Aluminum Oxide at the Monolayer Limit via Oxidant-free Plasma-Assisted Atomic Layer Deposition on GaN

Atomic layer deposition (ALD) is an essential tool in semiconductor device fabrication that allows the growth of ultrathin and conformal films to precisely form heterostructures and tune interface properties. The self-limiting nature of the chemical reactions during ALD provides excellent control over the layer thickness. However, in contrast to idealized growth models, it is experimentally challenging to create continuous monolayers by ALD because surface inhomogeneities and precursor steric interactions result in island growth during film nucleation. Thus, the ability to create pin-hole free monolayers by ALD would offer new opportunities for controlling interfacial charge and mass transport in semiconductor devices, as well as for tailoring surface chemistry. Here, we report full encapsulation of c-plane gallium nitride (GaN) with an ultimately thin (~3 {\AA}) aluminum oxide (AlOx) monolayer, which is enabled by the partial conversion of the GaN surface oxide into AlOx using a combination of trimethylaluminum deposition and hydrogen plasma exposure. Introduction of monolayer AlOx significantly modifies the physical and chemical properties of the surface, decreasing the work function and introducing new chemical reactivity to the GaN surface. This tunable interfacial chemistry is highlighted by the reactivity of the modified surface with phosphonic acids under standard conditions, which results in self-assembled monolayers with densities approaching the theoretical limit. More broadly, the presented monolayer AlOx deposition scheme can be extended to other dielectrics and III-V-based semiconductors, with significant relevance for applications in optoelectronics, chemical sensing, and (photo)electrocatalysis.


Introduction
Conformal and sub-nanometer thin dielectric layers can be grown by atomic layer deposition (ALD) at large scale and are essential for semiconductor device applications, including for gate dielectrics in field-effect transistors, [1] carrier-selective contacts in solar cells, [2] and corrosion protection layers in (photo)electrochemical cells, [3] as well as for sensing and catalysis. [4] While the properties of ALD films can be precisely controlled by substrate surface preparations, precursor chemistries, and external process parameters, [5] complex physical and chemical interactions lead to film and interface non-idealities. As a prominent example of this, ALD is often considered to proceed via layer-by-layer growth since the available surface binding sites react with a gas-phase reactant until saturation is reached, ideally resulting in the formation of one monolayer with every cycle. However, a combination of precursor steric effects, adsorption energetics and substrate surface inhomogeneities result in incomplete surface coverage during film nucleation. As a consequence, islands form during the nucleation phase and growth of additional layers on already formed islands is unavoidable during subsequent ALD cycles.
Thus, in contrast to the idealized concept of layer-by-layer growth, it is extremely challenging to create pin-hole free films at the monolayer limit by ALD. Overcoming this gap by growing continuous and conformal monolayers offers significant opportunities for creating atomically abrupt heterostructures, as well as for tailoring the electronic properties and chemistries of surfaces to achieve controlled functionalization or passivation.
In this work, we report the formation of an ultimately thin (~3 Å), yet continuous, aluminum oxide (AlOx) monolayer on gallium nitride (GaN) using an oxidant-free ALD process. To the best of our knowledge, this is the first experimental report of the formation of a closed single monolayer of AlOx on a non-metal surface. Although the strategy can be extended to other materials, GaN was selected due to its established technological importance. In particular, GaN is a III-V compound semiconductor that is industrially relevant for high frequency and highpower electronics because of its large breakdown field, thermal conductivity, thermal stability, and mobility. [6] In addition, it has a direct band gap (3.4 eV) and is broadly used in modern optoelectronics applications. However, a major challenge of III-V semiconductor technology is to precisely engineer interface properties, which is often hampered by high concentrations of surface states. This issue is especially pronounced for polar GaN surfaces due to polarizationcompensating surface charges. In this respect, dielectric passivation by ALD is promising since it has been demonstrated to significantly reduce interface state densities within GaN devices. [7] A major advantage of ALD is its ability to synthesize defined films at low temperature, which provides versatile process compatibility and synthetic access to amorphous dielectrics that have practical advantages as passivation and interfacial layers. [8] The most widely utilized and industrially relevant ALD process involves trimethylaluminum (TMA) and an oxidant (typically water), sequentially introduced into a reactor chamber, to form amorphous AlOx.
TMA is an electron-deficient [9] and thus highly reactive metal-organic compound that has been applied for ALD of AlOx at low temperatures, even below 25 °C. [10] In situ studies of the initial growth regime during ALD of AlOx have revealed that TMA can directly react with the native oxides of III-V semiconductors, resulting in the formation of aluminum oxide via interconversion and ligand exchange reactions. [11] This high reactivity of TMA with the substrate has been previously applied to passivate the surfaces of III-V semiconductors in situ before deposition of an ALD dielectric. [11a, 11b] In later work, the sequential exposure of III-V surfaces to a hydrogen (H2) plasma and TMA in an ALD process proved to be even more effective for passivation of the surface oxide layers. [12] Despite the intriguing chemical characteristics of oxidant-free ALD oxide interconversion reactions, neither the potential for creating ultrathin conformal oxides down to the monolayer limit nor the chemical functional characteristics of the resulting layers have been explored so far.
Understanding film nucleation is key for realizing ultimately thin coatings by ALD. Though the nucleation behavior is well established for thermal ALD of AlOx using TMA and water, [13] little is known about film formation in oxidant-free processes involving TMA and H2 plasma.
Here, we evaluate the AlOx growth evolution on free-standing c-plane GaN during cyclic exposure to TMA and atomic hydrogen by in situ monitoring of the ALD film thickness using spectroscopic ellipsometry (SE), as well as by complementary ex situ analysis using atomic force microscopy (AFM) and X-ray photoelectron spectroscopy (XPS). We find a window of self-limiting oxide growth resulting from the consumption of terminal oxygen moieties of the GaN native oxide layer during AlOx formation. Once monolayer coverage is achieved, the surface is deactivated for subsequent chemisorption of TMA, resulting in self-limited growth with no island formation. The thickness of the closed layer, independently measured by SE, AFM and XPS to be ~3 Å, agrees well with the theoretically predicted thickness for a single monolayer. [14] Creation of this single AlOx monolayer leads to a significant reduction of the work function of GaN by 0.38 eV. Furthermore, relative to the bare GaN surface, the monolayer AlOx provides additional chemical functionality, which is highlighted by its high reactivity with phosphonic acids to form self-assembled organic monolayers with a coverage that approaches the theoretical limit at room temperature. Thus, the GaN/monolayer AlOx system provides a novel platform for creating self-assembled monolayers with strong electronic coupling to the underlying semiconductor due to the absence of an extended interlayer. Not only does this work provide a new approach to creating dielectric films at the monolayer limit, but it also creates opportunities for engineering electronic and chemical properties of functional interfaces of key relevance in optoelectronics, (photo)catalysis, and energy storage applications.

Results and Discussion
As a starting point for elucidating the growth of AlOx on GaN substrates via the oxidant-free plasma-assisted ALD process, in situ spectroscopic ellipsometry was used to track film thickness dynamics during the different steps of each cycle, as well as over multiple successive cycles. As shown in Figure 1a, which provides a plot of film thickness as a function of time, three different growth regimes can be distinguished during 100 cycles of TMA and hydrogen plasma exposure. These regimes are characterized by (i) growth with an exponentially decaying rate, (ii) a saturated film thickness with a growth rate of zero, and (iii) slow growth with an approximately constant rate. Each of these is discussed in detail below.
In regime (i), the growth-per-cycle (GPC), which is defined as the measured change in layer thickness per cycle, follows an exponential decay and reaches saturation (i.e., GPC = 0 Å) after 8 cycles with a thickness of 2.8 ± 0.1 Å (Figure 1b). The origin of this saturation behavior can be understood by considering the detailed growth mechanism. As shown in Figure 1b, the first few cycles exhibit the largest GPC, which suggests a substrate-enhanced growth governed by the accessible binding sites on the GaN surface. [15] This stands in stark contrast to the growth characteristics of traditional thermal ALD of AlOx on GaN using TMA and H2O, in which growth is inhibited during the initial cycles (Figure 1b,c). XPS analysis revealed that the surface of GaN is coated with a ~1 nm thin native oxide layer that is terminated with OH groups (Supporting Information S5). The thickness of this native oxide layer was confirmed by X-ray reflectivity (XRR) ( Figure S2), and agrees with the literature. [16] Consistent with XPS analysis, static water contact angle measurements of these surfaces exhibit hydrophilic character, with contact angles of ~39° and ~34° obtained before and after the initial H2 plasma treatment (Table S2). This hydrophilicity indicates that the native gallium oxide surface is partially hydroxylated, thereby providing Ga-OH binding sites that can react with TMA through a ligand-exchange reaction, in a manner similar to TMA reacting with SiO2. [13a] As a consequence, exposure to TMA results in the formation of AlOx according to the following ligand exchange reaction: [17] Ga − OH + Al(CH 3 ) 3 → Ga − O − Al(CH 3 ) 2 + CH 4 (1).
Although Reaction 1 is self-limiting, the available binding sites cannot be completely consumed after one TMA half-cycle due to steric hindrances. In particular, binding sites are shielded by the methyl groups of neighboring chemisorbed TMA molecules and may also be blocked by physisorbed TMA molecules and clusters. While weakly bound TMA molecules are swept away from the surface during argon (Ar) purging, the ligands of chemisorbed TMA molecules will remain until chemically reactive species are introduced. As indicated by the red arrows of Figure 1b, subsequent H2 plasma exposure leads to a decrease in the adsorbate thickness, indicating that atomic hydrogen, which is used here instead of an oxidant, reduces the adsorbed TMA via the formation of methane. Consequently, Ga-OH binding sites that were blocked by methyl ligands after the first TMA half-cycle become accessible for subsequent reactions with TMA. This simple mechanistic model predicts an exponential decay of available binding sites [18] and explains the observed exponential decrease of the individual cycle growth rate over successive cycles.
A critical aspect of this oxidant-free AlOx ALD mechanism is that the surface is chemically deactivated as the Ga-OH sites are consumed, as indicated by the observed decrease of the GPC, which reaches zero after eight TMA/hydrogen plasma cycles. Since exposure to hydrogen plasma reduces the methyl ligands of chemisorbed TMA to methane, an O-Al* terminal surface, which may include partial coverage by O-Al-CH3 sites, is established. This surface composition inhibits further chemisorption of TMA and, thus, the growth of subsequent layers, thereby giving rise to the second regime, (ii), during which the thickness remains nearly constant ( Figure 1a,b). In addition, the in situ SE measurements reveal that the AlOx film is not chemically reduced upon exposure to several cycles of low power H plasma. Here, we note that the AlOx-coated GaN exhibits a water contact angle of ~8°, which indicates a strongly hydrophilic surface (Table S2) that is consistent with an Al-OH surface termination. However, these measurements require exposure to ambient atmosphere, which is expected to rapidly oxidize the -Al* termination, thereby introducing hydrophilic oxygen groups on the surface.
The proposed growth mechanism resembles a Frank-van der Merwe type of growth, in which TMA reacts preferentially with substrate surface sites, resulting in complete surface coverage.
However, in contrast to conventional growth modes, deposition of subsequent layers is suppressed due to chemical deactivation of the surface. Importantly, this chemical inactivity reduces the likelihood of island-formation [13b] and provides the intriguing opportunity for ideally self-saturating growth of AlOx down to the single monolayer limit. To explore this possibility, we consider the saturated AlO x film thickness and compare it to the predicted monolayer thickness, which can be estimated according to Equation 2, where M, ρ, and NA are the molar mass, density, and Avogadro constant, respectively. For an ultrathin layer of this type, it is difficult to determine the density. However, reasonable values can be obtained from prior reports of ALD alumina, which yield values in the range of 3.2 -3.5 g/cm 3 at similar substrate temperatures. [19] At the level of a monolayer, half the volume of the α-Al2O3 unit cell (i.e., a nominal composition of AlO1.5) is assumed, which yields an approximate thickness, hml, in the range of 2.89 to 2.98 Å. Thus, the film thickness in the saturation regime (ii) of 2.8 ± 0.1 Å, determined experimentally via in situ spectroscopic ellipsometry, is in excellent agreement with the predicted thickness of a single monolayer of alumina. This agreement is fully consistent with the proposed film formation mechanism and offers fascinating possibilities for the establishment of two-dimensional dielectric interlayers, as discussed below.
Although no growth is observed for several subsequent TMA/H2 plasma cycles, a third growth regime, indicated as the region (iii) in Figure 1a, emerges after 20 cycles. This regime is characterized by a constant, but slow, growth rate of 0.03 Å per cycle. According to the model discussed above, no such growth would be expected. However, in situ SE measurements suggest that repetitive exposure to H2 plasma and TMA pulses leads to accumulation of material on the surface, indicating film growth at a comparably slow rate. Although the origin of this growth regime is not conclusively known, it is apparent that TMA precursor adsorption on the surface must occur, but with very low coverages, suggesting a potential role of defects as binding sites. Chemical interactions of TMA and such non-idealities may yield dissociative chemisorption, [20] and the cumulative effect of repetitive H2 plasma exposure would then result in a complete reduction to elemental Al since no oxygen from the underlying oxide is accessible. To confirm the feasibility of metallic Al growth, we prepared a sample using 200 cycles of TMA/H2 plasma at a higher plasma power (300 W). The larger growth rate (GPC ≈ 0.15 Å) resulted in Al films that are sufficiently thick to not completely oxidize in air. XPS analysis of the resulting surfaces revealed an Al 2p component at a lower binding energy (~72.9 eV), indicative of metallic Al, alongside the higher binding energy feature associated with Al-O binding (~75.6 eV) ( Figure S10). This result, which is consistent with the prior report of metallic Al ALD using TMA/H2 plasma cycles, [21] lends additional credence to the proposed description of linear growth in regime (iii).  XPS was performed to determine the chemical changes to the surface arising from exposure of GaN to 20 cycles of TMA and H2 plasma. Figure 3a shows the Ga 3d core level region of the bare GaN surface, within which Ga-N, Ga-O, and Ga-Ga, as well as N 2s, components can be discriminated. Spectral fitting was achieved by constraining the spin-orbit splitting to be 0.43 eV [23] with an area ratio of 3:2 and equal FWHM for both components of the 3d doublets. For both the bare and TMA/H2 plasma-exposed sample, the magnitude of the shift of the Ga-O components relative to the main Ga-N photoemission is consistent with prior studies [24] and indicates the presence of a native oxide comprising primarily Ga2O3. Ga 3d photoelectrons emitted from Ga-O-Al environment are expected to shift towards lower BEs, i.e. closer to the Ga-N main peak because of the relatively lower electronegativity of aluminum. Indeed, XPS analysis indicates that the Ga-O/Ga-N ratio decreased after the TMA/H2 plasma process ( Figure   3a, inset). This finding suggests an AlOx film growth mechanism that includes the formation of Ga-O-Al bonds, [25] which is in agreement with the proposed Reaction 1 based on the real-time SE data, as well as previous reports of TMA reactivity on GaN [24,26] and AlGaN [27] during traditional thermal ALD using H2O as an oxidant.
The Al 2p core level spectrum is well described by reports. [28] The O 1s spectra of bare and AlOx-coated GaN substrates reveal two chemical components attributed to hydroxyl groups and metal-oxide bonds ( Figure S8). It is expected that the hydroxylated monolayer AlOx coating should enhance surface basicity (acid-base reactivity), thereby providing the opportunity to introduce new chemical functionality to the surface. To test this hypothesis, we investigated the reactivity of the ALDmodified GaN surface with phosphonic acids (PAs) under standard conditions. The PA head group binding sites, namely the acidic hydroxyls and the phosphoryl group, can react with the hydroxyl groups of the substrate via an acid-base heterocondensation. [29] While this reaction requires thermal activation for acidic and weakly basic (Lewis acidity) hydroxylated surfaces, such as the native oxides of Ga [30] and Si [31] , it can occur under standard conditions for basic hydroxylated substrates, such as aluminum oxide. [32] To test the reactivity of the ALD-modified surface, we performed an organophosphonic acid functionalization reaction with 11hydroxyundecylphosphonic acid (PA-C11-OH) via a modified immersion technique at room temperature and atmospheric pressure (Supporting Information S7).
The AFM topography (Figure 4a,b) demonstrates a uniform and smooth organic SAM with an rms roughness of 274 ± 9 pm, thereby providing further evidence of a continuous and pin-hole free AlOx coating. An average organic SAM thickness of ~5.8 ± 0.4 Å was determined from the step heights after removal of the organic layer in contact mode AFM (Figure 4a,b).
Considering the length of PA-C11-OH and the measured SAM thickness, a molecular tilt angle, θ, of 70.5 ± 2.8° with respect to the surface normal was determined. This value is relatively large compared to prior literature reports of PA-C11-OH SAMs. For example, (100) Si yields tilt angles of ~45°. [33] However, the tilt angle and thus the SAM thickness are related to the PA binding configuration, which is known to be affected by the deposition method, PA structure, and surface coverage. [34] In particular, the PA can interact with the underlying AlOx layer via P-O-Al bonding in mono-, bi-, or tridentate modes, depending on the number of PA functional groups that are involved, as indicated in Figure 4d. [29] Previously, bidentate bonding was reported to be dominant for GaN following functionalization at elevated temperature, [35] while a range of different binding motifs has been reported for aluminum oxide. [34] To probe the binding motif and quantify the molecular coverage, XPS measurements were performed on functionalized surfaces. Figure 4c shows a typical XPS spectrum of the P 2p core level region, which also includes the Ga 3p plasmon, for a GaN/monolayer AlOx/PA-C11-OH sample. Despite a theoretical spin-orbit splitting of 0.9 eV, [36] the main P 2p peak at 135 eV is broad (FWHM = 1.79 eV) and approximately symmetric, which is consistent with a mixture of binding modes. [36][37] The optimum fit of the P 2p spectrum (Figure 4c, red line) suggests a combination of two binding configurations, of which the mono-dentate binding motif is dominant (88 ± 3 %).
Prior reports attributed the evolution from higher to lower denticities to an increasing SAM surface coverage. [37][38] Our results are consistent with this finding since we observe a relatively large phosphorus atom density of (4.5 ± 0.3) × 10 14 cm -2 , corresponding to a PA surface coverage, n, of 4.5 ± 0.3 nm -2 , as calculated from the XPS data ( Figure 4c) according to the formalism introduced by Kim et al. [30] (Supporting Information S7). This high SAM surface coverage (4.5 ± 0.3 nm -2 ) on GaN/monolayer AlOx approaches the theoretical limits predicted for aluminum oxide (4.65 nm -2 ) [39] by DFT and estimated from the molar volume of phosphorus acid (4.25 nm -2 ), [40] and it is two times higher than reported for alkyl-PAs grafted on c-plane-Ga-polar GaN (2.3 nm -2 ). [30] These high coverages are consistent with the formation of a continuous AlOx coating, onto which the PAs molecules assume a predominantly mono-dentate binding motif in a high-density SAM. with XPS measurements, which are sensitive to changes of band bending but not work function: here, XPS revealed that the VB edge positions ( Figure S12) and core level BEs (Table S4) were not affected by the presence of the AlOx monolayer to within the measurement error. Thus, changes to the interfacial energetics can be explained in terms of a change of the surface dipole associated with monolayer AlOx deposition (Figure 5b). The polarity of this dipole, for which the positive endpoints outwards, is consistent with the smaller electronegativity of Al (1.7 [41] ) relative to Ga (2.4 [41] ). In addition, polarization-induced negative bound charges of the Gapolar GaN may facilitate interface dipole formation as these surface charges are compensated by the native GaOx layer, [42] rendering the GaOx at the GaN/GaOx interface more positive Assuming d =3 Å as the average distance between Ga and Al atoms at the Ga-O-Al interface, εAlOx ~ 4 for ultrathin ALD AlOx [18] and εGaOx ~ 9 for amorphous GaOx [43] yields σ = 3.5×10 13 cm -2 , which is in the range of polarization-induced charge densities of polar GaN. We note that in addition to the interface dipole, fixed charges of both polarities have been measured for thicker AlOx films by resonant XPS, [44] predicted by density functional theory (DFT) calculations, [45] and quantified by capacitance-voltage measurements, respectively. [46] [47] However, for thicker oxide layers, fixed charges would create an extended electric field and further influence the work function. , [48] where εPA is the relative dielectric constant, ε0 is the vacuum permittivity, and n is the areal density of adsorbed molecules (dipoles). Inserting the surface coverage of 4.5 × 10 14 cm -2 derived from XPS analysis, the measured change of the CPD, ∆ PA , of 0.18 V, and the reported alkyl SAM permittivity, εPA, of 3, [49] yields a surface dipole of +0.32 D. By correcting for the tilt angle, θ ~ 71° deduced from the measured SAM thickness (Figure   5b), the surface dipole of a vertically aligned PA-C11-OH monolayer would equal 0.95 D.
While this value is lower than the reported free molecule dipole moment (+1.46 D) predicted by density functional theory, [50] it has been established that depolarization within SAMs leads to a deviation between the free molecular dipole and the bound molecular dipole, of which the latter was found to correlate with observed changes in the work function. [51] Here, we find that surface energetics of the GaN are significantly modified by the introduction of a single monolayer of AlOx and can be additionally tuned by the introduction of a molecular monolayer, whose assembly is enabled by the chemical functionality of the ALD oxide.

Conclusion
The presented oxidant-free ALD process allows for the fabrication of conformal and pin-hole

Atomic layer deposition
GaN substrates were coated with AlOx in a hot-wall plasma-enhanced atomic layer deposition (PE-ALD) reactor (Fiji G2, Veeco CNT) in continuous flow mode. An in situ H2 plasma pretreatment (2 cycles, 3 sec, 100 W, 0.02 Torr) at 280 °C was performed to remove the approximately 5 Å thin adsorbate layer of air-exposed GaN ( Figure S4) prior to deposition of alumina. Conveniently, hydrogen treatment can also passivate dangling bonds on GaN. [52] AlOx films were grown using TMA (electronic grade, 99.999 %, STREM Chemicals) as the precursor and Ar (99.9999 %, Linde) as the carrier gas during the first half-cycle at ~0.09 Torr while the turbo pump was isolated. The pressure-time plot ( Figure S5) illustrates the ALD AlOx growth process, which has previously been implemented with similar parameters for the in situ preparation and passivation of III-V semiconductor surfaces before deposition of dielectrics. [12a, 53] The precursor and plasma doses were set such that the reactions with the GaN surface were

Preparation of self-assembled monolayers of phosphonic acids (SAMPs)
SAMPs of 11-hydroxyundecylphosphonic acid (≥ 99 % (GC); referred to as PA-C11-OH) (SiKÉMIA, Montpellier, France) were prepared via a modified immersion technique, which is described in detail in section S7 of the Supporting Information.

Spectroscopic ellipsometry
Changes in the GaN surface oxide layer thickness were monitored in real-time using an in situ  Figure S3).

X-ray photoelectron spectroscopy (XPS)
XPS spectra were acquired at pass energy of 10 eV with a Kratos Axis Ultra setup equipped with a monochromatic Al Kα X-ray source. Charge neutralization was not required as no binding energy shifts indicative of (differential) charging were observed for the free-standing not intentionally doped n-type GaN samples. The instrumental broadening (0.30 eV) was determined by fitting of the measured Ag 3d core level spectrum of a silver calibration sample with a Voigt function ( Figure S6). To facilitate the quantitative analysis of compositions and film thicknesses, the surface adsorbate layer (water and adventitious carbon) of the ambientexposed GaN samples was selectively removed by mild sputtering with Ar1000 + -ion clusters (10 keV, 60 s, 37 ° incidence angle) generated with a gas cluster ion source (GCIS). Comparison of GaN core level spectra before and after sputter cleaning demonstrates the attenuation effect of the carbon overlayer on GaN core level emission intensities ( Figure S7). The mild in situ Arion cluster sputter procedure was optimized to selectively remove the surface adsorbates without altering the chemical bonds in the GaN, as confirmed by the identical Ga 2p and Ga 3d core level peak shapes before and after sputtering.

Supporting Information
Additional information on the GaN substrate properties and surface preparation methods. Xray diffraction analysis data. Static water contact angle measurements. Spectroscopic ellipsometry model. Supporting XPS and AFM data. Description of SAMs preparation and analysis.
Author Contributions ⊥ These authors contributed equally. All authors have given approval to the final version of the manuscript.

S1. X-ray diffraction analysis of free-standing c-plane, Ga-polar GaN
Free-standing and unintentionally silicon (Si) doped (resistivity < 0.5 Ω×cm; donor densty ≤ 1×10 17 cm -  where λ is the wavelength of the source. Since the q-range extends up to ≈ 0.65 Å -1 , the scattering length density (SLD) distribution normal to the surface can be decomposed in layers with a resolution of ≈ 4 Å according to the Fourier sampling theory. [1] The experimental data was simulated with the Python-based reflectometry analysis package refnx (ver. 0.1.8). [2] SLD profiles were converted to theoretical reflectivity curves using Abeles formalism and a differential evolution algorithm. Furthermore, Bayesian Markovchain Monte Carlo sampling was used to estimate uncertainties of all fitting parameters (thickness, SLD, and roughness).

Figure S2a
shows the normalized X-ray reflectivity curve (gray squares) as a function of the momentum transfer for a representative bare GaN sample. Total reflection beneath a critical momentum transfer qc of 0.046 Å -1 is observed, which is in excellent agreement with values measured on MBE-grown GaN. [3] The inset of Figure S2a reveals periodically positioned maxima due to interference of waves reflected from the GaN/GaOx interface, where a finite discontinuity of the electron density occurs. Best fits (solid lines) were achieved by a three-layer model, considering GaN, GaOx, and an adsorbate overlayer. The corresponding depth-resolved, three-layer SLD profile is presented in Figure S2b, and the obtained fitting results are summarized in Table S1. The average thickness of the native gallium oxide (GaOx) layer is 11.3 ± 1.5 Å, and the corresponding SLD is 48.5 ± 1.5 10 -6 Å -2 (corresponding to an electron density of 1.72 ± 0.05 e Å -3 ), which agrees with reported values. [4] The electron density (0.25 ± 0.04) of the adsorbate layer is typical for hydrocarbon chains and suggests the presence of adventitious carbon on the GaOx surface, [5] which is observed in the X-ray photoelectron spectra prior to in situ cleaning (see Section 4).

S3. In situ thickness measurement by spectroscopic ellipsometry during ALD
A generic oscillator model was used to model the ellipsometric polarization change in the wavelength range between 210 nm to 400 nm ( Figure S3). The bandgap of GaN decreased to ~3.3 eV at the process temperature of 280 °C, in agreement with the literature. [6] Interferences are less pronounced in free-stand- ing GaN compared to GaN grown on a template such as sapphire. However, backside reflections from the GaN/chuck interface are present at energies below the band gap. Therefore, the absorbing region with wavelengths near and above the GaN band gap was considered for SE data modeling. The measured thickness of native gallium oxide (11 Å) by XRR and XPS is taken into account in the SE model. Figure S3. Spectroscopic ellipsometry measurement of Psi (black squares) and Delta (blue circles), and corresponding fits (grey lines) to a general oscillator model for wurtzite GaN with a layer of native oxide.

S4. GaN surface preparation by plasma-enhanced atomic layer deposition
The c-plane, Ga-polar GaN substrates were exposed to two cycles of remote plasma-generated atomic hydrogen (100 W, 3 sec) prior to AlOx growth in a plasma-enhanced atomic layer deposition reactor. The overlayer film thickness decreased by ~4 Å after H2 plasma exposure ( Figure S4). The thickness after overlayer removal was defined as the starting point prior to ALD. Notably, the surface adsorbate thickness (at 280 °C) determined by SE, agrees with the carbon overlayer thickness of ~7 Å estimated from XPS analysis of GaN before and after Ar-ion cluster sputtering (Section S5). A smaller thickness was determined by in situ SE, likely because of partial desorption of the adsorbate layer at 280 °C. Our results suggest that the H2 plasma facilitates the removal of carbon contamination on the GaN surface, previously demonstrated with forming gas for AlGaN. [7] Figure S4. In situ SE thickness measurements during H2 plasma treatment of an air-exposed GaN surface. The adsorbate thickness decreased after H2 plasma pretreatment.
The pressure-time plot (Figure S5) illustrates one cycle of the monolayer AlOx growth process. MΩ·cm at 25 °C, Merck Millipore) was dispensed with a rate of 1 μl/s from a high-precision syringe (Hamilton, DS 500/GT, gas-tight, 500 µl) on the sample surface and after ~3 s (reaching equilibrium) the side profile of the droplet was taken for further processing. SCAs from at least three different spots were determined to calculate a standard deviation (Table S2). Compared to the bare GaN substrate, the hydrophilicity of the monolayer AlOx-coated surface is increased by a factor of ~ 5 (Table S2). The OH density is a crucial parameter, e.g. for silanization reactions, [8] but less critical for the PA surface chemistry [9] of this work. Notably, the relatively small variation (7.8 ± 0.5°) indicates that the surface is uniform after monolayer AlOx deposition and that the surface hydrophilicity is persistent in air.

S5. X-ray photoelectron spectroscopy of bare and AlO x -coated c-plane, Ga-polar GaN
The energy resolution of the XPS system was determined from the fitting of the Ag 3d5/2 spectrum of a silver calibration sample using a Voigt profile and yielded a value of 0.30 eV (Figure S6). A natural linewidth of 0.33 eV for Ag 3d5/2 was used to deconvolute the contribution from instrumental broadening. Figure S6. Energy calibration of the XPS system with a sputter-cleaned silver sample. An instrumental broadening of 0.30 eV was determined by fitting with a Voigt function using a natural linewidth of 0.33 eV for Ag 3d5/2.
The substrates were cleaned by Ar1000 + -ion cluster sputtering (10 keV, 60 s) in the analysis chamber of the XPS setup to remove surface adsorbates, which presumably comprised physisorbed water and adventitious carbon from the air-exposed GaN samples prior to XPS analysis ( Figure S7). A BE energy shift of 0.2 eV towards lower energies is observed for the entire spectrum, indicating charging by the carbon overlayer. For larger sputtering energies (≥ 20 keV), we observed removal of the native gallium oxide (not shown here). However, preferential sputtering of oxygen and nitrogen resulted in a metallic Gaterminated surface as indicated by measured changes in Ga 3d and Ga 2p core level shapes and binding energies (> 1 eV), as well as enhanced photoemission from occupied band gap states at the Fermi level in the valence band spectrum. However, no such changes were observed with more mild sputtering at 10 keV, which was used for all subsequent pre-analysis surface preparations. Figure S7. XPS of (a) C 1s, (b) Ga 2p 3/2, and (c) Ga 3d core levels before and after in situ Ar1000 + -ion cluster cleaning of free-standing GaN surfaces.
The O 1s core level spectra (Figure S8) of the bare and the monolayer AlOx-coated c-plane GaN reveal the presence of O-H groups (even after Ar1000 + -ion cluster sputtering), which is consistent with a hydrophilic surface (see Supporting Information S4) and the proposed reaction mechanism (Reaction 1). Figure S8. O 1s core level spectra of bare and monolayer AlOx-coated GaN after in situ Ar1000 + -ion cluster cleaning.

Measurement of overlayer thicknesses by XPS
The thicknesses of the carbon overlayer, the native GaOx layer, and the monolayer AlOx film were estimated from XPS analysis (Table S3).
A carbon overlayer (ol) thickness, dol, of 7.7 ± 1.0 Å (4 samples considered) was obtained by comparing core level peak intensities before and after removal of the carbon overlayer by Ar1000 + -ion cluster sputtering. The overlayer thickness, dol, can be estimated from Equation S1, where I0 is the integrated core level peak intensity for a bare substrate, and I1 is the corresponding core level signal attenuated by the overlayer, θ is the electron take-off angle with respect to the sample surface (here θ = 90 °) and λol is the inelastic mean free path (IMFP) of the photoelectrons in the overlayer at the kinetic energy of the measured core level (here 368 eV, Ga 2p3/2). An IMFP for carbon (λol = 13.98 Å) for a density of 2.2 g/cm 3 (graphite), computed with the Tanuma, Powell, Penn (TPP-2M) formula, [10] was obtained from the QUASES-IMFP software.
The thickness, dox, of the native gallium oxide layer of 10.8 ± 0.8 Å (standard deviation out of 4 different measurement spots on three samples) was obtained from XPS analysis and is in good agreement with the oxide thickness measured by XRR. The oxide thickness was estimated for the carbon-free GaN substrate (after Ar1000 + -ion cluster sputtering) with the intensity ratio of the Ga-O and Ga-N components of the Ga 3d core level spectrum according to Equation S2, [11] = sin ln ( where Iox and Is are the integrated intensities of the oxide layer (ox) and substrate (s) component of the same core level (here Ga 3d). We note that in contrast to the Ga 2p peak shape, the Ga 3d peak is asym- with the TPP-2M formula [10] using the QUASES-IMFP software.
The AlOx thickness measured by SE (see Figure 1b) is in reasonable agreement with the thickness extracted from the relationship between the emission intensities of the Ga 2p and Ga 3d core levels. In particular, photoelectrons emitted from the Ga 2p and Ga 3d core levels probe the same element at different kinetic energies, [12] resulting in different IMFPs of λL = 12.45 Å and λH = 33.64 Å for photoelectrons moving through AlOx at relatively low (L) and high (H) kinetic energies, respectively. An AlOx layer thickness of dox = 3.3 ± 0.8 Å was calculated for GaN exposed to 20 TMA/H2 plasma cycles by using  We note that thickness measurements of sub-nanometer thin conformal coatings on corrugated (not atomically flat) surfaces is challenging with real-space techniques and is achieved here with SE and XPS.
Byproducts such as methane and CHx that are produced during ALD processes using TMA can potentially react with the sample surface and contaminate the ALD film. [13] However, comparison of the C 1s core level spectra before and after sputtering shows that the carbon content is just above the background noise level. The small signature that appears in the C 1s spectrum of AlOx-coated GaN most likely stems from carbon redeposition, though incorporation of low concentrations of carbon into the AlOx cannot be completely ruled out ( Figure S9). We note that substrate-enhanced growth has also been associated with precursor dissociation. [14] However, such a mechanism can be excluded in the present case since the dissociation would yield films with high carbon content. Here, XPS measurements indicate carbon signals that are indistinguishable for bare GaN and GaN coated with AlOx by thermal and H2 plasma/TMA ALD ( Figure S9). Figure S9. XPS of the C 1s core level after Ar-ion cluster sputtering of bare c-plane GaN (black squares), GaN coated with thermal AlOx following 20 cycles of TMA and H2O (black circles), and GaN coated with AlOx after 20 cycles TMA and H2 plasma (blue triangles).
The XPS data shown in Figure S10   Free-standing bare, c-plane GaN and bulk-like AlOx were characterized by XPS to estimate the valence band offset between monolayer AlOx and GaN using Kraut's method. [15] Therefore, a 25 nm thick (bulklike) AlOx layer was deposited on free-standing, c-plane GaN by thermal ALD at 280 °C. Figure S11 shows the analysis of valence band and core level spectra of the 25 nm thick AlOx/GaN and bare GaN sample, respectively. Analysis of the spectral features attributed to surface plasmons allowed to estimate the band gap of the ALD Al2O3 film. The valence band spectra show the characteristic signatures of GaN ( Figure S12) where the peaks labeled as A and C at BEs of ~ 5 eV and ~ 9.5 eV are assigned as Ga 4p-N 2p and Ga 4s-N 2p bonding, respectively. [16] The spectral region labeled as "B" signifies the BE range of a GaN hybridization state.
The valence band edge is at the same energy for both bare and monolayer AlOx-coated GaN. Figure S12. Valence band spectra of bare (black squares) and monolayer AlOx-coated c-plane GaN (blue circles). Table S4 summarizes the energies of the valence band edges and the binding energies of core level components for the different sample systems.

S6. Atomic force microscopy of bare and monolayer AlO x -coated c-plane, Ga-polar GaN
The power spectral density plots ( Figure S13) show distinct peaks at 74 µm -1 and 55 µm -1 for the bare and ALD-treated GaN, respectively, indicating repetitive patterns that are not discerned with the frequency density analysis (Figure 2b). These patterns can be attributed to atomic steps that are aligned in a preferred crystallographic orientation and result from uniformly misoriented growth seeds, commonly observed on the surfaces of epi-ready polished GaN [17] . There are no indications of threading dislocations on the length scale characterized in this study by AFM. Figure S13. Power spectral density plots derived from the AFM images of Figure 2.
By applying a tip-sample force of 3.25 µN with diamond-like carbon (DLC) tip in contact mode, the AlOx film could be (partially) removed off the GaN substrate ( Figure S14). Since DLC is harder than sapphire [18] it is suitable for abrading AlOx, thereby providing another method to confirm the formation of a continuous AlOx coating and the thickness of the film (3 ± 1 Å). The estimated step height measured with tapping mode (TM) AFM is consistent with the AlOx layer thickness determined from SE measurements and in agreement with the conclusion that a single monolayer of AlOx is formed. In addition, the contrast in the TM phase image revealed local differences in the tip-sample adhesion, which is indicative of different tip-sample interaction strengths ( Figure S14c) at the bare and AlOx-coated GaN surfaces in the scratched and non-scratched regions, respectively. In comparison, a smaller step height and no contrast in the phase image was observed after micro scratching the bare GaN, likely due to the removal of the surface adsorbate. GaN and β-Ga2O3 have a hardness of 12 ± 2 GPa and ~ 9 GPa, respectively, and are expected to be harder than amorphous AlOx. However, due to the atomically thin AlOx and the presence of a surface adsorbate, it is challenging to reliably determine the monolayer AlOx thickness by this micro-scratching experiment. Figure S15. Photoelectron spectra of O 1s, C 1s, P 2p, Al 2p, and Ga 3d of a monolayer AlOx sample (black line) and an AlOx/PA-C11-OH sample before (blue line) and after Ar-ion cluster sputtering (red line). Note that for the P 2p spectra no background correction was performed.

Calculation of PA surface coverage on monolayer AlO x /GaN substrates.
The number of phosphorus atoms and, therefore, PA molecules per unit area can be determined using Equation S4: [22]  (2.63 nm) [23] is the IMFP of Ga 3d photoelectrons in GaN (generated by Al Kα X-rays) is the thickness of the organic overlayer, and 2 , and 3 , are the IMFPs of P 2p and Ga 3d photoelectrons (generated by Al Kα X-rays) in the organic overlayer, respectively. θ is the take-off angle of photoelectrons with respect to the sample plane (θ = 90°). Note that the RSF values presented above can only be applied for a source-to-detector geometry of 54.7°.
Neglecting surface roughness and consistent with previous studies, [24] an empirical approximation for IMFPs in self-assembled monolayers can be expressed as: , [nm] = 0.9 + 0.0022•Ekin [eV], where X is the photoelectron transition of interest and Ekin [eV] is the kinetic energy in electron volts. [21] For the photoelectrons of interests, this corresponds to 2 , = 3.9 nm ≈ •sin ( ) ) ≈ 1, with an error of less than 5 % for < 4 nm. [22] Since the density of the AlOx intermediate layer is unknown but likely differs from stochiometric bulk Al2O3, the Ga 3d photoelectrons of underlying GaN substrate were taken as a reference signal and corrected for the attenuation effect of the AlOx intermediate layer following

Equation S5
: [25] 3 = The AlOx(ML) film presumably contains three different O sites. [26] O anions can act as electron-donating Lewis base sites and incompletely coordinated cations as electron-accepting Lewis acid sites, whereas OH anions can act as either a Lewis acid or base, but also as a proton-exchanging Brønsted acid-base sites. Van den Brand et al. [27] reported an XPS-based procedure to identify the character of O on oxide surfaces, however, for the present system higher resolution and higher signal-to-noise ratios, e.g. provided by synchrotron radiation, would be necessary to substantiate this claim.