SiO2 thin film growth through a pure atomic layer deposition technique at room temperature

In this study, less contaminated and porous SiO2 films were grown via ALD at room temperature. In addition to the well-known catalytic effect of ammonia, the self-limitation of the reaction was demonstrated by tuning the exposure of SiCl4, NH3 and H2O. This pure ALD approach generated porous oxide layers with very low chloride contamination in films. This optimized RT-ALD process could be applied to a wide range of substrates that need to be 3D-coated, similar to mesoporous structured membranes.


Introduction
Silicon dioxide (SiO 2 ) and more generally ultra-thin oxide lms have been extensively described as good components for modern nanotechnologies such as dielectric materials in silicon microelectronic devices, 1,2 anticorrosion lms 3 or nonexhaustive applications of nanoscale lms in catalysis. The environment-and human-friendly nature of SiO 2 induces its wide use in protective layers for antisticking, antifogging, selfcleaning or water repellency. [4][5][6][7] Various techniques such as chemical vapor deposition, 8 lithographic patterning, 9 electrochemical deposition 10 or sol-gel 11,12 were investigated to prepare superhydrophobic SiO 2 by tuning surface roughness or energy. SiO 2 is consistently known for its application in protective or gate insulator coatings 13 and interfacing high-k (ref. [14][15][16][17][18][19] or surface passivation materials. [20][21][22][23] The increased demand for transparent active materials at the nanoscale justify the need for a deposition technique compatible with sensitive pre-deposited underlying layers, exible plastic devices or high aspect ratio substrates. [24][25][26][27][28] Therefore, atomic layer deposition (ALD) is considered as one of the most suitable techniques for its performance in terms of sub-nanometer thickness control and penetration coating into deep trenches or mesoporous structures.
SiO 2 thin lms obtained through ALD have widely been described as binary surface reactions dealing with various types of precursors Si(C 2 H 5 O) 4 , 29 SiCl 4 (ref. 13 and 30-33) or Si(NCO) 4 (ref. 34) with H 2 O; CH 3 OSi(NCO) 3 (ref. 35 and 36) with H 2 O 2 ; Si(NCH 3 ) 2 ) 4 , ((CH 3 ) 2 N) 3 SiH 37 or SiH 2 (NEt 2 ) 2 (ref. 38)). Besides the necessity to work at high temperatures (i.e. >100-350 C), most of the reactions require large reactant exposure of $10 9 L (1 L ¼ 10 À6 Torr s) with a growth rate of 59 $1-2Å per cycle. 39 One approach to decrease the deposition temperature and the level of contamination is to use plasma-enhanced-ALD (PE-ALD). 40,58,59 It is possible to decrease the temperature below 50 C, and these SiO 2 lms are described as excellent candidates for thin lm encapsulation in organic devices or TFT gate insulators due to the absence of impurities and good electrical properties. [41][42][43]60 Currently, the use of amino ligands as precursors leads to promising results, even on large surfaces; however, a nal annealing step at 900-1000 C is required to decrease interface defects or carbon contamination. 44,45 Many efforts have been done, in terms of parameters and choice of precursors, in order to optimise ALD process, for the growth of SiO 2 coatings at high temperatures. Nevertheless, it is commonly agreed that there is a strong interest in the development of the process at room temperature. George et al. described the atomic layer-controlled growth using SiCl 4 and H 2 O many times. 13,31,44,46 They demonstrated that a reaction catalyzed using Lewis bases such as pyridine (C 5 H 5 N) or ammonia (NH 3 ) avoids large precursor ow rates and can only occur close to room temperature. Nevertheless, in these studies, C 5 H 5 N or NH 3 were never really considered as "precursors". The proposed mechanism, which took into account the hydrogen bonding between the Lewis base and either the SiOH* surface species or the H 2 O reactant, was studied by considering the global residual pressure of the continuous ow of the catalyst. Moreover, the secondary reaction of the catalyst reservoir (continuous ow), available in the reactor with HCl as the byproduct, drastically increased the probability of the inclusion of contaminants in the lm. Therefore, a sequential approach could enhance the quality of the lm and the understanding of the role of the catalyst. This paper describes a pure ALD study of SiO 2 using the optimized sequential exposure of SiCl 4 , H 2 O and NH 3(g) precursors at room temperature. The ALD mode was conrmed by tuning the exposition of each precursor and the related purges. The initiation of the exposition was followed by using the residual gas analysis (RGA) mass spectrometer. Atomic growth control was investigated by the in situ Quartz Crystal Microbalance (QCM) and X-ray Photoelectron Spectroscopy (XPS), Dynamic-Secondary Mass Ion Spectroscopy (D-SIMS) and X-ray Diffraction (XRD) post-characterizations. A comparison between our investigation and the state-of-the-art of low temperature ALD SiO 2 synthesis revealed the possibility to deposit ultra-thin lms with very low contaminations at room temperature. The lm conformality is shown and the capability of this optimized binary reaction, to be used on various types of temperature-sensitive supports with high aspect ratios, is conrmed.

Materials and methods
ALD processes were carried out in a TFS200-Beneq reactor in the planar conguration at a base pressure of 0.3 mbar. SiO 2 thin lms were then deposited on Si substrates, preliminary prepared by a standardized cleaning procedure established by Radio Corporation of America (RCA). The deposition reactor was equipped with a QCM (Neyco) for the gravimetric monitoring of the lm growth. The QCM was xed to the central part of the substrate holder. A quadrupole mass spectrometer, Vision-2000C, MKS-instrument, was assembled at the outlet of the deposition reactor to monitor the exhaust gas composition. SiO 2 thin lms were obtained at room temperature using SiCl 4 and H 2 O as precursors. The vaporized precursors were transferred to the ALD reaction chamber with N 2 as the carrier gas. SiCl 4 was purchased from Sigma Aldrich and used as-received. Both canisters containing the precursors were maintained at 19 C during deposition. NH 3 gas (<99.9%), used as a catalyst, was injected into the reactor under 1 bar pressure.
The morphology and thickness of the obtained samples were characterized using a FEI Heliosnanolab 650 Focused Ion Beam Secondary Electron Microscope (FIB-SEM). The structure of the lms deposited on dedicated Kapton tape was probed by smallangle X-ray scattering (SAXS) using an X-ray Diffractometer (X'Pert Pro (Panalytical)) equipped with a focusing mirror and a Pixcel 1D detector in the transmission mode. The elemental composition depth prole was assessed using D-SIMS (Cameca, IMSLAM); however, the quantication was performed by XPS (Thermo VG Scientic, MicroLab 350) using Al Ka source.

Results and discussion
Catalytic SiO 2 RT-ALD growth SiO 2 lms obtained at room temperature have already been prepared by the sequential exposure (ABAB.) of two reactants (A and B). Many well-known precursors require high deposition temperatures, plasma or highly reactive co-reactants such as ozone gas. 47 Despite a low enthalpy of reaction, SiCl 4 usually reacts with water (oxidant species) at high temperatures (>325 C). 13 The comparison of thermal ALD and room temperature processes reveals a higher growth rate/ALD cycle in favour of room temperature reactions ($2Å per cycle) (Fig. S1, ESI †).
The amount of contaminants integrated in RT-SiO 2 lms is inherently dependent on the way of tuning the surface exposure to precursors. Based on studies by George et al., 31 we investigated the growth of SiO 2 at room temperature (RT)-ALD by alternatively exposing the surface to SiCl 4 and H 2 O under a constant ow of NH 3 . The exposure time was xed at 90 s for SiCl 4 with a purge time of 1 min. H 2 O exposure was xed at 90 s with a purge time of 5 min to ensure a perfect saturation of the surface. 31 The in situ monitoring of the lm growth obtained by the QCM is shown in Fig. S2, ESI. † As already described, 31 the reduction reaction of SiCl 4 with water is depicted through the minimum of SiCl 4 weight gain (Fig. S2b, ESI †) and the longer reaction of the NH 3 -H 2 O mixture ( Fig. S2c-e, ESI †). Then, as shown in Fig. S2b, ESI, † the gain of mass that is observed in one ALD cycle is predominantly obtained from the half reaction of H 2 O. Nevertheless, the growth rate of $1.5Å per cycle (297 nm/ 2000 cycles), experimentally obtained through this process with a constant ow of NH 3 , tends to reach the 2Å per cycle value described by George et al. 31 XPS elemental analysis of the SiO 2 lm (deposited on Al 2 O 3 / Si) ( Fig. 1) shows the presence of carbon, nitrogen and chlorine in addition to silicon and oxygen within the SiO 2 deposited lm. A depth proling analysis reveals that the carbon is restricted to the surface of the lm.
The asymmetry of the C 1s peak suggests the presence of C-H, C-O-C or C]O bonds at the surface due to air exposure. The Si 2p peak of the Si-O layer appeared at 104.4 eV with a 1.77 eV (broadened up to 2.22 eV due to charge effects during the depth proling) full width at half maximum (FWHM). The sharp and symmetric Si 2p peak centred at $104 eV suggests an oxidation state of +1, which was attributed to the presence of SiO 2 . A high contribution of the byproducts of the reaction is detected through the presence of Cl and N elements in the lms. The Si/ Cl ratio increased from $1.3 at the surface to 3.2 in the bulk of the lm (Table 1).
Furthermore, the reaction of the lone pair of active -Cl with the hydrogen of water induced a signicant formation of HCl. The higher detection limit of D-SIMS was used to screen the chemical elements present in the "bulk" of the lm, particularly for light elements such as hydrogen. As shown in Fig. 2, all typical elements of the deposited lm, i.e. Si, O, Cl, N, H and C, are detected. The intensity of Cl is similar to that of Si and O while a difference in the intensity is observed for H, N and C. It can be seen from the depth prole analysis that the signal of C decreases rapidly, which is in agreement with the XPS results. As H% is roughly constant, the slow decrease in N and Cl tends to conrm that a part of the lm is composed of NH 4 Cl contaminants.

Contaminant inclusion mechanism
George et al. described the mechanism of a catalysed binary reaction that spontaneously takes place in the presence of pyridine or NH 3 as a Lewis base agent. 30,48 The hydrogen bonding between the Lewis base and SiOH* (surface species) or H 2 O allows the reaction to be performed at room temperature (Fig. S3, ESI †). Compared to high temperature processes that use large precursor exposures (>10 3 Torr s), SiO 2 RT-ALD takes place owing to the strong nucleophilic attack of the oxygen from (i) SiOH* on SiCl 4 and that from (ii) H 2 O on SiCl*. 46 Nevertheless, to the best of our knowledge, no specic data have been reported on the variation of the chemical composition and the morphology of such lms. Based on the catalytic effect of NH 3 , it can be clearly deduced that a constant ow of NH 3 statistically ensures a maximized reaction of -O on all -O-Si-(Cl) n available sites. Nevertheless, the perfect delimitation of the exposure windows at room temperature could be enhanced by working in a non-conventional high vacuum state (<10 À6 Torr). As it is not the case for standard ALD reactors, we attempted to understand and control the contaminant inclusion mechanism in the pulsed NH 3 regime. Thus, the state-of-the-art production of SiO 2 at RT using a constant ow of NH 3 has been compared to pulse NH 3 -catalysed RT-ALD. Inspired by the reactivity of chlorinated precursors described by Damyanov et al., 49 the amount of contamination could be cautiously explained by the functionality x of the adsorbed species at the surface explained hereaer: The injected precursor SiCl 4 reacted with the surface (^) hydroxyl species. Moreover, the competition between the single bond case (x ¼ 1) and multiple bonds (1 < x # 3) was directly linked to the stagnancy of precursors in the ALD regime. As far as the concentration of hydroxyl groups on the surface increased, the saturation of H 2 O directly enhanced the formation of HCl. Along with the constant ow of NH 3 , the $2.2 Si/N ratio measured by XPS in the bulk of the lm indicates a strong nitrogen contamination exceeding acceptable limits, especially through the inclusion of NH 4 Cl salt. As indicated by George et al., 31 this salt is formed as a result of the NH 3 catalyst complexing with HCl. Because of the vapour pressure of the NH 4 Cl salt (i.e. 4 Â 10 À5 Torr (ref. 50)), some quantity of the salt remained within the lm. In that context, note that compared to an inert gas, using NH 3 as a carrier gas may not contribute to a pure ALD process performed at RT. Indeed, a signicant contamination of the surface is attributed to the excessive dose of NH 3 . The contamination depicted here conrms the already described importance of adjusting the quantity of NH 3 to limit the reaction with HCl. 31 Thus, we considered that pulsing NH 3 similar to the other precursors could minimize unfavourable reactions at room temperature.

Low contamination SiO 2 growth under pulsed NH 3
Dense oxide under pure ALD regime. Based on the same chemistry used in the previous part, each chemical involved in the following process has been considered as a precursor. This indicates that an adequate separation of each pulsed chemical has been guaranteed. The purge of the reactor has been optimized using the appropriate ratio of carrier gas ow/reactor base pressure (<2 Torr). Any overlap between each precursor pulse has been prevented by checking the injection with the integrated RGA. Fig. 3 shows the ALD saturation curves at RT for SiCl 4 (a), H 2 O (b) and NH 3 (c) precursors. According to the diagrams, the saturation of all precursors occurs aer exposure for 90 s. The N 2 purging time between SiCl 4 and NH 3 precursors has been xed at 90 s. The appropriate purging time aer water exposure was then determined by RGA (H 2 O: m/z ¼ 18 uma) analysis using a systematic variation process (Fig. S4, ESI †). Aer 300 s purging time, water was completely removed from the reactor.
Based on the trends observed in Fig. 4, the growth of a SiO 2 lm in a pure ALD regime at RT has been investigated. SiCl 4 ,  Fig. S5a, ESI, † the 0.02 mg cm À2 per cycle weight gain is 30 times lower than the process with a constant ow of NH 3 (Fig. S2, ESI †). Nevertheless, the injection of NH 3 and H 2 O precursors signicantly contributes to a certain gain of mass ( Fig. S5 panel b, ESI †), and then a growth rate of 0.5Å per cycle is obtained for 500 deposition cycles. It can be observed that the high H 2 O mass adsorbed during the interaction of H 2 O molecules with active complexes at the surface ends through the efficient replacement of chlorine by hydroxyl groups (Fig. S3, ESI †). The XPS elemental analysis (Fig. 4) still shows the presence of chlorine, nitrogen and carbon in addition to silicon and oxygen. The amount of contaminants (Table 2) is nonetheless substantially decreased. Firstly, the Si/Cl ratio increased from $4 (surface) to $8.7 in the bulk of the lm. Secondly, compared to the lms obtained with a constant ow of ammonia, the Si/Cl ratio improved signicantly (threefold). Moreover, the Si/N ratio increased from 1.1 to 3.8 (2.2 to 5.9 inside the lm). This indicates a limited reaction between HCl and NH 3 to form NH 4 Cl. The best tting procedure of the high-resolution spectrum of N 1s reveals a single binding energy peak at 401.1 AE 0.3 eV, corresponding to NH 3 + . This conrms the formation of the NH 4 Cl salt, and the small amount of detected Al is attributed to the alumina sub-layer (i.e. SiO 2 /Al 2 O 3 /Si). The SIMS depth prole of the SiO 2 lm is shown in Fig. 5. The intensity of chlorine decreases $30 times faster than the process performed with the constant NH 3 ow. In fact, less than 100 s sputtering is needed to decrease the intensity below 1 Â 10 5 cnts per s compared to $2800 s for the NH 3 constant ow process. Moreover, the intensity of nitrogen seems to be in the same range of 10-100 cnts per s. Compared to the XPS results, this corroborates the formation of the NH 4 Cl salt. Note that the intensity of Si is higher than that of Al, conrming the coating process of SiO 2 on Al 2 O 3 . From the depth prole, we can estimate the SiO 2 lm thickness to be around 25 nm. Based on the XPS and SIMS results, it can be assumed that this RT process is optimized in terms of surface exposure. Nevertheless, the residual traces of HCl still react with NH 3 because of the difficulties to purge H 2 O or NH 3 at RT. Moreover, small quantities of byproducts, such as NH 4 Cl, were consequently integrated into the lm. Fig. 6a and b show top-view SEM images of the SiO 2 lm. We observed a rough layer with grain sizes of up to 200 nm. This roughness is highlighted in the 45 tilted view (Fig. 6c).
Furthermore, cross-section analyses evidence the presence of a compact lm (Fig. 6d). From these pictures, we can conclude that this pulsed NH 3 growth process leads to dense but rough SiO 2 thin lms. A thickness of 30 AE 5 nm is measured through the cross-section, close to the 25 nm value deduced from the SIMS analysis. This leads to a lower growth rate of $0.5 A per cycle related to the lower weight gain observed with QCM (i.e. 30Â lower than the SiO 2 lm processed under a constant ow of NH 3 ). Nevertheless, the irregular surface aspect reveals   that the process does not correspond to a pure ALD growth mode, as expected. [51][52][53] This peculiar non-homogeneous growth at RT suggests that the surface reaction is in competition with the integration of contaminants. The self-limiting process actually promotes the deposition of species onto the substrate and onto the deposits (e.g. islands) with equal probability. [54][55][56] The inclusion of contaminants at a sub-atomic growth rate (i.e. <1Å per cycle) could explain the morphology of the obtained lm. Moreover, the high amount of -OH surface groups could affect the dehydroxylation/rehydroxylation equilibrium (section: Contaminant inclusion mechanism), leading to the production of a higher quantity of HCl in the case of trifunctional bonds. Nevertheless, the oxide thin lm displays a signicant density in volume with limited inhomogeneity. This is in line with the sub-atomic growth rate mechanism surrounded by limited contamination. The tailoring of ALD parameters in this RT-SiO 2 growth process shows a substantial adaptability in terms of morphology and chemical composition. These results suggest that a tuning of the growth parameters could inuence the crystallisation. Hence, different types of SiO 2 layers could be processed at RT.
Porous oxide under limited ALD regime. As described in the previous section, less contaminated SiO 2 can be produced by adjusting the surface exposure of SiCl 4 , NH 3 and H 2 O precursors. Furthermore, the effect of limited exposure on the composition and the morphology of the lm has been investigated. Hence, the process has been tuned to maintain low level of contaminants in an ALD non-saturation regime. The precursor exposure has been decreased to a minimum value for SiCl 4 (i.e. 100 ms) in agreement with a low contamination strategy. Then, according to RGA results, the exposure time of NH 3 and H 2 O was xed to 2 s for both with a purge of only 10 s using 300 sccm of N 2 .
As shown in Fig. S6, ESI, † a growth rate of 1.54 mg cm À2 per cycle was obtained. Compared to the process described in the previous section, the exposure reaction used here generates $50 times higher weight gain. In order to disentangle any physicochemical inuence from the substrate, silicon oxide lms were grown on a pre-characterized barrier layer. Hence, SiO 2 growth was investigated on two different sub-layers, i.e. TiO 2 deposited by ALD and Si bulk.   This journal is © The Royal Society of Chemistry 2020 RSC Adv., 2020, 10, 18073-18081 | 18077 Fig. 7 shows the XPS experiment results. As expected, Cl, C, N elements were detected in both samples. For SiO 2 deposited on TiO 2 , the detection of Ti 2p before etching conrms the low thickness of the lm.
However, the percentage of chlorine was clearly maintained below the limit of 3% (obtained for the previous process with extended exposures) ( Table 3). The higher chlorine concentration observed aer etching (3.79 at%) was attributed to the chlorine inherent to the TiO 2 ALD process. This was conrmed by the low chlorine concentration for SiO 2 deposited directly on the silicon wafer (Table 3, SiO 2 /Si). In order to screen the composition and morphology of the lm, thicker SiO 2 layer (2500 cycles) were processed on a chlorine-free material, i.e. Al 2 O 3 (50 nm) on Si.
The SIMS depth proling of the synthesized SiO 2 thick lm exhibits a concentration of chlorine that rapidly decreases as a function of sputtering time (Fig. 8). Compared to the previous process, the intensity of Cl is starting a decade less, around 3.5 Â 10 4 cnts per s. Moreover, the amount of C and N is very low, which conrms the low level of NH 4 Cl contamination in a thick volume of SiO 2 .
Furthermore, the higher concentration of chlorine close to the surface of the lm indicated the slow dissociative chemisorption of water, which induced the desorption of HCl. This recombination clearly affected the growth mechanism of SiO 2 . As shown in Fig. 9, SEM analyses highlighted the porous state of the oxide lm. In addition to the 200-500 nm diameter aggregates on the surface of the layer, the top and tilted view (Fig. 9a) revealed a SiO 2 sponge-like structure. The porosity of the lm was conrmed by SAXS where a periodical arrangement of pores could be tted with an average radius of 130Å (std dev 30% and most frequent radius $ 110Å). The applied FIB cross-section (Fig. 9b) reveals the presence of 20-50 nm cavities (merging pores due to the preparation) and isolated pores of $15 nm. As explained by Puurunen in the ALD random deposition approach, 56 if the growth per cycle is not constant, the increase in the surface roughness should be fast at the beginning of the growth and slow thereaer. This naturally indicates that a smaller number of ALD reaction cycles are required to t a conformal deposition in a close-packed array as far as the growth rate is adjacent to an atomic monolayer.
By considering the growth rate of $0.11Å per cycle obtained in this process, it could explain why the SiO 2 lm is less "closed" as the one processed via the pure ALD approach. The spongelike porous structure growth may be related to the limited surface diffusion of the by-products (NH 4 Cl, HCl) generated during each half cycle. Indeed, the diffusion/desorption of byproducts is slower in the case of low reaction temperature  (here RT) and short purge time. In this case, residual water or byproducts (NH 4 Cl, HCl) are considered as surface fractions where the supplementary amount of the injected precursor will be adsorbed instead of the -Si-OH surface groups. This prevents the growth of SiO 2 and leads to non-uniform porous thin lms. Nevertheless, this peculiar structure is very attractive for applications that need to be processed at RT. Even if other techniques such as PE-ALD are able to produce SiO 2 dense thin lms with no impurities at reduced temperatures (50-80 C) for specic applications, [41][42][43]57 we evidenced our ability to fabricate a highly porous SiO 2 layer at room temperature with a signicant level of control of the contamination. This specic characteristic of this porous SiO 2 layer is clearly transferred to complex and temperature-sensitive 3D materials. Hence, ALD SiO 2 could be deposited in a porous anodic aluminum oxide (AAO) membrane, and the results are shown in Fig. 10. It clearly appears that the porosity obtained on planar substrates is perfectly transferable to 3D surfaces. This suggests that the ALD technique for SiO 2 thin lm synthesis could be applied to any 3D complex substrate. Further studies are in progress.

Conclusions
Porous SiO 2 thin lms have been produced by ALD using a sequential exposure of SiCl 4 , NH 3 and H 2 O at room temperature. The catalytic effect of ammonia has been exploited to optimize the saturation of the precursors and the extended purges. A relation between the signicant porosity and the chemical saturation of the surface has been indicated using QCM, XPS, SIMS and SEM. It has also been demonstrated that this optimized process exhibited a decrease in the inherent inclusion of contaminants such as NH 4 Cl and HCl in the lm.
As demonstrated on AAO membranes, the transferability of this 2D process to 3D structures could extend the use of SiO 2 lms in several domains such as complex or high aspect ratio materials.

Conflicts of interest
The authors disclose that there are there are no conicts to declare. This journal is © The Royal Society of Chemistry 2020 RSC Adv., 2020, 10, 18073-18081 | 18079