betaFeedback

WorksPhysics and Astronomy

Epitaxial formation of ultrathin HfO2 on graphene by sequential oxidation

Abstract

We demonstrate the formation of epitaxial, ultrathin hafnia (HfO2) on graphene. Monoclinic hafnia (m-HfO2) forms as the end of a series of sequential oxidation reactions. Starting from Hf metal grown epitaxially on graphene, oxidation leads first to an amorphous suboxide (a-HfOx), then to a crystalline, hexagonal suboxide (h-HfOx) in epitaxial relationship with the substrate, and finally to m-HfO2 that is also epitaxial. We use scanning transmission electron microscopy to characterize the epitaxial relationships and to investigate the structure of h-HfOx. We propose a series of displacive transformations that relate the different crystalline phases and are consistent with the observed epitaxial relationships with the graphene substrate. ReaxFF based reactive molecular dynamics simulations confirm our model of the oxide phase sequencing, and illustrate the role of graphene in promoting oxide crystallization. Our results suggest a way to achieve heteroepitaxial integration of high-performance, crystalline dielectrics with two dimensional (2D) semiconductors with an atomically sharp interface, which is also relevant to hafnia phase engineering.

Introduction

Ultrathin two-dimensional (2D) semiconductors are promising materials to enable continued scaling of microelectronics.[1, 2, 3] However, a major hurdle is the heterointegration of 2D semiconductors with oxide dielectrics. The oxides of most 2D semiconductors are not good dielectric materials (e.g., MoO3_3), so direct oxidation – which works so well for silicon – is not a generalizable solution. Direct growth of oxides on 2D semiconductors by established methods of atomic layer deposition (ALD) is challenging due to the low density of dangling bonds, which stymies the kinetics of ALD.[4, 5, 6] Research is underway into a variety of approaches to this problem, but there remains no solution that is clearly ready for future manufacturing.[7, 8, 9, 10]

One intriguing approach is that of depositing ultra-thin metal films on 2D materials, followed by controlled oxidation. This could produce an ultrathin dielectric layer, or provide a seed layer for subsequent ALD.[11, 12] Recent work, by ourselves and others,[13, 14, 15, 16, 17] illustrates that 3D metals (Au, Ti, and Nb) can be deposited epitaxially on 2D materials via quasi-van der Waals heteroepitaxy, provided that the 2D surfaces are sufficiently clean and that the kinetics of metal deposition is carefully controlled. Here we leverage epitaxial growth of ultrathin Hf metal films on graphene as the starting point to study controlled oxidation.

Hafnium oxidation is a complex process as there are at least eight polymorphs of HfO2_2. At low pressure, the equilibrium phase diagram includes the monoclinic phase (m-HfO2_2, space group P21/cP2_1/c) at ambient temperature, and tetragonal and cubic phases (c-HfO2_2) at high temperature that show an enhanced dielectric constant.[14] Two orthorhombic phases appear in the phase diagram at high pressure. There also exist three ferroelectric phases that do not appear in the equilibrium phase diagram, but can be stabilized through doping and finite-size effects.[18] The diversity of oxide phases contributes to the usefulness of hafnia for resistive switching devices, which make use of ferroelectric phases and conductive suboxides.[20, 21]

In this work we make and characterize yet another hafnium oxide, an oxygen-deficient (i.e., suboxide) hexagonal phase (h-HfOx_x) that is more electrically conductive than m-HfO2_2. We describe a sequence of oxidation transformations that start from epitaxial, ultrathin hexagonal close packed Hf (hcp-Hf) metal on graphene, and proceed through an amorphous suboxide (a-HfOx_x) and a hexagonal suboxide (h-HfOx_x), before terminating with m-HfO2_2. We find that all crystalline phases (hcp-Hf, h-HfOx_x, and m-HfO2_2) remain in epitaxial alignment with the underlying graphene. The intermediate phases may be stabilized by finite-size and substrate effects. Scanning near-field optical microscopy (SNOM) indicates that the intermediate oxide phases are more conductive than m-HfO2_2. These experimental observations are confirmed by ReaxFF molecular dynamics (MD) results, which illustrate the coexistence of crystalline (hcp-Hf, h-HfOx\mathrm{HfO}_{x}, m-HfO2\mathrm{HfO}_{2}) and amorphous (a-Hf, a-HfOx\mathrm{HfO}_{x}) phases, and suggest that graphene assists in the stabilization of the hcp-Hf and h-HfOx\mathrm{HfO}_{x} phases by templating crystallization.

Experimental Results

We deposit ultrathin Hf metal films (nominal thickness 0.86 nm, approximately 4 metal monolayers) on clean, few-layer (10-20 layers) graphene surfaces at elevated temperature. We allow the metal to oxidize under different conditions, and we use scanning transmission electron microscopy (STEM) to characterize the results. The details of graphene transfer, substrate cleaning, metal deposition in ultrahigh vacuum (UHV) and oxidation can be found in Methods.

Phases formed during native oxidation of ultrathin epitaxial Hf metal on few-layer graphene

To study the initial stages of native oxidation - i.e., spontaneous oxidation in ambient conditions – we deposit a Hf metal thin film in UHV, and then expose it to air for approximately 15 min before imaging by aberration-corrected STEM. We use a relatively high substrate temperature of 600 °C (nominal) for film deposition, to promote the growth of large, faceted crystalline islands by suppressing nucleation density and enhancing the adatom diffusion length. We observe oriented islands with triangular symmetry on the substrate (Fig. S1). The average island lateral size is 190 nm and the area coverage is 38%, implying an average island height of 2.3 nm calculated from the total deposited thickness and coverage. The few-layer graphene substrate is supported on a perforated, electron-transparent SiNx\mathrm{SiN}_{x} membrane. There are no apparent differences between the Hf islands that form on and off the holes in the membrane, presumably because the 2D crystal is sufficiently thick that roughness of the SiNx\mathrm{SiN}_{x} membrane does not affect its surface.1

After 15 min of air exposure, the islands appear mostly unoxidized as they still show the contrast and diffraction spots expected from hcp-Hf, Figure 1a-Figure 1c. A comparison between atomic resolution images of the island and substrate, Figure 1b, and an atomic model (space group P63/mmc\mathrm{P6}_{3}/\mathrm{mmc}) confirms that the (0001)(0001) planes of Hf are parallel to those of the graphene. In Figure 1c we confirm this with a selected area electron diffraction (SAED) pattern, in which we identify two distinct sets of reflections. The inner set (light blue) corresponds to hcp-Hf, and the outer set (red) corresponds to graphene. We observe two epitaxial orientation relationships: the majority of Hf islands appear with a 30∘30^\circ rotation between Hf (101ˉ0)(10\bar{1}0) and graphene (101ˉ0)(10\bar{1}0), and a minority of islands appear with (101ˉ0)Hf∥(101ˉ0)graphene(10\bar{1}0)_{\mathrm{Hf}} \parallel(10\bar{1}0)_{\mathrm{graphene}}.

Native oxidation of epitaxial Hf crystals on graphene

Figure 1. Native oxidation of epitaxial Hf crystals on graphene. (a-c) After 15 min of air exposure. (a) The appearance of a-HfOx\mathrm{HfO}_{x} at the island edges. (b) Crystal structure of hcp-Hf compared with atomic model as viewed along zone axis [1]. (c) SAED pattern. (d) The co-existence of a-HfOx\mathrm{HfO}_{x}, h-HfOx\mathrm{HfO}_{x} and m-HfO2\mathrm{HfO}_{2} during native oxidation. (e) Crystal structure of m-HfO2\mathrm{HfO}_{2} compared with

As well as hcp-Hf, the images also show a small amount of amorphous material visible at the island edge (Figure 1a). We assume that the amorphous material is a-HfOx\mathrm{HfO}_{x} and that it is also present as a thin layer covering the entire islands; this is not apparent in plan-view STEM but can be seen in cross-section imaging, as discussed below. Upon close inspection, we also find two crystalline oxide phases present near the edges, although in limited quantities (Fig. S2). Their crystal structures and chemical compositions are discussed in detail below.

atomic model with zone axis [1ˉ11][\bar{1}11]. (f) SAED patterns collected after one month of oxidation showing disappearance of sharp hcp-Hf spots and appearance of oxide spots spread into arcs, indicating a few degrees distribution of rotational orientation in-plane. (g) Three equivalent orientations of m-HfO2_2 domains. Inset is the corresponding FFT. (h) A summary of the epitaxy among graphene substrate, hcp-Hf and hafnium oxides.

We examined the same island after one month of native oxidation, during which the sample was stored in a vacuum desiccator under ambient temperature and sub-ambient O2_2 and H2_2O partial pressure conditions. We found that the Hf metal lattice had almost disappeared, replaced by a-HfOx_x and two crystalline oxide phases (Figure 1d and Fig. S3), that had been barely evident after only 15 min of air exposure. One of these is identified as [1ˉ11][\bar{1}11]-oriented m-HfO2_2, by comparing its real-space atomic structure with the atomic model (Figure 1e), and by comparing its fast Fourier transform (FFT) with the simulated diffraction pattern (Fig. S4a-b). The other crystalline phase shows a hexagonal pattern, but is inconsistent with hcp-Hf (Fig. S4c), and is also unlike previously-reported hafnium oxides. When visible in cross sectional images, as shown below, ABC-stacked layers can be seen. We therefore denote this phase as h-HfOx_x. SAED (Figure 1f) shows diffraction spots from both m-HfO2_2 and h-HfOx_x. Both are elongated into small arcs, indicating a few degrees of misorientation. Also visible is a diffuse ring that overlaps the 0.31 nm d-spacing of h-HfOx_x, and that we assign to a-HfOx_x. We suggest that a-HfOx_x may be a disordered precursor to h-HfOx_x, with short- but not yet long-range order. We finally note that the height of the oxide islands after one month of native oxidation is 3 nm at the edges and up to 6 nm in the center, as measured by atomic force microscopy (AFM, Fig. S5). The edge-to-center variation comes from a similar variation in the initial metal islands, and the overall thickness increase is due to volume expansion during oxidation.

We now discuss the orientation relationship between the m-HfO2_2 and the graphene substrate, in light of the imaging and diffraction data discussed above. The m-HfO2_2 diffraction spots corresponding to d(110)=0.37d_{(110)} = 0.37 nm show six-fold symmetry (Figure 1f), whereas the m-HfO2_2 crystal structure has two-fold symmetry. The diffraction pattern can be explained by the summation of three epitaxial domain orientations, a hypothesis that is confirmed by real-space images that show all three domain types (Figure 1g). In fig:1h we summarize the orientation relationships between the graphene substrate, the as-deposited hcp-Hf\mathrm{Hf}, and the m-HfO2\mathrm{HfO}_{2} that result from native oxidation.

The three oxide phases a-HfOx\mathrm{HfO}_{x}, h-HfOx\mathrm{HfO}_{x}, and m-HfO2\mathrm{HfO}_{2} remain present after an additional eight months of native oxidation (nine months in total). We conclude that, by one month of native oxidation, the process has self-limited. The typical microstructures observed after 4, 6 and 9 months of oxidation are presented in Fig. S6.

According to the equilibrium phase diagram, the stable oxide at room temperature and ambient pressure is m-HfO2\mathrm{HfO}_{2}.[22] We therefore attribute the unexpected appearance of a-HfOx\mathrm{HfO}_{x} and h-HfOx\mathrm{HfO}_{x} and their stability to finite-size effects in the ultra-thin regime, in which interface effects (including epitaxy) and electrostatic equilibration (including the Mott potential) become important. Finite-size effects resulting in varying phase outcomes during metal oxidation are widely studied. Calculations predict that amorphous oxides are thermodynamically stable over the corresponding crystalline oxides in the ultra-thin limit on metals including Cr, Al, Ti and Zr; for example, the critical thickness for amorphous oxide stability on Zr (0001) is 1 nm between 298 K and 900 K.[23, 24, 25] Experimental studies of low-temperature oxidation of Zr have revealed the formation of non-stoichiometric, amorphous oxides, with thickness between 1–2 nm.[26, 27, 28] For instance, synchrotron x-ray photoelectron spectroscopy reveals that exposing Zr (0001) to pure oxygen at 10−810^{-8} torr and 300 K for 20 min results in an amorphous oxide bi-layer, consisting of a stoichiometric oxide (ZrO2_2, 1 nm) atop a suboxide (ZrOx_x, 0.5 nm).[26]

Details of the sequential phase transformations

In this section we explore details of the sequential transformations from Hf\mathrm{Hf}, to a-HfOx\mathrm{HfO}_{x}, to h-HfOx\mathrm{HfO}_{x}, and finally to m-HfO2\mathrm{HfO}_{2}. We observe, although with different kinetics, similar trends in both native oxidation and thermal oxidation, a process we perform at elevated temperature to drive the system beyond the native oxide configuration. In Figure 2 we present STEM images recorded before and after thermal oxidation, at 400 °C in air for 10 min, of a sample that had previously been stored in a desiccator for 9 months after growth. We directly observe three conversion processes: (1) a-HfOx\mathrm{HfO}_{x} crystalizes into h-HfOx\mathrm{HfO}_{x} (Figure 2a, Figure 2d); (2) h-HfOx\mathrm{HfO}_{x} transforms into m-HfO2\mathrm{HfO}_{2} (Figure 2b, Figure 2e); and (3) growth of m-HfO2\mathrm{HfO}_{2} at the expense of surrounding a-HfOx\mathrm{HfO}_{x} and h-HfOx\mathrm{HfO}_{x}, resulting in more distinct grain boundaries between m-HfO2\mathrm{HfO}_{2} domains (Figure 2c, Figure 2f). These conversions are also confirmed from the SAED patterns, as the diffraction spots from m-HfO2\mathrm{HfO}_{2} intensify, while the h-HfOx_x spots and the a-HfOx_x ring decrease in intensity (Fig. S7).

Oxide phase transformations during thermal oxidation

Figure 2. Oxide phase transformations during thermal oxidation. Image pairs show the same regions before and after thermal oxidation in air at 400 °C for 10 min. (a, d) Crystallization of a-HfOx_x into h-HfOx_x. (b, e) Conversion from h-HfOx_x to m-HfO2_2. (c, f) The growth of m-HfO2_2 at the expense of surrounding h-HfOx_x and a-HfOx_x. Scale bar: 2 nm.

We identify a dependence of the oxidation sequence on film thickness, by comparing two areas on the same island but with different thickness (Fig. S8). In the thinner area located at the island edge (Fig. S8f), a-HfOx_x remains the predominant phase. In the thicker area located at the center, (Fig. S8e), the crystalline phases are more prominent. This observation is consistent with the overall phase sequence, with a-HfOx_x forming first, followed by crystallization, with effects such as strain possibly altering the reaction kinetics.

This sequence of phase transformations can be understood by considering the starting and final crystal structures and compositions, Figure 3. In Figure 3a we illustrate [1]-oriented hcp-Hf, with AB stacking along [1] direction, and an interlayer spacing of 0.253 nm. In Figure 3c we illustrate [1ˉ11\bar{1}11]-oriented m-HfO2_2. The [1ˉ11\bar{1}11] direction is tilted by 4.8° from the (1ˉ11\bar{1}11) normal. The Hf atoms are arranged with ABC stacking, and the d-spacing of the (1ˉ11\bar{1}11) planes is 0.314 nm. The transformation from hcp-Hf to m-HfO2_2 therefore requires a change of stacking sequence from AB to ABC, a 4.8° tilt of the Hf layers, and a 24.1% interlayer expansion of the Hf sublattice. We suggest that this substantial transformation is broken down into smaller steps by virtue of the intermediate phases. In particular, the h-HfOx_x structure that we infer from our images (Figure 3b, top; Fig. S9a) shows ABC stacking. Furthermore, the in-plane expansion of the Hf sublattice in h-HfOx_x, relative to hcp-Hf, is similar to that of m-HfO2_2 (Table 1); and the interlayer spacing of h-HfOx_x is measured as 0.267 nm (Table 1), while the spacing of the ABC-stacked (111) planes in c-HfO2_2 is 0.290 nm. We define a supercell in a single layer Hf sublattice and compare the corresponding in-plane lattice parameters (aa, bb, and θ\theta) in Figure 3d. As listed in Table 1, from hcp-Hf to h-HfOx_x there is a 7.6% expansion of the Hf sublattice. Therefore, for reasons of symmetry and of geometry, we propose that ABC-stacked h-HfOx_x is an immediate precursor to m-HfO2_2, consistent with observations in Figure 2b and e. The final conversion to m-HfO2_2 requires a martensitic transformation. This must be coupled to in-diffusion of oxygen, since electron energy loss spectroscopy (EELS, Fig. S10-11) shows that a-HfOx_x and h-HfOx_x are both sub-oxides, with oxygen content lower than that of m-HfO2_2, consistent with previous XPS studies of Hf oxidation. [29, 30, 31, 32] In-diffusion of oxygen for the final step of the transformation to m-HfO2_2 is analogous to what has been described in literature for Zr oxidation.[33]

Phase transformation model and discussion of the structure of h-HfO$_x$

Figure 3. Phase transformation model and discussion of the structure of h-HfOx_x. (a) Atomic arrangement of [0001][0001]-oriented hcp-Hf viewed along [0001][0001] and [112ˉ0][11\bar{2}0]. Three symmetry-equivalent directions for further distortions in hcp-Hf are indicated as O1, O2 and O3. (b) HAADF-STEM images showing the h-HfOx_x structure, in plan view and in side-view, imaged in a lamella cut from a sample after 17 days of native oxidation using focused ion beam preparation. (c) Atomic arrangement of [1ˉ11][\bar{1}11]-oriented m-HfO2_2 along [1ˉ11][\bar{1}11] and [110][110]. (d) In-plane transformation of single layer Hf sublattice from hcp-Hf to m-HfO2_2 via h-HfOx_x; aa, bb, and θ\theta label the in-plane supercell lattice parameters of the Hf sublattice. (e) Cross-sectional image showing the a-HfOx_x, h-HfOx_x and m-HfO2_2 on graphite. (f) Cross-sectional image showing an atomically sharp interface between m-HfO2_2 and graphite.

Phasea (nm)b (nm)θ\theta (°)Interlayer spacing (nm)
Experimental results
hcp-Hf0.6731.165900.253
h-HfOx_x0.72±0.020.72 \pm0.021.26±0.031.26 \pm0.0389.9±1.089.9 \pm1.00.27±0.010.27 \pm0.01
m-HfO2_20.75±0.010.75 \pm0.011.23±0.031.23 \pm0.0385.4±0.985.4 \pm0.90.31±0.010.31 \pm0.01
Simulation results
hcp-Hf ReaxFF minimized0.6391.10690.0000.252
hcp-Hf at 900 K0.639±0.0180.639 \pm0.0181.100±0.0201.100 \pm0.02089.988±1.51689.988 \pm1.5160.260±0.0080.260 \pm0.008
m-HfO2_2 ReaxFF minimized0.7311.27182.9000.319
h-HfOx_x (Hf/Gr/50 O2_2 at 900 K)0.633±0.0160.633 \pm0.0161.086±0.0131.086 \pm0.01389.983±1.61989.983 \pm1.6190.260 ±\pm 0.011 (AB)<br>0.262 ±\pm 0.024 (ABC)
h-HfOx_x (Hf/Gr/100 O2_2 at 900 K)0.649±0.0190.649 \pm0.0191.097±0.0181.097 \pm0.01889.977±1.96389.977 \pm1.9630.256±0.0150.256 \pm0.015
m-HfO2_2 (Hf/Gr/3072 O2_2 at 900 K)0.716±0.0380.716 \pm0.0381.090±0.0991.090 \pm0.09983.513±6.11283.513 \pm6.1120.257±0.0130.257 \pm0.013

Table 1. Table 1: Lattice properties of hcp-Hf and oxide phases from experimental observations and simulations. The columns shaded in gold, green and blue are from Materials Project, experimental observations and MD simulations respectively.

The presence of ABC stacking in the h-HfOx_x phase should cause the disappearance of diffraction spots with dd-spacing of 0.31 nm, as these are related to the {101ˉ0}\{10\bar{1}0\} planes (d1‾01‾0=0.28 nm)(d_{\overline{1}0\overline{1}0} = 0.28\ \mathrm{nm}) of hcp-Hf (Fig. S9b). However, we do observe diffraction spots with d=0.31 nmd = 0.31\ \mathrm{nm} in samples after both native and thermal oxidation. We infer that the change in stacking sequence from AB to ABC is incomplete, and that the h-HfOx_x phase has a mixture of AB and ABC stacking. Cross sectional imaging (Figure 3b, bottom) indeed shows that both AB and ABC stacking can be present. This is consistent with the hypothesis of a transformation from AB to ABC stacking, as proposed above, and with the diffraction data. The presence of ABC stacking implies a structure that is distinct from hexagonal suboxides reported previously (Hf6_6O and HfO0.7_{0.7}) that are thought to be ordered phases of oxygen dissolved into hcp-Hf.[34, 35, 36] Since the h-HfOx_x phase shares ABC stacking with c-HfO2_2, but differs from the cubic phase in stoichiometry and in its lattice constants, this phase may be thought of as a structurally-distorted form of the cubic ZrO2_2 phase, which forms during oxidation of Zr with (111) parallel to Zr (0001), and has a high concentration of O vacancies (45% on the oxygen sublattice).[37]

The h-HfOx_x phase identified here may be inextricably linked to its processing history and nanoscale boundary conditions. For all oxide islands that we examined in cross section, we observed that h-HfOx_x when present was at the graphene/oxide interface, amounting to at least part (but never all) of the interface area. We also observed that h-HfOx_x is always covered by a-HfOx_x. We also observe the formation of oriented m-HfO2_2 atop h-HfOx_x (Figure 3e). Combined with the observed transformation sequences (Figure 2), this suggests that h-HfOx_x forms at the interface between graphene and a-HfOx_x, that its stability may be enhanced by the graphene lattice, and that it templates the eventual epitaxy between m-HfO2_2 and graphene that forms during thermal oxidation (Figure 3f).

Making continuous, ultra-thin hafnium oxide films on graphene

The above results demonstrate that ultra-thin Hf films deposited on graphene can begin as epitaxial metal-on-graphene, and end as epitaxial m-HfO2_2-on-graphene, with appropriate processing history. To improve control over this transformation, it is helpful to separate more clearly the steps of metal film deposition and oxide formation. To do so, we carried out deposition of an ultra-thin Hf metal film (nominal thickness 1.7 nm, approximately 8 metal monolayers) on graphene at room temperature (RT), followed by native oxidation. It is known that refractory metals can deposit epitaxially on graphene if the graphene surface is sufficiently clean,[13, 17, 38, 39] and we obtained epitaxy by using a pre-clean involving resistive sample heating in UHV (Figure 4a, S12). Compared to the high-temperature (HT) deposition (nominally 600 °C) that was described above, RT deposition produces higher nucleation density due to lower Hf adatom diffusion length. The higher-resolution image in Figure 4b shows the presence of Hf even in the darker boundary regions between islands. As a result, whereas HT deposition results in isolated and relatively thick Hf islands, in RT deposition we achieve a continuous film, without observable pinholes, and the height distribution has a FWHM of 0.66 nm (Fig. S13 and the insert of Figure 4a).

Continuous and ultra-thin hafnium oxide thin film grown on graphene

Figure 4. Continuous and ultra-thin hafnium oxide thin film grown on graphene via room temperature deposition and native oxidation. (a) Low- and (b) high-magnification images showing the formation of continuous HfOx_x film fully covering graphene substrate after 15 minutes of native oxidation. The insert in a is the height distribution histogram (relative to average height) of this

sample measured from AFM after 45-day oxidation. (c) Comparing the SAED patterns collected after 15-min and 12-day oxidation. The diffraction spots from graphene, h-HfOx\mathrm{h\text{-}HfO}_x and m-HfO2\mathrm{m\text{-}HfO}_2 are indicated by red, blue and yellow circles respectively. (d) Comparing the oxide phase ratios after native oxidation of Hf metal films deposited at HT and at RT, and for different oxidation times. The phase ratios are estimated from HAADF images. For the HT film, we use images from the same island after 15 min and 1 month of oxidation. For the RT film, which has more uniform island distribution, we use randomly-selected low-magnification images.

After native oxidation we observe the same a-HfOx\mathrm{a\text{-}HfO}_x, h-HfOx\mathrm{h\text{-}HfO}_x and m-HfO2\mathrm{m\text{-}HfO}_2 oxide phases in the RT film that were described above in the HT film, but the transformation kinetics are faster for the RT film. After 15 minutes of native oxidation, all three oxide phases are visible, and the hcp-Hf phase has disappeared (Figure 4b), and after 12 days the m-HfO2\mathrm{m\text{-}HfO}_2 phase is predominant (Figure 4b-Figure 4c). In contrast, in the HT sample, the hcp-Hf metal is slower to oxidize, and the m-HfO2\mathrm{m\text{-}HfO}_2 phase does not predominate even after 1 month (Figure 4d). The RT and HT films have comparable nominal starting Hf thickness (approximately 2 nm), but the island lateral sizes are quite different: 190 nm for the HT film, and 25 nm for RT film. We suggest that the faster oxidation kinetics in the RT film result from this difference in island size and exposure of surfaces other than the flat Hf (0001).

Comparing the electrical properties of ultra-thin oxides of Hf

The usefulness of Hf oxides is connected to their electrical properties. To compare the electrical properties of the ultra-thin oxides that we observe here, we use scattering scanning near-field optical microscopy (s-SNOM) with mid-infrared (mid-IR, 10 μ\mum) illumination. Mid-IR s-SNOM enables semi-quantitative measurements of electrical conductivity with spatial resolution below 20 nm. We performed measurements on a sample with a HT Hf metal film after undergoing thermal oxidation in at 400 °C for 10 min. In Figure 5a-Figure 5b we present images of topography and second harmonic scattering amplitude S2(f)S_2(f) (see Methods). The topography shows the expected dendritic oxide islands that were also observed in the STEM and AFM data (Figs. S1, S3, S5, and S8). The S2(f)S_2(f) data shows clear contrast between the oxide islands and the conductive graphene substrate. The S2(f)S_2(f) data also show an inverse correlation with island thickness, reflecting weaker mid-IR scattering where the islands are thickest, as illustrated by a line scan at position #1 (Figure 5c). We also observe significant deviations from this trend in many line scans, with local maxima in S2(f)S_2(f) indicating more conductive regions within the oxide islands, for example in the line scans at positions #2 and 3 (Figure 5c).

AFM and s-SNOM amplitude images of hafnium oxides with line scan profiles

Figure 5. Local conductivity mapping of hafnium oxides by mid-IR s-SNOM. AFM (a) and s-SNOM amplitude (b) images of hafnium oxides on graphene after thermal oxidation. (c) Line scan profiles corresponding to the positions indicated in panels (a) and (b).

Based on the STEM analysis above, we expect that the a-HfOx_x, h-HfOx_x, and m-HfO2_2 phases coexist in this sample, with the suboxides being more prominent along the dendritic arms of the islands where the initial Hf was thinner. We therefore conclude that the locally enhanced S2(f)S_2(f) indicates that the suboxides are more electrically conductive than m-HfO2_2, consistent with previous experiments and theoretical predictions.[21, 34, 40, 41, 42]

Computational modelling of oxidation pathways

We use MD simulations to study the kinetics of oxidation on length and time scales that are inaccessible to direct experimentation, yet can probe the outcomes of the oxidation processes that we described above. In Figure 6a-Figure 6b we illustrate the model considered in our ReaxFF MD simulations: a thin Hf metal slab on a 4-layer graphene substrate, surrounded by a variable number of oxygen molecules. The criteria for identifying distinct phases within the MD data are described in Sec. S2.2.

The MD model and thermal oxidation behavior of Hf computed by ReaxFF, showing top and front views of the MD model and thermal oxidation data

Figure 6. The MD model and thermal oxidation behavior of Hf computed by ReaxFF. (a) Top and (b) front views of the MD model showing an Hf slab on few-layer graphene in an O2O_2 environment. (c), (d), and (e) illustrate the percentages of hcp-Hf (Hf-Hf coordination numbers falling within 2-12), oxidized Hf, and oxygen consumption during the thermal oxidation, with and without graphene, at 900 K for 2.0 ns. Solid square symbols represent cases with graphene, and cross symbols represent cases without graphene. Each data point is an average of 11 samples, and error bars indicate their standard deviations.

Thermal oxidation kinetics of epitaxial Hf metal on few-layer graphene

In Figure 6c-Figure 6e, we present the evolution in time of a thin Hf slab during oxidation at 900 K over 2.0 ns, comparing cases with and without graphene, and for varying oxygen content. With sufficient oxygen present, the hcp-Hf metal phase disappears entirely by the end of the simulation, as expected. At intermediate oxygen content, we observe an interesting rebound of the hcp-Hf metal phase fraction (Figure 6c). This is due to the formation, on very short timescales, of amorphous Hf during a period of rapid dissociation of O2O_2 molecules and Hf-O bond formation. These chemical transformations are net exothermic and therefore generate transient local heating well above the 900 K thermostat temperature (Fig. S15), which seems to result in the formation of a-Hf. When there is insufficient oxygen present to fully oxidize the slab, these regions of a-Hf reform to hcp-Hf within the simulation time window. In Fig. S17 we visualize this process in greater detail. We observe the re-emergence of hcp-Hf between time t=1−3000t = 1 - 3000 ps for the case of 50 O2O_2 molecules – although noting that this regime of controllable, ultra-low oxygen concentration is experimentally inaccessible.

We note that the presence of graphene stabilizes hcp-Hf. For instance, in the case of 200 O2O_2 molecules, the slab on graphene retains a significant hcp-Hf phase fraction, whereas in the absence of graphene, the hcp-Hf phase diminishes to nearly zero (Figure 6c). Two factors may contribute to this stabilizing effect: that the graphene physically shields one side of the Hf slab from the O2O_2 atmosphere, and that the graphene crystal lattice templates [0001][0001]-oriented hcp-Hf.

The percentage of oxidized Hf (Figure 6d) and the consumption of O2O_2 gas (Figure 6e) vary as expected with total O2O_2 content. For intermediate O2O_2 content, the presence of graphene slightly lowers the total extent of Hf oxidation. Since the kinetics are fairly stable after 1 ns of simulation, this is evidence for the role of graphene as an epitaxial template that stabilizes hcp-Hf, in addition to being a diffusion barrier for O2O_2 gas. The effects of the graphene substrate are further discussed in § 3.3.

Oxide phases and phase transformations

We now describe the sequence of oxide phases and their transformations. For cases with low O2O_2 content (2–50 O2O_2 molecules), we observe h-HfOx_x epitaxially aligned to the graphene substrate (Fig. S19). Interestingly, we observe both AB and ABC stacking patterns of h-HfOx_x, in the case with 50 O2O_2 molecules, confirming our experimental results for this previously unreported phase. As the O2O_2 content increases (100–400 O2O_2 molecules), we observe the coexistence of crystalline (hcp-Hf, h-HfOx_x) and amorphous phases (a-Hf, a-HfOx_x). At much higher O2O_2 content (200–800 O2O_2 molecules), the h-HfOx_x phase is suppressed (Fig. S21). At the highest O2O_2 content (3072 O2O_2 molecules), the Hf atoms are fully oxidized, with oxygen-to-metal ratio approaching HfO2_2.

We track the oxide phase transformation through the variation in Hf-O coordination numbers (CNs). In Figure 7a–Figure 7d we present the distributions of Hf-O CNs for cases with a graphene substrate and with 200–3072 O2O_2 molecules. The 5-coordinated Hf atoms start to emerge with 400 O2\mathrm{O}_2 molecules and become extensively populated with 800 O2\mathrm{O}_2 molecules. For the case with 3072 O2\mathrm{O}_2 molecules, we observe the appearance of 6- to 8-coordinated Hf atoms. Hf atoms with Hf-O CNs of 6–8 are likely associated with the formation of HfO2\mathrm{HfO}_2 in tetragonal, monoclinic, and cubic crystal phases, respectively.[18] The 5-coordinated Hf atoms observed in Figure 7c may correspond to incipient nuclei of these fully-oxidized phases.

Hf/O coordination numbers and phases observed during Hf thermal oxidation

Figure 7. Hf/O coordination numbers (CNs) and phases observed during Hf thermal oxidation. All data are shown after annealing at 900 K for 2.0 ns. (a–d) Hf-O CNs with graphene present for 200 O2_2, 400 O2_2, 800 O2_2, and 3072 O2_2 molecules. (e–i) Bond and angle distributions for the Hf-Hf bond, Hf-O bond, O-O bond, Hf-O-Hf angle, and O-Hf-O angle, illustrating the predominant formation of the m-HfO2_2 for the case with 3072 O2_2 (black curves), in contrast to the case with 800 O2_2 (gray curves). The transparent cyan and blue curves represent the characteristic bond and angle distributions for pristine hcp-Hf and m-HfO2_2 samples equilibrated at 900 K. The vertical lines in (e) mark the 1st Hf-Hf bond length peaks for hcp-Hf and m-HfO2_2, those in (f) mark the 1st and 2nd Hf-O bond length peaks for m-HfO2_2, and the line in (g) marks the 1st O-O bond length peak for m-HfO2_2 (j) Illustration showing the coexistence of amorphous and crystalline Hf and HfOx_x phases during the transformation sequence from hcp-Hf to m-HfO2_2 for the case with 400 O2_2. (k) O:Hf ratios for cases with and without graphene, and for varying O2_2 content. The transparent black bar shows the O:Hf ratio calculated from ReaxFF fbMC/MD accelerated simulations, with the ratio being closer to 2.0, compared to those from regular ReaxFF MD. Each bar represents the average of 11 samples.

To better identify the HfO2\mathrm{HfO}_2 phases that emerge at high oxygen content (800 and 3072 O2\mathrm{O}_2 molecules), we present in Figure 7e-Figure 7i the distributions of Hf-Hf bond length, Hf-O bond length, O-O bond length, Hf-O-Hf angle, and O-Hf-O angle. We also present data from an accelerated simulation with the hybrid fbMC/MD method for the case with 3072 O2\mathrm{O}_2 molecules, and we reference distribution curves from pristine hcp-Hf, m-HfO2\mathrm{HfO}_2, t-HfO2\mathrm{HfO}_2, and c-HfO2\mathrm{HfO}_2 (all equilibrated at 900 K in the NpT ensemble). For the case of 800 O2\mathrm{O}_2 molecules, none of the reference HfO2\mathrm{HfO}_2 phases are unambiguously present. We observe partial overlap of the Hf-Hf and Hf-O bond length distributions with those of pristine hcp-Hf and m-HfO2\mathrm{HfO}_2 (Figure 7e-Figure 7f), but the O-O bond length and angle distributions for this case exhibit significant deviations from the references (Figure 7g-Figure 7i). In contrast, the result with 3072 O2\mathrm{O}_2 molecules shows a peak in the Hf-Hf bond length distribution that is intermediate between hcp-Hf and m-HfO2\mathrm{HfO}_2 (Figure 7e). The first- and second-neighbor peaks of the Hf-O bond length distributions closely align with those of m-HfO2\mathrm{HfO}_2 (Figure 7f), and the contours of the O-O bond length and angle distributions also align with those of m-HfO2\mathrm{HfO}_2 (Figure 7g-Figure 7i). This suggests that m-HfO2\mathrm{HfO}_2 predominates in the case with 3072 O2\mathrm{O}_2 molecules, while coexisting with a small fraction of HfOx\mathrm{HfO}_x suboxides.

We present in Figure 7k the slab-averaged O:Hf ratios at the end of the simulation time window. As expected, the average oxygen content in the oxides formed increases with total O2\mathrm{O}_{2} system content. We also see a decrease in oxidation in simulations without graphene compared to with graphene, due to the formation of smaller and more scattered HfOx\mathrm{HfO}_{x} clusters in the absence of graphene. For the highest O2\mathrm{O}_{2} system content studied (3072 O2\mathrm{O}_{2} molecules), the O:Hf ratio is 1.82, which is consistent with a m-HfO2m\text{-HfO}_{2} predominant phase coexisting with suboxides.

In Table 1 we present the lattice parameters found in our simulation results for slabs on graphene. For the highest O2\mathrm{O}_{2} system content studied (3072 O2\mathrm{O}_{2} molecules), the lattice parameters fall between those of hcp-Hf and m-HfO2m\text{-HfO}_{2}, with aa and θ\theta being closer to those of m-HfO2m\text{-HfO}_{2}. This observation further supports the predominance of m-HfO2m\text{-HfO}_{2} at high oxygen content. The computed interlayer spacing for h-HfOxh\text{-HfO}_{x} is also close to the value determined by experiment.

Effects of the graphene substrate

The experimental data suggests a role of the graphene substrate in the formation of the h-HfOxh\text{-HfO}_{x} and m-HfO2m\text{-HfO}_{2} phases. In Figure 8, we present our computational analysis of the effects of the graphene substrate on the thermal oxidation of Hf. After annealing at 900 K for 2.0 ns, the pristine (i.e., not-yet-oxidized) Hf slab on graphene shows a layered structure near the substrate, compared to the case without graphene (Figure 8b and Figure 8e). Additionally, the layered structure exhibits a higher fraction of the hcp-Hf phase than the case without graphene, as seen from top, front, and bottom views (Figure 8a–Figure 8f). We suggest that these layers promote the subsequent formation of h-HfOxh\text{-HfO}_{x}. This is illustrated in Figure 8g–Figure 8l, which shows that the presence of graphene promotes the formation of h-HfOxh\text{-HfO}_{x} in cases of limited oxygen content (here, 100 O2\mathrm{O}_{2} molecules). The effects of graphene can also be seen in the spatial distributions of suboxides HfOx\mathrm{HfO}_{x}, that form in smaller, more irregular clusters without graphene (Fig. S24).

Visualization of Hf atom distributions in hcp-Hf and non-hcp-Hf phases, hcp-Hf/h-HfO$_x$ and non-hcp-Hf/h-HfO$_x$ phases, and O atoms

Figure 8. Analyzing the effects of graphene substrate on Hf oxidation. (a–c) Visualization of Hf atom distributions in hcp-Hf and non-hcp-Hf phases (top, front, and bottom views) for the not-yet-oxidized case without graphene. (d–f) Visualization of Hf atom distributions in hcp-Hf and non-hcp-Hf phases (top, front, and bottom views) for the not-yet-oxidized case with graphene. (g–i) Visualization of Hf atom distributions in hcp-Hf/h-HfOx_x and non-hcp-Hf/h-HfOx_x phases and O atoms (top, front, and bottom views) for the 100 O2_2 case without graphene. (j–l) Visualization of Hf atom distributions in hcp-Hf/h-HfOx_x and non-hcp-Hf/h-HfOx_x phases and O atoms (top, front, and bottom views) for the 100 O2_2 case with graphene.

In Figure 9, we plot the distributions of reacted O atoms along the out-of-plane direction for cases with 50 to 3072 O2_2 molecules, both with and without the graphene substrate. The decrease in total slab thickness, from time t=1t = 1 to 100 ps, is likely related to the disappearance of amorphous phases at short times. Up to 10 ps, the difference in oxygen distribution without and with graphene is clearly related to the role of graphene as a diffusion barrier (Figure 9b and Figure 9g). At longer times, the distributions are less clearly dependent on diffusion kinetics. For t=100t = 100 ps and longer, the presence of graphene leads to an enhanced oxygen content at the bottom of the slab. This supports our experimental finding that the graphene lattice promotes the formation of Hf crystalline oxides, both h-HfOx_x and m-HfO2_2, through epitaxial alignment. The graphene substrate therefore contributes to directing the phase sequence. These results therefore suggest the mechanism by which we achieve atomically sharp interfaces between a 2D material and an epitaxial oxide dielectric.

Distributions of reacted oxygen along the out-of-plane direction

Figure 9. Distributions of reacted oxygen along the out-of-plane direction. (a–e) Reacted oxygen deposition along the out-of-plane direction for the Hf slab at 1 ps, 10 ps, 100 ps, 1000 ps, and 2000 ps without graphene. (f–j) Reacted oxygen deposition along the out-of-plane direction for the Hf slab at 1 ps, 10 ps, 100 ps, 1000 ps, and 2000 ps with graphene.

Conclusions

We have demonstrated the synthesis of m-HfO2_2 on graphene with an epitaxial alignment between the two materials and an atomically sharp interface. We show that this can be achieved by the deposition of a thin Hf layer followed by controlled oxidation. Electron microscopy characterization shows that the phase sequence, beginning with ultra-thin hcp-Hf on graphene, proceeds through suboxides a-HfOx_x and h-HfOx_x, before terminating with m-HfO2_2. All crystalline phases (hcp-Hf, h-HfOx_x, and m-HfO2_2) form epitaxially on the substrate. The sequence can be modeled as a series of displacive transformations of the Hf lattice, coupled to oxygen in-diffusion. The h-HfOx_x intermediate phase appears to be similar to c-HfO2_2 with a high concentration of stacking faults and oxygen vacancies. The formation of this h-HfOx_x phase is likely to be tied to the process and boundary conditions employed here, including gentle oxidation and the presence of the graphene substrate.

We also demonstrate the epitaxial deposition of a continuous Hf film on graphene using low temperature deposition on a clean substrate. This therefore constitutes a low-temperature processing pathway to form an ultra-thin (approximately 2 nm), conformal, epitaxial HfO2_2 film on graphene.

ReaxFF based molecular dynamics simulations are consistent with the experimental results. These simulations illustrate the roles of kinetics, limited oxygen pressure, and the graphene substrate in promoting the formation of the crystalline h-HfOx_x and m-HfO2_2 oxide phases.

Our results suggest strategies to create atomically-sharp interfaces between crystalline 2D semiconductors and crystalline oxide dielectrics, as desired for the future integration of 2D materials in microelectronics. The dependence of the oxide phase on processing that we show here also suggests that there may be opportunities in engineering the oxide phases that form. We anticipate that this may enable us to adapt the strategies shown here to take advantage of the diversity of hafnium oxides for devices such as resistive switches.

Methods

Exfoliation and transfer of graphene substrate

We exfoliate few-layer graphene from natural bulk graphite (NGS Trading & Consulting GmbH, Germany) onto SiO2_2 (90 nm)/Si substrates using Magic Scotch tape (3M). We then use a cellulose acetate butyrate (CAB)-assisted method to transfer the exfoliated graphene flakes to TEM grids.[15, 43] To locate the graphene flakes, we used Location-Tagged Holey Nitride TEM Grids with 200 nm thick, 0.5 ×\times 0.5 mm SiNx_x windows containing 2 μ\mum diameter holes with spacing 3 μ\mum, supported on a 3 mm diameter silicon frame (Norcada, Canada). We transferred graphene flakes onto the SiNx_x membrane and recorded the images at regions of the flakes suspended over the holes. Both the SiO2_2 (90 nm)/Si substrate and the TEM grid were cleaned by oxygen plasma before use.

Epitaxial deposition of Hf on graphene in UHV

We bake the graphene substrate in UHV (pressure <10−9< 10^{-9} Torr) using a Joule-heating holder at 400 °C for 4 h, to remove polymeric residue from the transfer step and contamination picked up in air. We then raise the temperature to 500–600 °C and deposit Hf using electron-beam evaporation in UHV (pressure ≈10−9\approx10^{-9} Torr), during which the sample is located between 80–90 cm from the source. The deposition rate is set at 50 Hz min−1\mathrm{min}^{-1} (corresponding to ∼\sim0.05 nm min−1\mathrm{min}^{-1}) and a total of 1000 Hz is deposited, where Hz refers to the frequency shift of a quartz crystal microbalance (QCM). For the RT deposition, a total of 2000 Hz is deposited.

Oxidation

After the Hf metal deposition, we inspect samples in the STEM after a total air exposure time of 15 minutes. We then stored samples in a pumped desiccator for variable lengths of time. We refer to this composite process of air exposure, followed by storage in a desiccator, as “native oxidation”. We performed thermal oxidation in air at 400 °C using a box furnace with a heating rate of 30 °C/min, and faster cooling achieved by opening the furnace door.

STEM characterization

We collected HAADF-STEM images, SAED patterns and EELS of plan view samples using a probe-corrected Thermo Fisher Scientific Themis Z G3 60-300 kV S/TEM operated at 200 kV. We used a probe current of ∼\sim100 pA and a semiconvergence angle of 19 mrad for imaging. We prepared cross-sectional samples by FIB using an FEI Helios 600.

Scanning probe measurements

We performed mid-IR s-SNOM experiments at room-temperature using a s-SNOM system (Neaspec GmbH) equipped with continuous wave, mid-IR quantum cascade lasers (DRS Daylight Solutions Inc.), using an illumination wavelength of 10 μ\mum. We used a Pt-coated silicon tip with a radius of 20 nm, operating in tapping mode at f=240f = 240 kHz, and measured the scattered light using a HgCdTe detector. Pseudoheterodyne interferometric detection was used to extract the near-field signal. The background signal is suppressed by demodulation of the near-field signal at the second harmonic of the tapping frequency. The data S2(f)S_2(f) presented in Figure 5 therefore represents the background-suppressed electrical polarizability of the tip-sample system, which increases sub-linearly with sample conductivity.

To determine sample thickness, we use a Bruker Dimension Icon SPM in tapping mode to carry out AFM measurements on sample regions supported by the SiNx_x membrane.

Modeling and RMD simulations

To construct the MD model, we obtain the unit cell of hcp-Hf from the Materials Project database.[44] We expand the unit cell to a 16×16×616 \times16 \times6 supercell, resulting in a Hf slab containing 3072 Hf atoms. Then, the hexagonal lattice is converted into an orthogonal lattice, yielding a rectangular Hf slab with a thickness of 2.775 nm, which is comparable to the experimentally measured value of ~2.8 nm (corresponding to a nominal thickness of 0.85 nm with 30 % coverage). Next, we construct a 4-layer graphene substrate with AB stacking, with an initial interlayer spacing of 3.4 Å, to support the Hf slab. The 4-layer graphene structure contains 4800 C atoms. While the number of graphene layers in this model (4 layers) is fewer than the 10–20 layers typically used in experiments, this choice enhances computational efficiency without compromising the ability to investigate the role of graphene in stabilizing the metal and metal oxide phases during annealing simulations. In the last step of building the model, we assemble the Hf slab, the 4-layer graphene, and a variable number of O2O_2 molecules (2–3072 O2O_2) in an orthogonal simulation box with periodic boundary conditions (PBC). The initial distance between the Hf slab and the top graphene layer is set to 5 Å, with a ~24 Å gap between the lateral surfaces of the Hf slab at opposite ends of the simulation box, as determined by the PBC. The modeling parameters for 18 cases, with and without a graphene substrate and with varying initial O2O_2 densities, are summarized in Table S1.

We perform ReaxFF MD simulations and ReaxFF uniform-acceptance fbMC/MD hybrid simulations on the Amsterdam Modeling Suite (AMS).[45] For the ReaxFF MD simulations, we first perform energy minimization using the conjugate gradient method. Then, we equilibrate each system with varying numbers of O2O_2 molecules at 300 K in the NVT ensemble for 100 ps. The Berendsen thermostat is applied with a temperature damping parameter of 100 fs. Subsequently, we heat the system from 300 K to 900 K with a heating rate of 10 K/ps. To prevent temperature discrepancies across different components of the system, we apply three Berendsen thermostats separately to the Hf slab, 4-layer graphene, and O2O_2 molecules, each with identical damping parameters (100 fs) and identical start and end temperatures (300 K and 900 K). This approach is necessary because, in systems with high O2O_2 densities, using a single thermostat could lead to an extremely high temperature (up to 2000 K) for the Hf slab and a very low temperature (down to 300 K) for the graphene. During the equilibration and heating steps, the Hf/O interactions in the ReaxFF “ffield” file are disabled to prevent chemical reactions between the Hf slab and O2O_2 molecules. In the subsequent annealing step, the Hf/O interactions in the ReaxFF “ffield” file is enabled. We set a target temperature of 900 K in the NVT ensemble and use three Berendsen thermostats with varying damping parameters: 10,000 fs for the Hf slab and O2_2 molecules, and 100 fs for the 4-layer graphene. The lower damping parameter for graphene is chosen because it is strongly thermally coupled with the heating source. The higher damping values for the Hf slab and O2_2 molecules are selected to prevent the clustering of gas-phase Hf oxides and to dissipate excess energy from gas/solid surface interactions, thereby simulating a high-vacuum environment. The annealing simulations run for 2 ns. For all above simulations, the time step is set to 0.1 fs.

For the hybrid simulations, we employ the uniform-acceptance fbMC/MD method, as detailed in previous references. [46, 47, 48] The simulations run alternatingly between fbMC and MD steps. During the fbMC steps, the ReaxFF optimizations are alternated with fbMC steps to accelerate events such as O2_2 adsorption and diffusion into the subsurface of the Hf slab, as well as Hf oxide phase transformations associated with an increased concentration of O depositing deeper into the Hf slab. The parameters for the fbMC steps, including the maximum displacement (drmax), the fbMC frequency (imcfrq), the number of fbMC steps (imcstp), and the mass scaling factor (imcroo), are set to 0.1, 10,000, 10,000, and 4, respectively, in the “control” file.[48] In this setup, the AMS-ReaxFF engine runs alternatingly between 10,000 iterations of MD steps and 10,000 iterations of fbMC steps until the total number of MD steps (nmdit) is reached. During the MD steps, the system is re-equilibrated, and the temperature is gradually reduced to the target value of 900 K. We use the NVT ensemble during the MD steps, with three separate Berendsen thermostats assigned to the Hf slab, the 4-layer graphene, and the O2_2 molecules, each having temperature damping parameters set to 10,000, 100, and 10,000, respectively. The total numbers of MD iterations and fbMC iterations are each 36,000,000. With a timestep size of 0.1 fs, the MD steps correspond to a total equilibration time of 3.6 ns at 900 K in the NVT ensemble.

Further effects, such as direct bonding between the substrate and the film, and reconstruction of the graphene lattice, are beyond the scope of this study because the Hf/C/H/O ReaxFF force field has not been specifically trained for covalent interactions between graphene and Hf-metal or partially/fully oxidized HfOx_x.[49]

Acknowledgements

This work was carried out in part through the use of MIT.nano facilities. The work was funded in part by Semiconductor Research Corporation (SRC) under contract no. 2021-NM-3027. R. K. acknowledges support by the National Science Foundation Graduate Research Fellowship Program under Grant No. 2141064. R. K. was supported in part by the United States Department of Energy under grant number DE-NA0003965. Financial support was provided by the Two Dimensional Crystal Consortium–Materials Innovation Platform (2DCC-MIP) at The Pennsylvania State University (PSU) under NSF cooperative Agreements DMR-1539916 and DMR-2039351. Computations for this research were performed on the PSU’s Institute for Cyber Science Advanced Cyber Infrastructure (ICS-ACI).

References

  1. [1]Liu, Y.; Duan, X.; Shin, H.-J.; Park, S.; Huang, Y.; Duan, X. Promises and Prospects of Two-Dimensional Transistors. Nature 2021, 591 (7848), 43–53. https://doi.org/10.1038/s41586-021-03339-z.
  2. [2]Li, T.; Guo, W.; Ma, L.; Li, W.; Yu, Z.; Han, Z.; Gao, S.; Liu, L.; Fan, D.; Wang, Z.; Yang, Y.; Lin, W.; Luo, Z.; Chen, X.; Dai, N.; Tu, X.; Pan, D.; Yao, Y.; Wang, P.; Nie, Y.; Wang, J.; Shi, Y.; Wang, X. Epitaxial Growth of Wafer-Scale Molybdenum Disulfide Semiconductor Single Crystals on Sapphire. Nat. Nanotechnol. 2021, 16 (11), 1201–1207. https://doi.org/10.1038/s41565-021-00963-8.
  3. [3]Zhu, J.; Park, J.-H.; Vitale, S. A.; Ge, W.; Jung, G. S.; Wang, J.; Mohamed, M.; Zhang, T.; Ashok, M.; Xue, M.; Zheng, X.; Wang, Z.; Hansryd, J.; Chandrakasan, A. P.; Kong, J.; Palacios, T. Low-Thermal-Budget Synthesis of Monolayer Molybdenum Disulfide for Silicon Back-End-of-Line Integration on a 200 Mm Platform. Nat. Nanotechnol. 2023, 18 (5), 456–463. https://doi.org/10.1038/s41565-023-01375-6.
  4. [4]Das, S.; Sebastian, A.; Pop, E.; McClellan, C. J.; Franklin, A. D.; Grasser, T.; Knobloch, T.; Illarionov, Y.; Penumatcha, A. V.; Appenzeller, J.; Chen, Z.; Zhu, W.; Asselberghs, I.; Li, L.-J.; Avci, U. E.; Bhat, N.; Anthopoulos, T. D.; Singh, R. Transistors Based on Two-Dimensional Materials for Future Integrated Circuits. Nat. Electron. 2021, 4 (11), 786–799. https://doi.org/10.1038/s41928-021-00670-1.
  5. [5]Yang, S.; Liu, K.; Xu, Y.; Liu, L.; Li, H.; Zhai, T. Gate Dielectrics Integration for 2D Electronics: Challenges, Advances, and Outlook. Adv. Mater. 2023, 2207901. https://doi.org/10.1002/adma.202207901.
  6. [6]Illarionov, Y. Yu.; Knobloch, T.; Jech, M.; Lanza, M.; Akinwande, D.; Vexler, M. I.; Mueller, T.; Lemme, M. C.; Fiori, G.; Schwierz, F.; Grasser, T. Insulators for 2D Nanoelectronics: The Gap to Bridge. Nat. Commun. 2020, 11 (1), 3385. https://doi.org/10.1038/s41467-020-16640-8.
  7. [7]Price, K. M.; Najmaei, S.; Ekuma, C. E.; Burke, R. A.; Dubey, M.; Franklin, A. D. Plasma-Enhanced Atomic Layer Deposition of HfO ₂ on Monolayer, Bilayer, and Trilayer MoS₂ for the Integration of High-k Dielectrics in Two-Dimensional Devices. ACS Appl. Nano Mater. 2019, 2 (7), 4085–4094. https://doi.org/10.1021/acsanm.9b00505.
  8. [8]Mleczko, M. J.; Zhang, C.; Lee, H. R.; Kuo, H.-H.; Magyari-Köpe, B.; Moore, R. G.; Shen, Z.-X.; Fisher, I. R.; Nishi, Y.; Pop, E. HfSe₂ and ZrSe₂: Two-Dimensional Semiconductors with Native High-k Oxides. Sci. Adv. 2017, 3 (8), e1700481. https://doi.org/10.1126/sciadv.1700481.
  9. [9]Li, W.; Zhou, J.; Cai, S.; Yu, Z.; Zhang, J.; Fang, N.; Li, T.; Wu, Y.; Chen, T.; Xie, X.; Ma, H.; Yan, K.; Dai, N.; Wu, X.; Zhao, H.; Wang, Z.; He, D.; Pan, L.; Shi, Y.; Wang, P.; Chen, W.; Nagashio, K.; Duan, X.; Wang, X. Uniform and Ultrathin High-K Gate Dielectrics for Two-Dimensional Electronic Devices. Nat. Electron. 2019, 2 (12), 563–571. https://doi.org/10.1038/s41928-019-0334-y.
  10. [10]Lu, Z.; Chen, Y.; Dang, W.; Kong, L.; Tao, Q.; Ma, L.; Lu, D.; Liu, L.; Li, W.; Li, Z.; Liu, X.; Wang, Y.; Duan, X.; Liao, L.; Liu, Y. Wafer-Scale High-k Dielectrics for Two-Dimensional Circuits via van Der Waals Integration. Nat. Commun. 2023, 14 (1), 2340. https://doi.org/10.1038/s41467-023-37887-x.
  11. [11]Son, S.; Yu, S.; Choi, M.; Kim, D.; Choi, C. Improved High Temperature Integration of Al2O3 on MoS2 by Using a Metal Oxide Buffer Layer. Appl. Phys. Lett. 2015, 106 (2), 021601. https://doi.org/10.1063/1.4905634.
  12. [12]Zeng, D.; Zhang, Z.; Xue, Z.; Zhang, M.; Chu, P. K.; Mei, Y.; Tian, Z.; Di, Z. Single-Crystalline Metal-Oxide Dielectrics for Top-Gate 2D Transistors. Nature 2024, 1–7. https://doi.org/10.1038/s41586-024-07786-2.
  13. [13]Reidy, K.; Thomsen, J. D.; Lee, H. Y.; Zarubin, V.; Yu, Y.; Wang, B.; Pham, T.; Periwal, P.; Ross, F. M. Mechanisms of Quasi van Der Waals Epitaxy of Three-Dimensional Metallic Nanoislands on Suspended Two-Dimensional Materials. Nano Lett. 2022, 22 (14), 5849–5858. https://doi.org/10.1021/acs.nanolett.2c01682.
  14. [14]Thomsen, J. D.; Reidy, K.; Pham, T.; Klein, J.; Osherov, A.; Dana, R.; Ross, F. M. Suspended Graphene Membranes to Control Au Nucleation and Growth. ACS Nano 2022, 16 (7), 10364–10371. https://doi.org/10.1021/acsnano.2c00405.
  15. [15]Reidy, K.; Varnavides, G.; Thomsen, J. D.; Kumar, A.; Pham, T.; Blackburn, A. M.; Anikeeva, P.; Narang, P.; LeBeau, J. M.; Ross, F. M. Direct Imaging and Electronic Structure Modulation of Moiré Superlattices at the 2D/3D Interface. Nat. Commun. 2021, 12 (1), 1290. https://doi.org/10.1038/s41467-021-21363-5.
  16. [16]Zagler, G.; Trentino, A.; Mustonen, K.; Mangler, C.; Kotakoski, J. Interface Effects on Titanium Growth on Graphene. 2D Mater. 2023, 10 (4), 045025. https://doi.org/10.1088/2053-1583/acf22e.
  17. [17]Anton, R.; Schneidereit, I. In Situ TEM Investigations of Dendritic Growth of Au Particles on HOPG. Phys. Rev. B 1998, 58 (20), 13874–13881. https://doi.org/10.1103/PhysRevB.58.13874.
  18. [18]Schroeder, U.; Park, M. H.; Mikolajick, T.; Hwang, C. S. The Fundamentals and Applications of Ferroelectric HfO2. Nat. Rev. Mater. 2022, 7 (8), 653–669. https://doi.org/10.1038/s41578-022-00431-2.
  19. [20]Banerjee, W.; Kashir, A.; Kamba, S. Hafnium Oxide (HfO₂) – A Multifunctional Oxide: A Review on the Prospect and Challenges of Hafnium Oxide in Resistive Switching and Ferroelectric Memories. Small 2022, 18 (23), 2107575. https://doi.org/10.1002/smll.202107575.
  20. [21]Li, L.-H.; Xue, K.-H.; Yuan, J.-H.; Mao, G.-Q.; Miao, X. Hafnia for Analog Memristor: Influence of Stoichiometry and Crystalline Structure. Phys. Rev. Mater. 2022, 6 (8), 084603. https://doi.org/10.1103/PhysRevMaterials.6.084603.
  21. [22]Ohtaka, O.; Fukui, H.; Kunisada, T.; Fujisawa, T.; Funakoshi, K.; Utsumi, W.; Irifune, T.; Kuroda, K.; Kikegawa, T. Phase Relations and Volume Changes of Hafnia under High Pressure and High Temperature. J. Am. Ceram. Soc. 2001, 84 (6), 1369–1373. https://doi.org/10.1111/j.1151-2916.2001.tb00843.x.
  22. [23]Jeurgens, L.; Sloof, W.; Tichelaar, F.; Mittemeijer, E. Thermodynamic Stability of Amorphous Oxide Films on Metals: Application to Aluminum Oxide Films on Aluminum Substrates. Phys. Rev. B 2000, 62 (7), 4707–4719. https://doi.org/10.1103/PhysRevB.62.4707.
  23. [24]Reichel, F.; Jeurgens, L. P. H.; Mittemeijer, E. J. Thermodynamic Model of Oxide Overgrowth on Bare Metals: Relaxation of Growth Strain by Plastic Deformation. Phys. Rev. B 2006, 74 (14), 144103. https://doi.org/10.1103/PhysRevB.74.144103.
  24. [25]Reichel, F.; Jeurgens, L. P. H.; Mittemeijer, E. J. The Thermodynamic Stability of Amorphous Oxide Overgrowths on Metals. Acta Mater. 2008, 56 (3), 659–674. https://doi.org/10.1016/j.actamat.2007.10.023.
  25. [26]Ma, W.; Herbert, F. W.; Senanayake, S. D.; Yildiz, B. Non-Equilibrium Oxidation States of Zirconium during Early Stages of Metal Oxidation. Appl. Phys. Lett. 2015, 106 (10), 101603. https://doi.org/10.1063/1.4914180.
  26. [27]Lyapin, A.; Jeurgens, L. P. H.; Graat, P. C. J.; Mittemeijer, E. J. The Initial, Thermal Oxidation of Zirconium at Room Temperature. J. Appl. Phys. 2004, 96 (12), 7126–7135. https://doi.org/10.1063/1.1809773.
  27. [28]Lyapin, A.; Jeurgens, L. P. H.; Mittemeijer, E. J. Effect of Temperature on the Initial, Thermal Oxidation of Zirconium. Acta Mater. 2005, 53 (10), 2925–2935. https://doi.org/10.1016/j.actamat.2005.03.009.
  28. [29]Chourasia, A. R.; Hickman, J. L.; Miller, R. L.; Nixon, G. A.; Seabolt, M. A. X-Ray Photoemission Study of the Oxidation of Hafnium. Int. J. Spectrosc. 2009, 2009, 1–6. https://doi.org/10.1155/2009/439065.
  29. [30]Suzer, S.; Sayan, S.; Banaszak Holl, M. M.; Garfunkel, E.; Hussain, Z.; Hamdan, N. M. Soft X-Ray Photoemission Studies of Hf Oxidation. J. Vac. Sci. Technol. Vac. Surf. Films 2003, 21 (1), 106–109. https://doi.org/10.1116/1.1525816.
  30. [31]Morant, C.; Galán, L.; Sanz, J. M. An XPS Study of the Initial Stages of Oxidation of Hafnium. Surf. Interface Anal. 1990, 16 (1–12), 304–308. https://doi.org/10.1002/sia.740160163.
  31. [32]Govindaraj, R.; Sundar, C. S.; Kesavamoorthy, R. Atomic Scale Study of Oxidation of Hafnium: Formation of Hafnium Core and Oxide Shell. J. Appl. Phys. 2006, 100 (8), 084318. https://doi.org/10.1063/1.2360148.
  32. [33]Liu, J.; Yu, H.; Karamched, P.; Hu, J.; He, G.; Goran, D.; Hughes, G. M.; Wilkinson, A. J.; Lozano-Perez, S.; Grovenor, C. R. M. Mechanism of the α-Zr to Hexagonal-ZrO Transformation and Its Impact on the Corrosion Performance of Nuclear Zr Alloys. Acta Mater. 2019, 179, 328–341. https://doi.org/10.1016/j.actamat.2019.08.051.
  33. [34]Kaiser, N.; Vogel, T.; Zintler, A.; Petzold, S.; Arzumanov, A.; Piros, E.; Eilhardt, R.; Molina-Luna, L.; Alff, L. Defect-Stabilized Substoichiometric Polymorphs of Hafnium Oxide with Semiconducting Properties. ACS Appl. Mater. Interfaces 2022, 14 (1), 1290–1303. https://doi.org/10.1021/acsami.1c09451.
  34. [35]Zhang, Y.; Mao, G.-Q.; Zhao, X.; Li, Y.; Zhang, M.; Wu, Z.; Wu, W.; Sun, H.; Guo, Y.; Wang, L.; Zhang, X.; Liu, Q.; Lv, H.; Xue, K.-H.; Xu, G.; Miao, X.; Long, S.; Liu, M. Evolution of the Conductive Filament System in HfO2-Based Memristors Observed by Direct Atomic-Scale Imaging. Nat. Commun. 2021, 12 (1), 7232. https://doi.org/10.1038/s41467-021-27575-z.
  35. [36]Hirabayashi, M.; Yamaguchi, S.; Arai, T. Superstructure and Order-Disorder Transformation of Interstitial Oxygen in Hafnium. J. Phys. Soc. Jpn. 1973, 35 (2), 473–481.
  36. [37]Ma, W. Initial Oxidation of Zirconium: Chemistry, Atomic Structure, Transport and Growth Kinetics.
  37. [38]Domask, A. C.; Cooley, K. A.; Kabius, B.; Abraham, M.; Mohney, S. E. Room Temperature van Der Waals Epitaxy of Metal Thin Films on Molybdenum Disulfide. Cryst. Growth Des. 2018, 18 (6), 3494–3501. https://doi.org/10.1021/acs.cgd.8b00257.
  38. [39]Ganz, E.; Sattler, K.; Clarke, J. Scanning Tunneling Microscopy of the Local Atomic Structure of Two-Dimensional Gold and Silver Islands on Graphite. Phys. Rev. Lett. 1988, 60 (18), 1856–1859. https://doi.org/10.1103/PhysRevLett.60.1856.
  39. [40]Xue, K.-H.; Blaise, P.; Fonseca, L. R. C.; Nishi, Y. Prediction of Semimetallic Tetragonal Hf₂O₃ and Zr₂O₃ from First Principles. Phys. Rev. Lett. 2013, 110 (6), 065502. https://doi.org/10.1103/PhysRevLett.110.065502.
  40. [41]Rushchanskii, K. Z.; Blügel, S.; Ležaić, M. Routes for Increasing Endurance and Retention in HfO₂-Based Resistive Switching Memories. Phys. Rev. Mater. 2018, 2 (11), 115002. https://doi.org/10.1103/PhysRevMaterials.2.115002.
  41. [42]Kaiser, N.; Song, Y.-J.; Vogel, T.; Piros, E.; Kim, T.; Schreyer, P.; Petzold, S.; Valenti, R.; Alff, L. Crystal and Electronic Structure of Oxygen Vacancy Stabilized Rhombohedral Hafnium Oxide. ACS Appl. Electron. Mater. 2023, 5 (2), 754–763. https://doi.org/10.1021/acsaelm.2c01255.
  42. [43]Schneider, G. F.; Calado, V. E.; Zandbergen, H.; Vandersypen, L. M. K.; Dekker, C. Wedging Transfer of Nanostructures. Nano Lett. 2010, 10 (5), 1912–1916. https://doi.org/10.1021/nl1008037.
  43. [44]Jain, A.; Ong, S. P.; Hautier, G.; Chen, W.; Richards, W. D.; Dacek, S.; Cholia, S.; Gunter, D.; Skinner, D.; Ceder, G.; Persson, K. A. Commentary: The Materials Project: A Materials Genome Approach to Accelerating Materials Innovation. APL Mater. 2013, 1 (1), 011002. https://doi.org/10.1063/1.4812323.
  44. [45]te Velde, G.; Bickelhaupt, F. M.; Baerends, E. J.; Fonseca Guerra, C.; van Gisbergen, S. J. A.; Snijders, J. G.; Ziegler, T. Chemistry with ADF. J Comput Chem 2001, 22 (9), 931–967. https://doi.org/10.1002/jcc.1056.
  45. [46]Timonova, M.; Groenewegen, J.; Thijsse, B. J. Modeling Diffusion and Phase Transitions by a Uniform-Acceptance Force-Bias Monte Carlo Method. Phys. Rev. B 2010, 81 (14), 144107. https://doi.org/10.1103/PhysRevB.81.144107.
  46. [47]Mees, M. J.; Pourtois, G.; Neyts, E. C.; Thijsse, B. J.; Stesmans, A. Uniform-Acceptance Force-Bias Monte Carlo Method with Time Scale to Study Solid-State Diffusion. Phys. Rev. B 2012, 85 (13), 134301. https://doi.org/10.1103/PhysRevB.85.134301.
  47. [48]Bal, K. M.; Neyts, E. C. On the Time Scale Associated with Monte Carlo Simulations. J. Chem. Phys. 2014, 141 (20), 204104. https://doi.org/10.1063/1.4902136.
  48. [49]Urquiza, M. L.; Islam, M. M.; Van Duin, A. C. T.; Cartoixà, X.; Strachan, A. Atomistic Insights on the Full Operation Cycle of a HfO₂-Based Resistive Random Access Memory Cell from Molecular Dynamics. ACS Nano 2021, 15 (8), 12945–12954. https://doi.org/10.1021/acsnano.1c01466.

Paper details

vlnc:W_hs9t9b9q
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents