity being localized over only a small fraction of the deposition region. For smaller pattern dimensions, uniformity is higher when x d is large, consistent with a more gradual change in film thickness over the width of the pattern. The lowest uniformity is obtained when pattern dimension is similar to x d , because in this case the film thickness is changing significantly over a large portion of the pattern. Thus, the best uniformity is obtained with smaller x d values for large patterns, and with larger x d values for small patterns. We note the geometric constraint in the observational model as x d approaches zero that causes a significant drop in uniformity (resulting in a local minimum) for very small pattern dimensions. This result demonstrates the potential negative impact of shadowing effects during etching for very small feature sizes and the limitations of the current model. However, the model provides meaningful insight for larger x d values or for small x d values with practical pattern dimensions (i.e., larger than 1 nm half-distance between features). Figure 6c shows the thickness profiles from TEM images (Figure 3) calculated from Equation (2) and using thicknesses from Table 2 and Figure S4 (Supporting Information). From Figure 6c and Table S4 (Supporting Information), we see that film uniformity decreases significantly after etching for samples processed with only one supercycle (e.g., etch periodicity = 1). In contrast, when increasing the etch periodicity (i.e., after three supercycles with a more frequent, shorter etch step), the uniformity is maintained and slightly improves after the final etching step. The improvement in uniformity for samples processed with more periods of passivation/deposition/etch supercycles is attributed to fewer particles on the sidewalls during this processing, leading to a larger area of the bottom film that is continuously exposed to the etch (i.e., resulting in a similar etch rate across the entire film). On the other hand, increased TiO 2 sidewall deposition (e.g., for longer deposition cycles between etching, Figure 3a,b) results in shadowing of a larger portion of the bottom film during the etching process, causing a lower effective etch rate at the feature edge compared to the feature center. Therefore, to achieve a more uniform film with high selectivity, supercycles with more frequent etching (i.e., higher periodicity) are desirable, consistent with the TEM observations for one versus three supercycles (Figure 3).
Overall, these results indicate the large impact of pattern dimension and processing on uniformity. Note that future work could relate the uniformity model discussed above to supercycle processing conditions to predict the deposited film thickness profile, for example using a nucleation and growth model. As one example, t ∞ and t 0 could be calculated from the ALD growth rate as well as the etch rate at the feature center and edge, respectively. Similarly, x d could be related to the difference between t ∞ and t 0 , the feature width/height, and etch periodicity. This predictive modeling could facilitate supercycle process design to optimize uniformity without compromising selectivity.

Recovery of Selectivity by Supercycles with DMA-TMS and Selective Etching
Understanding the impact of etching on the nongrowth surface (including effect of etch reactants, etch time, residue, and changes to surface chemistry and active site density) is crucial for design and optimization of supercycle ASD processes. In this work, etching the DMA-TMS passivated surface with TiO 2 nanoparticles not only removes the TiO 2 nanoparticles but also degrades the TMS layer to expose the original Si-OH surface termination, as indicated by the drop in contact angle after etching observed in Figure 2 and Figure S2 (Supporting Information). This could occur, for example, by etching of the TMS layer by HBr/BCl 3 via formation of volatile etched species such as SiCl x (CH 3 ) y , SiBr x (CH 3 ) y , SiCl x , SiBr x , CCl x , or CBr x , resulting in exposure of the Si-OH substrate surface. The smaller contact angle (Figure 2 and Figure S2, Supporting Information) after reapplying the inhibitor to the etched surface is attributed to residual etch contaminants (e.g., B) or slight damage to the slowly etched SiO 2 surface (e.g., decreased OH content), as observed by XPS in Figure S1 (Supporting Information), which could provide fewer sites for TMS groups to bind and passivate the surface. In order to make the repassivation more efficient, one could consider adding an in situ post-etch surface conditioning aiming to remove residual etch contaminants and/or rehydroxylate the surface. Further investigation into the affect of HBr/BCl 3 etching on the TMS-terminated SiO 2 surface would benefit the repassivation effectiveness and hence improve selectivity. Importantly, the etch process reliably removes TiO 2 nanoparticles on the passivated surface without completely removing the deposited, conformal film on the growth surface, as confirmed by XPS, TEM, and EDS results in Figure S1, Figures 3, and 4, respectively. TEM (Figure 3), TEM EDS (Figure 4), and SEM ( Figure 5) images are consistent in demonstrating significantly less TiO 2 on the passivated surfaces (i.e., SiO 2 top-surface and sidewalls) for depositions that incorporated periodic short etches instead of one long etch. This is expected if the etch removes smaller particles more easily than larger particles, and explains the lower TiO 2 particle coverage on SiO 2 lines (TEM, Figure 3) and the thinner SiO 2 linewidths (SEM, Figure 5) during the three supercycle process compared to the single supercycle process. While higher periodicity has demonstrated reduced defects and improved selectivity in this case, it is also possible that the composition of the deposited film (i.e., TiO 2 ) is affected by the supercycles (as evidenced by the contrast layers in TEM in Figure 3c,d). It will be important to further investigate the potential impact of supercycles on film properties (e.g., dielectric constant, refractive index, etch resistance, or crystallinity) for applications.
Our previous work demonstrated that the TiO 2 nuclei are continuously generated on DMA-TMS passivated SiO 2 during ALD, and hypothesized that periodic removal of generated defect sites could improve selectivity. [38] Together, the controllable TiO 2 etch and selective DMA-TMS passivation step are well-suited for such a cyclical ASD process. Our current work confirms this hypothesis, as the periodic, load-dependent etching removes the generated TiO 2 particles while they are still small. This enables successful repassivation of the surface and leads to a 2× improvement in selectivity in nanopatterns compared to previous TiO 2 ASD in similar patterns. [23,31] The ≈8 nm TiO 2 film deposited in this work is similar in thickness to other TiO 2 supercycle ASD processes, while we expand our results to smaller pattern sizes and quantify uniformity in the patterns. [6] Our results emphasize the importance of cycling the DMA-TMS passivation, deposition, and etch steps in improving selectivity, and provide an important example of how greater understanding on selectivity loss can improve ASD strategies in sub-50 nm patterns. Further optimization of the etch step and supercycle periodicity are desired to further reduce defectivity.

ASD Mechanism and Uniformity in Nanopatterns Versus Blankets
The overall deposition and etching processes described here behave somewhat differently on 3D patterns versus planar blanket substrates. This has important implications in transferring research insights to commercial applications, thus we specifically discuss relevant selectivity mechanisms on the nanopatterns. For the patterned samples shown in Figure 3, the same net deposition and etching time result in approximately the same TiO 2 film thickness on TiN, regardless of using one or three supercycles. Thus, the deposition and etch rates on the patterns (Figure 3) are consistent with those measured on blanket wafers ( Figure 1). However, for both patterned samples, the TiO 2 film on the TiN region is not completely uniform, with a slightly thicker film on the edges adjacent to the SiO 2 lines.
Since the "U"-shape is more pronounced after the etching step, the nonuniformity is attributed to slower TiO 2 etching in the region adjacent to the SiO 2 due to shadowing effects, rather than diffusion of adsorbed species from the passivated sidewalls down to the TiN growth surface. [41] The lack of diffusion is further supported by the lack of a TiO 2 "depletion zone" near the SiO 2 /TiN interface (TEM images in Figure 3), which would be expected if Ti-adspecies were diffusing away from the passivated SiO 2 sidewalls and onto the TiN (thereby causing the TiO 2 film thickness to be larger at the pattern edges). Results from Figure 6 demonstrate how this nonuniform profile depends greatly on the pattern dimensions (feature spacing and height). Depending on the x d value, uniformity may either increase or decrease with smaller pattern dimensions. Thus, characterizing the thickness profile and determining x d for a given process will be an important indicator for the success of further device scaling.
We can also use CD measurements from SEM images to compare TiO 2 etch rates on the sidewalls during the one and three supercycle process, as supported by consistent line width measurements between TEM ( Figure 3) and SEM ( Figure 5). We note the ongoing challenges in direct quantitative comparisons due to limitations in metrology for ASD processes. For example, slight tapering of the SiO 2 lines is observed in the TEM images (Figure 3), which may cause deviations in CDs measured between TEM ( Figure 3) and SEM ( Figure 5).

Conclusion
This work demonstrates that periodic (passivation + deposition + etch) ASD processing can be extended to industrially relevant 45 nm half-pitch patterns and the resultant film uniformity in nanopatterns can be quantified. We leverage the unique capabilities of the DMA-TMS inhibitor to develop a passivation + deposition + etch supercycle process that greatly enhances existing selectivity and reduces LER on patterns. We report ≈8 nm of selective TiO 2 ALD with 88% uniformity and 100% selectivity on 45 nm half-pitch SiO 2 /TiN line/space patterns, which is a factor of 2× improvement in TiO 2 film thickness at this scale, with lower defectivity on the pattern sidewalls, top surfaces, and corners compared to previous reports. Furthermore, we demonstrate that uniformity varies significantly with feature scale, and is expected to improve with shrinking patterns for the TiO 2 ASD process presented here. Thus, our work establishes the importance of understanding feature-dependence of uniformity to improve selectivity for electronic device-scale patterns. For the TiO 2 process described here, additional refining of the passivation, deposition, and etch parameters, including investigating nanolaminate behavior, could further improve selectivity and uniformity control on patterned substrates and 300 mm wafers. The supercycle process not only affects selectivity and uniformity, but it can also affect the composition, structure, and as such, properties of the deposited material. The TiO 2 film may be interesting for potential etch stop layer applications, but further work is required to characterize the etch properties of this film. These results are expected to extend to additional industrially relevant materials. The observational uniformity model provides a starting point for guidance on supercycle design, suggesting that higher periodicity of passivation/deposition/etch sequences will result in better selectivity and uniformity. Future work is needed to fully inte-grate known process parameters (e.g., growth rate, etch rate, and feature size) into the model and enable predictive capabilities for uniformity and selectivity.

Experimental Section
Materials and Processing: In this work, all processing steps were performed on 300 mm industry-compliant tools. Silicon oxide substrates were prepared using plasma-enhanced ALD (PEALD) to deposit 17 nm SiO 2 in an ASM Eagle 12 reactor at 75°C on 300 mm Si(100) wafers. The resulting dielectric surface had 2.5 OH sites per nm 2 . [43] For DMA-TMS exposure, a Tokyo Electron Ltd (TEL) LK-R chamber installed on a Tactras platform was first filled to 5 Torr with 500 sccm DMA-TMS and 350 Torr N 2 (DMA-TMS partial pressure = 3 Torr), as described previously. [31] Then, samples were exposed to this DMA-TMS environment for 300 s at 250°C (resulting in ≈900 Langmuir total DMA-TMS exposure). TiO 2 ALD was performed in a Polygon 8300 EmerALD chamber at 150 or 300°C at a pressure of 5 Torr using TiCl 4 and H 2 O. TiCl 4 and H 2 O pulse and purge times were used that resulted in well separated, saturated reactions for TiO 2 ALD on regular flat substrates such as SiO 2 . Wafers were degassed for 1 min before deposition started. TiO 2 etching occurred in a TEL T4 chamber installed on the same TEL Tactras platform as the LK-R chamber used for the passivation step. Etching was conducted by exposing the wafers to BCl 3 /HBr plasma (100 W, applied to the top electrode) at 50°C chuck temperature and a pressure of 100 mTorr. Gas flow during etching consisted of 150 standard cubic centimeters per minute(sccm) Ar, 36 sccm BCl 3 , and 30 sccm HBr. For process optimization, etch rates were determined by means of TiO 2 and SiO 2 coupons (3 × 3 cm 2 ) glued on a SiO 2 -coated 300 mm carrier wafer. Subsequent process transfer to full wafers (pattern tests) did not indicate major changes in etch rates.
Patterning: Patterns were constructed from TiN and SiO 2 lines following a previously reported procedure. [44] Briefly, this process entails physical vapor deposition of 15 nm TiN from sputtering a Ti target in N 2 ambient onto a Si wafer. Then, 75 nm SiO 2 was deposited via PEALD on top of the TiN substrate, followed by i-line immersion lithographic patterning. The final patterned structure resulted in alternating 40 nm SiO 2 lines (≈70 nm tall) with 50 nm TiN trenches, giving an overall pitch of ≈90 nm. During passivation/deposition/etch processing on these wafers, TiO 2 thicknesses were predicted based on the expected steady GPC and etch rate on blanket TiO 2 (for TiN spaces) or passivated wafers (for SiO 2 lines). For the passivated SiO 2 lines with 225 cycles deposition, the expected thickness was interpolated between the data points at 200 and 300 cycles.
Characterization: The TiO 2 films were characterized with several physical and chemical techniques. RBS was conducted using a 1.523 MeV He + incoming ion beam at a scattering angle of 170°, a tilt angle of 11°, and 20 nA beam current to quantify total Ti content deposited. WCA measurements were performed on a Dataphysics OCAH 230 tool using 1 μL deionized (DI) water droplets, with the average value of five measurements reported and the standard deviation used as error. Etch depth of the TiO 2 and SiO 2 layers was determined using spectroscopic ellipsometry (SE) for 5 points on a KLA Tencor F5-SCD spectroscopic ellipsometer.
XPS was used to analyze surface chemistry after passivation and etching. This was conducted in angle resolved mode using a QUANTES instrument from Physical electronics at an angle of 65°from the surface. Measurements utilized a 1486.6 eV monochromatized Al K X-ray source with a 100 μm spot size. Each XPS scan was calibrated to the C 1s peak at 284.8 eV. Sensitivity factors specific for each instrument were used to convert peak areas to atomic concentrations. As a result of this, it is possible that the concentrations deviate from reality in the absolute sense (generally not more than 20% relative). Unless otherwise specified, all data presented are for the 45°exit angle.
SEM was used to visualize TiO 2 growth in patterns and verify dimensions of the patterned substrates. This was performed on an FEI Helios 460 microscope with 3 kV beam energy and 0.10 nA beam current. TEM was performed on a Tecnai F30 ST (FEI) tool with an FEG electron source at 300 kV in TEM mode. Prior to TEM imaging, samples were prepared with a dual beam FIB/SEM Nova600i (FEI) and Ar Ion miller PIPS (Gatan).

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.