Nucleation studies of high-κ aluminum oxide and hafnium oxide thin films on silicon carbide by plasma-enhanced atomic layer deposition

Modern methods for the fabrication of inorganic materials are developing toward a bottom-up approach characterized by the arrangement of matter on the atomic scale. In this context, atomic layer deposition (ALD)^1,2 certainly represents a controlled synthesis technique based on chemical vapor deposition mode having high precision on thickness control, highly uniform and conformal growth, as well as reproducibility. All these key factors are important requirements for the industrial implementation.³ Indeed, ALD has attracted an increasing attention in the semiconductor industry because of its suitability for the growth of binary oxides and nitrides to be used as high permittivity (high-κ) dielectrics. In particular, their implementation on wide bandgap semiconductors, such as silicon carbide (4H-SiC) and gallium nitride (GaN), is particularly promising for power and high-frequency device applications.⁴ In this context, the silicon carbide scientific community is currently devoting significant efforts on the evaluation of the potential benefits of high-κ materials as alternative to the commonly used silicon dioxide (SiO₂) as gate insulator.^4–8 

Several review papers^4,7 have been dedicated to high-κ dielectrics growth on 4H-SiC, but their insulating quality has been evaluated to be strictly depending on the fabrication technique. Specifically, the main efforts have been surely focused on ALD approaches thath have shown to potentially provide better high-κ/4H-SiC interface with respect to other deposition techniques. Moreover, the plasma-enhanced atomic layer deposition (PE-ALD) approach has shown proper process parameters such as lower deposition temperature and higher growth rates as well as beneficial effects on the high-k material properties such as higher film density and lower impurities content with respect to the analogous classic thermal ALD method.⁹ 

PE-ALD can be used to deposit a wide range of materials, especially high-κ oxides, and among them, aluminum oxide (Al₂O₃) is the most studied high-permittivity (κ ∼ 7–10)^4,7 insulator mainly because of its high critical electric field (∼9–10 MV/cm)¹⁰ and its large bandgap (∼7 eV), which determines good conduction and valence band offset to 4H-SiC.⁷ Furthermore, it is an amorphous and thermally stable material (its crystallization occurs only at temperatures higher than 800 °C).^11,12 The aluminum oxide surely possesses many advantages but also some drawbacks for microelectronics application, such as its slight sensitivity to water and/or atmospheric humidity. On the other hand, hafnium oxide (HfO₂) is another appealing insulator material especially because of its very high dielectric constant, ranging from 11 to 25 according to synthesis techniques and conditions,^4,7,13,14 but its use as a single gate insulator on 4H-SiC wafers demonstrated to be not so promising especially because of the high leakage current. In this regards, it may be interesting to combine both materials in more complex systems (e.g., multilayers, laminates, etc.) to optimize the physical and dielectric properties. Recently, Lo Nigro et al.^15,16 used PE-ALD to obtain Al₂O₃- and HfO₂-laminated and nanomixed systems as gate dielectric on both silicon and silicon carbide. Their combination demonstrated to improve the total dielectric constant values as well as hafnium oxide’s thermal stability compared to their intrinsic physical properties.^15,16 However, neither the single high-k materials nor their combinations have shown better insulator/4H-SiC interface quality than the SiO₂/SiC system. Moreover, some concerns exist about the nucleation phenomena taking place during the initial cycles of the PE-ALD process and their effects on the final properties of the deposited dielectrics. Indeed, a conformal layer-by-layer mechanism is just the ideal growth regime for ALD processes: A wide range of possible deviations are anyway likely to happen, achieving submonolayer growth after each ALD cycle. This is often ascribed to the steric effects from the precursor’s ligands blocking active sites or sparse nucleation sites on the surface, affecting the nucleation phenomena during early growth stages.^2,17 For instance, Schilirò et al.¹⁸ showed that the presence of a thin interfacial layer of SiO₂ on 4H-SiC results in higher density Al₂O₃ films and better physical properties, such as higher permittivity and breakdown voltage values, as well as lower leakage current density.¹⁸ 

However, the best way to monitor and study the nucleation stage is the implementation of an in situ technique during the ALD growth, such as spectroscopic ellipsometry (in situ SE).¹⁹ Nucleation investigation of Al₂O₃ growth on silicon has been already carried out,¹⁹ and recently, Morales et al.²⁰ also studied the ALD process of Al₂O₃ layers on metallic surfaces by in situ SE and in situ x-ray photoelectron spectroscopy, revealing a different behaviors at early growth stages due to interactions between the film and growth surface. Nevertheless, to our best knowledge, no detailed study on the nucleation on silicon carbide has been reported to date. In this work, the PE-ALD early growth stages of high-κ dielectrics of interest such as Al₂O₃ and HfO₂ on 4H-SiC have been studied by in situ spectroscopic ellipsometry, which has been used as a nondestructive and real-time characterization method to investigate the nucleation of high-κ oxides on the SiC surface. In particular, ALD “half-cycles measurements” have been successfully introduced to discriminate the effect of chemisorption and plasma reactant step during the growth process. Moreover, conductive-atomic force microscopy (C-AFM) has been used to investigate thin film uniformity in terms of insulator behavior at the nanoscale. Finally, the nucleation phenomena at the 4H-SiC surface were addressed and compared with silicon, demonstrating that the different functional sites density can determine some nucleation delay.

Depositions of Al₂O₃ and HfO₂ thin films were carried out on a commercially available (0001) Si substrate and n-type 4° off-axis (0001)-oriented 4H-SiC epitaxial layers grown onto a heavily doped 4H-SiC substrate. 4H-SiC and Si wafers were previously cleaned with piranha solution (H₂SO₄:H₂O₂ = 1:3) and diluted hydrofluoric acid (H₂O:HF = 10:1) to remove potential carbon contaminations and residual native oxide. ALD processes were conducted on a PE-ALD LL SENTECH Instruments GmbH reactor, equipped with a remote capacitively coupled plasma source, excited by a 13.56 MHz RF generator via a matchbox with a power supply of 200 W. The Al₂O₃ film depositions were performed at 250 °C using trimethylaluminum (TMA) as a metal precursor (0.06 s per ALD cycle) and O₂ plasma (150 SCCM, 1 s per ALD cycle) as an oxidant agent. Different samples were deposited by repeating this PE-ALD process per 20 cycles (P20) and 50 cycles (P50) on both silicon and silicon carbide. The HfO₂ deposition was conducted at 290 °C, and tetrakis-dimethylamino hafnium (TDMAHf) was used as a metal precursor (0.06 s per ALD cycle) and O₂ plasma as an oxidant agent (150 SCCM, 5 s per ALD cycle). During both processes, nitrogen was the carrier and purging gas (40 SCCM) and the total pressure of the reactor chamber was 20 Pa as baseline value that increases during the pulsing time up to 22 Pa.

In order to study the film surface roughness, atomic force microscopy (AFM) was performed by a PSIA XE-150 microscope in tapping mode with highly doped silicon tips. A DI 3100 AFM equipment by Bruker with Nanoscope V electronics and a TUNA module with diamond-coated Si tips was used for current mapping by C-AFM in order to evaluate insulating behavior uniformity. Ellipsometric measurements were performed via a Sentech SE 801 in situ spectroscopic ellipsometer operating in the partial UV wavelength range (between 240 and 390 nm) at 70° single angle of incidence, acquiring psi (Ψ) and delta (Δ) spectra. Measurements were acquired repeatedly during the same deposition run. SENTECH Spectraray/4 software was used for data modelling and regression analysis. Fit quality was monitored as deviation between measured and modelled spectra as root mean square deviation (rMSD). To determine the optical parameters of the dispersion model, thick layers of different thicknesses were fitted simultaneously. The parameters of the dispersion model were then assumed to be constant. The dispersion models obtained in this way were subsequently used to investigate dielectric thickness on thinner layers and the early growth stages of the analyzed ALD processes. Finally, the roughness has been modeled as a thin layer whose index of refraction intermediated between that of the deposited materials and air. This model is called “effective medium approximation (EMA) layer.” It is important to be sure that EMA causes a significant decrease in the fitting error (rMSD) or by contrast it can be possible not to use it.

In the case of Al₂O₃ fitting, the EMA roughness layer produced a marked decrease in rMSD (>0.5), so it was added to the final optical model. This was not the same for HfO₂, where very low rMSD decrease did not justify the addition of this fitting parameter.

First, the surface morphology and uniformity of insulating properties have been evaluated after the early growth stages of the PE-ALD Al₂O₃ process. Two sets of Al₂O₃ thin films have been simultaneously deposited on 4H-SiC and Si by 20 (P20) and 50 (P50) cycles, respectively, and they are related to nominal thickness of ∼1 and ∼5 nm.

The effective thickness values have been measured after lithographic patterning and selective chemical etching to fabricate linear steps on the Al₂O₃ layers deposited on silicon. Hence, AFM analyses have been used to measure the heights of the fabricated steps, and the founded thicknesses are quite close to the nominal values, i.e., 1.1 and 4.9 nm for P20 and P50 samples, respectively.

The surface morphology was imaged via AFM mapping on 2 × 2 μm² areas, and the film surface roughness values have been compared to the one of the bare 4H-SiC epitaxial layers. Representative morphological maps are reported in Figs. 1(a)–1(c), and the measured roughness (RMS) values are summarized in Table I. After 20-cycle PE-ALD process, the roughness increases from RMS = 0.180 nm of the bare 4H-SiC surface to RMS = 0.474 nm of the P20 sample. Interestingly, upon increasing film thickness up to 4.8 nm obtained by 50 PE-ALD cycles, the measured surface roughness slightly decreases to RMS = 0.305 nm. The morphological evolution can be better described by the histogram distribution reported in Figs. 1(d)–1(f), where it is evident that a narrower height distribution is present in the P50 sample [Fig. 1(f), FWHM = 0.42 nm] when compared with the P20 sample [Fig. 1(e), FWHM = 0.65 nm]. This trend can be attributed to a complete covering of the epi 4H-SiC surface.

The morphological results have been correlated to further characterization at the nanoscale, using C-AFM, to study the uniformity of insulating behavior on 10 different samples per process under study. The investigation has been carried out at a constant 5 V voltage applied between the AFM tip (V_tip) and the sample surface and by probing the current flow across the thin Al₂O₃ films [see the schematic illustration of this configuration in Fig. 2(a)]. The representative acquired current maps, reported in Fig. 2(b), show that the 4H-SiC epitaxial layer forms a Schottky contact with the C-AFM tip, exhibiting large conduction. This is also evident in the related current distribution in Fig. 2(c) where a broad current peak is visible and centered at a high value of 0.13 pA. On the other hand, P20 and P50 Al₂O₃ films show weaker tunneling-like conduction mechanisms because of their insulating properties. In particular, sample P20 has a nonuniform insulating behavior since conductive spots are not homogeneously distributed on the investigated area. Conversely, sample P50 did not possess evident spots but a uniform insulating layer. The histograms of the current distributions are reported in Fig. 2(c) and show that the P50 sample possesses the narrowest current distribution, which is also shifted to the lowest current values. The C-AFM investigation provides evidence of nonuniform film growth during the initial ALD cycles, likely due to surface phenomena inducing a delay nucleation on the 4H-SiC surface. The surface discontinuities are then covered and smoothed out upon increasing the cycles’ number.

For comparison, the same measurements have been acquired on films fabricated by the same ALD processes on the silicon substrate. The current maps in Fig. 2(d) showed that the insulating surface uniformity of the P20 sample on silicon appears only slightly lower than P50. Moreover, the current distribution in Fig. 2(e) discloses that the P20 sample deposited on silicon shows remarkably lower values and narrower current distribution when compared with the P20 sample on silicon carbide. Physical characteristics recorded on the deposited films are summarized in Table I, and all the collected data pointed out to the evidence of different Al₂O₃ nucleation phenomena due to the different surface chemical nature of silicon and silicon carbide.

Hence, the AFM investigation clearly demonstrated that the nucleation stage of high-κ films on 4H-SiC deserves to be monitored, and in this perspective, the growth of two different oxides, i.e., Al₂O₃ and HfO₂, has been monitored by in situ spectroscopic ellipsometer (SE) investigation. In this context, it should be noted that the AFM investigation has not been carried out on HfO₂ thin layers because they possess poorer insulating behavior. In fact, it has been demonstrated that even thicker HfO₂ films (about more than 10 nm) directly grown on silicon carbide epilayers generally showed high leakage current values. This behavior, which has been extensively reported in literature, could compromise the accuracy of the C-AFM measurements and the determination of the insulating homogeneity.

On the other hand, the in situ SE has been chosen as timesaving, nondestructive, and real-time metrology method to monitor the ALD growth process; nevertheless, its application for measurements on double-sided polished 4H-SiC substrates is not trivial and much more complicated than in the case of silicon. In fact, in the case of silicon carbide or other wide bandgap materials, the acquired signal in the spectral range below the bandgap of 3.26 eV (higher than 380 nm) is strongly affected by anisotropic and doping-dependent optical properties, as schematically shown in Fig. 3(a). The incident polarized beam splits into two beams because of the 4H-SiC anisotropy, while the two penetrating beams are reflected at the back of the wafer. They travel different lengths and interfere with each other, providing anisotropy-related information. The optical behavior of 4H-SiC, however, can be studied in the UV range, above the bandgap of 4H-SiC without being influenced by backside reflections. This greatly reduces the sensitivity of ellipsometry to anisotropy, which can be neglected, and 4H-SiC can be described as an isotropic material in the optical model.

In this context, a sufficiently large UV wavelength range (240–380 nm) was first used to acquire and fit in situ SE spectra of the bare 4H-SiC, as can be clearly seen in Fig. 3(b).

Second, sufficiently thick (>100 nm) Al₂O₃ and HfO₂ layers have been separately deposited on 4H-SiC for modelling purposes. In situ SE measurements have been acquired in real-time mode, twice per ALD cycle, i.e., after the precursor purge and the plasma-reactant purge steps, during the whole deposition process. Regression analysis has been then applied to recorded spectra to calculate the film thicknesses. Data fitting was then performed using the Tauc–Lorentz dispersion equations²¹ for 4H-SiC modelling. The possible presence of native silicon dioxide layer and/or formed during PE-ALD depositions has been considered for the fitting procedure using the Cauchy dispersion model with constant refractive index (n = 1.4610 at 633 nm) and constant thickness (t = 0.5 nm).^22,23 The same procedure has been used for Al₂O₃ films,²⁴ while surface roughness has been modelled using an EMA optical model,²⁵ as it is shown in Fig. 4(a). On the other hand, the Tauc–Lorentz dispersion model has been applied in the case of the amorphous and slightly absorbing HfO₂ films^14,26,27 as illustrated in Fig. 4(b). The optical parameters have been fitted using the multithickness approach extensively described in the experimental section, and it has been applied to all the progressively acquired data during the same deposition run. The results reported in Fig. 4(c) for n > 100 ALD cycles show a linear growth characterized by a constant increase of film thickness per cycle (GPC). This trend has been observed for both Al₂O₃ and HfO₂ processes, as characteristic feature of the ALD self-limiting technique and generally taking place at high number of cycles.^28,29 In particular, the Al₂O₃ growth is characterized by a GPC = 1.2 Å/cycle and a refractive index value at λ = 633 nm of 1.6 which is in good agreement with bulk Al₂O₃ materials.²⁸ In the case of the HfO₂ growth, a GPC = 1.4 Å/cycle has been calculated and the measured value of the refractive index (n = 2.1 at λ = 633 nm) is well aligned with literature reports for HfO₂ films deposited at the same 290 °C temperature on different substrates.²⁷ 

The recorded data show that the optical models have been successful and can be also used to fit measurements from early growth stages of both Al₂O₃ and HO₂ processes. In Fig. 5, the film thickness trend measured via in situ SE is reported for the 1–100 range cycles. In both cases, the root mean square deviation (rMSD) values (lower than 0.5) indicate optimal fitting reliability. In Fig. 5(a), Al₂O₃ thickness evolution confirms that nucleation issues affect the initial growth. In particular, no growth at all is observed during the initial three cycles and nonlinear growth takes place until ∼30 ALD cycles. Afterward, the deposition runs with the same constant GPC value as it was already reported at higher number of cycles. These observations are consistent with experimental evidence from AFM and C-AFM previously reported in Figs. 1 and 2.

In this context, the nonuniform P20 insulating properties can be attributed to growth occurring in the nonlinear regime, as clearly visible in Fig. 5(a). By contrast, the uniform morphology and surface conductivity measured in the P50 sample is due to the GPC linear growth regime.

Similarly, the HfO₂ growth, reported in Fig. 5(b), demonstrates an absence of nucleation for the first 10 cycles before the starting of the linear growth regime. Therefore, PE-ALD processes for different insulating oxides exhibit nucleation-related challenges when deposited on 4H-SiC.

The role of chemical nature of the dangling bonds on the SiC surface and their relationship with the observed nucleation delay has been evaluated by comparison of in situ SE measurements carried out on the silicon substrate. In Figs. 6(a) and 6(b), the data acquired during the deposition on silicon have been compared to those recorded on silicon carbide for Al₂O₃ and HfO₂, respectively. Neither nucleation delay nor nonlinear growth occurred on the silicon substrate, indicating that the silicon surface possesses higher amount and/or more active nucleation sites than the 4H-SiC surface, and this issue has a strong role on the nucleation delay.

Finally, further insights on the surface mechanisms involved in the 4H-SiC nucleation delay have been obtained by “half-cycle” in situ SE measurements after each precursor purge and plasma reaction steps. This investigation is a valid method to distinguish between the film thickness variation due to organometallic precursor chemisorption and the one after the plasma reaction during every PE-ALD cycle. The results are shown in Figs. 7(a) and 7(b) for Al₂O₃ and HfO₂, respectively. A relatively high thickness increase is observed after precursor chemisorption while a slight decrease is given by the plasma reaction due to the metal-oxygen bond formation and the release of the ligand groups.²⁰ This expected trend is typical of ALD step-like behavior²⁰ since both the used organometallic precursors (TMA for Al₂O₃ and TDMAHf for HfO₂) have greater size than a complete monolayer of the related deposited oxides.²⁰ In Fig. 7(a), it is evident that thickness variation due to TMA precursor adsorption happens since the first cycle of the Al₂O₃ PE-ALD process, while after plasma reaction no evident thickness variation is recorded: This could be partly ascribed to the competitive effect between chemisorption and physisorption of TMA on surface sites during the first cycles of the process, with physiosorbed species removed after plasma exposure. Then, the small number of nucleation sites on the 4H-SiC surface prevents the coalescence of adsorbed precursors during early growth stages, and an island-growth-like mechanism is probably dominant. This is likely the reason for low thickness variation per cycle during the early part of the nonlinear regime growth, before coalescence phenomena begin to take place enhancing the growth rate. The half-cycles in situ SE measurements on HfO₂ PE-ALD reported in Fig. 7(b) show a different situation. The TDMAHf pulse does not affect the film thickness for several cycles during early growth stages. This different precursor chemisorption at early growth stages is probably caused by the greater steric hindrance of TDMAHf, making ligand-exchange chemisorption mechanism less favorable. However, chemisorbed precursor molecules size favors proximity during plasma reaction, probably contributing to linear growth.

This study investigated the nucleation phenomena at early growth stages of PE-ALD of two different high-κ dielectric materials on 4H-SiC, i.e., Al₂O₃ and HO_2. In particular, very thin layers were studied by combining ex situ surface characterization techniques such as AFM and C-AFM, demonstrating the lack of morphological and conductive uniformity for the thinnest sample (∼1 nm). This effect is much more evident when 4H-SiC is used as the growth surface compared to silicon. In contrast, upon increasing the film thickness to ∼5 nm, slightly lower roughness and uniform insulating behavior were observed on the whole dielectric surface. The film thickness evolution was monitored via in situ SE measurements to initially investigate the PE-ALD deposition of thick Al₂O₃ and HfO₂ layers on 4H-SiC. Because of the complex optical behavior of double-side polished 4H-SiC, only data acquired in the UV wavelength range were considered for fitting. The dielectric optical modelling at high thickness returned reasonable optical properties and GPC values. The results at early growth stages confirmed nucleation issues during the first few cycles of Al₂O₃ films, and similar results were observed for the related HfO₂ process. In both cases, GPC stabilizes into the ALD linear growth mode as the number of cycles increases, further confirming the previous observations on Al₂O₃ uniformity evolution. This behavior was related to the different distribution and reactivity of surface functional groups in 4H-SiC, which plays a critical role in the nucleation and growth dynamics, particularly when compared to silicon substrates where such delays were not observed. The “half-cycle” in situ SE measurements introduced in this work finally provided a high signal-to-noise ratio, a valuable method to observe thickness variation during the chemisorption and plasma reaction steps of the PE-ALD process, highlighting the differences between Al₂O₃ and HfO₂ in terms of precursor interaction and nucleation on 4H-SiC. These results showed a fundamental approach to study nucleation phenomena at early growth stages of ALD processes by nondestructive and operando in situ SE. Moreover, measurements acquired during a single run of the PE-ALD process represent a less expensive, time saving alternative to ex situ characterization techniques. Evidence on Al₂O₃ and HfO₂ films shows the role of the growth surface functionalization on chemisorption and precursor reactivity during the reactant step of PE-ALD processes. However, a comprehensive compositional study of reaction intermediates and by-products using in situ and operando chemical characterizations techniques could offer further insights into the mechanisms and phenomena occurring during film nucleation and the early stages of ultrathin growth, presenting an opportunity to deepen the understanding of ALD processes on 4H-SiC.

The authors would like to thank Filippo Giannazzo (CNR-IMM) for the fruitful scientific discussions. Moreover, they acknowledge the European Union (NextGeneration EU), through the MUR PNRR project SAMOTHRACE (Grant Nos. PNRR-M4C2 and ECS00000022), for the partial financial support to this research.

The authors have no conflicts to disclose.

B. Galizia: Data curation (equal); Investigation (equal); Writing – original draft (equal). E. Schilirò: Conceptualization (equal); Validation (equal); Writing – review & editing (equal). P. Fiorenza: Data curation (equal); Investigation (equal); Writing – review & editing (equal). S. Peters: Formal analysis (equal); Software (equal); Validation (equal); Writing – review & editing (equal). J. Zessin: Data curation (equal); Writing – review & editing (equal). F. Roccaforte: Funding acquisition (equal); Project administration (equal); Writing – review & editing (equal). R. Lo Nigro: Conceptualization (equal); Data curation (equal); Supervision (equal); Writing – review & editing (equal).

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