In-situ observation of nucleation and property evolution in films grown with an atmospheric pressure spatial atomic layer deposition system

Atmospheric pressure—spatial atomic layer deposition (AP-SALD) is a promising open-air deposition technique for high-throughput manufacturing of nanoscale films, yet the nucleation and property evolution in these films has not been studied in detail. In this work, in situ reflectance spectroscopy was implemented in an AP-SALD system to measure the properties of Zinc oxide (ZnO) and Aluminum oxide (Al2O3) films during their deposition. For the first time, this revealed a substrate nucleation period for this technique, where the length of the nucleation time was sensitive to the deposition parameters. The in situ characterization of thickness showed that varying the deposition parameters can achieve a wide range of growth rates (0.1–3 nm/cycle), and the evolution of optical properties throughout film growth was observed. For ZnO, the initial bandgap increased when deposited at lower temperatures and subsequently decreased as the film thickness increased. Similarly, for Al2O3 the refractive index was lower for films deposited at a lower temperature and subsequently increased as the film thickness increased. Notably, where other implementations of reflectance spectroscopy require previous knowledge of the film’s optical properties to fit the spectra to optical dispersion models, the approach developed here utilizes a large range of initial guesses that are inputted into a Levenberg-Marquardt fitting algorithm in parallel to accurately determine both the film thickness and complex refractive index.


Introduction
Atmospheric pressure-spatial atomic layer deposition (AP-SALD) is an emerging technique that is capable of producing conformal, pinhole-free films in open-air. Whereas conventional atomic layer deposition (ALD) operates by sequentially inserting two chemical precursor gases into a vacuum chamber, with evacuation and purge steps in between (precursors are separated in time), AP-SALD separates the precursors in space. Figure 1 illustrates the close-proximity AP-SALD approach, where a substrate is placed very close (∼100 micrometers) to a reactor head that contains parallel slits with flows of two precursor gases. The substrate can be oscillated back and forth underneath the reactor to replicate the sequential gas exposure steps and deposit a film one atomic layer at a time. The AP-SALD approach eliminates the evacuation and purge steps that make conventional ALD slow, and it does not require a vacuum chamber, which makes it scalable and potentially less costly. Furthermore, by adjusting the reactor-substrate spacing and/or flow rates, the system can deposit by chemical vapor deposition('CVD mode') [1], where some mixing of the precursors occurs in the gas phase. This atmospheric pressure chemical vapor deposition (AP-CVD) generally results in accelerated film deposition rates, while still producing conformal, pinhole-free films [2]. Growing interest in spatial ALD has led to several review papers in recent years [3][4][5][6], and companies such as Levitech, Beneq, and SoLayTec have commercialized spatial ALD systems. Both metal oxides and metals have been deposited [4,5,7], and AP-SALD and AP-CVD films have been utilized in thin film transistors [8,9], metal-insulator-metal diodes [10], photovoltaic devices [2,[11][12][13][14][15][16][17], and LEDs [18][19][20]. Yet despite the promise of these techniques, little has been done to characterize the nucleation and property evolution of these nanoscale films. This may follow, in part, from the expectation that the chemistry and film formation is identical to conventional ALD, which neglects the influence of atmospheric conditions (i.e. open-air). It was recently shown, for example, that the periodic exposure of Al:ZnO films to oxygen during deposition in an AP-SALD system results in a higher trap density at grain boundaries, which limits the electron mobility [21]. As for other nanoscale film deposition methods, an understanding of film nucleation and property evolution is necessary to permit the fabrication of AP-SALD and AP-CVD films with specific properties and desired thickness and morphology.
In-situ characterization techniques are typically used to study film nucleation and evolution. Quartz crystal microbalances which are commonly used to measure film deposition rates, are unsuitable for the AP-SALD arrangement shown in figure 1, as are reflection high energy electron diffraction (RHEED) methods employed in ultra-high vacuum environments [22][23][24][25]. Conventional ALD systems typically employ spectroscopic ellipsometry for the determination of film thickness and optical properties [26]. However, the arrangement of the optics involved tend to be 'bulky' and is not ideal to make multiple real time measurements across the substrate area. Other optical spectroscopy techniques can be more easily integrated into AP-SALD systems. Mione et al [27] recently used infrared and optical emission spectroscopy to study the gas-phase chemistry of aluminum oxide (Al 2 O 3 ) deposition in an atmospheric-pressure plasma-enhanced SALD system. Yersak et al [28] used in situ reflectance spectroscopy to monitor the growth of insulating Al 2 O 3 films deposited by a continuous web AP-SALD system. A reflectance probe was mounted above the polymer web substrate and continuously captured reflectance spectra throughout the growth. Known Al 2 O 3 refractive index values were inputted into commercial software which used the Cauchy dispersion model to determine the thickness of the deposited film, although accurate thicknesses below 20 nm could not be measured and hence this technique could not be used to study the nucleation behavior of films.
In this work, an advanced reflectance spectroscopy characterization tool is developed and implemented into an AP-SALD system. Using this characterization tool, the nucleation and property evolution of insulating (Aluminum Oxide, Al 2 O 3 ) and semiconducting (Zinc Oxide, ZnO) thin films deposited with an AP-SALD Figure 1. Schematic overview (not to scale) of the implemented method consisting of thin film deposition, measurement, analysis, and determination of film properties. Deposition in open-air was performed by AP-SALD utilizing a 'close-proximity' reactor head. The film is grown on the substrate that oscillates underneath the reactor head. Concurrently, reflectance spectra of the growing film are captured by the mounted reflectance probe and spectrometer. Each spectrum is fitted to optical models to obtain the film thickness and the refractive index (n) and extinction coefficient (imaginary part of refractive index, k). system (operated in AP-CVD conditions) are observed for the first time. Superior measurements at low thicknesses enable the direct observation of a substrate nucleation phase prior to film deposition, as well as the effect of different deposition parameters on this nucleation phase. The evolution of the complex refractive index can be observed throughout film deposition, and it is found that by changing the deposition parameters, the film nucleation can be controlled to produce ZnO films with different bandgaps and Al 2 O 3 films with different refractive indices.

Experimental
Film deposition was performed using a custom-built AP-SALD system, which employs a close-proximity reactor head configuration as described elsewhere [4,6,29]. Briefly, liquid precursors are bubbled with an inert gas (Nitrogen, N 2 ) that carries the precursor vapors to the reactor head. Before they arrive at the reactor head, they combine with additional streams of inert gas at higher flow rates that serve to push the precursors to the reactor (carrier lines). The ratio between the flow rate of the bubbler and the flow rate of the carrier line is used to represent the concentration of the precursor in the line, however, the actual concentration would be lower. As illustrated by the 'Open-Air Deposition' section of figure 1, flows of metallic and oxygen precursors are fed into the reactor head which subsequently distributes the flows into an alternating sequence of spatially separated parallel channels. In between the precursor channels exists N 2 purging and exhaust (not shown) channels. A vacuum pump was connected to the exhaust channels to actively exhaust any unreacted precursor. The reactor head is held a set distance above the heated substrate stage. Larger reactor-substrate separation heights reduce the effectiveness of the purging and exhaust channels and therefore allow intermixing of the precursors before deposition onto the substrate, analogous to chemical vapor deposition (AP-CVD mode). A single deposition cycle by AP-SALD consists of one complete oscillation (back and forth movement) of the substrate stage. This represents 2 conventional ALD cycles.
Insulating Al 2 O 3 films were deposited on silicon wafers, using trimethylaluminum (TMA) as the metallic precursor and water as the oxygen precursor. The outlet channel flow rate was set to 125 SCCM with a metal precursor concentration of 30% and an oxygen precursor concentration of 60%. The substrate temperature was varied from 100°C-250°C in increments of 50°C and the oscillation speed was varied from 15-50 mm s −1 . Semiconducting ZnO films were deposited on borosilicate glass, using diethylzinc (DEZ) as the metallic precursor and water as the oxygen precursor. The outlet channel flow rate was set to 150 SCCM with both the metallic and oxygen precursor concentrations set to 15% unless otherwise specified. These deposition conditions are similar to those previously reported to result in AP-CVD film growth [2]. The substrate temperature was varied from 50°C-200°C in increments of 50°C and the substrate oscillation speed was varied from 5-50 mm s −1 . The substrates were rinsed with isopropyl alcohol and dried with compressed air before placing on the heated substrate stage for 2min to arrive at the correct deposition temperature. In both cases, the reactor-substrate separation was set to 100 μm.
As shown in the 'Measurement' section of figure 1, a bifurcated reflectance probe is mounted above the substrate stage at normal incidence (Ocean Optics QR600). A UV-Visible light source (Ocean Optics DH-2000) coupled to the reflectance probe was used to illuminate the surface of substrate. Reflected light is directed back to the reflectance probe and captured by the UV-Visible spectrometer (Ocean Optics HDX-UV-vis) with a range of 250-800nm. Calibration of the reflectance measurement requires the measurement of the background (bare substrate), the noise in the system (dark measurement) and the measurement of a known standard (silicon). For measurements of films grown on borosilicate glass, a matte black tape was used to cover the metallic stage to remove unwanted backside reflections. The spectrometer acquisition time was set using a modified exponential search algorithm to optimize the light intensity range for the spectrometer's detector [30], resulting in acquisition times that were typically less than 6 milliseconds. Reflectance spectra were captured every two deposition cycles in the same spot.
The determination of film properties is done by fitting the measured reflectance spectrum to an optical reflectance model. This model (figure S1 in the supplementary material is available online at stacks.iop.org/ NANOX/1/010045/mmedia) consists of a three-layer stack (air-film-substrate) whose total reflectance is governed by the Fresnel reflectance equations (equations (1)- (8) in the supplementary material). Parameters of importance are the complex refractive indices of each layer and thickness of the growing film [31]. The complex refractive index is comprised of the real component n and the imaginary component k, also referred to as the extinction coefficient. The refractive indices of air and the substrate are known parameters, while the wavelength l ( )-dependent complex refractive index and the thickness of the film are unknowns. For Al 2 O 3 and other non absorbing films, the refractive index can be described by the Cauchy dispersion model [31] where l n( ) is a function of 3 fitting parameters and l = k 0 ( ) (equations (9)-(10) in the supplementary material). Including thickness, there are a total of 4 fitting parameters used to fit the measured spectrum to the reflectance model. The Tauc-Lorentz dispersion model is used for polycrystalline semiconductors such as ZnO [32,33]. Here, 9 fitting parameters are used to determine the film's complex dielectric function from which l n( ) and l k ( ) can be obtained and results in a total of 10 fitting parameters to fit the measured reflectance spectrum (equations (11)- (12) in the supplementary material).
A Levenberg-Marquardt (LM) fitting algorithm is used to fit the previously unknown parameters in the reflectance models to the measured reflectance spectra to simultaneously determine both the film's complex refractive index and thickness. The built-in function lsqcurvefit in MATLAB was used to perform the Levenberg-Marquardt (LM) fitting [34]. The function returns a goodness of fit (GOF) which is simply the R-square value. Where other implementations of reflectance fitting require previous knowledge of the film's optical properties to be used as initial guesses, the implementation discussed here makes use of several guesses in a range inputted in parallel into the LM algorithm, such that previous knowledge of the film's optical properties is not used. For both the Cauchy and Tauc-Lorentz fit parameters an initial guess value of 1 was used. For the thickness parameter, several guesses in a range centred around the deposition cycle number were inputted in parallel to the lsqcurvefit function, from which the fit with the highest GOF was selected. This was done to account for the tendency of LM fitting programs to minimize to the local minima instead of the global minima resulting in erroneous values. This method is found to be quite successful in measuring the overall growth behavior of a deposited film. However, inaccurate fitting results occasionally occur and are represented by large spikes in the thickness values that deviate from the overall growth rate. For these cases, the incorrect thickness is corrected by refitting the spectrum using the fit parameters of the deposition cycles before and after the incorrect point. Furthermore, to improve accuracy of low thickness fits, a smaller range of parallel guesses was used (±5 nm) until the film thickness reached 30nm after which the range was extended (± 50 nm). The thickness of films, as determined by reflectance spectroscopy, matched closely with the thickness measured by spectroscopic ellipsometry (J.A. Woollam M-2000DI). This validated the method described herein for the in situ thickness measurement of films deposited by AP-SALD. As shown in the 'Film Properties' section of figure 1, the LM fitting algorithm returns the film's thickness, and refractive index, as well as the extinction coefficient for semiconductors. Consequentially, the absorption coefficient a ( ) can be determined and the optical bandgap identified using the well known Tauc equation [35,36] (equations (16)- (17) in the supplementary material). A direct bandgap ( = r 1 2 / ) was assumed for ZnO films [37][38][39]. The effect of film roughness can be accounted for by introducing an additional layer in between the air and film layers of the optical model. The thickness of this layer is equated to the RMS roughness of the film, while the optical properties of the layer are described by the Bruggeman effective medium approximation [40,41]. Due to the increased complexity of the four-layer model and its longer computation fitting time, the effect of roughness was omitted in this work. This was justified by comparing the fitting results between the four-and three-layer models (shown in figure S2 of the supplementary material), which showed negligible differences. Furthermore, it was previously shown that the optically determined roughness of the film often does not reflect the true roughness of the film as determined by atomic force microscopy [41,42].

Results and discussion
Video S1 and S2 (available online at stacks.iop.org/NANOX/1/010045/mmedia) included in the supplementary material show the measured and fitted reflectance spectra for ZnO and Al 2 O 3 films during their deposition, and the evolution of the film thickness and optical properties. A summary of these videos is illustrated in figure 2 Measured reflectance spectra are shown in figures 2(a) and (d) for ZnO and Al 2 O 3 , respectively. The thickness of the film at each measured cycle is plotted (figures 2(b) and (e)) and can be used to determine the growth rate per cycle (GPC). The goodness of fit is also shown. The real part of the refractive index at a wavelength of 600nm is shown as a function of deposition cycles and is relatively consistent throughout the deposition (figures 2(c) and (f)). For ZnO, the bandgap of the growing film is also shown (figure 2(c)). Figure 3(a) shows the obtained thickness of ZnO films deposited at three different substrate oscillation speeds (15, 30 and 50 mm s −1 ). Here, substrate oscillation speed is inversely related to the precursor dose or exposure time that is typically referenced in conventional ALD literature. The substrate stage moving at a lower speed gives the precursors more time to adsorb and react on the surface of the substrate. As depicted in figure 3(a), film growth is found to occur in two stages: a nonlinear surface nucleation/incubation stage and a linear growth stage. This growth behavior is consistent with that of conventional vacuum based ALD, which has been verified by in situ crystal microgravimetry(QCM) [22], scanning electron microscopy(SEM) [43], atomic force microscopy(AFM) [44], and surface profilometry [45]. The lack of growth during the initial deposition cycles is attributed to the nucleation time to form ZnO nuclei on the bare substrate [22]. The density of ZnO nuclei created is related to the amount of adsorbed hydroxyl groups on the surface. The precursors will readily adsorb and react with the ZnO nuclei (active sites) resulting in an island-like growth. This continues until the islands eventually coalesce forming a complete layer, after which layer by layer growth behavior occurs in a linear fashion. This two-stage growth mechanism has been extensively reviewed by R.L. Puurunen who states that many ALD growth studies inaccurately claim a full monolayer per cycle growth from the start but that this has never been experimentally verified [46]. Methods to decrease nucleation time have included substrate pretreatment with oxygen plasma [43] or acid cleaning [45] to increase the hydroxyl density on the surface. More commonly, increasing the length of the precursor dose time results in an increase in ZnO nuclei formation consequently decreasing the nucleation time [44]. This effect has been reproduced by AP-SALD here, where it is seen in figure 3(a) that depositions done with quicker oscillation speeds and hence lower precursor dose times,  require more deposition cycles to populate the surface with ZnO nuclei that subsequently grow into a continuous film and enter the linear growth regime.

Film Nucleation and Growth
The goodness of fit of the obtained reflectance spectra is shown in figure 3(b) where a similar trend is observed in that quicker oscillation speeds require more deposition cycles to obtain a high GOF (>0.9) corresponding to the formation of a complete film layer on the substrate. These results show that linear growth rates cannot be assumed from the first deposition cycle, which becomes particularly releveant in thin film applications such as passivation layers and transparent coatings. The in situ reflectance spectroscopy technique introduced in this work enables the deposition of nanoscale films with accurate thickness.
Control of the ZnO film nucleation and growth rates was further explored by introducing the deposition temperature as an additional parameter. ZnO films were deposited on borosilicate glass at temperatures of 50, 100, 150 and 200°C, with substrate oscillation speeds of 5,15, 30 and 50 mm s −1 for a total of 16 ZnO samples in the study. Each film was deposited for 200 cycles, with in situ reflectance measurements taken every 2 cycles. To obtain the same dataset by an ex situ measurement technique would require the deposition of 1600 individual ZnO samples. Figure 4 compares the measured thicknesses of ZnO films deposited at different temperatures for an oscillation speed of 5 mm s −1 . Similar thickness plots for other oscillation speeds are presented in figure S4 in the supplementary material. The growth behavior results of the total study are summarized in table 1 (nucleation period) and table 2 (linear growth rates). Table 1 shows that an increase in deposition temperature results in a decrease in the number of cycles to saturate the surface of the substrate. This is in contrast to the nucleation behavior of ZnO in conventional chamber-based ALD, where higher temperatures result in desorption of the precursors from the substrate and therefore require longer nucleation times [46]. The deviation from this behavior is attributable to the highpressure open-air environment of AP-SALD which has an abundance of oxygen and hydroxyls, as well as the use of parameters that are expected to result in chemical vapor deposition. In agreement with the results obtained here, AP-CVD of ZnO films by Biswas et al showed that substrate coverage can be significantly improved by increasing growth temperatures [47]. This is because the elevated temperatures accelerate precursor gas phase reaction in air, rather than adsorption on the substrate, resulting in shorter nucleation times and quicker growth rates [48].   58 44 The data presented in table 2 demonstrate that a wide range of growth rates of 0.1-2 nm/cycle (corresponding to 0.05-1 nm per conventional ALD cycle) are achievable. An increase in deposition temperature generally results in an increase in growth rate per cycle. This is in agreement with the shorter nucleation times observed at higher temperature. The growth rate is seen to decrease as the substrate oscillation speed increases (dose time decreases). Despite operating with parameters that are expected to result in AP-CVD, the growth rates at oscillation speeds of 30 and 50 mm s −1 are comparable to ZnO growth rates achieved by conventional ALD of 0.1-0.3 nm per ALD cycle [4,[49][50][51]. Decreasing the oscillation speed to 15 or 5 mm s −1 resulted in an increase in growth rates up to 2 nm/cycle (1 nm per ALD cycle), consistent with AP-CVD, where the slower oscillation speed allows for a longer AP-CVD reaction [48]. There are a few minor deviations from the noted trends in table 2: GPC at 100°C versus 150°C for an oscillation speed of 5mm s −1 , and, 150°C versus 200°C for oscillation speeds of 30 mm s −1 and 50mm s −1 . Since many of these depositions occurred one after another, one possible explanation to the higher growth rate at a lower temperature can be due to an accumulation of precursor in the line. Additionally, the spacing between the stainless-steel reactor head and the substrate was manually set to 100μm at the start of the study. It is possible that at the lower temperature the substrate thermally expanded less and therefore resulted in a spacing that was larger than 100μm and promoted more precursor mixing resulting in the higher growth rate. Another possible explanation lies in classical ALD theory, where higher deposition temperatures outside of the 'ALD window' result in desorption of the precursor from the substrate [52][53][54][55]. This phenomena may be more relevant to higher growth rates at 150°C than 200°C for both 30 and 50 mm s −1 oscillation speeds, whereas the anomaly observed at 100°C versus 150°C is most likely due to the previous two explanations.
A similar study to examine the influence of deposition parameters was done for Al 2 O 3 films. Al 2 O 3 films were deposited at temperatures of 100, 150, 200 and 250°C with oscillation speeds of 15, 30 and 50 mm s −1 on silicon substrates (12 samples in total, 100 cycles each). Figure 5 compares the measured thickness of films deposited at 200°C for different oscillation speeds. Similar thickness plots for other deposition temperatures are presented in figure S4 in the supplementary material. The growth behavior results of the total study are summarized in table 3 (nucleation period) and table 4 (linear growth rate). While the growth behavior for ZnO  closely followed the expected trends, this is not the case for Al 2 O 3 . For example, as shown in figure 5, the nucleation period is larger for an oscillation speed of 15mm s −1 compared to 30 mm s −1 , where the opposite would be expected. There are no clear trends in the nucleation period, as presented in table 3. Additionally, the rates presented in table 4 do not exhibit clear trends in growth behavior. As reported previously, the atmospheric ('open-air') deposition of Al 2 O 3 using the highly pyrophoric precursor TMA, can be challenging to reproduce [56]. The ambient oxygen and water in the air (humid conditions) can act as an additional oxygen source that can increase growth rates [57]. Human error in adjusting the reactor head spacing can lead to an overmixing of the precursors (AP-CVD) and deposition with high rates [5]. Jur and Parsons concluded that the high pressures of atmospheric deposition will typically increase Al 2 O 3 growth rates by ∼25%-30% [58]. More importantly, Mousa et al noted that at high atmospheric pressures the growth of Al 2 O 3 with TMA is largely independent of temperature [59]. It is therefore difficult to observe the influence of temperature and oscillation speed on the Al 2 O 3 growth. However, it strongly supports the need for an in situ thickness characterization tool for Al 2 O 3 grown in atmospheric conditions using this technique. Added benefits of monitoring the film thickness in situ include the ability to monitor for deposition errors and to control the deposition rate. Figure 6(a) depicts an example of error monitoring during the deposition of an Al 2 O 3 thin film. The linear growth regime starts after 20 deposition cycles and after 36 cycles the vacuum pump for the exhaust channels was turned on which immediately reduced the growth rate. In this case, the lack of vacuum exhaust during the initial growth period resulted in a higher amount of precursor mixing and consequentially a higher growth rate. Figure 6(b) illustrates the ability to control the deposition rate of a ZnO thin film by varying the precursor (DEZ) flow rate. The initial 100 deposition cycles were done at a concentration   of 15% to first saturate the surface and subsequentially enter the linear growth regime. After 100 cycles the concentration was increased to 25% resulting in a stark increase in the growth rate per cycle. After another 50 cycles the concentration was decreased to 5% resulting in a decrease in growth rate which was followed by an increase in growth rate once the concentration was set back to 15%. The large variation in growth rate is consistent with AP-CVD type growth as discussed earlier.

Evolution of optical properties
As previously described, the bandgap E g ( ) of semiconducting films can be obtained from in situ reflectance spectroscopy and thus can be monitored over the deposition period. Figure 7 shows the bandgap evolution of growing ZnO films deposited at different temperatures. The E g data was taken after the completion of the nucleation period and when high GOF was achieved. As observed in the figure, the obtained bandgaps range from 3.39-2.95 eV, where films deposited at a lower temperature have a higher initial bandgap. This is consistent with previous reports, where three reasons are presented. The first attributes the bandgap temperature dependence to the thermal expansion of the ZnO lattice and its direct effect on the relative position of the valence and conduction band [60]. The second is the temperature dependence of the electron-phonon interactions and their affect on the band energies [61]. Lastly, it is known that the higher deposition temperatures result in an increased crystallinity for ZnO films grown by this method, and the decrease in bandgap has been attributed to the improved crystallinity [62]. The bandgap then decreases from the initial value as the film grows which is attributed to the size of the grains as well as the porosity in the nanocrystalline film. The early stages of film growth will consist of smaller crystallite sizes corresponding to higher film porosity. As the film grows the crystallite sizes become larger, the film becomes more compact and the overall film porosity decreases. Like quantum dots, the optical bandgap of a nanocrystalline ZnO film will decrease as the grain size increases [39,63]. Similarly, the bandgap is expected to decrease as the porosity of the film decreases as well [64,65]. The general trend of decreasing bandgap as deposition temperature increases has been shown for ZnO films grown by conventional ALD methods [66,67]. However, values as low as 2.9 eV (deposited at 200°C) have not, to our knowledge, been reported for ZnO grown by ALD. As described by the Varshini equation [68], the bandgap of semiconductors tends to decrease as a function of the measurement temperature. The sample grown at 200°C was re-measured at room temperature and had a bandgap of 3.07eV which is still low when compared to that of ALD grown ZnO. This may be attributed to the deposition done in atmospheric conditions, as AP-CVD grown ZnO has reported bandgaps as low as 2.8eV [69].
The refractive index of a thin film can give an indication of film quality and density. Figure 8 shows the refractive index (at 600 nm) of Al 2 O 3 films deposited at different temperatures. Values obtained before the completion of the nucleation period were ignored. The initial formation of the films is observed as the refractive index starts at zero or near zero values and increases to a value of 1.4-1.5, which is the expected refractive index range for Al 2 O 3 grown by conventional ALD [70,71]. It is clear that the refractive index of the film grown at 100°C is lower compared to the films grown at higher temperatures. This is attributed to the decreased density and increased impurity levels of Al 2 O 3 grown at lower temperatures [71][72][73]. It is expected that higher temperatures would promote a higher degree of crystallinity and higher density resulting in an increase in refractive index approaching a value of 1.76, that of sapphire (α-Al 2 O 3 ).

Conclusions
An in situ reflectance spectroscopy tool was developed and implemented with an atmospheric pressure-spatial atomic layer deposition system. A reflectance probe mounted above the deposition stage was able to capture reflectance spectra every two deposition cycles of growing insulating (Al 2 O 3 ) and semiconducting (ZnO) metal oxide thin films. The spectra were fitted to optical reflectance models in real-time to obtain properties such as thickness, refractive index and optical bandgap (for semiconductors). Goodness of fit values for each fitted spectrum were obtained to indicate the level of confidence in the obtained properties.
The approach works well to study the nucleation and property evolution of metal oxide films deposited with this open-air technique. A two-stage growth mechanism was observed, consistent with previous reports of an initial nucleation period to saturate the surface with metal oxide nuclei that eventually grow into a continuous film after which, linear growth occurs. In-situ thickness characterization is therefore crucial to enable to the accurate deposition of films with a desired thickness, as linear growth rates cannot be assumed from the start.
The influence of deposition parameters such as substrate oscillation speed, temperature and precursor concentration, on the growth behavior of the films was studied. The length of the nucleation period for ZnO depositions was sensitive to the oscillation speed and temperature. Faster oscillation speeds resulted in a longer nucleation period and higher deposition temperatures shortened the nucleation period, which was consistent with AP-CVD growth. Similarly, a wide range of growth rates were achievable (0.1-3 nm per AP-SALD cycle) by varying the deposition parameters. Furthermore, the high reactivity of pyrophoric precursors deposited in open atmosphere can result in errors that appear as sudden changes in growth rate during a deposition. The real time monitoring of growth rate allows the user to ensure film growth is occurring as expected.
The evolution of optical properties such as bandgap and refractive index of the growing film was observed and followed expected behavior relating to the crystallinity and density of the thin films. It was clear the deposition temperature influenced the optical properties of the film. The bandgap of ZnO films decreased as a function of deposition cycles and increasing temperature. The low bandgap of ZnO obtained at higher deposition temperatures was consistent with reported atmospheric CVD grown ZnO. The trend of increasing Al 2 O 3 refractive index as a function of deposition temperature was similar to that of films grown by conventional ALD. An ex situ characterization technique such as x-ray photoelectron spectroscopy [10] or ion beam analysis [74,75] is warranted to study variation in the film's chemical composition and its effect on the optical properties.
The nucleation and property evolution for films deposited with an AP-SALD system has not been studied in detail previously. By implementing in situ reflectance spectroscopy, the growth behavior and film properties can be obtained and monitored in real time during a deposition. The observed growth behavior and optical properties of the films were comparable to values reported for atmospheric pressure CVD.

Acknowledgments
K Mistry would like to thank V H Nguyen and D Muñoz-Rojas for sharing their knowledge, guidance and training in the initial stages of this project. K Mistry acknowledges the support of the Mitacs Globalink research

Conflicts of interest
There are no conflicts to declare.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.