Growth per cycle of alumina atomic layer deposition on nano- and micro-powders

Use of nanopowders for technological applications is growing because of their size specific properties, such as high surface area to volume ratio and quantum confinement effects, which can lead to unique physical, optoelectronic, catalytic, mechanical, and chemical properties when compared to their bulk counterpart. Often times the particles of powders are engineered via coating thin films of secondary materials as active or protective layers for tailoring or preserving the properties of the core particles. For example, TiO₂ particles used as pigments in paints are coated with an alumina (Al₂O₃) thin film to protect against weathering.¹ Similarly TiO₂ nanopowder used as an ultraviolet B (UVB) and UVA absorber in sunscreen is coated with a thin ZnO film to enhance absorbing efficiency of UVA.² In both of these applications and common in applications of core/shell particles, the thickness of the shell on a core particle plays a critical role in the overall properties of the composite core-shell. With an increase of alumina shell thickness, the resistance of TiO₂ particles to weathering is improved and for thicknesses of a ZnO shell beyond ∼4 nm, UVA absorption ability of TiO₂ does not increase significantly.^1,2 Therefore, knowledge about growth rate/growth per cycle (GPC)³ is crucial to be able to control precisely the minimum required thickness of shells.

For uniform and complete coating of a thin film on a powder substrate, particle atomic layer deposition (pALD), to date, is the most reliable technique.⁴ In the growth of thin-films via ALD, alumina is one of the most widely studied materials, with studies typically focusing on flat surfaces using various combinations of precursors including the most common one; trimethylaluminum (TMA)/water.^5–9 It is well known that for flat surfaces, the film GPC of alumina using TMA/water chemistry is ∼1 Å at a growth temperature ∼150 °C,^10–12 but studies on ALD growth of alumina on powders are not as extensive as that on flat surfaces. Interestingly, prior works on the growth of alumina on different nano and micron size powders have unanimously reported a GPC of ∼2 Å, which is larger by a factor of ∼2 as compared to the typical GPC on flat surfaces.^13–20 However, for powders of bigger sizes, such as 5 and 150 μm, growth rates were reported to be more similar to that on a flat surface, ∼1.1 Å/cycle.²¹ While much work has reported on the variation of GPC as a function of growth temperature, precursor pulse, and purge duration,^12,22–25 there are no systematic studies on GPC as a function of particle size despite the obvious trends already observed in the literature.

For the study of GPC, we selected to grow alumina films on tungsten (W) powder because (1) ALD chemistry of alumina is well known and studied, (2) this substrate/film material combination is perfect for core–shell analysis by electron microscopy since the large difference in molar mass between alumina and tungsten provides high contrast between the core and shell in the micrographs for easy determine of film thickness and hence GPC, and (3) alumina, which has a very close matching thermal expansion coefficient of ∼6 × 10⁻⁶ K⁻¹ (Ref. 26) compared to ∼4.5 × 10⁻⁶ K⁻¹ for tungsten,²⁷ is an excellent match as a preserving coating for tungsten, which is itself a technologically interesting and important material. Tungsten and tungsten-compounds are used in a wide range of modern, traditional, and military technological applications including use as lighting filaments, electrocatalyst for fuel cell, electrical contacts, electronic heat sinks, kinetic energy penetrators, cathodic electrochromic materials, cathode matrixes, as well as various chemical uses.^28–30 Tungsten is also the main constituent in cemented carbides (i.e., hard metals), which are formed by powder metallurgy. The wide range of application is due to its outstanding physical and chemical properties, such as its very high density and melting point, high strength, low reactivity/toxicity, and low thermal expansion coefficient.

In this work, the GPC during atomic layer deposition is compared for different batches of powder with average particle sizes ranging from nanometer to micrometer. Samples prepared after depositing thin alumina films (from ∼10 to 15 nm) on tungsten powders using pALD were investigated with x-ray photoelectron spectroscopy (XPS), scanning electron microscopy (SEM), and transmission electron microscopy (TEM).

Coating of powders was carried out in a custom pALD Nanosolutions, Inc., rotary reactor integrated for glove box use. The reactor is constructed from stainless steel parts. The main chamber is a standard 6 in. vessel, which can be heated via a jacket heater up to 200 °C. As shown in Fig. 1(a), the reactor chamber is integrated for glovebox access at one side of the glovebox, while at the other side of the glovebox, a high temperature reactor is integrated for hydrogen reduction of the powders prior to coating thin film shells on them [Fig. 1(a)]. Powders are loaded in a reactor drum of size ∼5 in. long by 1.5 in. diameter [Fig. 1(b)], which fits to a rotary shaft via screw. The reactor drum is magnetically coupled to a stepper motor for controlled rotation [Fig. 1(b)] and allowing the ALD reactor to be completely isolated from the glove box environment. A base pressure of better than 1 × 10⁻²Torr is maintained using a mechanical pump with a pumping speed of 8 m³/h. A thermocouple, which makes physical contact with the outer body of the drum, measures the temperature of the reactor drum. Reactants are fed into the chamber via ¼ in. stainless steel tubing [Fig. 1(b)]. The reactants are in a manifold, which has two separate lines side by side to protect from mixing of oxidizing (water) and reducing precursors (metal-organic). All the feeding lines of precursors, the manifold, and valves can be heated to prevent condensation of precursors using control heating arrangements.

Four tungsten (W) powders with different average particle sizes were purchased from U.S. Research Nanomaterials, Inc., MTI Corporation, and Alfa Aeser. The manufacturer specific particle size distributions for the powders are indicated in the Table I, and the purity of all the powders are ≥99.5%. The surface of tungsten typically oxidizes to WO₃ and WO₂ upon exposing to an ambient environment;³¹ therefore, all as-received W-powders were reduced in the high temperature reactor in flowing hydrogen environment. During hydrogen reduction, the temperature of the reactor was ramped to 1000 °C in 2 h and held at 1000 °C for 10 min, followed by furnace cooling down to RT. Particles size distributions of the samples used for analysis were determined from SEM images of the cleaned powders [for example, Figs. 2(a) and 2(b)]. For each tungsten powder, the particle size distributions were determined from around four images each consisting of 200–300 particles taken at different regions in the samples. The size distributions for the powders in Figs. 2(a) and 2(b) are shown in Figs. 2(c) and 2(d). Based on the analysis from SEM images, the particle size range indicated in Table I for all the samples represent reasonably a large number of the particles (i.e., “majority”) in the respective powder. The SEM determined majority particle size distributions are presented alongside the manufacturer specification in Table I. Hereafter, the powders will be referred to by the median of their majority size distribution in Table I; 54, 85, 340, and 970 nm. Powders 54 and 85 nm had reasonable numbers of spherically smooth particles while 340 and 970 nm had mostly polyhedron-type particles where particle surfaces were made from many flat surfaces [Figs. 2(a) and 2(b)]. For ALD studies, 2.5 ± 0.3 g of each cleaned powder were used. Prior to each ALD experiment, the powders were heated for at least 2 h at the coating temperature of 150 ± 10 °C with the drum rotating at 140 rpm to ensure thermal equilibrium of the powders, drum, and reactor.

TMA (97%, Sigma-Aldrich, in a Sure-Pac) and in-house DI H₂O (18.2 MΩ in a 25 cm³ stainless steel canister, prepared from a “Barnstead Thermolyne NANOpure Diamond Analytical” ultrapure water system) were used as metal and oxidation precursors, respectively. Both precursors were held at RT and fed into the reactor by means of its own vapor pressure. For purging, 99.999% purity N₂ was used. Static exposure of precursors was adopted for ALD process. To minimize condensation of precursors on the walls of precursors' lines and measuring/controlling components, all connecting tubing not heated by the reactor were heated to ∼110 °C.

The reactants were dosed using the “micropulse”¹⁸ scheme, where a number of small doses of a precursor were repeated in a row until a substrate surface was saturated (see pressure versus time plot in the  Appendix). Numbers of microdoses of a reactant needed to saturate the sample were determined from observation of a signature pressure transient. The reactant exposure sequence for each cycle was as follows: (1) TMA microdoses (3 s each), (2) TMA reaction time after each microdose (40 s each), (3) pump out excess TMA and reaction products (10 s), (4) N₂ purge (180 s), (5) pump out N₂ (10 s), (6) H₂O microdoses (3 s), (7) H₂O reaction time after each microdose (40 s), (8) pump out excess H₂O and reaction products (10 s), (9) N₂ purge (240 s), and (10) pump out N₂ (10 s). For both TMA and water, the numbers of microdoses were from three to four and partial pressures were ∼0.4 Torr. These steps were repeated in sequence to deposit Al₂O₃. Precursor doses and pump/purge times were not rigorously optimized.

XPS and TEM analysis were done ex situ. For XPS analysis of the coated powders, pellets were made from the powders inside the glovebox using a powder die and a laboratory bench press. Porosity of the pressed pellets was ∼35%. Samples were XPS characterized using a Thermo Scientific K-alpha. The XPS spectra presented were taken after 500 eV Ar⁺ ion sputtering for 45 s to get rid of adventitious carbon and oxygen. A monochromatic aluminum Kα (1.487 keV) x-ray source with a spot size of 400 μm was used. All the survey (two scans) and high resolution (HR) elemental XPS (five scans) were taken at 200 and 20 eV analyzer pass energy, respectively. All the XPS were taken with a low energy and current [<1 eV and 130 μA, e⁻ (plus Ar⁺)] flood gun on. All high resolution spectra were referenced to the Au 4f_7/2 peaks at 84.0 eV. The TEM samples were made by suspending small amounts of coated tungsten powders in chloroform and drop-casting a small volume (<1 μl) onto lacey carbon TEM grids. TEM images were obtained on a JEOL 2200FS transmission electron microscope operating at 200 kV. The CCD camera on which the images were recorded was calibrated with a gold lattice magnification standard.

Figures 3(a)–3(d) show TEM micrographs of particles of 54, 85, 340, and 970 nm W-powders after 52, 80, 82, and 83 cycles, respectively, of TMA/H₂O at 155 ± 10 °C. Due to high differences in mass-thickness contrast, the coated samples show clear distinction between the ALD coated alumina and W-substrate in TEM. The center dark cores are the substrate W-particles and the light colored shells are alumina films. As seen in Fig. 3, all the particles in the clusters are coated with alumina films. The coated alumina films on the particles are conformal and uniform around particles independent of their shapes and sizes. The average film thickness measured on about ten particles are 10.1, 15.4, 13.6, and 13.2 nm for powders samples 54, 85, 340, and 970 nm, respectively (e.g., see Table II). High magnification TEM images of the sample did not show crystallinity in alumina film implying amorphous films (see the  Appendix), which is in agreement with the quality of the ALD alumina films deposited under similar conditions.^14,22,32 Also, in the high magnification micrographs, there is an obvious distinction between W and alumina at the interface suggesting abrupt interfaces consistent with prior observations^33,34 and XPS data explained below.

Calculation of stiochiometry and evaluation of chemical cleanliness of the grown alumina shell and interface chemistry were done from analysis of XPS spectra. Figure 4 shows high resolution XPS spectra of (a) tungsten (W 4f), (b) carbon (C 1s), (c) aluminum (Al 2p), and (d) oxygen (O 1s) for the 54, 85, 340 and 970 nm tungsten powders after ALD. The survey spectra (see the  Appendix) confirm that the composition of coated powder is mainly tungsten, aluminum, and oxygen. A weak carbon peak in survey spectra is visible as small broad peaks in the high resolution XPS [Fig. 4(b)].

To determine stoichiometry, relative concentrations (RCs) of the overlayer and substrate elements are calculated from the area intensities of their high resolution spectrum and their XPS sensitivity factors³⁵ (Table III). The RC of an element is calculated assuming the sum of the RCs of all the detected elements is 100%. For all the samples, the RCs of oxygen and aluminum are ∼57 and ∼37, respectively, yielding a shell stoichiometry Al₂O_{3.0 ± 0.1}. The correctness of this stoichiometry is further verified from the binding energy (BE) spacing of the O 1s and Al 2p peaks.³⁶ In this experiment, the BE spacing of these peaks is 531.7 ± 0.1− 75.0 ± 0.1 = 456.7 ± 0.2 eV, which is in agreement with the spacing 531.2 − 74.5 = 456.7 eV for the electronic grade sapphire of Al₂O₃ stoichiometry (see the  Appendix). The XPS for the sapphire were done using the same conditions as that for core–shell samples. Furthermore, the 456.7 ± 0.2 eV is in good agreement with the BE spacing for O 1s and Al 2p peaks for ALD prepared alumina reported in the literature.^{33,34,37–41} Therefore, we conclude that the stoichiometery of the prepared film is Al₂O₃, within a small error.

The RCs of tungsten and carbon are ∼1% and ∼4%, respectively. It is known that in XPS, a very small number of photoelectrons from an element at a depth more than three times its effective attenuation length contributes in a spectrum.⁴² Therefore, the small RC of ∼1% for W implies that film thicknesses for the samples are well above three times the effective attenuation length of W 4f electrons. A 1% RC of W also suggests that the overlayer films are uniform, have the same thickness on all size particles in a powder, and are pinhole free in agreement with the TEM results explained above. The RC of ∼4% for carbon is slightly higher than a typical ∼1% for ALD grown alumina on flat surfaces from TMA/H₂O.^{11,32,43–46} We attribute the high RC for carbon here to the adventitious carbon trapped within the porous pellet which are not removed by Ar⁺ ion sputter. This attribution is supported by the measured BEs for the center of the broad C 1s peaks at 285.0 ± 0.3 eV, which are in good agreement with the value for the adventitious carbon.^37,38 Furthermore, intensities of C 1s spectra are fairly constant for all the samples despite that the thicknesses of the films on them are not the same [Fig. 4(b)]. If the carbon is due to inclusion in the film due to incomplete ALD, then just as observed for aluminum and oxygen, carbon would also be a component of the films and should be everywhere in the film. Therefore, for the sample 85 nm, which has the thickest film (∼15 nm), intensity of the C 1s peaks should be appreciably increased, as the intensities of O 1s and Al 2p [Figs. 4(c) and 4(d)]. In the report of ALD growth of Al₂O₃ on ZrO₂ nanoparticle, a very large C 1s peak corresponding to an adventitious carbon was observed in the XPS survey spectrum¹⁸ and it is probably because the sample was not sputtered prior to the XPS. Therefore, we believe that grown alumina overlayer films are low in carbon impurity inclusion, as reported by others.

As said above, abruptness of alumina-tungsten surfaces at the core–shell interface can be gleaned from the analysis of the XPS peaks. The measured BEs of 31.5 and 33.7 eV for W 4f agree quite well with the BE of that of W-metal.^47–49 This unaffected W 4f BEs, despite the alumina films on the top of W-surfaces, is indicative of an abrupt interface and is consistent with the sharp distinct surfaces of tungsten and alumina observed at the interface in the high magnification TEM image (see the  Appendix). This observation is also consistent with reports on the growth of alumina via ALD, including Al₂O₃ on Si(001)³³ and Al₂O₃ on WO₃ nanoparticle.³⁴ 

Figures 5(a) and 5(b) are plots of GPC (Ref. 50) as a function of the samples and the size of the particles in the powders. Figure 5(a) shows a decrease in GPC with an increase in average particle size distribution. The GPC for both powders 54 and 85 nm is 2.0 ± 0.1 nm, for 340 nm the GPC is 1.7 ± .01, and for 970 nm, it is 1.5 ± 0.1 nm (see also Table II). Importantly, Fig. 5(b) plots the GPC for different particles sizes in a sample and demonstrates that the GPC of the alumina film is same for a range of sizes of particles in a single batch of a powder (see also Table II). The identical GPC for different particles sizes in a single batch is also apparent in Figs. 3(a) and 3(b), where the measured alumina thickness is the same for different particle sizes.

TMA and H₂O are model reactants for growth of alumina. It is well known that when using this chemistry, GPC on a flat surface is ∼1 Å/cycle at growth temperature ∼150 °C.^10–12 Though studies of ALD growth of alumina on powder is not as extensive as that on flat surfaces, several studies have shown varying GPCs. The GPC was determined to be 2 Å/cycle on 21 nm TiO₂,¹³ ∼1.8 Å/cycle on 0.1–0.5 μm BN,^14,15 2 Å/cycle on 40 nm SiO₂ and 21 nm TiO₂,¹⁶ 1.2 Å/cycle on 40 nm SiO₂,¹⁷ ∼2 Å/cycle for 62 nm ZrO₂,^18,19 and ∼1.6 Å/cycle on 50–80 nm Fe.²⁰ 

Similar to previous studies, we also observed GPC ≥ 1.5 Å/cycle [Figs. 5(a) and Table II], though there appears to be a distinct trend of GPC on particle size. In some prior studies, higher GPC have been attributed to the easy reach of gas precursors on active sites due to high aspect ratios (i.e., high curvature) and to the higher surface areas adsorbing additional amounts of the oxidizing precursor that leads to some Al₂O₃ chemical vapor deposition (CVD).^13,19 Such an explanation would expect smaller particles to always have a higher GPC than large particles, which contradicts the current experimental observation. For example, Fig. 5(b) shows that the largest particle in sample 85 nm and smallest particle in sample 340 nm are roughly the same size, but the GPCs are different. This observation is more dramatic when comparing samples 340 and 970 nm, where the largest particle in sample 340 nm is bigger than the smallest particle in sample 970 nm but the GPC of the smaller particle is lower.

Contradictions to the claim that high GPCs on smaller particles is due to high curvatures providing easy access to precursors is also apparent when considering a carefully designed study by Goldstein et al., where ALD growth of alumina films from TMA/O₃ on ∼50 nm ZrO₂ powder pressed into a tungsten grid reported GPC of alumina 1.1 Å/cycle for a wide window of temperatures.⁵¹ If the high GPC were simply because of easy access for precursors to active sites in high aspect ratio particles, then the observed GPC by Goldstein et al. should also be >1.5 Å/cycle as metal precursor is the same, O₃ is structurally same as H₂O, and substrate particle size is ∼50 nm. Therefore, it is not likely that the cause for higher GPC in most powders studied to date is the high curvature of nanoparticles.

Notably, in early ALD studies of the growth of ZrO₂ film on soda lime glass using ZrCl₄/H₂O chemistry,²² Ritala and Leskeläin reported that the purge duration following the water pulse has a significant effect on the GPC of films grown on flat substrates. In that work, a ∼13% increase in GPC was observed when the purge time after the water cycle was reduced by 50%. The authors attributed this dramatic change in GPC to CVD-like growth. Furthermore, previous studies¹⁰ reported slight increases of GPC when pumping between precursor dosing was inadequate and suggested that H₂O pressures <3.8 × 10⁻⁶Torr were needed to avoid Al₂O₃ CVD on Si(100) to maintain GPC 1.1 Å/cycle.¹⁰ However, Ferguson et al. in their studies of growth of alumina on submicron BN powder reported GPC of ∼1.8 Å/cycle despite a maintained pressure <2 × 10⁻⁶Torr (lower than 3.8 × 10⁻⁶Torr suggested by Ott et al.¹⁰) between TMA and H₂O exposure.¹⁴ In this case, the authors reported difficulty in desorbing water from within the powder substrate. The above examples indicate that surface water is difficult to remove and that there can be a reasonable amount of residual H₂O in powder substrates that does not considerably contribute to chamber pressure but increases GPC. Indeed, water is viewed as an enabler for CVD because of its ability to form a hydrogen bonded network on the top of the surface -OH groups, which can persist at temperatures >400 °C.^22,52,53 Therefore, the distinctly higher GPC on particles compared to flat surface in the present study can be attributed to residual H₂O in the powder, which enables small amounts of conformal CVD during TMA exposure via a 2Al(CH₃)₃ + 3H₂O → Al₂O₃ + 6CH₄ reaction.

If the increased ALD growth per cycle in powder substrates is due to residual H₂O, as suggested above, the GPC should vary with the amount of the residual water per unit powder mass, where the powder mass consists of particles of a given size distribution. It is pointed out that the amount of residual water per unit powder mass is related to total surface area of the powder, which in turn is a function of the sizes of the particles in the powder. As seen in Fig. 6, the particle number/gram and the surface area (cm²)/gram decreases as the size of the particles increase.⁵⁴ Therefore, the quantity of residual water molecules per unit mass of the powder should decrease with larger size particles resulting in a smaller GPC. In good agreement with this argument, the results in Fig. 5(a) demonstrate a low GPC of the alumina films on the particles of the powder with the larger average particle size. The GPC for all particles of all sizes in a single powder is the same within a small error [see Fig. 5(b)] because all the particles, regardless of the size of each particle, see the same amount of water molecules, which is a function of the average size of the majority particles in the particle size distribution. Therefore, as shown in Fig. 5(b), the smallest particle in sample 970 nm has a lower GPC than the biggest particle in sample 340 nm because the particle size distribution of sample 970 nm has a larger average particle size.

We used XPS, TEM, and SEM to investigate ALD of alumina films on a set of tungsten powders where the particle size distribution of each powder is different. The growth per cycle showed an inverse relationship with the average size of the particles in a powder. The highest growth per cycle of the films is 2 Å/cycle for powders with particle sizes in the nanometer range, such as 54 nm powder, which is comprised of particles ranging in size from ∼27 to ∼81 nm, and the lowest growth per cycle is 1.5 Å/cycle on powders with a distribution covering 460–1480 nm. However, all particles within a powder are shown to receive the same ALD shell thickness and consequently have the same growth per cycle. The variation of growth per cycle as a function of the average particle size and the same growth per cycle for all particles in a powder is determined to be due to the presence of residual water molecules in the powder left during the ALD process, which are not removed during purging.

The authors gratefully acknowledge the Naval Research Laboratory basic research program, Office of Naval Research and DARPA for support of this work. The authors also gratefully acknowledge Dev Palmer, Baruch Levush, and Fritz Kub for their support of this work and Rhonda Stroud for help in TEM. In addition, K.M. would like to thank the American Society for Engineering Education for his postdoctoral support.

See Fig. 7 for a pressure versus time plot; see Fig. 8 for high magnification TEM images of the sample that did not show crystallinity in alumina film implying amorphous films; see Fig. 9 for the survey spectra; see Fig. 10 for the electronic grade sapphire of Al₂O₃ stoichiometry.