Heterogeneous Local-Epitaxial Growth Behavior of Ultrathin HfO₂/Al-Doped TiO₂ Bilayer Dielectrics for Dynamic Random-Access Memory Capacitor Applications

Abstract

This study systematically examines the growth behavior and interfacial properties of a HfO₂/Al-doped TiO₂ (ATO) bilayer deposited on a Ru thin-film bottom electrode. Local-epitaxial growth of ATO into the rutile structure proceeds in two distinct, thickness-dependent stages that ultimately influence the crystallinity and interfacial integrity of the subsequently grown HfO₂ capping layer. At low ATO thicknesses (<4–4.5 nm), a coherent interface with strained rutile ATO is stabilized by the dominant influence of the underlying in situ-formed RuO₂. Conversely, at higher ATO thicknesses (>4–4.5 nm), an incoherent interface with relaxed rutile ATO becomes energetically preferred due to the increased volumetric energy contribution of the ATO. On coherently grown rutile ATO, the HfO₂ capping layer stabilizes in the higher-k tetragonal phase by efficient local lattice matching between aligned tetragonal HfO₂ and rutile ATO planes. In contrast, incoherent rutile ATO promotes disorder/intermixing, and monoclinic HfO₂ can emerge as the HfO₂ thickness increases. An appropriate r-ATO thickness region preserves insulating integrity and minimizes interfacial low-k layer effects. A Pt/HfO₂/ATO/Ru capacitor achieves a minimum equivalent oxide thickness of 0.62 nm at a minimum physical thickness of 5.9 nm, while maintaining sufficiently low leakage current level (<10^–7 A/cm² at 0.8 V).

1. Introduction

Dynamic random-access memory (DRAM), the main working memory in modern computers, is continuously scaled down toward a design rule of around 10 nm, where scaling of the cell capacitor comprises the most demanding task. − Each layer in the metal–insulator–metal (MIM) cell capacitor must be scaled to a physical thickness below 5 nm while maintaining reliable device performance for the significantly scaled pillar-type capacitor structures. The ZrO₂/Al₂O₃/ZrO₂ (ZAZ) dielectric film with TiN electrodes has been used in DRAM capacitors. These capacitors are now evolving into higher-k dielectric films that incorporate Hf. Although the high bandgap of ZrO₂ and HfO₂ (5–6 eV) can provide sufficiently low leakage current, the unavoidable presence of oxygen vacancies (V_O) induces an n-type doping effect. This adverse effect can be suppressed by introducing the Al₂O₃ layer, which mitigates the n-type doping effect by the counter-doping effect of Al in ZrO₂, enabling thickness scaling below 5 nm. However, ensuring a sufficient sensing margin for DRAM operation becomes more challenging as devices are further scaled down, due to the low dielectric constant of the HfO₂–ZrO₂ film (∼30–50, depending on the Hf:Zr ratio).

In this regard, TiO₂ with a rutile phase attracts a great deal of attention because it provides a sufficiently high dielectric constant (∼80–120), even at the modest postdeposition annealing (PDA) temperature of only ∼400 °C, reducing the risk of structural degradation. , High-quality TiO₂ films can be readily grown by a simple atomic layer deposition (ALD) process at lower growth temperatures (∼250–300 °C). However, the rutile structure is not thermodynamically stable at the processing temperature and pressure, where the lower-k (∼40) anatase structure is favored. Nonetheless, rutile-phase TiO₂ (r-TiO₂) can be achieved through an O₃-based ALD process on Ru substrates, where an in situ-formed rutile RuO₂ (r-RuO₂) interfacial layer during the TiO₂ ALD enables the r-TiO₂ formation through local-epitaxial growth. Similarly, sputtered or ALD RuO₂ electrodes can also induce r-TiO₂. The Ru and RuO₂ electrodes have higher work functions (∼4.7 eV for Ru and ∼5.1 eV for RuO₂) than TiN (∼4.4 eV), thereby helping suppress leakage current in the capacitor. However, capacitors based on r-TiO₂ with Ru-based electrodes still exhibit inferior leakage current characteristics compared to the (Hf,Zr)­O₂-based ones due to a low Schottky barrier height at the electrode interface, caused by a relatively narrow bandgap (∼3.1 eV) and the n-type nature of TiO₂ due to V_O presence. Al-doping can help mitigate this issue during the TiO₂ ALD process. − Still, the Al-doped TiO₂ (ATO) layer requires a minimum physical oxide thickness (POT) of about 9–10 nm to achieve a sufficiently low leakage current level (<10^–7 A/cm² at 0.8 V). This thickness, however, significantly exceeds the practical scaling target (<5 nm) necessary for advanced DRAM applications.

In this context, combining a higher-bandgap thin film with the r-TiO₂ film has been explored to decrease the POT while still suppressing leakage current and maintaining high capacitance. Among the options, HfO₂ is particularly promising due to its relatively high bandgap (∼5.5 eV) and well-established semiconductor manufacturing processes. Seo et al. reported a bilayer structure with HfO₂ capping on r-TiO₂, grown by ALD on a Ru thin-film electrode. They discovered that a tetragonal HfO₂ (t-HfO₂) phase with a higher dielectric constant (∼29) was formed on the r-TiO₂ layer, instead of a thermodynamically stable monoclinic HfO₂ phase (m-HfO₂) with a lower dielectric constant (∼16). This stabilization of the tetragonal phase (t-phase) was related to local multiunit-cell matching between the t-HfO₂ and the underlying r-TiO₂. Also, the trap density within the TiO₂ layer could be decreased during the oxidative HfO₂ ALD process, thereby helping suppress trap-assisted tunneling (TAT)-induced leakage current. As a result, a minimum equivalent oxide thickness (EOT) of ∼0.78 nm, meeting DRAM leakage current requirements, was achieved with the bilayer at a POT of ∼10 nm. However, this POT is still too thick for practical use, and the detailed interfacial and crystalline behaviors within the oxide and electrode layers were not fully examined. Therefore, a systematic and reliable investigation of the r-TiO₂-based dielectric at low oxide thicknesses (≪10 nm), including growth, crystalline properties, and electrical performance, is crucial for enabling further scaling for practical DRAM capacitor development.

In this regard, the present work aims to provide a mechanistic understanding of ultrathin high-k dielectric integration under a DRAM-relevant scaling region by systematically examining the growth behavior and interfacial characteristics of the HfO₂/ATO bilayer. The ATO film, rather than TiO₂, was used to improve device reliability in the scaled thickness region by suppressing leakage current and dielectric loss. Also, this work carefully assessed the crystallographic alignment at interfaces, a property that has rarely been studied in ultrathin TiO₂-based films. The local-epitaxial relationships at the rutile-ATO (r-ATO)/RuO₂ and HfO₂/r-ATO interfaces were specifically examined, since the degree of plane-to-plane alignment critically influences both phase stability and dielectric performance. Because even small variations at these interfaces can cause structural relaxation, defect formation, or phase transitions in ultrathin films, a thorough analysis of interfacial relationships was performed to mitigate interfacial low-k layer effects and address reliability issues during aggressive scaling.

The two-stage local-epitaxial growth of the r-ATO was confirmed by experimental results and further supported by ab initio calculations, which evaluated interfacial and volumetric energy contributions as well as unit-plane strain across possible matching configurations. Then, the effects of this two-stage r-ATO growth on the crystallization of the HfO₂ capping layer were studied, revealing their crucial role in determining whether the HfO₂ capping layer forms a t- or m-phase. By combining experimental observations with computational predictions, the proposed matching-dependent growth framework offers a physically consistent guide to engineering interfacial stability and controlling phase evolution in ultrathin stacked dielectric films. The optimized HfO₂/r-ATO bilayer structure serves as a model system for understanding how interfacial coherence can improve dielectric performance while diminishing interfacial degradation, thereby providing a scientific basis for dielectric design relevant to scaled DRAM capacitors and broader ultrathin dielectric heterostructures in advanced nanoelectronic devices.

2. Experimental Methods

2.1. Sample Preparation

A 50 nm-thick sputtered Ru bottom electrode (BE) was deposited on a substrate consisting of ZrO₂ (10 nm)/SiO₂ (100 nm)/Si, where the ZrO₂ film improved the adhesion of the Ru film to the SiO₂ layer. A 2 × 2 cm² coupon wafer was used for the experiments. ATO films were deposited using a traveling-wave-type ALD reactor (CN-1 Co., Plus-100) at a wafer temperature of 250 °C. Titanium isopropoxide (Ti­(OC₃H₇)₄, TTIP), trimethylaluminum (Al­(CH₃)₃, TMA), and ozone (O₃ concentration: 250 g/m³) were used as the Ti and Al precursors and oxygen source, respectively, to grow the ATO film. Al-doping was achieved by adding one Al₂O₃ ALD cycle for every 99 TiO₂ ALD cycles (i.e., a 1:99 ratio), resulting in an Al concentration of ∼5 atomic percent in the ATO film. For thinner TiO₂ films with fewer than 99 cycles, one Al₂O₃ ALD doping cycle was incorporated, since one cycle is the minimum ALD cycle, thereby increasing the Al-doping level. HfO₂ films were deposited using a traveling-wave type ALD reactor (CN-1 Co., Atomic Classic) at 280 °C. Tetrakis­(ethylmethylamido)­hafnium (Hf­[N­(C₂H₅)­CH₃]₄, TEMAHf) and ozone (O₃ concentration: 190 g/m³) were used as the Hf precursor and oxygen source, respectively.

To fabricate MIM capacitors, 50 nm-thick Pt top electrodes (TE) were deposited by direct-current (DC) sputtering through a circular shadow mask with a 300 μm diameter. Postmetallization annealing (PMA) was carried out at 400 °C for 30 min in a tube furnace under a nitrogen (N₂ 95%) and oxygen (O₂ 5%) atmosphere.

2.2. Sample Characterization

Film thickness was estimated using cross-sectional transmission electron microscopy (Cs-TEM; JEOL, JEM-ARM200F), spectroscopic ellipsometry (SE; J.A. Woollam, M-2000), and X-ray fluorescence spectroscopy (XRF; Thermo Fisher, ARL Quant’X), based on the correlation between observed film thickness and areal element density. Crystallinity of the films was analyzed by X-ray diffraction (XRD; PANalytical, X’pert Pro) with Cu Kα radiation, and additional structural analysis was performed with Cs-TEM accompanied by fast Fourier transform (FFT) analysis. The surface morphology of the films was observed using field-emission scanning electron microscopy (FE-SEM; Carl Zeiss, SUPRA 55VP). Chemical bonding structure of the films was analyzed by X-ray photoelectron spectroscopy (XPS; Kratos, AXIS SUPRA). Leakage current and capacitance of the MIM capacitors were measured with Hewlett-Packard 4140B and 4194A, respectively. Capacitance was measured under a DC bias sweep from −1 to 1 V with an alternating current signal amplitude of 0.05 V at 10 kHz, during which the capacitance remained constant; the capacitance at 0 V DC bias was used for analysis. The EOT of the measured devices was then calculated from the extracted capacitance as EOT = ε₀ε_SiO2 × (A/C), where ε₀, ε_SiO2 , A, and C represent the vacuum permittivity, SiO₂ permittivity (3.9), electrode area, and measured capacitance at 0 V, respectively.

2.3. Ab Initio Calculation

Density functional theory (DFT) calculations were performed using the Vienna Ab initio Simulation Package, employing the local density approximation alongside Blöchl’s projector augmented-wave method. − A 600 eV cutoff energy was used for the plane-wave basis set. Structural relaxations continued until the residual Hellmann–Feynman forces on all atoms were less than 0.02 eV/Å. Surface energies were calculated from slab models with more than eight atomic layers, with a vacuum spacing exceeding 1 nm to prevent spurious interactions between periodic images.

3. Results and Discussion

3.1. Two-Stage Local-Epitaxial Growth of Rutile ATO and Associated Effects on the HfO₂ Capping Layer

The crystalline properties of HfO₂ and ATO films within HfO₂/ATO bilayer structures on Ru substrates were examined. Figure (a) shows grazing incident X-ray diffraction (GIXRD) patterns of ∼8 nm-thick HfO₂ films depending on the ATO seed layer thickness, where the samples were annealed at 400 °C for 30 min in an N₂ (95%)/O₂ (5%) atmosphere. Samples that were not annealed remained in nearly amorphous states with negligible differences (data not shown). Since the ATO films grown on Ru substrates using an O₃-based ALD process were expected to exhibit a rutile phase, these films are therefore referred to as r-ATO. The HfO₂ film without an r-ATO seed layer (on r-ATO 0 nm sample) shows distinct peaks at approximately 28.5°, 31.8°, and 35.8°, corresponding to the m-phase with (1̅11), (111), (002), and (002)/(200) planes, respectively. It is well-known that m-phase tends to form in thicker fluorite-structured films because its bulk thermodynamic stability dominates over factors that may stabilize the t-phase at thinner thicknesses. − The m-phase typically develops above ∼5–8 nm in HfO₂ films deposited via ALD. In contrast, samples with 2.5 and 4.5 nm-thick r-ATO seed layers show significant suppression of the m-phase, with a peak at 35.4° corresponding to the t(110) plane as the primary crystalline feature. Notably, it is also noted that the HfO₂ film reverts to the m-phase when the underlying r-ATO film is 10 nm thick. In addition, a peak at ∼38.6° corresponding to the r-ATO (200) plane becomes detectable, consistent with the increased r-ATO thickness, providing sufficient diffracted signal. These GIXRD results suggest that a specific thickness range of the r-ATO seed layer not only suppresses m-phase formation but also promotes t-phase formation in HfO₂ films. The inset of Figure (a) shows GIXRD patterns of ∼8 nm-thick HfO₂ films on ATO seed layers of different thicknesses deposited on Si substrates, not Ru substrates. Since ATO films grown on Si substrates are typically amorphous or crystallize into the thermodynamically stable anatase phase due to the absence of lattice match between the ATO film and Si substrate (with an interfacial SiO₂ layer), , these ATO films are referred to as amorphous-ATO (a-ATO). All HfO₂ films on a-ATO seed layers exhibit the m-phase instead of the t-phase, regardless of the a-ATO thickness. These findings suggest that the phase evolution of the HfO₂ films on the ATO seed layer is strongly influenced by physical factors, such as lattice structure and atomic arrangement associated with the rutile frame.

1.

(a) GIXRD patterns of ∼8 nm-thick HfO₂ films depending on the r-ATO seed layer thickness. (b) EOT vs POT plots for HfO₂ films on r-ATO seed layers with different thicknesses. Cross-sectional TEM images and corresponding FFT results of HfO₂/r-ATO bilayers on Ru substrates with (c) 2.5 nm-thick, and (d) 10 nm-thick r-ATO seed layers.

Figure (b) shows EOT versus POT plots for HfO₂ films on r-ATO seed layers with different thicknesses. The dielectric constants (k-values) were calculated from the slope of the best linear fit (EOT = POT × (3.9/k) + EOT_i, where EOT_i is the interfacial component of the capacitor). For relatively thin HfO₂ films (<∼8 nm), similar slopes are observed regardless of r-ATO seed layer thickness, indicating consistent t-phase crystallinity (k ∼28–37). It is also observed that thin HfO₂ films on the bare Ru substrate also have k ∼37, suggesting that thinner HfO₂ films tend to form the t-phase even without the assistance of the r-ATO seed layer. For HfO₂ thicknesses above ∼8 nm (up to 15 nm in this study), films grown on bare Ru, and 4.5 nm and 10 nm-thick r-ATO seed layers show increased slopes, indicating the development of the m-phase (k ∼12–18). Notably, the HfO₂ films on the 2.5 nm-thick r-ATO seed layer remain in the t-phase (k ∼32) across the entire film thickness range (4–15 nm), demonstrating the m-phase suppression effect. This result may originate from a potential local-epitaxy between the t-HfO₂ and r-ATO layers, which could further stabilize the t-phase and prevent m-phase formation, as discussed later. The difference in phase evolution between thin (2.5 nm) and thick (10 nm) r-ATO seed layers, despite their identical rutile crystal structure, can be attributed to differences in the HfO₂/r-ATO interfacial characteristics that depend on r-ATO thickness. Kwon et al. reported a two-stage local-epitaxial growth mechanism for r-TiO₂ films, in which the influence of the in situ-formed r-RuO₂ diminishes as the TiO₂ film thickness increases. During the initial growth stage, thin r-TiO₂ maintains coherent grain growth with the underlying r-RuO₂ because of strong local-epitaxy. However, intrinsically different lattice constants between r-TiO₂ and r-RuO₂ induce strain accumulation in the thicker r-TiO₂ layer. Once the r-TiO₂ thickness exceeds a critical thickness, this strain is relieved by a transition to incoherent growth with lattice mismatch, dislocation formation, and more randomly oriented grains. In a similar manner, the thickness-dependent transition from coherent to incoherent growth in r-ATO would modify the surface properties, thereby affecting the crystallization behaviors of the subsequently deposited HfO₂ capping layer in this study because the rutile structure of the ATO film is expected to remain preserved at the Al concentration used in this work (∼5 at %). , In addition, the EOT_i values corresponding to the y-intercepts for these samples are 0.54, 0.42, 0.43, and 0.92 nm for r-ATO 0, 2.5, 4.5, and 10 nm cases, respectively. However, these EOT_i values include not only the HfO₂-induced interfacial low-k layer contribution, but also the EOT contributions of the r-ATO seed layer itself. Considering that the EOT contribution of the r-ATO seed layer is expected to be 0.2–0.6 nm, depending on its thickness, as can be inferred from Figure (a), the intrinsic HfO₂-induced interfacial low-k layer penalty is expected to remain lower for all cases employing an r-ATO seed layer (EOT_i < 0.4 nm) than for the case without r-ATO (EOT_i ∼0.54 nm). These results suggested that introducing an r-ATO seed layer can mitigate interfacial degradation at the HfO₂ interface, possibly due to improved structural compatibility between the Ru bottom electrode and the overlying HfO₂ layer, mediated by the r-ATO seed layer. Figure (c),(d) displays cross-sectional TEM images of HfO₂/r-ATO bilayers on Ru substrates with different r-ATO thicknesses. The right panels of each figure show the crystallographic planes from FFTs of the red square regions. On the 2.5 nm-thick r-ATO seed layer, the HfO₂ film shows t-phase with grain orientation matching between r-ATO (200) and t-HfO₂ (110) planes, indicating additional local-epitaxy. However, the crystallographic coherence from direct plane-to-plane matching between the r-ATO (200) and t-HfO₂ (110) planes is expected to be low due to their substantially different atomic configurations. Therefore, the observed local-epitaxy appeared to originate from another mechanism, as discussed later. In contrast, the HfO₂ film on the 10 nm-thick r-ATO seed layer displays m-phase with randomly oriented grains, consistent with the absence of additional local-epitaxy between r-ATO and HfO₂. These findings suggest that the coherent growth of r-ATO seed layer, enabled by sufficiently low r-ATO thicknesses, effectively promoted t-phase stabilization and suppressed m-phase development in the HfO₂ layer, extending this effect to at least ∼15 nm-thick HfO₂ film. In contrast, this effect lessened with thicker r-ATO seed layers. The bare Ru substrate also cannot provide a local-epitaxial relationship, thereby leading to m-phase formation in thicker HfO₂ films.

6.

(a) EOT vs total POT plots for HfO₂ single-layer, ATO single-layer, and HfO₂ (2.1 nm)/ATO bilayer capacitors. (b) J-V plots for HfO₂/r-ATO (3.8 nm) capacitors with different HfO₂ capping layer thicknesses. (inset) J–V plots of the same samples measured up to an applied voltage of 5 V. (c) J-EOT plots for ATO single-layer, ZAZ layer, and HfO₂/ATO bilayer capacitors.

The growth behavior of HfO₂ films on r-ATO seed was studied. Figure (a) shows the HfO₂ film thickness as a function of the number of HfO₂ ALD cycles for different r-ATO seed layer thicknesses. Without an r-ATO seed layer (on r-ATO 0 nm sample), HfO₂ films exhibit slightly faster initial growth compared to other cases, due to extra oxygen supply from the interfacial RuO _x layer. Nonlinear growth of oxide films such as TiO₂, SrO, and GeO₂ on Ru substrates during an O₃-based ALD process has been reported, driven by redox reactions between Ru and the in situ-formed RuO _x , which supply additional oxygen to the growing films. − During the HfO₂ ALD process in this study, O₃ injection oxidizes the exposed Ru surface at the initial growth stage, forming RuO _x (0 < x ≤ 2). In subsequent Hf precursor injection steps, the in situ-formed RuO _x can readily supply oxygen atoms to the Hf precursors due to its facile reducibility, thereby accelerating HfO₂ film growth. After ∼10 ALD cycles, the growing HfO₂ film itself suppressed further oxygen supply from the bottom RuO _x to the surface, reaching a steady-state growth mode with a saturated growth-per-cycle (GPC) of 1.41 Å/cy. With thin (2.5 and 4.5 nm) r-ATO seed layers, HfO₂ films show nearly linear growth from the start with saturated GPC of 1.46 and 1.41 Å/cy, respectively, because of adequate oxygen blocking by the r-ATO seed layer. Lee et al. also reported a similar oxygen-blocking effect of TiO₂ on Ru substrates, where an ∼3 nm-thick TiO₂ interlayer suppressed the accelerated growth of oxygen-scavenging SrTiO₃ film growth by an ALD at a substrate temperature of ∼370 °C. Conversely, with thick (10 nm) r-ATO seed layers, HfO₂ films show inhibited initial growth, indicating a delay in nucleation. After the nucleation period, the HfO₂ films reach a saturated growth with a GPC of 1.45 Å/cy. This nucleation delay would promote island growth via the Volmer–Weber growth mode, contributing to the formation of surface protrusions in the HfO₂ films. ,, Figure (b) shows SEM images of the surface morphology and the corresponding grain size distributions of ∼8 nm-thick HfO₂ films on r-ATO seed layers with different thicknesses. The grain size was estimated over a 1400 × 975 nm² surface area using the watershed method by Gwyddion software. The HfO₂ film grown directly on the bare Ru substrate (on r-ATO 0 nm) shows a relatively uniform surface with smaller grains, indicating fluent HfO₂ nucleation. As the thickness of the r-ATO seed layer increases, the surface morphology of the HfO₂ film is degraded, showing clearly distinguishable protruding and larger grains, accompanied by a broadened grain size distribution shifted toward larger radius. This degradation was attributed to nucleation delay at the heterogeneous interface followed by island-type growth and coalescence of HfO₂ nuclei, whereas the r-ATO seed layers themselves did not show noticeable protruding grains, as discussed in the following section. In particular, on the 10 nm-thick r-ATO seed layer, the HfO₂ film exhibits a significantly rough surface morphology, with the broadest distribution extending toward larger grain radii. For the 0 nm- and 10 nm-thick r-ATO seed layers, m-phase evolution of HfO₂ was observed, as shown in Figure . In the 0 nm-thick r-ATO case, this is attributed to excessive oxygen supply from the in situ-formed RuO _x layer, which decreases oxygen vacancy concentration in the HfO₂, whereas in the 10 nm-thick r-ATO case, it is associated with excessively large grain formation of the HfO₂. ,, In contrast, despite nearly identical HfO₂ grain size distributions arising from similar oxygen-blocking effects of the 2.5 nm and 4.5 nm-thick r-ATO seed layers, m-phase evolution was more effectively suppressed in the case of the 2.5 nm-thick r-ATO layer, as shown in Figure (b). These observations indicate that the primary origin of t-phase stabilization and m-phase suppression in the HfO₂ layer was the coherently grown r-ATO seed layer, which enabled effective local-epitaxial interactions at the HfO₂/r-ATO interface. Such coherency could be preserved only for sufficiently thin r-ATO films below 4.5 nm. When the r-ATO layer became incoherent at larger thicknesses, the local-epitaxial interactions at the HfO₂/r-ATO interface were weakened, and the phase evolution of the HfO₂ became dominated by the effects of grain size and oxygen concentration effects rather than by the interfacial structural effect. Overall, these results suggest that the r-ATO seed layer serves as a growth template, strongly influencing nucleation behavior, grain evolution, and surface morphology of the HfO₂ film, thereby determining phase evolution within the multilayer stack. Among the r-ATO thicknesses examined, the coherently grown 2.5 nm-thick r-ATO layer effectively stabilized the t-phase and suppressed m-phase evolution by establishing a coherent epitaxial framework, in which the grain size and oxygen concentration of the HfO₂ films were optimally balanced.

2.

(a) HfO₂ film thickness as a function of the number of HfO₂ ALD cycles for different r-ATO seed layer thicknesses. (b) SEM images of the surface morphology and the corresponding grain size distributions of ∼8 nm-thick HfO₂ films on different r-ATO seed layer thicknesses.

The growth behavior of HfO₂ films on r-ATO seed and their interfacial characteristics in the ultrathin regions were further investigated. Figure (a) presents SEM images of the surface morphology of HfO₂ films on r-ATO seed layers with different thicknesses. The r-ATO films themselves do not show distinguishable protruding grains up to a thickness of 6 nm in this figure, which can be attributed to the crystallographic continuity between r-RuO₂ and r-ATO and fluent nucleation of ATO due to the additional oxygen supply from the in situ-formed RuO _x interfacial layer during the ATO ALD process. In contrast, the surface morphology of the HfO₂ films is strongly dependent on the thickness of the underlying r-ATO layer. At low r-ATO thicknesses (2.5 and 4 nm) where the coherent growth mode was preserved, smooth surfaces without protrusions are observed across the ultrathin HfO₂ thickness range (1.3–2.7 nm). It indicates that a relatively flat and continuous HfO₂/r-ATO interface was formed from the initial growth stage, with HfO₂ fully covering r-ATO, thereby further promoting potential local-epitaxy between r-ATO and t-HfO₂, as supported by the TEM results in Figure (c). In contrast, at a higher r-ATO thickness (6 nm), where the growth mode transitioned to incoherent, protruding grains appeared in HfO₂ films due to the nucleation delay followed by island-type growth. This behavior would form a discontinuous HfO₂/r-ATO interface in some regions during the initial growth stage, thereby further disrupting the potential local-epitaxy between r-ATO and t-HfO₂.

3.

(a) SEM images of the surface morphology of HfO₂ films on different r-ATO seed layer thicknesses. (b) Hf 4f XPS spectra of 1 nm-thick HfO₂ films on 2.5 nm-thick and 10 nm-thick r-ATO seed layers.

Moreover, disruption of local-epitaxy between the r-ATO and HfO₂ layers would result in roughened interfaces rather than distinct, coherent boundaries, thereby affecting interfacial diffusion. Figure (b) displays Hf 4f XPS spectra of 1 nm-thick HfO₂ films on the thin (2.5 nm) and thick (10 nm) r-ATO seed layers. The spectra were calibrated using the C–C bonding peak at 284.5 eV as a reference. Both samples exhibit the main Hf–O chemical bonding structure with the Hf 4f_7/2 peak at approximately 16.3 eV. − However, the HfO₂ film on the 10 nm-thick r-ATO seed layer shows a relatively strong Hf–Ti–O peak at approximately 21.5 eV, indicating the formation of a (Hf, Ti)­O₂ solid solution instead of a chemically distinct HfO₂/r-ATO interface. This intermixing behavior was linked to the roughened interface caused by disrupted local-epitaxy, which facilitated atomic diffusion between the two layers. In contrast, such intermixing was hardly observed in the thinner r-ATO seed layer case, suggesting that the coherency enabled by local-epitaxy effectively preserved chemical separation between the two layers. Overall, these findings suggest that thin r-ATO seed layers promoted coherent growth and well-defined interfaces between r-ATO and HfO₂, whereas thicker r-ATO seed layers led to incoherent growth and rougher, less distinct interfaces, resulting in intermixing.

3.2. Modeling of Two-Stage Local-Epitaxial Growth of the Rutile ATO Layer

Figures – show that the local-epitaxial growth of r-ATO films on Ru substrates exhibits a two-stage growth behavior depending on the r-ATO thickness, which in turn affects the crystalline characteristics of the subsequent HfO₂ capping layer. The first growth mode of r-ATO is coherent, with strong crystallographic alignment between the r-ATO and r-RuO₂ planes. In contrast, the second growth mode is incoherent, with a degraded alignment between adjacent planes leading to the formation of defects and grain boundaries.

To further clarify the growth modes of r-ATO, the local-epitaxy between r-RuO₂ and r-ATO was theoretically examined using the DFT calculation. Cell parameters of r-TiO₂ were used in this calculation, since those of r-ATO were expected to be nearly identical due to the feasibility of substitutional Al-doping at a low concentration (∼5 atomic %) without clustering or significant lattice distortion. , In addition, a local-epitaxy model based on unit-plane area matching was used because this approach is more suitable for polycrystalline systems than direct lattice-parameter matching, which is mainly applicable to single-crystalline films. By analyzing the cell parameters along with the surface and volume energies of each unit plane for both r-ATO and r-RuO₂, the most energetically favorable growth mode of r-ATO was determined.

The total system energy of growing r-ATO grains for both coherent (strained) and incoherent (relaxed) growth modes was estimated according to the following equations. By comparing the system energies of the coherent and incoherent cases as a function of the number of r-ATO layers (i.e., thickness of r-ATO layers), the energetically favorable growth mode and the corresponding critical thickness can be determined.

$${\mathit{E}}_{\mathrm{grain},\mathrm{coherent}\mathrm{r}‐\mathrm{ATO}}={\mathit{E}}_{\mathrm{bulk},\mathrm{coherent}\mathrm{r}‐\mathrm{ATO}}+\frac{{(\mathit{\gamma }}_{\mathrm{r}‐{\mathrm{RuO}}_2/\mathrm{r}‐\mathrm{ATO}\mathrm{interface}})\mathit{A}+{(\mathit{\gamma }}_{\mathrm{strained}\mathrm{r}‐\mathrm{ATO}\mathrm{top}\mathrm{surface})}\mathit{A}}{\mathrm{number}\mathrm{of}\mathrm{r}‐\mathrm{ATO}\mathrm{layers}}$$ 1

$${\mathit{E}}_{\mathrm{grain},\mathrm{incoherent}\mathrm{r}‐\mathrm{ATO}}={\mathit{E}}_{\mathrm{bulk},\mathrm{incoherent}\mathrm{r}‐\mathrm{ATO}}+\frac{{(\mathit{\gamma }}_{\mathrm{GB}})\mathit{A}+({\mathit{\gamma }}_{\mathrm{relaxed}\mathrm{r}‐\mathrm{ATO}\mathrm{top}\mathrm{surface}})\mathit{A}}{\mathrm{number}\mathrm{of}\mathrm{r}‐\mathrm{ATO}\mathrm{layers}}$$ 2

$${\mathit{\gamma }}_{\mathrm{GB}}=\mathit{\alpha }\times (\mathrm{surface}\mathrm{energy})$$ 3

where γ represents the interface energy at junctions or surfaces within the model and A is the junction area for each unit plane. Grain boundary energy (γ _GB) was approximated from the surface energy using a scaling factor (α), which was set to 0.5 in accordance with previous reports on thin films, as direct atomistic calculations of γ _GB are computationally prohibitive. −

From these equations, the energy difference between growing grains and the bulk state, which indicates the thermodynamic tendency for grain formation, was calculated for both growth modes across different crystallographic planes of the rutile phase. Among the possible orientations, the (110), (101), and (200) planes, which are commonly reported in experiments, were analyzed because other planes tend to have higher surface energies and are less likely to dominate the grain surface. Figure (a)–(c) displays the energy differences between growing grains and the bulk state for both growth modes of r-ATO layer on r-RuO₂ as a function of the number of r-ATO layers for these three planes, with the corresponding energetics shown in Table . In all cases, coherent growth of the r-ATO layer was energetically preferred at low thicknesses due to strong local-epitaxy with the underlying r-RuO₂. As the number of r-ATO layers increased, incoherent growth became more favorable, with the critical transition thickness varying with crystallographic plane. For the (110) plane, the transition from coherent to incoherent growth occurred between 1.9 and 2.6 nm (Figure (a)); for the (101) plane, between 4.3 and 4.9 nm (Figure (b)); and for the (200) plane, between 6.3 and 6.8 nm (Figure (c)). Since r-ATO films are polycrystalline, this transition would occur across various grain orientations, resulting in a gradual shift from coherent to incoherent growth within the thickness range of 1.9 to 6.8 nm. These findings are consistent with experimental results showing a transition near approximately 4–4.5 nm, as shown in Figures (a)–(b) and (a).

4.

Energy differences between growing grains and the bulk state for coherent and incoherent growth modes of r-ATO layer on r-RuO₂ as a function of the number of r-ATO layers for (a) rutile (110), (b) rutile (101), and (c) rutile (200) planes.

1. Energetics for Calculation in the Local-Epitaxial Growth of r-ATO on r-RuO₂ .

	bulk E [meV/f.u.]		surface E [J/m2]
rutile plane	coherent ATO	incoherent ATO	coherent ATO	incoherent ATO	RuO2-ATO Interface E [J/m2]
(110)	16	0	0.85	0.83	0.191
(101)	22	0	1.41	1.40	0.215
(200)	8	0	1.08	1.11	0.254

3.3. Modeling of the Growth of the HfO₂ Capping Layer on the Rutile ATO Layer

Based on the experimental and computational results, a two-stage growth mode of r-ATO films was confirmed. Then, the local-epitaxy between the r-ATO and t-HfO₂ layers was examined in detail. Figure (a) shows θ–2θ XRD patterns of ∼8 nm-thick HfO₂ films depending on the r-ATO seed layer thickness, with deconvoluted peak components near 35° presented in the right panel. These θ–2θ XRD data reveal preferred grain orientations, inferred from the alignment of plane-normal vectors along the out-of-plane direction. On thin r-ATO seed layers (2.5 and 4.5 nm), HfO₂ films display a t(110) peak at approximately 35.2°, indicating that the normal vector of the t(110) plane was oriented nearly along the out-of-plane direction. Combined with the GIXRD data for the same samples, which also show the t(110) peak, it can be inferred that the normal vector of the t(110) plane in the HfO₂ grains is aligned within ∼17.6° of the surface normal. Conversely, samples without r-ATO or with a 10 nm-thick r-ATO seed layer exhibit a m(002)/m(200) peak at around 35.5°, rather than a t(110) peak. This variation reflects diminished local-epitaxy between the r-ATO and t-HfO₂ layers and enhanced m-phase development, consistent with Figure . Meanwhile, a (200) XRD peak from the r-ATO film at approximately 38.5° is observed for the 10 nm-thick r-ATO seed layer case, where the r-ATO film was sufficiently thick to generate detectable diffracted beam intensity. This indicates that the normal vector of the (200) plane in r-ATO grains was also preferentially oriented along the out-of-plane direction. Although the preferred orientations of thin and thick r-ATO films may differ, the presence of <200>-preferred orientation of the normal vector along the out-of-plane direction was also supported for the thin r-ATO case by the TEM image in Figure (c). Taken together, these results suggest that the local-epitaxy between r-ATO and t-HfO₂ was primarily mediated by alignment between the r-ATO (200) and t-HfO₂ (110) planes.

5.

(a) θ–2θ XRD patterns of ∼8 nm-thick HfO₂ films depending on the r-ATO seed layer thickness (left panel), with deconvoluted peak components near 35° (right panel). (b) Cation configurations and areal densities of matching candidates planes for HfO₂ and ATO. (c) Schematic matching relationships between r-ATO and t-HfO₂ films, consisting of (i) t-HfO₂ (110)/r-ATO (200), and (ii) t-HfO₂ (002)/r-ATO (101) planes, respectively. (d) System energies of a single HfO₂ (002) slab (t- or m-phase) on coherent or incoherent r-ATO (101) planes.

Figure (b) shows the cation configurations and areal densities of candidate planes for HfO₂ and ATO. Since the ATO and HfO₂ films in this study have inherently different crystallographic structures (i.e., rutile for ATO and tetragonal or monoclinic for HfO₂, respectively), the cation configurations of each crystallographic plane at the interface must be carefully considered to evaluate possible matching relationships. Although the θ–2θ XRD results (Figure (a)) suggested a likely alignment between the r-ATO (200) and t-HfO₂ (110) planes, a direct plane-to-plane match was excluded due to their substantially different cation configurations. Instead, domain-matching epitaxy based on the in-plane periodicities of r-ATO (200) and t-HfO₂ (110) planes can be considered. ,, In addition, alternative matching relationships were evaluated between planes perpendicular to r-ATO (200) and t-HfO₂ (110) planes. Because the normal vectors of the r-ATO (200) and t-HfO₂ (110) planes were slightly tilted from the surface normal direction, as inferred from experimental results, planes perpendicular to these orientations can also contribute to local-epitaxy. Accordingly, r-ATO (011) (or (101)), (020) (or (200)), and (002) planes, together with the t-HfO₂ (002) and (1̅10) (or (110)) planes, were considered as additional candidates. Among these candidates, the r-ATO (101) and t-HfO₂ (002) planes were selected for estimating quantitative thermodynamic tendencies based on their favorable cation configurations and matching areal densities Figure (c) shows schematic matching relationships between r-ATO and t-HfO₂ films, consisting of (i) t-HfO₂ (110)/r-ATO (200), and (ii) t-HfO₂ (002)/r-ATO (101) planes, respectively. For case (i), five r-ATO (200) unit planes and three t-HfO₂ (110) unit planes can form a domain-matching relationship. The estimated lattice mismatches are 8.6% for δ([1̅10])_t and 3.9% for δ([001])_t, where δ([1̅10])_t and δ([001])_t denote lattice mismatches along the [1̅10] and [001] directions of the t-HfO₂, respectively. For case (ii), the estimated lattice mismatches between these planes are 8.6% and −8.6% for δ([1̅10])_t and δ([110])_t, respectively. Although case (ii) shows a larger lattice mismatch than case (i), both epitaxial scenarios can be plausible in a polycrystalline thin-film system. Case (i) involves domain-matching epitaxy that benefits from longer-range interfacial registry, whereas case (ii) represents one-to-one matching that can be accommodated through short-range registry. Accordingly, both (i) and (ii) may contribute to the observed local-epitaxial relationship.

For the quantitative analysis of phase stability between t- and m-phases of HfO₂ arising from the local-epitaxy, the system energies of the HfO₂ layer stacked on either coherent or incoherent r-ATO (101) planes were estimated using the following equations

$${\mathit{E}}_{\mathrm{grain},\mathrm{coherent}{\mathrm{HfO}}_2}={\mathit{E}}_{\mathrm{bulk},\mathrm{strained}{\mathrm{HfO}}_2}+{\mathit{E}}_{\mathrm{vib}}+\frac{{(\mathit{\gamma }}_{\mathrm{GB},x}{\mathit{A}}_x+{\mathit{\gamma }}_{\mathrm{GB},y}{\mathit{A}}_y+{\mathit{\gamma }}_{\mathrm{GB},z}{\mathit{A}}_z)}{\mathrm{number}\mathrm{of}{\mathrm{HfO}}_2\mathrm{layers}}$$ 4

$${\mathit{E}}_{\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n},\mathrm{i}\mathrm{n}\mathrm{c}\mathrm{o}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e}\mathrm{n}\mathrm{t}{\mathrm{H}\mathrm{f}\mathrm{O}}_2}={\mathit{E}}_{\mathrm{b}\mathrm{u}\mathrm{l}\mathrm{k},\mathrm{r}\mathrm{e}\mathrm{l}\mathrm{a}\mathrm{x}\mathrm{e}\mathrm{d}{\mathrm{H}\mathrm{f}\mathrm{O}}_2}+{\mathit{E}}_{\mathrm{v}\mathrm{i}\mathrm{b}}+\frac{({\mathit{\gamma }}_{\mathrm{G}\mathrm{B},x}{\mathit{A}}_x+{\mathit{\gamma }}_{\mathrm{G}\mathrm{B},y}{\mathit{A}}_y+{\mathit{\gamma }}_{\mathrm{G}\mathrm{B},z}{\mathit{A}}_z)}{\mathrm{n}\mathrm{u}\mathrm{m}\mathrm{b}\mathrm{e}\mathrm{r}\mathrm{o}\mathrm{f}{\mathrm{H}\mathrm{f}\mathrm{O}}_2\mathrm{l}\mathrm{a}\mathrm{y}\mathrm{e}\mathrm{r}\mathrm{s}}$$ 5

where E _vib represents the electronic and vibrational free energy contribution. , As in eq , the γ _GB were calculated using the surface energies for each plane, which are included in Table . The surface energies of coherent and incoherent cases correspond to the values calculated without and with strain, respectively. In addition, the m-HfO₂ (002) plane was included in the analysis due to its suitable areal density and cation configuration among the m-phase planes, enabling evaluation of phase stability and comparison of the energetic favorability between the t- and m-phases of HfO₂. Unlike the r-ATO/r-RuO₂ interface system, where direct one-to-one matching between identical rutile structures allows a thickness-dependent energetic analysis, the local-epitaxy between r-ATO and HfO₂ involves heterogeneous crystal structures, where both direct and domain-matching relationships can coexist. Because the relative contributions of these two matching mechanisms cannot be readily separated as a function of thickness, thickness-dependent evolution was not explicitly considered here. Instead, a single HfO₂ slab model was employed to evaluate the intrinsic phase-stabilization tendency arising from local-epitaxy. Figure (d) shows the system energies of a single HfO₂ (002) slab (t- or m-phase) on coherent or incoherent r-ATO (101) planes. Among the configurations considered, the t-HfO₂ (002) slab on the coherent r-ATO (101) plane shows the lowest system energy (2.34 eV/f.u.), indicating that t-phase stabilization of HfO₂ was further favored on the coherent r-ATO surface, potentially enabling retention of the t-phase even at increased HfO₂ thicknesses. These computational findings are consistent with experimental observations, in which t-phase formation was promoted, and m-phase evolution of HfO₂ was suppressed on coherent r-ATO seed layers. Overall, these results reveal the phase-evolution tendency of HfO₂ on r-ATO under varying interfacial coherency, and further suggest that a multilayer stack composed of t-HfO₂ and r-ATO films with carefully optimized thicknesses can offer superior electrical performance by promoting the t-phase, suppressing the m-phase in HfO₂, and minimizing interfacial mismatches through well-aligned crystallographic planes.

2. Energetics for Calculation in the Local-Epitaxial Growth of HfO₂ on r-ATO (101) Plane.

	surface E [J/m2]
HfO2 plane	coherent	incoherent
t(110)	1.74	1.75
t(002)	1.94	1.97
m(200)	2.63	2.26
m(020)	2.24	2.39
m(002)	1.86	1.90

3.4. Electrical Characteristics of t-HfO₂/r-ATO Bilayer-Structured Capacitor

Using the HfO₂/ATO bilayer, a MIM capacitor with a Pt TE/HfO₂/ATO/Ru BE structure was fabricated, and its electrical properties were evaluated. For comparison, MIM capacitors with HfO₂ single-layer and ATO single-layer, both with identical TE and BE, were also fabricated. Figure (a) displays EOT versus total POT plots for HfO₂ single-layer, ATO single-layer, and HfO₂ (2.1 nm)/ATO bilayer capacitors. The HfO₂ single-layer showed a dielectric constant of about 37 (k ₁) at relatively low thicknesses (<∼8 nm), corresponding to the t-phase. This value decreased to approximately 12 (k ₂) in thicker films, reflecting a transition to the m-phase. Its EOT_i (y-intercept ∼0.5 nm) was relatively high, indicating interfacial degradation at the HfO₂/Ru BE and Pt TE/HfO₂ interfaces. Considering that the EOT_i values for other oxide stacks (ATO and HfO₂/ATO) were substantially lower (∼0.1 nm) despite similar interfacial degradation from sputtered Pt TE, it suggests that local-epitaxy between t-HfO₂ and in situ-formed r-RuO₂ was not effectively established in the HfO₂ single-layer capacitor, unlike in the t-HfO/r-ATO case. Conversely, the ATO single-layer exhibited a high dielectric constant of ∼77, consistent with the rutile structure. Its EOT_i (∼0.1 nm) was much lower, indicating improved interfacial quality due to effective local-epitaxy between r-ATO and in situ-formed r-RuO₂. For the HfO₂/ATO bilayer, the HfO₂ capping layer thickness was fixed at 2.1 nm, so the slope in the plot primarily reflects the dielectric constant of the underlying r-ATO layer. The r-ATO layer in the bilayer capacitor exhibited a dielectric constant of about 42 (k ₁) up to a ∼4.4 nm (corresponding to a total POT of ∼6.5 nm), which was notably lower than that of the ATO single-layer (k ∼77). This decreased dielectric constant can be attributed to the diffusion of Hf into the r-ATO matrix, which can distort the rutile lattice of ATO due to the larger ionic radius of Hf. Such performance degradation due to structural disturbance has been reported in previous studies involving Ti-doped HfO₂ and HfO₂–TiO₂ mixed films. , At higher r-ATO thicknesses, the dielectric constant increased to around 69 (k ₂), approaching the value of the ATO single-layer case. It was likely due to the decreased influence of Hf diffusion within the thicker r-ATO film. Due to the effective local-epitaxy at both r-ATO/Ru and t-HfO₂/r-ATO interfaces, the EOT_i (∼0.1 nm) contribution in the HfO₂/ATO bilayer remained low despite multiple interfaces within the capacitor. Given that sputtering-induced damage from the Pt TE was expected to be comparable across all capacitor stacks, the substantially lower (vs HfO₂ single-layer case) and comparable (vs ATO single-layer case) EOT_i further indicates that the interfacial degraded layer at the t-HfO₂/r-ATO interface was minimal in the thickness range.

Furthermore, excellent leakage current suppression was achieved by introducing the HfO₂ capping layer. Figure (b) shows leakage current density (J) versus applied voltage (V) plots for HfO₂/r-ATO capacitors with different HfO₂ capping layer thicknesses, while the r-ATO thickness remained fixed at 3.8 nm. As the HfO₂ capping layer thickness increased, the leakage current decreased significantly due to the high bandgap energy of HfO₂ (5–6 eV). Additionally, forming a well-defined t-HfO₂/r-ATO interface, rather than an interfacial low-k layer caused by a solid-solution state, further decreased leakage current by minimizing interface degradation. The inset of Figure (b) shows J–V plots of the same samples measured up to an applied voltage of 5 V to observe hard breakdown characteristics. Hard breakdown was observed at 2.8 V (5.5 MV/cm), 3.5 V (5.9 MV/cm), and 4.4 V (6.7 MV/cm) for HfO₂ capping layer thicknesses of 1.3, 2.1, and 2.7 nm, respectively. These results indicate that the proposed dielectric stack has sufficient voltage tolerance well above the typical DRAM operating voltage range (∼1.0–1.2 V). , This improvement was also aided by trap density curing in the ATO layer during the oxidative HfO₂ ALD process.

Finally, the electrical performance of the HfO₂/ATO bilayer oxide was compared with other promising dielectric oxides. Figure (c) shows the J-EOT plots for ATO single-layer, ZAZ layer, and HfO₂/ATO bilayer capacitors, where ZAZ data were reproduced from a previous study. Although the ATO single-layer achieved the lowest minimum EOT (0.52 nm) while maintaining sufficiently low J levels (<10^–7 A/cm² at 0.8 V), its relatively large POT (9.5 nm) made it less suitable for current DRAM design rules. This limitation was due to the low bandgap energy (∼3.1 eV) of r-ATO, despite its high dielectric constant (∼77). For the ZAZ layer, which is currently used in industry, the minimum POT achieved was thinner (5.2 nm), enabled by the high bandgap energy (5–6 eV) of the ZrO₂ matrix and the leakage current suppression effect of the Al₂O₃ interlayer. However, its minimum EOT (0.80 nm) remained relatively higher than those of other dielectrics due to the intrinsically low dielectric constant (30–40) of t-ZrO₂ and further degradation caused by Al-doping. In contrast, the HfO₂/r-ATO bilayer capacitors with a 2.1 nm-thick HfO₂ capping layer achieved both a low minimum EOT (0.62 nm) and a comparably low POT (5.9 nm), while maintaining a low J level. This was possible through the synergistic combination of the relatively high dielectric constants of r-ATO (∼42) and t-HfO₂ (∼36), along with leakage current suppression by Al-doping and the high bandgap energy of HfO₂. Moreover, the improved electrical characteristics were enabled by establishing a thickness window in which interfacial coherency between r-ATO and t-HfO₂ was preserved, thereby minimizing adverse interfacial-layer effects and allowing the complementary advantages of the two layers to be fully realized. When a thinner HfO₂ capping layer (1.3 nm) was employed, a thicker r-ATO layer was required to satisfy the leakage current criterion. The resulting incoherence in the thicker r-ATO film disrupted local-epitaxy at the t-HfO₂/r-ATO interface and promoted the formation of an adverse interfacial layer, thereby limiting further EOT improvement. In addition, the protrusions on the surface of the HfO₂ capping layer on the incoherently grown r-ATO, as shown in Figure (a), can induce local electric-field concentration under bias, which further degraded the leakage current characteristics for this thickness combination. Consequently, a relatively large total POT (∼7.3 nm; ∼6 nm-thick r-ATO with a 1.3 nm-thick HfO₂ capping layer) was required to ensure sufficiently low leakage current. In contrast, with a thicker HfO₂ capping layer (2.7 nm), a thinner r-ATO layer could be allowed while still satisfying the leakage current criterion (total POT of ∼5.8 nm; ∼3.1 nm-thick r-ATO with a 2.7 nm-thick HfO₂ capping layer). However, excessively thin r-ATO layers could not function as effective insulating layers due to increased defect density. Thus, the r-ATO thickness and, consequently, the total POT cannot be drastically reduced in this configuration. Moreover, the dielectric constant of such ultrathin r-ATO would be significantly degraded due to the increased Hf diffusion impact into the r-ATO matrix, thereby limiting further EOT improvement. Overall, these results suggest that the t-HfO₂/r-ATO bilayer capacitors designed with a suitable thickness window that preserves interfacial coherency can effectively harness the complementary advantages of t-HfO₂ and r-ATO while minimizing common drawbacks of multilayer integration, such as interfacial low-k layer formation. This is achieved through efficient local-epitaxial relationships at both the t-HfO₂/r-ATO and r-ATO/Ru BE interfaces. These findings therefore provide a mechanistic basis for designing ultrathin stacked dielectric films with diminished interfacial degradation under a DRAM-relevant scaling region.

Conclusion

The sequential crystalline behaviors of the ATO seed and HfO₂ capping layers on Ru substrates were examined, focusing on the heterogeneous local-epitaxial relationships among r-RuO₂, r-ATO, and t-HfO₂ within the HfO₂/ATO bilayer structure. The r-ATO layer exhibited a two-stage growth behavior: an initial coherent regime with aligned grain orientations and a strained lattice continuous with the underlying r-RuO₂ template, followed by an incoherent regime in which strain relaxation led to more randomly oriented grains. This coherence transition in r-ATO critically governed the interfacial integrity of the t-HfO₂/r-ATO interface and the electrical performance of the capacitor.

Within a specific r-ATO thickness window (approximately 3–4.5 nm), coherent r-ATO enabled robust local-epitaxy with the HfO₂ capping layer, forming a sharp, well-matched t-HfO₂/r-ATO interface with diminished adverse interfacial-layer contributions and suppressed interdiffusion. When r-ATO was excessively thin, its insulating property became insufficient, and Hf incorporation/diffusion into the r-ATO matrix became more pronounced, degrading the dielectric response of r-ATO and limiting further EOT improvement. Conversely, when r-ATO exceeded the coherency threshold (>4–4.5 nm), incoherent r-ATO weakened local-epitaxy with t-HfO₂, leading to inferior interfacial registry and increased disorder/intermixing, thereby enhancing adverse interfacial-layer effects and degrading device performance.

Overall, these competing constraints defined a coherency-preserving thickness window in which heterogeneous local-epitaxial relationships were continuously maintained across both r-ATO/r-RuO₂ and t-HfO₂/r-ATO interfaces, thereby minimizing interfacial penalties while maximizing the functional benefits of the bilayer. As a result, the optimized t-HfO₂/r-ATO stack demonstrated excellent electrical performance, achieving an EOT of 0.62 nm at a POT of 5.9 nm while maintaining low leakage current. These findings provide a structural basis for the interfacial and crystallographic engineering of ultrathin, stacked dielectric films in a DRAM-relevant scaling region and offer useful design insights for dielectric stacks for next-generation DRAM capacitors.

The manuscript was written through the contributions of all authors. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.