Effects of deposition temperature on the mechanical and structural properties of amorphous Al – Si – O thin films prepared by radio frequency magnetron sputtering

Aluminosilicate (Al


Introduction
Aluminosilicates are composed of the 2nd and 3rd most abundant elements in the Earth's crust, Al and Si [1], and is thus one of the most important materials in geoscience [2].It is also among the most important materials in glass technology [3].Because of its properties, high chemical stability and refractory character, it is also of potential use as corrosion-and wear-resistant coatings [4] and gate insulator in metal oxide semiconductor (MOS) devices [5][6][7].The deposition rate, composition, structure, morphology and properties of aluminosilicate thin films are influenced by their synthesis conditions [8].Aluminosilicate thin films have previously been prepared by a variety of methods including sol-gel methods [9,10], chemical vapor deposition (CVD) [11,12], and physical vapor deposition (PVD) [7,[13][14][15][16], which all have their advantages and disadvantages [17].PVD magnetron sputtering has become widely used in industry because of its versatility and proven scale-up capabilities for industrial production [18].
However, there are also suggestions that 2, 3 or 4-fold coordinated O may be present to fulfill the charge compensation [45,[47][48][49]].Löwenstein's avoidance rule suggests that the framework is constructed by Si-O-Al bridges [50] which implies that a [AlO 4 ] unit is surrounded by four [SiO 4 ] units, a concept that generally applies for aluminosilicates, e.g., zeolites.Löwenstein's avoidance rule may be violated if the Al/Si ratio is >1 [51,52].Vibrational spectroscopy is a useful tool to study chemical bonding in glass.The vibrational spectroscopy band assignments of silica-rich amorphous compositions are well-studied [53][54][55][56] but band assignments of Al 2 O 3 -rich compositions are less known [57][58][59].
This work is a follow-up study to our previous report on reactive radiofrequency magnetron co-sputtering synthesis of Al-Si-O coatings and their optical properties [14].In bulk, Al 2 O 3 -SiO 2 , have shown to have good crack resistance [60] which makes the investigation of Al-Si-O thin films interesting for damage resistance, in addition to wear and corrosion resistance.In this study, we analyze the mechanical properties and the structure of Al-Si-O thin films by indentation and infrared spectroscopy.Furthermore, we explore the possibility of decreasing the coating process temperature as a route towards energy efficient magnetron sputtering by investigating the deposition temperatures 500, 300 and 100 • C.

Materials and deposition
The substrate was a 1 mm float glass substrate in the form of conventional microscope slides.The typical composition (given in wt%) is 72.3 % SiO 2 , 0. Al-Si-O thin films were deposited on the atmospheric side of float glass by radio frequency (RF) magnetron reactive co-sputtering using elemental targets of Al and Si in ultra-high vacuum (UHV).The deposition system is described in detail elsewhere [61].An oxygen and argon mixture were kept constant with a total gas flow of 40 mL/min (O 2 mL/min and Ar mL/min) at standard temperature and pressure, STP, conditions.The Al target power was kept constant at 100 W and the Si target power was varied between 40, 60 and 80 W. In addition, the substrate temperature was varied between 100, 300 and 500 • C giving a sample matrix of 9 samples, see Table 1.Complete details of the physical vapor deposition processing are given in ref. [14].

Nanoindentation
Nanoindentation with a Berkovich tip was used for measuring hardness (H) and reduced elastic modulus (E r ) according to the Oliver and Pharr method [62], where H is defined by H = Fm Ap , with F m is the maximum applied load and A p the projected contact area.A p is calculated by polynomial fitting: , where C x is the indenter specific factors with C 0 being 24.56 for a perfect Berkovich tip and h c is the real contact depth that compensates for the sink-in effect by h c = h m − ε Fm S , where h m is the maximum penetration depth, ε is a tip factor for a Berkovich tip (ε = 0.72) and S is the stiffness defined by the slope upon unloading, S = ∂P ∂h .The reduced elastic modulus is defined as Ap √ where β is a geometrical tip factor, which is β = 1.034 for a Berkovich tip.
The samples were cleaned with ethanol and laboratory tissue before measurements.The nanoindenter instrument was an Anton Paar NHT 2 and the measurements were run with the loads 1, 5, 10, 15, 25, 50 and 75 mN.20 indents were made for each load.In some cases, one indentation outlier datapoint, or for the 1 mN load, sometimes two datapoints, was removed from the analysed data due to unrealistic scattering.The measurement settings were the following: 10 Hz acquisition rate, loading/unloading rate two times the load per minute, 10 s holding time at max load, 0.2 µm/N frame compliance, and 500 µN/µm stiffness threshold.The Berkovich tip geometry was calibrated using a standard reference material of fused silica and the measurements were performed at ambient pressure at a temperature of 22±2 • C and a relative humidity of 30 ± 15 %.

Microindentation and crack resistance
The crack resistance was determined from microindentation measurements using a Vickers indenter tip that were performed using a Micro-Combi Tester from CSM Instruments.15 indents for each load were made using a Vickers diamond tip.The microindentations were run with the following settings: 10 Hz acquisition rate, loading/unloading rate two times the load per minute, 15 s holding time, 8 µm/min approach speed, 16.6 µm/min retract speed, 30 mN contact force, and 25 mN/µm contact stiffness threshold.The Vickers indenter tip was calibrated using a standard steel reference material.
The crack resistance method is described in detail elsewhere [20], and follows the original procedures [63,64].From the microindentation imprints the probability of radial crack initiation (PCI) was calculated from counting the radial corner cracks and the results were fitted to the Weibull sigmoidal function, where x is the load, x c the characteristic values and m is the Weibull modulus.The crack resistance, CR, is then defined as the load when the PCI is equal to 0.5 (50 % probability).All microindentations were performed in an environment with a temperature of 22±1 • C and a relative humidity of 20 ± 10 %.
The error of the Weibull fit was calculated by the corresponding load deviation as given by the root-mean-square deviation, RMSD, of the PCI, Table 1 Target power (P), substrate temperature, chemical composition, thickness (d) (all data reproduced from [14]), as well as nanoindentation mechanical properties: hardness (H), reduced elastic modulus (E r ), H/E r , U E /U T and H 3 /E r 2 of Al-Si-O thin films.
Si P (W) where PCI real is the experimentally determined PCI from the crack resistance test, PCI fit is the fitted PCI, N is the number of different loads tested in the series and i is the specific load tested in the series.The error of CR is estimated to be in the range 10-20 % [65].

Infrared spectroscopy
Specular Reflectance FT Infrared Spectroscopy was measured in the range from 600 to 4000 cm − 1 using a Bruker IFS 66v/S employing a liquid N 2 cooled MCT detector.The angle of incidence was 30 • .Spectra were averaged over 33 reflectance scans at a resolution of 2 cm − 1 .A microscope slide was used as background.The reflectance spectra were cut to the region of interest, 600-1400 cm − 1 , and were then baselinecorrected with an adaptive baseline (coarseness 75 and offset 0) which was individually applied on all spectra using Spectragryph [66].Reflectance spectra were converted to Kubelka-Munk (KM) function by [67].OriginPro version 10 was used for deconvolution of the baseline-corrected KM transformed spectra.

Hardness and reduced elastic modulus
For measuring hardness (H) and reduced elastic modulus (E r ) of coatings [68] as rule of thumb, the penetration depth should not exceed >10 % of the coating thickness to avoid substrate effects [69].H and E r was measured using nanoindentation methodology, which is fairly stable down to loads of the order 1 mN for hard materials such as amorphous oxides.Loads up to 75 mN were measured to explore whether the results are consistent also at higher loads.The results are presented in the Supplementary, Figs.S1 and S2.H increases with the substrate temperature as a function of the Al concentration, see Fig. 1a and Table 1.Fig. 2a also shows H at different substrate temperatures.E r on the other hand increases linearly for Si target powers of 80 and 60 W as a function of Al concentration (Fig. 1b), and at 40 W E r has a maximum for 300 • C. In Fig. 2b E r is also presented at different substrate temperatures, where it follows an increasing trend with the Al concentration.H is less easily described in series of the substrate temperatures where it is less obvious with a trend as a function of the Al-concentration, see Fig. 2.
Table 1 presents the parameters H/E r , U E /U T (elastic energy of recovery) and H 3 /E r 2 used for assessing the tribological properties where, as a rule of thumb, materials having H/E r >0.1 are wear resistant [39,70].H/E r can then also be directly connected to the elastic recovery energy (U E /U T ) in the indentation process given by where κ is a proportionality factor, κ ≈ 5.17 in the range 0.08-0.12for H/E r [71].H 3 /E r 2 is a measure of the resistance to plastic deformation [65,72].
The synthesized Al-Si-O thin films exhibit good tribological properties and for those prepared at a substrate temperature of 500 • C exhibit excellent tribological properties, i.e., H/E r ≈ 0.1 [39].The trends of H/E r , U E /U T , and, H 3 /E r 2 follow the same as H as a function of Al concentration, and thus increases with increasing substrate temperature.
H/E r and H 3 /E r 2 can in general be seen as a proxy for the fracture toughness of hard coatings [73], but it is unclear how the ductility affects the proxy.

Crack resistance
The Crack resistance (CR) as a function of Al concentration is shown in Fig. 3 and the probability of crack initiation (PCI) as a function of load is shown in Fig. S3.In contrast to H, E r and the tribological parameters, CR decreases with increasing Al concentration.CR can be seen as a measure of the ductility, thus its ability to protect the coated glass from hard contact damage.Ductility covers both densification and plastic deformation (the latter is given by UP UT = 100 − UE UT , see Table 1).Since the trend is opposite to the tribological parameters the CR is in this case governed by densification and thus a low Al/Si ratio facilitates a more open structure that can densify.The Al-Si-O coated glass substrates frequently showed delayed cracking, which is a relatively common feature for bulk glasses [74].We recorded both the immediate CR, and the delayed CR which reduced the CR by about 5 to 10 %, see Table 2.

FT-IR spectroscopy
Fig. 4 shows specular FTIR spectra for thin films prepared at different substrates temperatures.Raw FTIR spectra are shown in Fig. S4.The trends in Fig. 4 are very clear at all substrate temperatures, although specular FTIR spectra are sensitive to the surface roughness [75], however, the surface roughness range (2.0-0.9 nm) for our coatings [14] is small compared to our observed spectral changes in Fig. 4, and does not follow a sample treatment trend.The broad peak centered around 1000 cm − 1 consists of at least two different vibrational modes, which can be assigned to the asymmetric stretching vibrations of Si-O-Si/Al [51,[76][77][78] at 1000 cm − 1 and Al-O-Al [57,58] at 1050 cm − 1 .The intensity of the 1000 cm − 1 band increases drastically with increasing Si target power, and the relative FWHM/Height ratio decreases with Fig. 1. a) H and b) E r in Si target Power series as a function of the Al concentration in at% (as taken from [14]).The solid lines are guides for the eyes.
S. Karlsson et al. increasing substrate temperature, i.e., the peak gets more narrow, both trends suggesting increased structural order on a local molecular scale, even though the thin films are X-ray amorphous [14].The smaller peak at 800 cm − 1 is assigned to Si-O-Si bending and/or stretching vibrations [76] and increases with increasing Si target power.The band at 1200 cm − 1 which is assigned to asymmetric Al-O-Al stretching modes, increases with increasing Si target power and generally red-shifts with increasing power [57].At 80 W Si target power, a band at 1350 cm − 1 consistently appears at all substrate temperatures, and is tentatively assigned to O-Al --O-chains [59].Based on these band assignments we have chosen to deconvolute the wavenumber range 600-1400 cm − 1 , see Fig. S5, using the band assignments given in Table 3.The deconvoluted Fig. 2. a) H and b) E r in substrate temperature series as a function of the Al content in at% (as taken from [14]).The solid lines are guides for the eyes.

Table 2
Direct and delayed crack resistance (CR) including Weibull fitting parameters, the characteristic crack resistance (x c ) and Weibull modulus (m) (see Eq. ( 1)).The error represents the Weibull fit error.

Direct Crack resistance
Delayed Crack resistance data is used and discussed in relation to the mechanical properties in Section 4.

Discussion
The composition-property trends are relatively clear.Al 2 O 3 -SiO 2 in bulk has shown increasing trends of H and E with increasing Al concentration [60], consistent with our results.In contrast, bulk CR has shown to increase with the Al 2 O 3 concentration up to 60 mol%, in contrast to our work where the CR is reduced with increasing Al concentration.This rule out an otherwise possible explanation, i.e., that the lower bond energy of Al-O would cause the lower CR with increasing Al content.Instead, this can be explained by the variable density of our thin films in combination with lower thin film thickness with increasing the Al concentration, see Table 1 and ref [14].Absolute values of CR of bulk and Al-Si-O thin films can thus not be compared since thin films to a large extent is governed by the substrate beneath, in our case a soda-lime-silicate with CR ≈ 0.7 [20,79].
The trends are nonlinear and sometimes more linear and the guiding solid lines should be seen as more or less deviation to linearity, see Figs. 1-3.Possible explanations could be the process differences [14], a mix of two network formers (i.e., the network forming effect [80]) or that Si preferentially forms tetrahedral units [81,82].Al on the other hand is known to exist in 4, 5 or 6 coordination alternatives, and it is reasonable to assume that the average coordination increases with increasing Al concentration [3,83,84].We hypothesize that the increase of the 1350 cm − 1 IR band indicates an increase of -O-Al=O-chains at higher substrate temperature and higher Si target power, indicating a possible increase of the local order and/or degree of crystallinity.From the measured composition given in Table 1 we calculate that the mismatch to stoichiometry, i.e., missing O atoms, is in the range from 2.5 % to 10 %.The non-stoichiometry, that depends on the deposition conditions, increases with substrate temperature but there is no clear trend with the Si target power.
The compositions and IR intensity trends are clear, especially the 1000 cm − 1 band assigned to the Si-O-Si/Al asymmetric stretching.The deconvoluted intensity data for this band has a quite linear correlation with the Al concentration, see Fig. 5a, but the linear fits should not be over-interpreted.Furthermore, the deconvoluted 1050 cm − 1 Al-O-Al band shift relatively linearly with Al concentration, giving a blue-shift of the peak with increasing Al concentration, see Fig. 5b.Therefore, the Si concentration as well as H, E r and CR correlates linearly with both the 1000 cm − 1 intensity and 1050 cm − 1 band shift.The blue shift can be observed as the Al concentration increases in the respective series (Fig. 5) indicating a decrease of the average bond length [53], thus the thin films become denser, as is manifested by the increase of H (Fig. 1a).On the other hand, the red-shift of the 1050 cm − 1 band that can be observed with increasing substrate temperature (see Fig. 5), can be explained by an increasing average coordination number of Al that counteracts the supposed blue-shift as the PVD process temperature increases the residual stress and thus also the density of the thin film [85], which we can see in the increase of the H and E r .An increase in the Al average coordination number would in fact decrease also the CR as it becomes a less adaptive network [86].In conclusion, we hypothesize that an Al-Si-O thin film structure develops at low Al/Si ratio that follows the Löwenstein's avoidance rule comprising Al-O-Si bridges [51,52].However, with increasing Al/Si ratio Löwenstein's avoidance rule is violated, and AlO x -rich clusters are formed that promote crystallization and increase the Al average coordination number.

Conclusions
Al-Si-O amorphous thin films were synthesized using radio frequency magnetron sputtering with different Si target power and substrate temperatures giving a 3 × 3 matrix.The thin films having Al/Si ratios of approximately 0.6 to 3 exhibits trends in both their mechanical properties as measured by indentation and in their IR spectra.Higher substrate temperature and low Si target power gives a higher Al/Si ratio and a denser structure that results in higher hardness and reduced elastic modulus but less crack resistance.A strong vibrational band at 1000 cm − 1 that is assigned to the Si-O-Si/Al asymmetric stretching, whose intensity increases with increasing sputtering target power, can be interpreted in terms of increased structural ordering.A vibrational band assigned to the Al-O-Al shift around 1050 cm − 1 shifts to higher frequency with increasing Al concentration We hypothesize that it is due to a decreased average bond length, manifesting a denser structure that is evidenced by the increase of the hardness.The Al-O-Al band around 1050 cm − 1 red-shifts with increasing substrate temperature, indicating an increase of the Al average coordination number.We emphasize that IR band assignments at high Al/Si ratios are not well established in literature.Therefore, further studies are needed to complete the

Declaration of Competing Interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Per Eklund, Jens Birch and Sharafat Ali are co-founders and coowners of a startup company, HPViCo AB, aiming to commercialize hard transparent coatings.

Fig. 3 .
Fig. 3. Direct Crack resistance (CR) as a function of Al concentration in a) series of substrate temperature and b) in series of Si target power.The solid lines are guides for the eyes.

Fig. 4 . 3 − 1 S
Fig. 4. FTIR spectroscopy spectra of reactive co-sputtered Al-Si-O thin films prepared with substrate temperatures a) 100 • C, b) 300 • C and c) 500 • C. Solid lines are spectra after baseline-correction and KM-transformation, see Section 2.4, and dashed lines are deconvolutions of the 1000, 1050 and 1200 cm − 1 bands.

Fig. 5 .
Fig. 5. a) Deconvoluted 1000 cm − 1 band intensity and b) deconvoluted 1050 cm − 1 band shift as a function of Al concentration.See deconvoluted spectra in Fig. S5.The solid lines are linear fits.