Suppressing substrate oxidation during plasma-enhanced atomic layer deposition on semiconductor surfaces Open Access

Atomic layer deposition (ALD) is a versatile method that enables generation of conformal thin films with precise control over thickness, physical properties, and functional characteristics. Consequently, ALD is broadly used across various applications, including for advanced electronic devices incorporating high-k gate dielectrics,^1–3 charge selective contacts and surface passivation layers,^1,3–6 and chemically and mechanically protected interfaces.^3,4,6,7 Thermal ALD is a gentle approach since it relies on self-limiting surface reactions at relatively low temperatures. While this feature allows for growth on chemically sensitive substrates, it limits the range of suitable metalorganic precursors. The resulting films are often characterized by non-ideal stoichiometries, foreign atom impurities from incomplete ligand removal, and comparatively low deposition rates.⁸ In contrast, plasma-enhanced ALD (PE-ALD) is a powerful variant that overcomes these limitations. For metal oxide deposition, plasma-generated oxygen radicals additionally promote combustion-like ligand removal reactions, leading to improved film density and stoichiometry compared to thermal ALD.^9,10 These properties are particularly important for producing high-k gate dielectrics with low leakage currents, chemically resistant protection layers, and coatings with reproducible optical properties.

Most PE-ALD processes employ remote inductively coupled plasma (ICP) sources operating at powers ranging from 200 to 600 W.¹¹ While the remote nature of the plasma minimizes surface damage by suppressing ion bombardment, the high flux of radicals can result in substrate surface reactions during the initial stages of nucleation and growth, which is particularly pronounced for deposition of oxides using oxygen plasma (OP).^12–15 Due to these in situ oxidation processes, poorly controlled substrate/film interfaces, reduced gate capacitances, and high interface state densities can occur.^12,13 Furthermore, for applications requiring efficient interfacial charge transport, such as protection layers on semiconductor photoelectrodes or tunneling contacts, substrate oxidation can reduce performance by introducing energetic barriers.^16–19 

In this study, we demonstrate that greatly reduced plasma powers can suppress substrate and interface oxidation during PE-ALD, without sacrificing the high optoelectronic quality of the dielectric coatings. To establish the broad applicability of this approach, we investigate PE-ALD of titania (TiO_x) on technologically relevant group IV (Si) and III–V (InP) semiconductor surfaces, comparing low power (5 W) and conventional high power (300 W) PE-ALD with thermal ALD. The resulting InP/TiO_x and Si/TiO_x structures are of specific relevance as charge selective contacts and chemical protection layers in photovoltaic and photoelectrochemical applications. Using remote OP powers as low as 5 W results in significantly reduced substrate oxidation, approaching the levels achieved by corresponding thermal ALD processes. Importantly, characterization of PE-ALD films deposited at low and high plasma power reveals that the compositional, optical, and electrical properties of the TiO_x films are practically indistinguishable from one another. In terms of application, the benefits of the low plasma power process are highlighted for InP/TiO_x photocathodes and Si/TiO_x photoanodes, both of which exhibit significantly improved onset potentials for light-driven water splitting reactions. These findings are confirmed for the most commonly used metalorganic precursors for ALD TiO_x, titanium isopropoxide (TTIP) and tetrakis(dimethylamino)titanium (TDMAT), highlighting the generalizability of this approach. Thus, low power PE-ALD not only enables improved control of semiconductor interface properties but also expands the compatibility of energy-enhanced ALD with chemically sensitive substrates.

Here, we compare three different oxidation half-cycle conditions: (i) thermal ALD with H₂O, (ii) conventional PE-ALD with 300 W O₂/Ar plasma (OP₃₀₀), and (iii) low power PE-ALD with 5 W O₂/Ar plasma (OP₅), using two different metalorganic precursors, TTIP and TDMAT. These processes were performed on Si and InP surfaces at a substrate temperature of 200 °C in a hot-wall Veeco Fiji G2 reactor equipped with a remote rf ICP source (13.56 MHz). Complete experimental details, including aspects of process optimization (Figs. S1 and S2), are provided in the supplementary material. The Si surfaces were prepared with a thin chemical oxide, and InP was used with its native oxide intact to ensure a reproducible starting condition. Such surface preparations prior to ALD have been reported to yield optimal performance for semiconductor photoelectrodes investigated for solar fuel applications.^18,20–22

To gain insight into the influence of H₂O and OP exposure on film growth and semiconductor surface oxidation during ALD, we performed in situ spectroscopic ellipsometry (SE) during each process. Through optical modeling and fitting of these data, we simultaneously determined the thickness of the growing ALD TiO_x layer and the interface changes due to substrate oxidation as a function of ALD cycle number and process parameters. Figure 1(a) illustrates the results of this analysis for TiO_x deposition on Si using TDMAT as the metalorganic precursor. All three oxidants result in the expected linear growth of TiO_x with cycle number, though the growth per cycle (GPC) is affected by the oxidant type. Consistent with previous reports, the H₂O and OP₃₀₀ processes lead to GPC values of 0.39 and 0.50 Å/cycle, respectively.²³ Importantly, linear ALD film growth is observed even for the low power OP₅ process, though the GPC is reduced to 0.44 Å/cycle. Examination of in situ ellipsometry data obtained with high time resolution within averaged individual ALD cycles [Fig. S3(a)] reveals that the lower GPC is due to slower oxidation,⁹ which manifests as both a smaller effective thickness increase after the precursor pulse and slower ligand removal after OP exposure. The slower oxidation is likely caused by a reduced flux of oxygen radicals to the growth surface at low OP power, as indicated by optical emission spectroscopy (OES) that confirms an approximately two order of magnitude decrease in emission from oxygen radicals (O I) and ionic species (O II) for OP₅ compared to OP₃₀₀ (Fig. S4). Similar results are obtained for growth using the TTIP precursor on Si, with GPCs of 0.39, 0.29, and 0.25 Å/cycle for the OP_300, OP₅, and H₂O processes, respectively [Fig. S3(b)]. For all cases, the GPC values for TTIP are smaller than for TDMAT processes, which is consistent with the lower reactivity of this precursor.^9,23

As a control experiment, we also tested the behavior of the TDMAT process with pure O₂ gas without plasma ignition and compared it to the corresponding OP₅ process. After a similar thickness increase within the first ten cycles due to the consumption of surface OH-groups, also known as substrate-enhanced growth,²⁴ the GPC of the plasma-free process decreases to <0.01 Å/cycle (Fig. S5), indicating that plasma ignition is crucial for film growth.

While reduced plasma power results in slower film growth, it offers a distinct advantage by minimizing substrate oxidation. As depicted in Fig. 1(a), in situ SE measurements reveal that the OP₃₀₀ process leads to significant oxidation of the Si surface during early stages of growth. The oxidation rate is especially pronounced during the first ten PE-ALD cycles, during which the growing TiO_x film is not yet closed and lacks sufficient thickness to protect the underlying substrate. However, substrate oxidation is self-limiting and leads to an increase in the interfacial silicon oxide by 7 Å, from 1.0 to 1.7 nm [Fig. 1(b)]. In contrast, both the thermal and OP₅ processes yield significantly less oxidation of the Si interface. Comparative analysis of the thermal and PE-ALD processes on solvent-cleaned Si substrates with fully formed native oxide layers (2.3 nm initial thickness) instead of the chemical (RCA) surface treatment yielded a similar suppression of interfacial oxidation for the OP₅ compared to OP₃₀₀ process (Fig. S6).

X-ray photoelectron spectroscopy (XPS) provides further insight into the role of oxidation half-cycle on oxidation of Si. Figure 1(c) shows Si 2p XPS data obtained from the as-prepared Si surface, following deposition of 2 nm thick TiO_x layers. All spectra exhibit a spin–orbit split doublet near 99 eV, characteristic of Si from the substrate. The spectral contribution from the surface Si oxide is observed between 102 and 103 eV, indicative of Si⁴⁺ in SiO₂. Consistent with in situ SE measurements, deposition of TiO_x results in a slight increase in the Si⁴⁺ contribution for both the H₂O and the OP₅ processes, and a more significant increase for the OP₃₀₀ process. Overall, the relative intensities of the oxide components measured by XPS indicate increasing SiO₂ interfacial layer growth with increasing oxidation power during ALD (OP₃₀₀ > OP₅ ≈ H₂O), as shown in Fig. 1(b). Thus, interface oxidation can be significantly suppressed via reduction of the oxygen plasma power, with the OP₅ sample exhibiting considerably less SiO₂ than the OP₃₀₀ sample.

We note that there is some discrepancy between in situ SE and ex situ XPS measurements regarding the relative degree of interfacial oxidation. These differences are attributed to limitations of in situ SE, where correlation between modeling parameters of the two layers introduces uncertainty when dealing with changes at the few Å level.²⁵ Therefore, we consider XPS to be more reliable for characterizing the nature of the interface. Nevertheless, both techniques indicate suppression of substrate oxidation for the OP₅ compared to the OP₃₀₀ process.

Similar results are observed on InP surfaces, as exemplified by XPS data obtained following analogous ALD processes using TTIP (Fig. S7). Analysis of In 3d and P 2p core level spectra indicates the presence of an In(P)O_x interface layer, which is characteristic of InP.^19,26–29 As with the Si substrates, the OP₃₀₀ process leads to significant substrate oxidation, which can be suppressed by employing the OP₅ process. This finding highlights that low power PE-ALD offers a generally applicable strategy for improving interface quality compared to traditional processes performed with plasma powers of 200–600 W.¹¹ 

Having established the effectiveness of low power OP to suppress substrate oxidation during ALD, we now focus on the properties of the deposited layers. For this purpose, the number of cycles was adjusted to achieve uniform 10 nm thick films. No evidence for crystalline peaks was observed from Raman spectroscopy or x-ray diffraction (Figs. S8 and S9), indicating that all deposited films are amorphous, as expected for such thin TiO_x films deposited at 200 °C. In addition, the threshold temperature and thickness for crystallization should increase with lower oxygen plasma power, as less energy from impinging ions and photons is transferred to the film.^30,31 Consistent with their amorphous structure, all films exhibit smooth and conformal morphologies without any discernible grain-like structures or pinholes, as shown in the atomic force micrographs (AFM) in Figs. S10 and S11 on Si and InP substrates, respectively.

XPS analysis of the ALD films reveals that the composition and impurity content of TiO_x are not measurably affected by reducing the plasma power from 300 to 5 W. Figure 2(a) shows Ti 2p core level spectra from each of the TDMAT-derived films deposited on Si, along with their spectral fits. The corresponding data from films deposited using TTIP are provided in Fig. S12. For all films, the primary contribution to the Ti 2p_3/2 peak near 458.8 eV can be assigned to Ti⁴⁺ within TiO_x. In addition, a low binding energy shoulder is observed near 457.4 eV, which can be attributed to Ti³⁺ states generated by oxygen vacancies or donor impurities.^32–36 While such states are frequently observed for ALD TiO_x films,^32,33 the relative spectral contribution by Ti³⁺ is ∼2% for both OP₃₀₀ and OP₅ samples, which is significantly lower than the value of ≥4% observed for thermally deposited films using TDMAT, as highlighted by the difference spectra in Fig. 2(b). This is consistent with recent results on InP substrates and indicates that improved stoichiometry is achieved by PE-ALD and that this stoichiometry is preserved regardless of plasma power.¹⁹ Furthermore, we find no evidence for increased impurity concentrations in OP₅ compared to OP₃₀₀ samples, as indicated by their comparable C 1s spectra (Fig. S13) and the lack of any N 1s signal (Fig. S14). From these results, we conclude that the reduced oxygen plasma power does not introduce additional compositional non-idealities.

Previous studies have demonstrated that Ti³⁺ centers possess electronic states deep within the bandgap of TiO_x.^33,34,37,38 Valence band (VB) spectroscopy enables the detection of such occupied electronic states and provides a connection between chemical and optoelectronic properties of the films. As shown in Fig. 2(c), the onset of VB photoemission is ∼2.9 eV below the Fermi level for all films, indicating native n-type character. However, thermal deposition using H₂O yields significant concentrations of occupied electronic states within the bandgap, ∼0.6 eV below the Fermi level and ∼2.3 eV above the VB, which is characteristic of deep Ti³⁺ states [Fig. 2(d)].^{17,19,32,33,39,40} In contrast, no gap states are observed for the OP₅ and OP₃₀₀ samples, consistent with the low Ti³⁺ bonding contribution obtained from core level analysis. This finding indicates that high electronic quality films are formed via PE-ALD, including the low power OP process reported here, while thermal ALD results in larger gap state concentrations.

We now turn to the optical characteristics of ALD TiO_x films, which provide additional insight into their defect properties and dielectric character. Figure 3 shows the refractive indices (n) and extinction coefficients (k) determined via variable angle spectroscopic ellipsometry (VASE) for TiO_x films deposited on Si. The details of the optical model are provided in the supplementary material. While significant differences in optical properties are observed for films formed using the thermal compared to the OP processes, all PE-ALD films exhibit similar optical constants. In particular, both the OP₃₀₀ and OP₅ samples are characterized by large refractive indices that are indicative of dense TiO_x layers with minimal ligand incorporation and improved stoichiometry compared to the thermally deposited samples. Analogous TiO_x films deposited via PE-ALD on InP possess equivalent optical constants (Fig. S15).

Consistent with the VB spectra, VASE indicates a clear sub-bandgap optical response for the thermally deposited sample using TDMAT, which is absent for all OP- and thermally derived TTIP samples. We recently reported such a sub-gap response, which is associated with electronically active Ti³⁺ defect states within the bandgap of TiO_x.¹⁹ To determine if OP power affects the concentrations of such defects, we performed photothermal deflection spectroscopy (PDS) measurements, which enable sensitive characterization of weakly absorbing sub-bandgap states. Measurements were performed on TiO_x films deposited on fused silica (SiO₂) to enable quantification of the absorption coefficient, α, of the film without contributions from the substrate. As shown in Fig. 4(a), PDS measurements reveal that both OP₃₀₀ and OP₅ samples possess negligible sub-bandgap absorption down to the detection limit of ∼10¹ cm⁻¹, regardless of the utilized metalorganic precursor. In contrast, the thermal ALD films possess considerable sub-bandgap absorption, consistent with our recent findings.¹⁹ 

The Tauc analysis of the PDS data indicates similar bandgaps between 3.24 and 3.30 eV for all PE-ALD films, regardless of OP power, whereas thermally deposited samples possess larger bandgaps of approximately 3.4 eV (Fig. S16). The smaller values for the PE-ALD films are in agreement with previous reports for dense and stoichiometric amorphous titania thin films.^41–44 Likewise, both OP₅ and OP₃₀₀ samples are characterized by moderate Urbach energies of ∼100 meV, suggesting similar tail state concentrations and distributions, while the thermal films are characterized by larger Urbach energies.¹⁹ Overall, these results highlight the advantage of PE-ALD for achieving films with minimal concentrations of gap states compared to thermal ALD at the same temperature. The optical quality is preserved even for the low power OP₅ process, indicating that high quality dielectric coatings can be produced while simultaneously suppressing detrimental deposition-induced substrate oxidation.

Electrical transport measurements further reveal that PE-ALD films feature high in-plane resistivities that are consistent with low in-gap defect concentrations. To quantify film resistivities, transfer length method (TLM) measurements were performed following deposition of Ohmic Ti/Au (20 nm/80 nm) contacts (Fig. S17) on fused silica substrates. As shown in Fig. 4(b), all PE-ALD films are insulating and possess resistivities of >2 × 10⁵ Ω cm. Such high resistivities are consistent with prior reports, which have attributed increased electrical conductivities to the presence of Ti³⁺ states within the bandgap.^32,33,45,46 Consistent with this interpretation, both thermally deposited films possess significantly lower resistivities that are correlated with their increased sub-bandgap optical absorption coefficients.

To demonstrate the benefits of suppressed substrate oxidation by low plasma power compared to conventional PE-ALD, water-splitting photoelectrodes coated with TiO_x corrosion protection layers were fabricated and tested. The interfacial oxide represents a tunneling barrier and thus impedes out-of-plane charge transport, which is expected to yield inferior electrode performance. Figure 5 shows the photoelectrochemical (PEC) characteristics of p-InP/TiO_x/Pt photocathodes and n-Si/TiO_x/Ni photoanodes. In both cases, significant shifts in the photocurrent onset potentials between photoelectrodes coated with OP₅ and OP₃₀₀ TiO_x films are observed. In particular, we observe beneficial anodic and cathodic shifts for the hydrogen [Fig. 5(a)] and oxygen [Fig. 5(b)] evolution reactions, respectively. Consistent with prior findings for InP photocathodes, the larger photocurrent gradients indicate reduced series resistance (R_s) with suppressed interface oxidation,¹⁹ while the improved photovoltages (V_ph) can be attributed to a lower interface state density.⁴⁷ Together, reduced R_s and increased V_ph imply improved charge transport and electrical driving force, respectively, and thus enhanced device performance. While these findings are demonstrated for solar fuel-generating electrodes, they are also of relevance to other devices that involve out-of-plane charge transport, including solar cells^5,48 and sensors.²³ 

In conclusion, we have demonstrated that low power PE-ALD can be used to create TiO_x films possessing high optical and electronic quality, while also minimizing substrate oxidation that is typically associated with plasma-based deposition approaches. Although the reduced plasma power comes at the cost of reduced GPC, the resulting film properties are effectively indistinguishable from those generated with much higher plasma powers and are superior to those grown by thermal ALD in terms of stoichiometry and defect content. Thus, this milder PE-ALD approach enables improved interface quality without sacrificing film quality. Such advantages were realized across different ALD precursors (TTIP and TDMAT) and semiconductor surfaces (Si and InP), thereby highlighting that the approach is generalizable. The benefit of low power PE-ALD was demonstrated by TiO_x-protected water-splitting photoelectrodes, yielding significant performance enhancement over conventional processes. Overall, we expect that low plasma powers can enable PE-ALD on a broader range of sensitive substrates, providing improved processing compatibility with diverse materials.

See the supplementary material for details of materials and methods; ALD process parameters; in situ spectroscopic ellipsometry and optical emission spectra as a function of plasma power; analysis of interfacial oxidation on native oxide-coated Si; XPS of InP samples; Raman spectra, XRD, AFM, and XPS of each type of ALD TiOx film; optical constants and Tauc analysis; and solid state IV characteristics.

This work has received support by the Federal Ministry of Education and Research (BMBF, Germany) project number 033RC021B within the CO2-WIN initiative and from TUM.Solar in the context of the Bavarian Collaborative Research Project Solar Technologies Go Hybrid (SolTech).

The authors have no conflicts to disclose.

Oliver Bienek: Conceptualization (supporting); Formal analysis (lead); Investigation (lead); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Tim Rieth: Formal analysis (supporting); Investigation (supporting); Methodology (supporting); Writing – original draft (equal); Writing – review & editing (supporting). Julius Kuehne: Formal analysis (supporting); Investigation (supporting); Methodology (supporting); Writing – review & editing (supporting). Benedikt Fuchs: Investigation (supporting); Writing – review & editing (supporting). Matthias Kuhl: Investigation (supporting); Writing – review & editing (supporting). Laura I. Wagner: Investigation (supporting); Writing – review & editing (supporting). Lina M. Todenhagen: Investigation (supporting); Writing – review & editing (supporting). Lukas Wolz: Investigation (supporting); Writing – review & editing (supporting). Alex Henning: Investigation (supporting); Methodology (supporting); Supervision (supporting); Writing – review & editing (supporting). Ian D. Sharp: Conceptualization (lead); Funding acquisition (lead); Investigation (supporting); Methodology (equal); Supervision (lead); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.