Solid‐State NMR Spectroscopic Investigation of TiO2 Grown on Silica Nanoparticles by Solution Atomic Layer Deposition

Atomic layer deposition in solution (sALD) is just emerging as a technology for the preparation of thin films. Unlike ALD from the gas phase, it allows for mild reaction conditions in a solvent phase and at room temperature, thus decreasing the energy requirements of the process and widening the range of accessible precursor molecules. In this work, the deposition of thin films of titania on silica is investigated using titanium(IV) isopropoxide (TTIP) and water as precursors, which are alternatingly brought into contact with the support in a home‐built plug flow reactor. The mechanism of covalent grafting of the precursor to the surface, subsequent hydrolysis, and reaction to a layer of titania are investigated in detail using magic angle spinning (MAS) solid‐state nuclear magnetic resonance (NMR) spectroscopy. TTIP preferentially reacts with Q2 groups of condensed silica. 2D solid‐state NMR spectra allow to clearly show the successful grafting of this compound to the support by the appearance of a characteristic signal at −107 ppm, which is tentatively attributed to silicon nuclei in a SiOTi bond, and to reveal the presence of titanol groups on the emerging TiO2 film.


Introduction
Thin film materials with adjustable physical and chemical properties are of high interest for the fabrication of microelectronic DOI: 10.1002/admi.202202131 devices. [1] Among other techniques, thin films have been prepared by atomic layer deposition (ALD) in the gas phase since the 1990s. [2] ALD is achieved by exposing a substrate surface to alternating gaseous precursors. In between the pulses of the precursors, the reaction chamber is flushed with an inert purge gas (usually nitrogen). Thanks to the self-limiting nature of the process, ALD yields atomically controlled films of homogeneous thickness, good conformality, and trenchfill capability. [3,4] However, the technique also requires costly equipment, strict vacuum conditions, and high temperatures, making the process energy-intensive. It is furthermore limited to precursors that are volatile in a suitable temperature range without decomposition. [5] ALD in solution (sALD) can overcome these limitations. In sALD, diluted solutions of the precursors in an inert solvent are brought into contact with the substrate, applying alternating flows of the different solution at room temperature. As in gas-phase ALD, two half-cycles are necessary for one full monolayer deposition. sALD was introduced in 2009 for vanadium-doped TiO 2 and ZrO 2. [6] We have implemented this technology for a wide range of materials which are hardly or not processable by gas-phase ALD, such as TiO 2 , SiO 2 , MgO, [7] and PbS, [8] but also for methylammonium lead iodide perovskites. [8] In sALD, high temperatures are avoided, which opens up a much wider range of possible precursors than gas-phase ALD. At the same time, the advantages of conventional ALD, including the self-limited nature and the homogeneous film thickness are retained.
However, although an idealized model of the surface deposition process in sALD exists, more detailed insights into the processes at the surface are still lacking, in particular with respect to the grafted surface species and the interaction between nongrafted species or solvent molecules with the surface. The goal of sALD is the formation of a homogeneous, closed film after one deposition cycle, but this may not be the case for all substrates and precursors.
TiO 2 is chemically and physically stable and one of the most widely used semiconductors [9] with applications in photovoltaics, [10] sensors, [11] and photocatalysis. [12] We have recently achieved homogeneous deposition of 30 nm thin films of TiO 2 on flat silicon wafers over a SiO 2 layer, using alternating flows of titanium(IV) isopropoxide (TTIP) and water in diethyl www.advancedsciencenews.com www.advmatinterfaces.de ether. [7] An idealized reaction scheme is shown in Figure 1. However, many questions remain open in this process: How do TTIP molecules bind to the silica? Does TTIP during a single half-cycle completely cover substrate? If not, what is the coverage fraction? What is the impact of the pre-treatment of silica in this research model? The detailed elucidation of the reaction will help the optimization of this and other sALD processes.
In this work, we therefore investigate the deposition of titania on silica nanoparticles by solid-state nuclear magnetic resonance spectroscopy under magic angle spinning (MAS NMR), inductively coupled plasma optical emission spectroscopy (ICP-OES), and ellipsometry. By 1D and 2D 1 H and 29 Si MAS NMR techniques, we follow the deposition of the precursor titanium isopropoxide onto silica with different degrees of hydroxylation. The chemical binding of TTIP to the surface, as well as the further reaction toward a TiO 2 layer, can be clearly traced in MAS NMR spectra. We identify a so far unknown silicon species with a chemical shift of −107 ppm, which we attribute to Si-O-Ti bonds. The spectral features of TiOH groups become visible after a full sALD cycle. After 20 cycles, a closed multilayer of TiO 2 has developed on the silica particles.

Experimental Section
All substances were used as received, if not indicated otherwise. Diethyl ether (Sigma-Aldrich, ≥ 99.7%) was recycled by rotational evaporation in-house and stored over molecular sieve. The particle size of silica nanoparticles (Sigma-Aldrich, 99.5%) is in the range of 5-20 nm. The calcination of silica was carried out in a muffle furnace in a flat layer in an open vessel at 500°C (heating ramp 2 K min −1 ) for 8 h, then cooled to 120°C with a rate of 1.5 K min −1 and kept at 120°C. The hot sample was transferred into the glove box and stored under nitrogen.

Pre-Treatment of Silicon Wafers
Silicon wafers (Silicon materials) were cut into small pieces of 0.5 cm 2 and then ultrasonicated for 7 min each first in acetone, then in isopropanol, and then in ethanol, and dried under ambient conditions.

sALD in Batch Mode
In a typical batch sALD process, 0.100 g (1.6 mmol) of silica powder was dispersed in 30 mL anhydrous diethyl ether. 18 μL (0.06 mmol) of TTIP (Acros Organics, ≥99%) were added to the mixture and shaken, to achieve a concentration of 2 mmol L −1 TTIP in the suspension. The suspension was then centrifuged with 6000 rpm (2000 g) for 40 s. After removing the supernatant, the suspension was washed again with anhydrous diethyl ether and centrifuged again. The washing step was repeated ten times. A dry powder was obtained, which corresponds to a "half sALD cycle" sample. To obtain a sample with a complete sALD cycle, the obtained sample was reacted in the same way with a solution of 2 mmol L −1 water in anhydrous diethyl ether, followed by the same washing procedure. All operations are carried under nitrogen atmosphere in a glove box.

sALD Plug Flow Reactor
In a typical preparation using a home-built plug flow reactor, solutions of all precursors as well as the solvent anhydrous diethyl ether were prepared in advance in a nitrogen atmosphere and stored in Schlenk flasks. Water in diethyl ether was ultrasonicated for 20 min prior to use. A plug flow reactor was designed in-house (Figure 2): A silicon tube was assembled with a fitting, followed by a silicon wafer, a layer of cotton, and a filter paper at each end of the tube. Then the tube was filled with ≈180 mg of silica powder, slightly compacted, but not tightly pressed. The increase in thickness of the growing TiO 2 layer was determined on the silicon wafers at both ends of the reactor by ellipsometry. After the reaction, the silica powder was removed from the reactor for analysis. To prevent the precursors from reacting with each other in the pipeline, three branchings were installed in front of the reactor (Figure 2a and Figure S1, Supporting Information). The four channels of the peristaltic pump controlled the entry of the precursors and solvent into the reactor. The inner diameter of the peristaltic tube of the peristaltic pump was fixed to 1.52 mm. The feed duration and flow rate in each channel were controlled using a pump control software by Ismatec. Each cycle consisted of four stages: t 1 , the pulse duration of TTIP feed, t 2 , the purging time after the TTIP flow, t 3 , the pulse time of the water feed, and t 4 , the time of purging after water the feed (for details, see Figure S1, Supporting Information). All subsequent cycles were performed in the same manner.
After the reaction, the tubes were cleaned using subsequently diethyl ether, isopropanol, deionized water, 5% aqueous potassium hydroxide solution, deionized water, and isopropanol, at a rate of 3 mL min −1 for 6 min each.

Solid-State NMR Spectroscopy
MAS NMR spectra were acquired on a 500 MHz (11.7 T) DD2 Agilent wide-bore NMR spectrometer in 1.6 mm zirconia rotors, with the exception of CP build-up experiments, which were recorded on a 400 MHz (9.4 T) Bruker Avance spectrometer in a 4 mm zirconia rotor. Samples were packed into the rotor under an Argon atmosphere in a glove box. For packing of slurries, the rotors were additionally sealed with silicone plugs.

1 H Spectra
1 H spectra were recorded at a MAS rate of 30 kHz using the DEPTH sequence for background suppression, starting with a www.advancedsciencenews.com www.advmatinterfaces.de /2 pulse of 2.5 μs followed by 2 pulses of 5 μs phase-cycled according to a combined "EXORCYCLE" and "CYCLOPS" pulse scheme. [13][14][15] 16 scans were recorded, and the recycles delays were set to five times the 1 H longitudinal relaxation time, determined in saturation recovery experiments. Spectra were referenced to sodium trimethylsilylpropanesulfonate. 13 C cross polarization (CP) spectra were recorded at a MAS rate of 15 kHz. After a /2 pulse of 2.5 μs on 1 H, the RF of 1 H was ramped from 75 to 100 kHz while the RF on 13 C was kept constant at 75 kHz during a contact time of 2 ms (first spinning sideband condition). 100 kHz spinal-64 decoupling was applied on 1 H during acquisition. Recycle delays were set to 5 s, 6000 to 10 000 scans were recorded depending on the sample. For CP experiments on dried and calcined silica, the powders were wetted with diethyl ether prior to measurement. 29 Si CP spectra were recorded at a MAS rate of 15 kHz. After a /2 pulse of 2.5 μs on 1 H, the RF of 1 H was ramped from 55 to 76 kHz while the RF on 29 Si was kept constant at 75 kHz during a contact time of 2 ms (Hartmann-Hahn match). 100 kHz spinal-64 decoupling was applied on 1 H during acquisition. Recycle delays were set to 2 s, 10 000 to 40 000 scans were recorded. Direct excitation spectra were recorded at a MAS rate of 14 kHz, using a /2 pulse of 5.25 μs, an interscan delay of 300 s and around 400 scans depending on the sample. 29 Si heteronuclear correlation (HETCOR) spectra were acquired at a MAS rate of 15 kHz. After /2 pulse of 2.5 μs on 1 H, the RF of 1 H was ramped from 55 to 76 kHz while the RF on 29 Si was kept constant at 75 kHz during a contact time of 2 ms (Hartmann-Hahn match). 700 increments during the evolution period were recorded, while applying 100 kHz FSLG decoupling. 100 kHz spinal-64 decoupling was applied on 1 H during acquisition. 1600 scans were accumulated in every increment with a recycle delay of 2 s.

Ellipsometry
Spectroscopic ellipsometry data were collected on a SENPro spectroscopic ellipsometer from SENTECH with range from 370.0 to 1050.0 nm under a 70°incidence angle. The fit model is based on Cauchy layers for the "TiO 2 -SiO 2 -Si" stack, [16] which is implemented in the data evaluation software Spectra Ray 3.

Energy-Dispersive X-Ray Spectroscopy
Energy-dispersive X-ray (EDX) spectra were acquired on a JSM 6400 by JEOL with an acceleration voltage of 10.00 kV and a magnification of 10 000.

Nitrogen Adsorption-Desorption Isotherms
Textural properties of the calcined silica was assessed by a N 2 adsorption-desorption isotherm at 77 K recorded on an ASAP 2000 from Micrometrics, Inc.: 100 mg of sample were used and outgassed under vacuum at 250°C for 12 h before analysis. The specific surface area of the samples was calculated using the Brunauer-Emmett-Teller (BET) method within the relative pressure range of 0.05-0.2.

Scanning Electron Microscopy
Scanning electron microscopy (SEM) images were recorded with an SE2 detector in the electron microscope ULTRA55 (Carl Zeiss MST AG).

Dynamic light scattering
The dynamic light scattering (DLS) measurements were made with a Zetasizer Nano ZS (Malvern Instruments) with anhydrous diethyl ether as solvent.

Results and Discussion
A silica powder with primary particle size of 5-20 nm was used in this work. To track possible changes in the silica morphology by calcination, the material was characterized after calcination by SEM, N 2 adsorption-desorption isotherms and DLS in anhydrous diethyl ether ( Figures S2-S4, Supporting Information). N 2 adsorption-desorption isotherms show that the calcined silica has a surface area of 630 m 2 g −1 , determined by the BET ( Figure  S2, Supporting Information). During calcination, sintering of the primary particles occurred, yielding a size distribution of particle agglomerates centered at 600 nm ( Figure S4, Supporting Information) in DLS measurements.
To achieve sALD of thin layers of titania on powdered silica, two different processes were developed in this work. sALD was carried out in a batch mode in vials, by adding to the dry silica powder a solution of TTIP in anhydrous diethyl ether (called here first half sALD cycle), alternated with a solution of water in anhydrous diethyl ether (second half sALD cycle). The process was carried under a nitrogen atmosphere in a glove box and yielded small amounts of ≈20 mg of silica powder. Losses during washing by centrifugation could not be completely avoided, therefore, this process was only used for experiments in which one half or one full sALD cycle was carried out. By 1 H MAS NMR investigation, ten washing steps and a concentration of 2 mm TTIP and 2 mm water in diethyl ether were found to be optimally suited (see below). For thicker TiO 2 layers produced by multiple sALD cycles, a larger plug flow reactor process was developed, which allowed for sample masses of ≈180 mg of powdered silica. In a plastic tube, the dry powder was packed between filter papers and layers of cotton ( Figure 2a). Solutions of TTIP in anhydrous diethyl ether, water in anhydrous diethyl ether, and pure anhydrous diethyl ether for purging were then flown alternatingly through the packed bed. The operating parameters of the previously constructed experimental pump set-up were based on the existing work by Wu et al. (Table 1). [7] The concentration of precursors was optimized before in batch mode experiments to 2 mm.
The precursor pulse time was optimized so as to achieve a monolayer of TiO 2 in every sALD cycle. Since the thickness of the titania layers on a powdered silica could not directly be obtained, the silicon wafer at the inlet and the outlet of the reactor were measured by ellipsometry, as TTIP readily forms titania on this support (Figure 3). With a pulse duration of about 15 s, a growth rate per cycle (GPC) of 0.3 Å, indicative for the formation of a single TiO 2 layer, was obtained. [7,17] 15 s pulses were therefore retained for further experiments. Larger pulse du- rations lead to a higher GPC, which may be caused by an incomplete removal of precursor. An extension of the purge durations would be therefore necessary for high pulse durations to avoid uncontrolled TiO 2 formation from accumulated precursor. The GPC on the silicon wafer on the inlet were always slightly higher than the GPC on the silicon wafer at outlet of the reactor. A possible explanation is a concentration gradient along the axial direction of the tube reactor, due to the desired reaction of TTIP with the powdered silica. To verify the reproducibility of the experimental setup, the same experiment was repeated 12 times. GPCs were largely reproducible, with a mean value of 0.3 Å per cycle ( Figure S5 and Table S1, Supporting Information).
The thickness increase of the titania layer was also determined as a function of the number of completed sALD cycles (Figure 4). After the first and after ten cycles, the thickness increase differs from the expected values, that is, 0.3 Å after one and 3 Å after ten cycles. After higher cycle numbers, the thickness of the TiO 2 layer is very close to the expected values (6 Å after 20, 12 Å after  40 cycles). This variation in the first few cycles can be attributed to a limited accuracy of the ellipsometry, since the method is close to the detection limit here. To estimate in a semi-quantitative way the amount of titania directly in the powder, the titanium and silicon content of the dried powder were determined by ICP-OES. Note that due to the low TiO 2 content, the amount of Ti is close to the quantification limit of the instrumental setup. However, we observe that for more than ten cycles, the molar ratio of Ti to Si increases linearly with the thickness increase of the titania layers on the silicon wafer. 1 H, 13 C, and 29 Si solid-state NMR spectroscopy under MAS was carried out on samples prepared under various selected conditions. Initially, we studied a series of model samples: the calcined silica, the same support treated with pure TTIP after one half sALD cycle and treated with water saturated diethyl ether after one full sALD cycle. Calcined silica shows a signal around 1.8 ppm, which is known to correspond to isolated silanol groups on the surface, but may also arise from non-accessible silanol groups in ultramicropores (with diameters <1 nm) or occluded in the bulk silica (Figure 5b). [18][19][20] A broad, weak signal around 4-4.5 ppm can be found in some spectra, which is attributed to adsorbed water (Figure 5a). Note that adsorbed water on silica can occur in a range of 1 H chemical shifts between 2 and 4 ppm. Grünberg et al. have explained this phenomenon with binding of adsorbed water to isolated silanol groups and a fast av-erage of protons, which leads to a chemical shift which is weightaveraged according to the proportion of water and silanol groups. At 5 ppm, water is considered to be present in water clusters or as bulk water. [21] Calcined silica was impregnated first with pure TTIP and the slurry was packed into the NMR rotor. The spectrum of this sample is dominated by the two resonances of CH and CH 3 groups of liquid TTIP at 1.1 and 4.5 ppm (Figure 5c). Note that NMR resonances from liquids are usually much narrower than resonances from solids, as the mobility of the liquid phase averages out dipolar couplings between spins, residual chemical shift anisotropies, and chemical shift distributions due to distributions of chemical environment. After washing of excess TTIP and drying, some residual diethyl ether remains on the silica, evidenced by resonances at 1.1 and 3.4 ppm (Figure 5d,e). Note that we did not expose the sample to any low pressure or high temperature treatment after sALD, in order not to alter the surface chemistry, for example, thermal elimination of diisopropyl ether or water from surface species (see below). The resonances of diethyl ether are not sharp like liquid TTIP, but significantly broader, indicating that the sample contains diethyl ether molecules strongly adsorbed on the surface, possibly by hydrogen bonds. It is thus possible to use the signals as an internal chemical shift reference. Unfortunately, the methyl resonance of diethyl ether at 1.1 ppm overlaps with that of TTIP, making it difficult to assess the presence of TTIP in the further experiments by this signal. Interestingly, after one half cycle, two resonances appear at 4.1 and 4.5 ppm, which may be assigned to TTIP and residual water on the surface, but the respective attribution is not unambiguous here (Figure 5d). Both resonances vanish after the treatment of the surface with water second half sALD cycle (Figure 5e), indicating that the hydrolysis of isopropyl groups was successful and no more detectable free or surface bound TTIP remains in the sample. We therefore tentatively assign the resonance at 4.1 ppm to the CH groups of a TTIP molecule that is covalently bound to the surface to form a Si-O-Ti(O i Pr) 3 moiety. Surface-binding may lower the 1 H chemical shift of these hydrogen atoms compared to the free TTIP (CH group at 4.5 ppm, see Figure 5b). 1 H- 29 Si heteronuclear correlation (HETCOR) spectra will give further evidence for such surface-bound species (see below).
The effect of the contact of TTIP and H 2 O in the two half cycles with the silica surface can be followed in 13 C CP MAS NMR spectra. Here too, resonances of ether can be detected in the spectra after each single sALD step (Figure 6, marked in yellow and red). As expected, the presence of TTIP is detected after one half and one and a half sALD cycles by resonances at 76 (CH group, marked in green) and 23 ppm (CH 3 group, marked in blue, Figure 6c). [22] These peaks disappear after the second half sALD cycle and two full sALD cycles (Figure 6b,d), that is, after the hydrolysis of TTIP with water. However, a signal at 29 ppm remains, which may be attributed to residual isopropanol which has been cleaved, but not completely washed away.
We further investigated the impact of calcination of the support on the TiO 2 deposition. In non-calcined silica, a strong, broad resonance between 4 and 8 ppm corresponds to a layer of hydrogen-bonded OH groups, and the resonance at 4.5 ppm to clustered water on the surface (Figure 7a,b). In calcined silica, this broad signal of hydrogen-bonded silanols on the surface is much weaker and has shifted to 6-7 ppm, which is  characteristic for smaller amounts of hydrogen-bonded silanol groups (Figure 7c,d), in line with the more complete drying of the material. [19] In samples treated with both 2 mm and with pure TTIP, again, a resonance at 4.1 ppm occurs (Figure 7c,d).
Since the signal reappears in samples after a half cycle TTIP treatment, we tentatively assign this resonance to the CH groups of a TTIP molecule that is covalently bound to the surface to form a Si-O-Ti(O i Pr) 3 moiety. Surface-binding may lower the 1 H chemical shift of these hydrogen atoms compared to the free TTIP (CH group at 4.5 ppm, see Figure 5b). 1 H- 29 Si HETCOR spectra will give further evidence for such surface-bound species (see below).
The optimum concentration of TTIP was then varied using 0.2, 2, and 20 mm TTIP in anhydrous diethyl ether. Signals of TTIP deposited on silica are visible in the 1 H MAS NMR spectra after one half cycle starting from a concentration of 2 mm ( Figure S6, Supporting Information), so this concentration was retained for subsequent experiments both in the batch process and in the plug flow reactor process. Likewise, the number of washing steps was varied from 5 to 20. After ten washing steps, no more residual TTIP remained in 1 H MAS NMR spectra ( Figure S7, Supporting Information), so that this condition was judged to be sufficient for further sample preparations in the batch mode.
Covalent grafting of TTIP to the silica surface will consume silanol groups by the condensation reaction of TTIP with a hydroxyl group and release of a molecule of isopropanol. We have tried in the following to quantify the percentage of surface silanol groups thus consumed, to determine the optimum sALD conditions and in order to probe the existence of preferential reaction sites. The grafting of TTIP was assessed by 29 Si direct excitation (DE) spectra ( Figure S8, Supporting Information). In amorphous silica, Q 2 , Q 3 , and Q 4 species, corresponding to Q n = Si(OH) 4−n (OSi) n , can be readily discerned by their respective chemical shifts around −92, −101, and −110 ppm. DE spectra quantitatively reproduce the relative intensities of Q 2 , Q 3 , and Q 4 sites in a silica material. However, since most of the silica consists of completely condensed Q 4 groups in the bulk material, the relative intensities of Q 2 and Q 3 groups are low, and spectral deconvolution remained ambiguous. As a tendency, the spectra show a decrease of relative intensity of Q 2 and Q 3 sites from pure, calcined silica to silica treated with 2 mm TTIP to silica treated with pure TTIP.
Therefore, the effect of TTIP grafting was further followed by 29 Si CP MAS NMR spectra of the powdered samples before and after the sALD reaction (Figure 8). Note that CP spectra do not reflect the true abundance of each group in the silica. The signal intensity of each silica species depends on the CP buildup time constant T HSi and the 1 H relaxation time constant in the rotating frame T 1 , leading to a characteristic CP build-up curve. [23][24][25] In order to verify if, here, CP spectra are indeed suitable to quantify the amount of consumed Q 2 and Q 3 groups, CP build-up curves were recorded on four different samples: noncalcined silica, calcined silica, calcined silica after one half cycle reaction with 2 mm TTIP, followed by ten washing steps with diethyl ether and after one half cycle reaction with pure TTIP and ten washing steps ( Figures S9 and S10, Supporting Information). Note that for these experiments, the pristine silica powders were wetted with diethyl ether, too, in order to ensure a comparable density of protons on the surface of the reference samples and the samples after half cycle sALD reactions. Within the error, CP build-up curves of Q 2 , Q 3 , and Q 4 groups show very similar slopes up to a contact time of 8 ms ( Figure S10, Supporting Information). This somewhat untypical behavior is certainly due to the high proton density on the surface, ensuring good dipolar interactions between protons and all near-surface silica sites.
Pristine calcined silica shows the highest ratio of Q 3 /Q 4 compared to the samples after sALD (Figure 8a). After one half sALD cycle reaction, samples treated with pure TTIP and with 2 mm TTIP show much lower Q 2 and Q 3 intensities, the material treated with pure TTIP having the lowest remaining content of Q 2 and Q 3 groups. Clearly, pure TTIP leads to a more Adv. Mater. Interfaces 2023, 10, 2202131 Table 2. Relative signal areas of Q 2 , Q 3 , and Q 4 groups after deconvolution of 29 Si CP MAS spectra in Figure 8, and the fraction of reacted Q 2 groups ( Q2 ), Q 3 groups ( Q3 ), and overall fraction of reacted hydroxyl groups ( ) after one half cycle. Two different scenarios are considered: Q 2 groups react with only one hydroxyl group, yielding Q 3 species, or Q 2 groups react with both hydroxyl residues, yielding Q 4 species.

Sample
Relative signal area / % complete coverage of the surface after one half sALD cycle. However, the surface is not fully covered after one half cycle, in which case a nearly complete disappearance of Q 2 and Q 3 signals would be expected. Very small signals of Q 2 and Q 3 sites might remain since silica often contains inaccessible silanol groups in ultramicropores or occluded in the bulk material. [18,19] In non-calcined silica, the decrease of Q 2 and Q 3 groups is less pronounced than in calcined silica (Figure 8b). Here, the presence of water and higher amounts of hydrogen-bonded silanol groups, as observed in the 1 H NMR spectra (Figure 5a), seem to impede TTIP grafting onto the surface, or may lead to intermolecular TTIP homocondensation without any involvement of surface SiOH groups. Q 2 , Q 3 , and Q 4 resonances were deconvoluted and the intensity ratios were analyzed quantitatively. We use here a mathematical model established by Pallister et al. for the grafting of a gallium complex on silica in a gas ALD process. [26] From the peak intensities, the coverage and the fractions of Q 2 and Q 3 silanol groups which have reacted with TTIP were calculated ( Table 2). For details of the calculation, see chapter 1 in the Supporting Information. For the implementation of this model, following Pallister et al., the assumptions were made that Q 2 and Q 3 sites have a similar CP build-up behavior (compare Figure S10, Supporting Information) and that all compounds reacting with the silica will react at the hydroxyl surface sites. [26] The reaction mechanism of Q 2 groups with TTIP can progress via two pathways. In a first scenario, only one of the two silanol groups might react with TTIP (monodentate binding), leaving the other present on the surface, thus Q 2 would transform into Q 3 groups. In a second scenario, both silanol groups on Q 2 species will react with a single TTIP molecule (bidentate binding), cleaving two equivalents of isopropanol, and resulting in a Q 4 group with a fourmembered Si-(O) 2 -Ti ring. Alternatively, on Q 2 groups, the two silanol groups might react with two different TTIP molecules. We therefore consider these two scenarios in the calculation of the fractions of Q 2 and Q 3 groups: 1) Only one hydroxyl group of Q 2 sites reacts with one precursor molecule, yielding Q 3 groups, and 2) a reaction of both hydroxyl groups of Q 2 sites, yielding Q 4 sites. Obviously, in both cases, Q 3 groups are supposed to react to Q 4 groups upon TTIP grafting. Importantly, since the calculation of the coverage of Q 3 groups depends also on the transformation of Q 2 → Q 3 , it is not possible to state which scenario is occurring, or if both cases are competing.
CP build-up curves, discussed above, were recorded on a second set of samples (compare Figures S9 and S10, Supporting Information). Here, likewise, the fractions of reacted Q 2 and Q 3 groups and the overall fraction of reacted silanol groups were calculated individually for spectra with 2, 4, and 8 ms CP contact time (Table S3, Supporting Information). The percentages of reacted groups are in very good agreement when determined from spectra with different contact time. Between the two sets of samples, the percentages of reacted groups are also in good agreement, demonstrating the robustness of the method.
In both calcined and non-calcined samples, Q 2 groups on the surface of silica account for a very small percentage of the total number of hydroxyl groups. However, we find that in both scenarios, Q 2 groups are more prone to react with TTIP (Table 2) than Q 3 groups, since the final fraction of reacted Q 2 groups is significantly higher. This is true for both non-calcined and calcined silica, and both for 2 mm and pure TTIP as a precursor. The most complete coverage is obtained by pure TTIP grafted on calcined silica, where around 80% of Q 2 hydroxyl groups react upon deposition of pure TTIP in scenario 1 (91% in scenario 2), versus 23% of the Q 3 groups (14% in scenario 2). We note that absolute quantification of different silanol species from 29 Si CP spectra is difficult and thus, the values for Q2 , Q3 , and in Table 2 might still be subject to a systematic error. Yet, the trend observed here for the different reactivities of Q 2 and Q 3 groups remains noteworthy. The finding that Q 2 species react more readily is also consistent with the study by Pallister et al. In that study, the authors attributed the higher reactivity of Q 2 groups to a bidentate binding mode and the formation of fourmembered Si-(O) 2 -Ga rings, thus to Q 4 groups, since two ethyl ligands can be cleaved from the Ga complex. Four-membered Si-(O) 2 -Ti rings have been shown to exist in titania-containing zeolites, although with low stability. [27] They may thus explain the preferential reactivity of Q 2 groups toward TTIP. We consider the covalent grafting of two individual TTIP molecules to a single Q 2 site to be less likely, since the first TTIP molecule with its three bulky isopropyl groups would impose a strong steric hindrance for the covalent bonding of a second precursor. Yet, it is possible that while Q 2 species covalently bind to TTIP with one silanol group, the second silanol develops a hydrogen bond to the oxygen atom of another isopropoxyl moiety, thus stabilizing the grafted complex. This may explain the higher reactivity of Q 2 compared to Q 3 groups even in scenario one, considering only monodentate binding.
To obtain a deeper understanding of the silica surface, 2D 1 H- 29 Si HETCOR MAS NMR experiments were carried out. These spectra offer additional information thanks to the correlation of the 1 H and the 29 Si dimension, and increased resolution of the 1 H dimension compared to 1D 1 H spectra due to dedicated pulse schemes which decouple proton dipolar interactions during the acquisition. Calcined silica only shows one correlation of protons in hydrogen bonding around 6 ppm to Q 3 (−101 ppm) and Q 2 (−92 ppm) silica groups (Figure 9a, and schemes in Figure 9f,g). In the samples after one half or one full sALD cycle with 2 mm or pure TTIP, correlations between the methyl protons of diethyl ether with silicon atoms of Q 4 and between protons in CH 2 groups of diethyl ether with Q 3 silica sites can be found. We attribute this to surface-adsorbed diethyl ether, probably hydrogen-bonded to Q 3 sites (schematically shown in Figure 9h). Note that the correlation highlighted in red might also be due to hydrogen atoms of TTIP with Q 4 silica species. Since the chemical shifts of 1 H of methyl groups in both molecules are very close together, these two possibilities cannot be disentangled.
After one half sALD cycle with pure TTIP on calcined silica (Figure 9b), a new correlation appears at ( 1 H) ≈ 4.0 ppm and ( 29 Si) ≈ −107 ppm, which is marked in green. This 29 Si chemical shift is situated between −102 ppm, where Q 3 species are located, and −110 ppm, where Q 4 groups are located. We tentatively assign this signal to a Si-O-Ti species, explaining this unusual 29 Si chemical shift in analogy to Si(1Al) groups in zeolites, in which silicon is bound in a Q 4 surrounding, and in which one silicon neighbor is replaced by aluminium. [28] By replacing one silicon neighbor by titanium, the same effect is observed here. Along these lines, a downfield chemical shift of +7 ppm was also observed between Si(0Ti) and Si(1Ti) sites in a microporous titanogallosilicate by Rocha et al. [29] The resonance is in the following denominated as Q 4 ′. Its 1 H chemical shift is close to the chemical shift of the proton of the CH group of the TTIP molecule, that is, ( 1 H) = 4.0 ppm. The feature can also be seen in the HETCOR spectrum of calcined silica after one half sALD cycle with 2 mm TTIP (Figure 9d). Since the correlation of this peak appears in all samples after one half sALD cycle, we attribute this peak to a surface-bound Si-O-Ti(O i Pr) 3 or bidentate Si-(O) 2 -Ti(O i Pr) 2 species (Figure 9h). This is the first time that a Q 4 -like silicon species bound to one titania atom has been discerned by its 29 Si chemical shift in amorphous silica, an observation made possible by the enhanced resolution of 2D 1 H- 29 Si HETCOR spectra. In the 1D 29 Si CP MAS NMR spectra in this work, as well as in 29 Si (CP) MAS NMR spectra in the work by Pallister et al., [26] this signal was not discernible due to the overlap with the broad neighboring resonances of Q 3 and Q 4 groups.
In the sample after one full sALD cycle, the Q 4 ′ resonance does not appear anymore (Figure 9c). We do however observe a correlation signal at a 29 Si chemical shift of −109 ppm and a 1 H chemical shift of 3.2 ppm (brown). A similar signal at 3.2-3.4 ppm has been observed by Crocker et al. in titania supported on silica, and has been attributed to titanol (TiOH) groups on tetrahedral Ti sites, which are only observed in thin layers of amorphous titania. [30] In contrast, isolated titanol groups on octahedrally coordinated Ti sites, as found in anatase or rutile, typically resonate at 2.0-2.3 ppm, hydrogen bridged titanol groups around 6.7 ppm. [30] This resonance therefore clearly shows here the hydrolysis of TTIP toward TiOH groups, and the correlation between silica Q 4 ′ sites and these titanol groups proves their existence on the surface as Si-O-Ti-OH groups. Note that since lone or geminal titanol groups have the same chemical shift, we cannot discern from the spectra between different (SiO) n Ti(OH) 4−n groups, thus titanium may be grafted to the silica surface via one, two, or three covalent bonds. Moreover, the presence of only the TiOH-Q 4 ′ correlation and no TiOH-Q 3 ′ correlations implies that the scenario 2, that is the reaction of Q 2 into Q 4 ′ groups, occurs preferentially.
The spectrum in Figure 9e was recorded on non-calcined silica instead of calcined silica. Here, the aspect of the spectrum strongly changes, involving several signals which are more difficult to attribute. A new, strong correlation between Q 3 silicon atoms and CH hydrogen atoms of TTIP is found, marked in pink. This correlation reflects the high density of hydroxyl groups on the silica surface in this sample. It may be attributed either to surface-grafted TTIP, in which CH groups are close to neighboring Q 3 hydroxyl species, or to TTIP which is only adsorbed on the surface and not covalently bound (schemes in Figure 9j,k).
The spectrum of non-calcined silica after one full sALD cycle (Figure 10a) again shows hydrogen atoms of diethyl ether correlated for Q 4 and Q 3 species (red and yellow), as well as a broad signal of unreacted Q 3 hydroxyl groups on the silica surface. Importantly, no more signals of TTIP can be found, evidence for the desired complete hydrolytic cleavage and washing of all organic ligands in the second half cycle. The spectrum after 20 sALD cycles (Figure 10b) no longer shows any correlation for Q 4 and Q 4′ species, and the correlation between Q 3 species and ether is also very weak. This is most likely due to the presence of a closed multilayer of TiO 2 on the surface of the silica substrate, which weakens the CP effect, so that the resonance of Q 4 and Q 4′ species disappears. The correlation signal at 4 ppm arises from remaining physisorbed water, which may cover small residual spots of remaining TTIP inaccessible silica surface, or which may be occluded in the TiO 2 film.

Conclusions
We have developed here a first set-up for the ALD from solution of a thin film of titania onto silica. For this study, porous silica particles were used as a model support, their high surface area simplifying further analytical characterization. Furthermore, for the first time, an sALD deposited material was investigated by solidstate NMR spectroscopy. This technique is extremely well suited to track the organic and inorganic surface-deposited species, to discern bound from physisorbed species and to gain deeper insight into the reaction proceeds and its mechanism.  A tubular reactor was developed as a simple benchtop set-up to produce around 180 mg of titania-coated silica powder. The addition of silicon wafer at the entrance and at the end of the tubular reactor allowed to verify the thickness increase of the TiO 2 layer in every cycle by ellipsometry. By optimizing the operating parameters of sALD, including the precursor concentration and the pulse duration of precursors, a stable growth rate per cycle of 0.3 Å thickness increase was achieved. By 1 H MAS NMR spectroscopy, the state of the surface before sALD treatment, after one half, and one full sALD cycle, the adsorption of diethyl ether as solvent as well as the effect of TTIP concentration and washing steps could be analyzed, allowing for the optimization of these parameters. Comparing two different supports, calcination of the silica support reduces the amount of hydrogen bonded silanol groups on the surface, leading to a higher proportion of deposited TTIP, this support treatment is thus necessary for future optimizations. 29 Si cross polarization spectra showed the presence of Q 2 , Q 3 , and Q 4 groups on the silica surface and their relative change upon sALD with TTIP. TTIP molecules prefer to react with Q 2 hydroxyl groups compared to Q 3 groups, which we tentatively explain by the formation of Si-(O) 2 -Ti four-membered rings, or to stabilization via hydrogen bonding of a TTIP molecule that is bound in a monodentate way to a Q 2 group. Interestingly, 2D 29 Si HETCOR solid-state NMR spectra further show a unique signal at a 29 Si chemical shift of −107 ppm, which we tentatively attribute to a Si-O-Ti(O i Pr) 3 species. After 20 full sALD cycles, 2D HETCOR spectra show only weak correlations of silica with small amount of residual solvent or water, thus indicating the generation of a closed multilayer of titania on the surface.
In the future, the sALD set-up in the in-house designed plug flow reactor will be further improved, for example, by optimizing the contact between the silica and the fluid. Both the reactor developed here and the analytical insight can be extended to other sALD systems to produce well-controlled thin film on variable supports, while understanding their formation mechanisms on a molecular scale.

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