Ferroelectricity in (Hf,Zr)O₂ thin films has attracted substantial academic and industrial interest since its first report in 2011^[1], driven by its potential for next-generation high-density semiconductor memories and emerging computing hardware. Hf_0.5Zr_0.5O₂ (HZO) is highly compatible with complementary metal-oxide-semiconductor (CMOS) processing, and atomic layer deposition (ALD) enables uniform ultrathin films in which remanent polarization (P_r) can exceed ~ 15 μC∙cm^-2 even at thickness below 5 nm^[2-5]. Together with switching endurance that has been improved toward and beyond 10¹² cycles, Hf_0.5Zr_0.5O₂ has become a leading candidate for ferroelectric random-access memory (FeRAM), ferroelectric field-effect transistors (FeFETs), and ferroelectric tunnel junctions (FTJs)^[6-9]. In particular, FeRAM and FeFET concepts are being actively pursued as promising future technologies for advanced DRAM and 3D NAND^[3-5,9-13].

A key integration challenge is that ferroelectricity in vacuum-processed Hf_0.5Zr_0.5O₂ films is strongly governed by the bottom electrode^[14-16]. The electrode crystal structure, work function, and microstructure, together with its chemical properties that control interfacial redox reactions, have all been reported to affect Hf_0.5Zr_0.5O₂ phase stability, oxygen-vacancy (V_o) distributions, wake-up behavior, and endurance.

Mo has emerged as a next-generation metal for advanced interconnects as dimensional scaling proceeds and the shortened electron mean free path penalizes incumbent low-resistivity conductors such as Cu and W in ultrathin regimes. Notably, W word lines in 3D NAND flash are already being replaced by Mo^[17-19]. This transition is motivated by the lower resistivity of Mo in ultrathin films, the potential to reduce or avoid barrier-layer use and thereby lower line resistance, and the feasibility of conformal film formation in three-dimensional nanostructures by ALD. Consequently, Mo and Mo-based compounds, including MoO_x and molybdenum nitrides (MoN_x), are increasingly considered not only as line/interconnect materials but also as electrode candidates for advanced device stacks^[20-23].

Mo is also attractive as an electrode for ferroelectric (Hf,Zr)O₂^[24-26]. Its relatively high work function (~ 4.7-4.9 eV^[27,28]) is sufficient to suppress leakage current. In addition, prior studies have suggested that Mo does not strongly scavenge O from (Hf,Zr)O₂; rather, because Mo-O bonding is weaker than Hf-O or Zr-O bonding, Mo can supply O to (Hf,Zr)O₂ and decrease V_o concentration. This behavior can suppress interfacial tetragonal-phase formation driven by high V_o concentrations and significantly mitigate wake-up. However, during ALD of (Hf,Zr)O₂, strong oxidants such as ozone can oxidize Mo and form a defective MoO_x interlayer, degrading leakage characteristics and switching endurance.

Mo-based compound electrodes, such as MoO_x and MoN_x, are therefore promising alternatives for addressing the limitations of elemental Mo. Since these compounds remain Mo-based, oxygen scavenging and the subsequent V_o formation in (Hf,Zr)O₂ are expected to be minimal. Furthermore, the incorporated O or N may improve resistance to further surface oxidation during exposure to ALD oxygen sources. Among Mo oxides, MoO₂ is potentially advantageous because of its high work function and expected favorable interfacial compatibility with HZO^[24,29]. However, additional oxidation can yield MoO₃; because MoO₃ is a wide-bandgap semiconductor^[30], its formation can substantially degrade electrode functionality.

MoN_x compounds offer a particularly compelling route because they can preserve the weak oxygen-scavenging characteristics of Mo while mitigating ALD-induced surface oxidation. Importantly, MoN_x provides a broad stoichiometric window with closely related crystal structures. As illustrated in Figure 1A, B1-MoN (Mo:N = 1:1) adopts a rock-salt-type ($$ Fm\bar{3}m $$) structure with a lattice parameter of ~ 0.436 nm and metallic character. γ-Mo₂N (Mo:N = 2:1) has a closely related structure in which only ~ 50% of the anion sites are occupied, with a lattice parameter of ~ 0.416 nm. Thus, within Mo:N ratios from ~ 2:1 to 1:1, MoN_x can remain metallic while enabling lattice-parameter tuning from ~ 0.416 to ~ 0.436 nm. Reported work functions for γ-Mo₂N and B1-MoN span 4.4-5.3 eV^[31-36], suggesting that electrode energetics may also be tunable via stoichiometry.

In this study, we propose MoN_x as a compound-electrode material platform that enables both (i) interfacial engineering through oxidation resistance and redox/transport control and (ii) potential texture engineering through stoichiometry-dependent lattice-parameter tuning. By sputtering MoN_x electrodes with different Mo:N ratios and integrating them as bottom electrodes in ferroelectric HZO capacitors, we demonstrate that the defective interfacial layer (IL) formed during ALD can be substantially reduced. Suppressed oxygen scavenging and partial O/N supply effects reduce non-ferroelectric monoclinic-phase formation, lower the coercive field (E_c), mitigate wake-up, improve endurance, and measurably alter the HZO microstructure. The MoN_x bottom-electrode system introduced here provides immediate performance gains and a scalable pathway for further optimization through compositional and microstructural engineering.

As shown in the top panel of Figure 1A, elemental Mo adopts a body-centered cubic structure, whereas N incorporation stabilizes rock-salt-derived MoN_x phases. In this family, B1-MoN corresponds to full anion-site occupancy, while γ-Mo₂N can be viewed as an anion-vacancy derivative with partial anion-site occupancy. In this work, we systematically control the Mo:N ratio in sputtered MoN_x electrodes and investigate how electrode stoichiometry governs the interfacial microstructure and electrical performance of ferroelectric HZO capacitors.

This approach addresses limitations intrinsic to MoO_x electrodes. As the O content increases in MoO_x, MoO₃ can form. Because MoO₃ is semiconducting, it may introduce parasitic capacitance, generating depolarization fields, and/or degrade carrier-injection barriers at the Mo/HZO interface, thereby increasing leakage. In contrast, MoN_x generally maintains metallic behavior over a broad composition window, enabling stable use as a conductive electrode while providing an additional compositional knob for interfacial chemistry control.

Figure 1B shows Bragg-Brentano XRD patterns of four MoN_x films deposited by Ar/N₂ sputtering with controlled N₂/(Ar + N₂) flow ratios of 0%, 5%, 20%, and 50%. The corresponding compositions are x = 0.00, 0.05, 0.52, and 0.79, denoted as Mo, 05MoN, 52MoN, and 79MoN, respectively. The composition-extraction procedure is provided in Supplementary Figure 1 and Supplementary Table 1. Mo exhibits strong (110) preferred orientation. The low-N film (05MoN) shows very broad diffraction features, consistent with reduced crystallite size and/or mixed-phase character. Because both Mo and γ-Mo₂N are metallic, 05MoN is still expected to operate as an electrode. In contrast, 52MoN and 79MoN exhibit well-developed cubic rock-salt-type diffraction peaks with pronounced (111) preferred orientation, suggesting increased crystallographic compatibility with low-index HZO planes^[38] [Figure 1C].

Figure 1D and E summarize resistivity and work function, respectively. Resistivity is lowest for elemental Mo and generally increases with N content. Importantly, even 52MoN, which shows the highest resistivity among the measured films, remains below ~ 200 μΩ∙cm, which is sufficiently low for electrode operation. The work function increases from 4.70 eV for Mo to 4.94 eV for 52MoN and then decreases to 4.76 eV for 79MoN. Although the bulk composition evolves monotonically toward the N-rich rock-salt MoN_x regime, the UPS-derived work function does not follow a simple linear trend with the average stoichiometry. Instead, the measured value is highly sensitive to the near-surface electronic structure. The reduced work function of 79MoN suggests that in the highly N-rich regime, surface-sensitive factors such as surface termination, local phase constitution, nitrogen-vacancy population, and residual surface oxidation play dominant roles in determining the effective work function. The UPS data are provided in Supplementary Figure 2.

Electrode microstructure was further examined by planar SEM and AFM. Grain-size distributions extracted from SEM images using Gwyddion [Supplementary Figure 3] yield average lateral grain radii of ~ 3.9-4.7 nm assuming cylindrical grains, with a modest increase as N content increases. AFM analysis confirms well-optimized sputtering conditions, showing that all electrodes exhibit smooth surfaces with root-mean-square roughness (R_q) values below 0.55 nm [Supplementary Figure 4].

After establishing the electrode series, symmetric MoN_x/HZO/MoN_x capacitors were fabricated by depositing 8 nm HZO by ALD, followed by sputtered top electrodes with the same MoN_x composition. To assess how electrode stoichiometry affects IL formation during ALD and subsequent crystallization annealing, cross-sectional STEM-EDS and STEM-EELS analyses were performed.

Figure 2A and B present cross-sectional high-magnification HRTEM images of the Mo/HZO/Mo and 52MoN/HZO/52MoN capacitor stacks, respectively. To quantitatively analyze the IL formed at the bottom interface, depth-dependent HRTEM intensity profiles were extracted from the indicated regions. The IL thickness was determined by calculating the FWHM from a Gaussian fit of the bright IL intensity peak located between the electrode and the HZO bulk. As supported by the statistical analysis [Figure 2C], the pure Mo stack exhibits a prominent IL with an average thickness of 2.20 ± 0.21 nm. In sharp contrast, the 52MoN stack in Figure 2B shows a significantly suppressed IL with an average thickness of only 1.15 ± 0.07 nm. This corresponds to a 47.7% reduction in IL thickness for 52MoN relative to the Mo electrode.

Figure 2D and E present STEM-EDS elemental maps for these stacks. In both stacks, an IL is observed at the bottom electrode/HZO interface, where oxygen signals overlap with electrode elements. The observed difference in thickness is consistent with the higher oxidation resistance of N-containing MoN_x under ozone exposure during ALD. In addition, N is detected within the HZO film in Figure 2E, indicating that N can incorporate into HZO, most plausibly during the post-deposition anneal given diffusion kinetics. Thermodynamically, the strong stability of nitrides such as HfN and ZrN relative to Mo₂N (formation free energies at 700 K: -638.8 and -595.6 kJ∙mol^-1 for HfN and ZrN, respectively, vs. -52.4 kJ∙mol^-1 for Mo₂N) provides a driving force for N transfer into the HZO interfacial region^[39,40].

The EELS line-scan positions across the Mo/HZO/Mo stack are indicated in Figure 2F, and the corresponding O-K edge spectra are shown in Figure 2G. Similarly, the scan positions for the 52MoN/HZO/52MoN stack are presented in Figure 2H, with the corresponding N-K and O-K edge spectra displayed in Figure 2I and J, respectively. The line-scan colors match the indicated positions. Within HZO, the O-K edge exhibits the expected double-peak features in the 532-533 eV and 536-537 eV ranges [Figure 2G and J]. Ideally, if the electrodes remain unoxidized, no O-K peak would be observed in the electrode region. However, in Mo/HZO/Mo (green circles in Figure 2G), both the top and bottom Mo electrodes exhibit an O-K feature near ~ 532 eV, indicating electrode oxidation and the presence of a MoO_x IL^[41]. Stoichiometric MoO₃ typically exhibits a clearly separated doublet, whereas oxygen-deficient MoO_x tends to show less distinct splitting. Therefore, the observed interfacial oxide is more consistent with oxygen-deficient MoO_x than with fully stoichiometric MoO₃. This structurally defective and relatively less dense oxygen-deficient MoO_x layer is consistent with the bright contrast observed in the HRTEM images, because it permits higher electron transmission than the adjacent dense Mo and HZO layers. Notably, the bottom interface shows O-K peaks in the three spectra closest to the interface (green circles in Figure 2G). If the line-scan spacing is ~ 2 nm, this corresponds to an oxide thickness of approximately 4-6 nm. This chemical oxygen-penetration depth is larger than the structural IL thickness (~ 2.20 nm) observed by HRTEM. While HRTEM visualizes the severely amorphized, structurally defective region through contrast variation, the EELS captures deeper oxygen diffusion that forms an oxygen-deficient MoO_x layer without completely collapsing the underlying Mo crystalline lattice. In contrast, the top Mo electrode shows an O-K peak in only one spectrum, implying weaker oxidation. This asymmetry indicates that Mo oxidation likely occurs predominantly during the highly reactive ozone pulses used in the initial ALD growth of HZO on the bottom electrode at 280 °C, rather than during the subsequent RTP step, consistent with earlier observations in Mo-based HZO stacks.

A markedly different behavior is observed for 52MoN/HZO/52MoN. In Figure 2I, a strong N-K peak appears near ~ 398 eV within the 52MoN electrode region, consistent with typical transition-metal nitride signatures (~ 397-400 eV)^[42]. Importantly, a weak N-K signal is also detected within HZO near the bottom interface (purple circle in Figure 2I), directly evidencing partial N incorporation into the HZO interfacial region. Moreover, the O-K line-scan analysis [Figure 2J] shows that electrode oxidation is strongly suppressed: only the spectra immediately adjacent to the interfaces exhibit an O-K peak, indicating that the oxidized region is confined to a much thinner interfacial zone than in Mo/HZO/Mo. Together, these results experimentally confirm that higher N content in MoN_x significantly mitigates ALD-induced bottom-electrode oxidation and reduces defective IL formation.

To elucidate the origins of the electrical differences and isolate bottom-interface reactions, interfacial chemistry and O transport were examined using ToF-SIMS and XPS [Figure 3A and B]. Simplified stacks consisting of 8 or 2.5 nm HZO deposited on Mo or MoN_x were fabricated to focus on the bottom interface.

The ToF-SIMS depth profiles reveal a clear contrast between the Mo and MoN_x series. In these profiles, the shaded interfacial boundaries are schematically highlighted as a visual guide to indicate the chemical transition zone, encompassing the broad peak distribution of the HfO₂^- secondary ion signal, which acts as a chemical marker due to the matrix effect^[44]. In the Mo reference sample, O penetrates deeply into the electrode during HZO growth and subsequent annealing, forming a relatively thick oxide interfacial region and an extended MoO_x zone associated with strong interfacial damage (red circle in Figure 3A). Notably, the low-N 05MoN electrode exhibits an even thicker oxide IL despite N incorporation. This result indicates that introducing only a small amount of N can degrade Mo crystallinity and facilitate deep O penetration. At higher N contents, the behavior changes qualitatively: 52MoN and 79MoN show pronounced N accumulation near the interface and within the HZO side (green circles in Figure 3A), suggesting that a nitride-rich interfacial region functions as a barrier against excessive oxidation.

These coupled O/N transport processes are further supported by XPS [Figure 3B]. Upon annealing, the Mo 3d spectra show the reduction of MoO₃ toward MoO₂ (red arrows in Figure 3B) and a concurrent shift of the Mo-N component toward a higher binding energy. These results imply that MoO₃ can act as an oxygen reservoir supplying O to both HZO and the underlying electrode, while N in MoN_x diffuses into the interfacial region of HZO and modifies local bonding configurations. The observed chemical-state evolution is consistent with the Gibbs free-energy hierarchy shown in Figure 3C and D. Figure 3E summarizes the maximum secondary-ion intensities of MoO^-, MoO₂^-, and HfN⁻ fragments extracted from the ToF-SIMS profiles. Collectively, these data indicate that N incorporation is established predominantly during HZO ALD and/or early thermal steps, whereas annealing primarily redistributes V_os rather than contributing to substantial additional N diffusion.

Figure 4A shows GIXRD patterns of 8 nm HZO films crystallized on Mo, 05MoN, 52MoN, and 79MoN electrodes. For these measurements, MoN_x/HZO/MoN_x stacks were annealed, and the top electrode was removed by wet etching to avoid top-electrode contributions. Electrode diffraction peaks such as 110_Mo and 111_MoN are observed across samples, consistent with the Bragg-Brentano XRD patterns in Figure 1C.

HZO diffraction peaks are denoted as hkl_x, where x indicates the orthorhombic (O, Pca2₁), tetragonal (T, P4₂/nmc), or monoclinic (M, P2₁/c) phase. In all samples, a dominant peak near 30.5° 2θ is observed, corresponding to overlapping 111_O/101_T reflections. However, HZO films on Mo and 05MoN show clear monoclinic peaks at ~ 28.5° and ~ 31.7° (-111_M and 111_M). Peak deconvolution of the 27°-33° region [Supplementary Figure 6] enables estimation of the M phase fraction [Figure 4B]. The M phase fraction decreases systematically with N content: Mo yields the highest M phase fraction (~ 21%), whereas 52MoN and 79MoN reduce the M phase fraction to < 5%.

Quantitative separation of the T phase and O phase fractions by conventional XRD is challenging because these structures are similar and their peaks overlap. Nevertheless, because the T phase peak is typically located near 30.8°^[45,46] and the O phase peak near 30.4°^{[38, 47]}, the position of the combined 111_O/101_T peak can provide insight into their relative contributions. As N content increases, the peak position shifts from ~ 30.55° for Mo to ~ 30.45° for 79MoN [Figure 4C]. Although this shift toward lower angles suggests a tendency to favor orthorhombic contribution, it may also be influenced by residual strain arising from lattice expansion in the underlying MoN_x electrodes. Therefore, rather than indicating a large increase in the orthorhombic phase alone, these results suggest that higher-N MoN_x electrodes effectively suppress the non-ferroelectric monoclinic phase and stabilize an overall ferroelectric-compatible structural configuration. This interpretation is consistent with the stable double remanent polarization (2P_r) behavior observed in Figure 5 and the mitigated wake-up effect observed in the electrical measurements.

Because the HZO 002_O reflection overlaps with the diffraction signals from the MoN_x electrodes [Supplementary Figure 7], the texture analysis focused on the 111_O/101_T region. Figure 4D shows azimuthal cuts of the GIWAXS intensity around this peak, as indicated by the red dashed arcs in Supplementary Figure 7. An azimuthal angle of 90° corresponds to the substrate normal, whereas 0° and 180° correspond to in-plane directions. Although GIWAXS has limited access to the exact substrate-normal region, the 90° intensity was reconstructed by extrapolation, consistent with prior approaches. Overall, strong preferential orientation is not observed for HZO on any MoN_x electrode; if anything, the 111 fiber texture slightly decreases with increasing N content. TEM diffraction analyses [Supplementary Figure 8] likewise do not show strong texture.

Figure 4E summarizes the lateral grain-size analysis for HZO on MoN_x, extracted from planar SEM images using the watershed method in the Gwyddion software [Supplementary Figure 9]. The mean lateral grain radius (assuming cylindrical grains) shows a modest increase with N content but remains within ~ 3.8-4.4 nm. AFM topography of HZO [Supplementary Figure 10] reveals R_q below 0.65 nm for all films, indicating that MoN_x stoichiometry does not introduce significant roughening or morphological instability in the HZO layer.

Figure 5A-D show P-E hysteresis loops of Mo/HZO/Mo, 05MoN/HZO/05MoN, 52MoN/HZO/52MoN, and 79MoN/HZO/79MoN capacitors in the pristine state and after wake-up cycling. Wake-up cycling was performed at 3 MV∙cm^-1 and 100 kHz, for 10⁴ cycles. Extracted 2P_r and double coercive field (2E_c) values are summarized in Figure 5E and F, respectively; additional P-E loops measured at various fields are provided in Supplementary Figures 11 and 12. To ensure statistical reliability and account for device-to-device variation (DTDV), ferroelectric parameters were extracted from 10 randomly selected devices for each condition. As summarized in Figure 5E, the average pristine 2P_r values are 57.02 ± 0.41, 66.69 ± 0.80, 57.02 ± 0.83, and 53.23 ± 0.78 μC∙cm^-2 for Mo, 05MoN, 52MoN, and 79MoN, respectively. After wake-up cycling, the corresponding average values are 58.66 ± 0.51, 67.13 ± 1.19, 57.56 ± 0.89, and 54.10 ± 0.84 μC∙cm^-2. These statistical averages demonstrate that the relative 2P_r change upon wake-up remains exceptionally low (≤ 3.0%) across all optimized MoN_x stacks, confirming that the near-wake-up-free behavior is a robust intrinsic property rather than a measurement artifact.

A localized statistical anomaly is observed in the 2E_c of the pristine 05MoN stack [Figure 5F]. Although the 2P_r values remain statistically stable, pristine 05MoN capacitors exhibit a considerably larger standard deviation in 2E_c than the other conditions. This wide DTDV directly reflects the microstructural degradation observed in the chemical analyses [Figures 1 and 3]. The low-N 05MoN film has a mixed-phase character with degraded crystallinity, which facilitates severe and non-uniform oxygen penetration during ALD and forms a highly defective, thick MoO_x interfacial layer. Unlike 2P_r, which is primarily governed by the volume-averaged phase fraction, 2E_c is highly sensitive to localized pinning sites and interfacial defect distributions. The structurally non-uniform interface therefore introduces substantial local variation in the initial defect distribution, leading to a broad distribution of switching barriers across devices. After wake-up cycling, the electric-field-driven redistribution and stabilization of these interfacial defects effectively homogenize the switching paths, resulting in a noticeably narrower 2E_c variance. Moreover, the change in 2E_c is limited to ≤ 8.0% across all stacks [Figure 5F], indicating a stable switching barrier and a consistent ferroelectric response with minimal electrical conditioning. This near-wake-up-free behavior is attractive for ferroelectric memory applications because it reduces variability and eliminates the need for extensive post-fabrication field training.

Figure 6A-C show the evolution of 2P_r under bipolar cycling at 2.4, 2.0, and 1.6 V (100 kHz). Figure 6D-F summarize the evolution of 2E_c and imprint field [E_imp = (E_c⁺ + E_c^-)/2]. Endurance strongly depends on electrode N content. At 2.4 V, the highest field condition, Mo and low-N MoN_x (05MoN, corresponding to 5% N₂ flow) exhibit relatively early breakdown near ~ 10⁵ cycles, whereas higher-N electrodes (52MoN and 79MoN, corresponding to 20% and 50% N₂ flow) maintain stable switching up to ~ 10⁶ cycles [Figure 6A]. In this high-field regime, the increase in polarization during the initial 10⁴ cycles is only ~ 0.20-1.70 μC∙cm^-2 (≤ 3.1%) for all electrodes, consistent with strongly suppressed wake-up.

Reducing the cycling voltage to 2.0 V expands the endurance window by roughly two orders of magnitude [Figure 6B]: Mo and 05MoN fail near ~ 10⁶ cycles, 52MoN remains switchable to ~ 10⁷ cycles, and 79MoN sustains cycling to ~ 10⁸ cycles. Under this condition, Mo, 05MoN, and 52MoN retain more than ~ 68% of their initial polarization at their respective maximum cycle counts. In contrast, 79MoN shows a stronger reduction in apparent polarization to approximately one-third of its initial value by 10⁸ cycles, consistent with its higher coercive field and increasingly incomplete switching at 2.0 V. At 1.6 V, the applied voltage is below 2E_c for all capacitor stacks [Figure 6F]. Therefore, the extracted polarization corresponds to partial or minor-loop switching, and the decrease in 2P_r under this condition should be attributed mainly to incomplete polarization reversal rather than catastrophic dielectric breakdown. Nevertheless, the capacitors sustain very high cycle counts: Mo, 05MoN, and 52MoN remain functional to at least ~ 10⁸ cycles, and 79MoN endures up to ~ 10⁹ cycles. In this sense, the 1.6 V result represents a low-voltage survival window, whereas the 2.0 V data are more representative of practical switching endurance. Overall, 79MoN provides the longest endurance but at the cost of reduced apparent polarization under limited-voltage operation, whereas 52MoN maintains higher polarization with excellent cycling stability. Thus, 52MoN is the most favorable compromise between polarization and endurance for practical operation.

The central outcome of this work is that MoN_x stoichiometry provides a single-layer lever for controlling the interfacial microstructure that governs ferroelectric HZO performance. By systematically varying N content, we simultaneously tune (i) oxidation resistance and O permeability of the bottom electrode during ozone-based ALD and (ii) N availability for interfacial incorporation into HZO. This coupled chemical and microstructural control directly translates into phase stabilization, wake-up suppression, and improved cycling reliability.

STEM-EDS/EELS and ToF-SIMS consistently show that oxygen penetrates deeply into Mo during HZO ALD and annealing, forming a thick MoO_x region associated with interfacial damage. Increasing N content dramatically confines this oxidation to a thinner region, culminating in the high-N electrodes (52MoN and 79MoN), for which O-K edge signals in the electrode appear only in the spectra closest to the interface. This behavior indicates that N-rich MoN_x effectively acts as a redox/transport barrier under ALD conditions.

A particularly important and practically relevant observation is that low-N MoN_x (05MoN) can form an even thicker oxidized IL than pure Mo. This result implies that oxidation resistance cannot be inferred solely from the presence of N; rather, insufficient N may degrade crystallinity and open fast diffusion pathways, thereby facilitating O penetration. Thus, MoN_x does not follow a simple monotonic trend, and a threshold N content is required to form a stable, oxidation-resistant, nitride-rich interfacial microstructure.

In high-N stacks, a weak but measurable N-K signal appears within HZO near the bottom interface, and ToF-SIMS shows nitride-related fragments on the HZO side. These results support a scenario in which MoN_x supplies N to the interfacial HZO region during ALD and/or annealing. Because V_os in HZO are widely associated with deep traps that can facilitate Poole-Frenkel conduction and trap-assisted tunneling^[52], N incorporation provides a plausible pathway for V_o compensation and trap-state modification. Furthermore, steady-state leakage-current measurements [Supplementary Figure 13] reveal that the pristine macroscopic leakage levels remain comparable across the electrode variations. This indicates that the dramatic enhancement in cycling stability is not driven by a reduction in initial global leakage but rather by the suppression of localized leakage-current paths during prolonged electric-field cycling^[52,53]. Although direct trap spectroscopy is outside the present scope, the concurrent suppression of defective MoO_x growth, which typically acts as a vulnerable reservoir for defect clustering, together with evidence of N incorporation, provides a coherent microstructure-based explanation for the inhibition of defect-mediated percolation paths and the resulting reliability improvements observed in endurance tests.

GIXRD reveals that increasing N content strongly suppresses the monoclinic fraction, from ~ 21% on Mo to < 5% on 52MoN/79MoN, while the 30.5° peak position shifts in a manner consistent with an increased orthorhombic contribution. This structural evolution aligns with the electrical signature of near-wake-up-free switching (≤ 3.0% change in 2P_r after 10⁴ cycles). In conventional TiN-electrode HZO, wake-up is often attributed to field-driven redistribution of V_os and gradual conversion of non-ferroelectric regions into ferroelectric domains. Here, the small wake-up indicates that the as-fabricated stacks already possess a more favorable defect/phase configuration, consistent with reduced interfacial oxidation and modified O/N chemistry.

To further validate the advantages of the MoN_x system, we benchmarked its ferroelectric properties and reliability against other reported stoichiometry-controlled electrode materials, including TaN_x, RuO_x, and TiN_x [Figure 6G-I]. As shown in Figure 6G, MoN_x-based capacitors exhibit high pristine 2P_r values exceeding 47.5 μC∙cm⁻² across the entire investigated stoichiometric range, outperforming the comparative systems, which generally show lower polarization or stronger composition dependence. Figure 6H further highlights the wake-up immunity of our devices. While other material systems often show substantial wake-up, with up to a 76.8% increase in 2P_r after cycling, the MoN_x electrodes maintain negligible polarization changes regardless of N content. We acknowledge that some prior studies on HZO-based capacitors have reported superior absolute endurance values exceeding 10¹¹ cycles under highly optimized device and testing conditions^[5,11,54]. However, when benchmarking specifically against variable-stoichiometry systems such as TiN_x and RuO_x, as illustrated in Figure 6I, high reliability is often highly sensitive to stoichiometric variations; even slight deviations can lead to a drastic reduction in cycle life. In contrast, the MoN_x platform provides a broader and more robust reliability window, enabling predictable endurance enhancement across a wide compositional range. Moreover, because endurance is known to degrade exponentially as HZO thickness is scaled down, achieving stable operation up to 10⁸-10⁹ cycles in an 8 nm thin film is significant. This comparison underscores that the MoN_x platform offers an optimal balance of robust ferroelectricity, immediate reliability, and sustained cycle life, thereby mitigating the trade-offs commonly observed in other variable-stoichiometry electrodes.

Beyond the immediate performance improvements, MoN_x is attractive because it enables interfacial engineering without multilayer electrode stacks, which can complicate thickness scaling and introduce additional series resistance. Future work can expand this platform by correlating IL-thickness distributions with breakdown statistics, directly probing trap spectra, and exploring process variables such as N content, deposition temperature, and surface termination to tune E_c without sacrificing oxidation resistance. The ability to modulate stoichiometry over a wide range while maintaining weak oxygen-scavenging behavior is an intrinsic advantage of MoN_x compared with other transition-metal compounds examined to date. TiN and TaN can also retain rock-salt structures over broad stoichiometric ranges; however, their interfacial redox chemistry is strongly affected by the oxygen-scavenging nature of the metal constituents because of strong metal-oxygen bonding. Although hexagonal MoN is thermodynamically more stable than the B1-MoN phase, this work shows that B1-type MoN_x can be reliably formed in thin-film form suitable for semiconductor electrode applications despite its metastability.

Although this study demonstrates the excellent interfacial stability of MoN_x electrodes using an 8 nm HZO model system, dielectric-thickness scaling introduces more severe challenges, including enhanced depolarization fields and direct tunneling currents. To explore this scaling behavior, we evaluated the ferroelectric properties of capacitors with HZO thickness scaled down to 6 nm [Supplementary Figure 14]. The ferroelectric response remains observable at 6 nm, but the hysteresis loops exhibit significant rounding, indicating that macroscopic leakage begins to dominate the electrical response. In such ultrathin regimes, suppression of the defective MoO_x interfacial layer by MoN_x is expected to be even more critical for maintaining acceptable leakage levels. However, at scaled thicknesses of ≤ 5 nm, macroscopic leakage currents heavily dominate the response, and intrinsic ferroelectric switching becomes completely masked, making it difficult to reliably decouple the purely interfacial benefits of the electrode. Process conditions such as annealing temperature and oxidant dose are also expected to strongly influence the delicate balance between bulk HZO crystallization and interfacial oxidation. For example, more aggressive oxidation conditions would likely exacerbate the degradation of elemental Mo and further necessitate the robust barrier properties of MoN_x. Therefore, future studies focused on co-optimizing these thermal and chemical parameters will be essential to fully unlock the potential of MoN_x electrodes for ultrathin (≤ 5 nm) ferroelectric memory applications.

In summary, we demonstrated that stoichiometry-tunable MoN_x can serve as a scalable single-layer electrode platform for ALD-grown ferroelectric HZO, enabling deterministic control of interfacial chemistry and device reliability. By sputter-depositing MoN_x electrodes with x = 0.00, 0.05, 0.52, and 0.79 and integrating them into symmetric MoN_x/HZO/MoN_x capacitors, we established a clear composition-microstructure-property relationship. Structural characterization confirmed that increasing N content drives the electrode toward a rock-salt-type MoN_x phase with (111) texture while retaining electrode-grade conductivity.

Importantly, we elucidated the role of nitrogen incorporation in modifying the interfacial redox dynamics during the aggressive ozone-based ALD process. N-rich MoN_x electrodes function as robust transport barriers, markedly suppressing ALD-induced electrode oxidation and confining the defective oxide layer to a thin interfacial zone. This controlled stoichiometry also facilitates partial nitrogen incorporation into the HZO layer near the bottom interface. The resulting interfacial chemistry provides crystallographic benefits by stabilizing a ferroelectric-compatible structure and significantly suppressing non-ferroelectric M phase formation without requiring complex multilayer electrode engineering.

From an electrical perspective, this microstructural optimization directly addresses critical reliability bottlenecks in HZO-based devices. The fabricated MoN_x stacks exhibit near-wake-up-free switching and extended endurance stability across a broad compositional window, distinguishing this platform from conventional variable-stoichiometry electrodes. In particular, 52MoN is identified as a best-balanced composition that combines strong oxidation resistance, minimized wake-up, robust polarization, and excellent cycling stability. Overall, this work shows that composition-engineerable MoN_x electrodes provide a versatile and scalable pathway for interfacial defect and redox engineering, advancing the viability of high-performance ferroelectric memory technologies.