¹Semiconductor Frontier Research Center, National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba Central 2, 1-1-1 Umezono, Tsukuba, Ibaraki 305–8568, Japan

²NEC-AIST Quantum Technology Cooperative Research Laboratory, National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba Central 2, 1-1-1 Umezono, Tsukuba, Ibaraki 305–8568, Japan

Ferroelectric HfO₂ is attractive for next-generation devices because it retains ferroelectricity in nanometer-scale films and is compatible with semiconductor processing. Most reports ascribe ferroelectricity to a metastable orthorhombic (Pca2₁) phase stabilized by oxygen vacancies, and vacancy migration under electric field is regarded as the origin of remanent-polarization instability (wake-up, fatigue). This study introduces a design strategy for vacancy-independent ferroelectricity: valence-complementary codoping. Substituting Hf⁴⁺ with equimolar Y³⁺ and Nb⁵⁺ (6% each) maintains the average cation valence at 4 + while intentionally creating local charge inhomogeneity that induces internal fields and lattice strain to stabilize a ferroelectric phase. The resulting Y_0.06Nb_0.06Hf_0.88O₂ (YNHO) is identified as noncentrosymmetric tetragonal P4mm by transmission electron microscope, electron diffraction, and grazing-incidence X-ray diffraction. The tetragonal phase persists after annealing in air at 600°C for 100 h, indicating near-thermodynamic stability. TaN/YNHO/TaN capacitors endure 10¹⁰ polarization-switching cycles at 3.3 MV cm^− 1 without detectable wake-up or fatigue, indicating that polarization stability does not rely on vacancy migration. Unlike Pca2₁ ferroelectric HfO₂, P4mm YNHO contains no non-polar sublayers (spacers), suggesting a distinct ferroelectric HfO₂. These findings suggest valence-complementary codoping as a practical strategy for realizing intrinsically reliable ferroelectric HfO₂ and outline a pathway to next-generation logic and memories.

Ferroelectric HfO₂ has attracted considerable attention as a promising platform for high-density nonvolatile electronics, as it allows polarization switching even in nanometer-scale thin films while remaining compatible with advanced CMOS process flows^1,2. In particular, ferroelectric FETs (FeFETs) are regarded as promising devices for in-memory computing architectures suited to machine learning applications, driving active research efforts toward the practical implementation of HfO₂-based ferroelectrics³.

Most reports indicate that ferroelectricity in doped HfO₂ arises from a polar orthorhombic (Pca2₁) phase that is metastable with respect to the monoclinic (P2₁/c) ground state^4–6. However, because oxygen vacancies are highly mobile, they migrate and redistribute under an electric field, inducing field-driven phase transitions between the monoclinic and orthorhombic phases^7–15. The resulting changes in remanent polarization—manifested as wake-up and fatigue—pose a critical reliability challenges for capacitors and transistors utilizing ferroelectric HfO₂. Therefore, whether HfO₂ can exhibit robust ferroelectricity independent of oxygen vacancies, with polarization stability decoupled from defect chemistry, remains an important question.

Here, we propose a new design strategy to stabilize a ferroelectric phase of HfO₂ without relying on oxygen vacancies. We adopt a valence-complementary codoping approach, in which Hf⁴⁺ sites are substituted with paired acceptor–donor cations to maintain an average cation valence of 4+, while introducing local charge inhomogeneity. This suppresses the net charge imbalance that would otherwise require compensation by oxygen vacancies, while deliberately creating local charge inhomogeneity. This approach is inspired by perovskite relaxor ferroelectrics, where mixed B-site valences generate random local electric fields and nanoscale polar order^16–19. We hypothesize that charge disorder in HfO₂ could give rise to internal electric fields and lattice distortions, thereby stabilizing a polar structure independently of oxygen vacancies. Unlike oxygen vacancies, cations do not readily migrate under an electric field. Therefore, if a ferroelectric phase can be stabilized by cation-induced internal fields, it is expected to exhibit robust ferroelectricity.

In perovskite-type oxides, the effects of valence-complementary codoping on physical properties —such as the relaxor characteristics—have been investigated. In contrast, little research has been conducted on how valence-complementary codoping influences binary oxides. Recently, studies on ferroelectric HfO₂ have begun to draw attention to valence-complementary codoping; however, these investigations have primarily focused on the suppression of oxygen vacancies and have not examined the impact of local charge inhomogeneity as observed in relaxor ferroelectrics^20–23.

In this study, we sought to induce ferroelectricity in HfO₂ by codoping Y³⁺ and Nb⁵⁺ in equimolar proportions (6% each), thereby introducing local charge inhomogeneity. Y and Nb were selected to provide strong valence complementarity while maintaining chemical compatibility with the fluorite lattice and keeping the average cation radius close to that of Zr. The total doping concentration of Y and Nb was set to 12% (approximately 1/8), because this corresponds to about one dopant per two fluorite-type HfO₂ unit cells, which is expected to generate a moderate local electric field.

We show that Y_0.06Nb_0.06Hf_0.88O₂ (YNHO) adopts a noncentrosymmetric tetragonal P4mm structure and exhibits robust ferroelectric switching. The tetragonal phase persists after prolonged annealing in air at 600°C for 100 h, indicating near thermodynamic stability. Electrical measurements show that YNHO capacitors withstand 10¹⁰ cycles without detectable wake-up or fatigue, consistent with a reduced role of oxygen vacancies in polarization evolution. Together, these results establish a vacancy-independent route to ferroelectricity in HfO₂ and define a generalizable design principle—valence-complementary codoping—for intrinsically reliable ferroelectric materials. In the remainder of the paper, we describe the structural identification of the P4mm phase, the stability of the ferroelectric phase, the exceptional endurance of YNHO, and benchmarking against state-of-the-art HfO₂-based ferroelectrics.

Transmission electron microscopy (TEM) was used to examine YNHO/TaN/n-Si and TaN/YNHO/TaN/n-Si structures and to analyze the YNHO lattice.

For the YNHO/TaN/n-Si sample, YNHO and TaN thicknesses were 45 nm and 10 nm, respectively. Crystallization was performed by annealing at 600°C for 10-min in Ar + O₂ (21%) at 1 atm, followed by slow cooling at 6°C min^− 1 (see the Experimental Section/Methods). Figure 1a–d shows cross-sectional annular bright-field scanning transmission electron microscopy (ABF-STEM) images; the corresponding high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images are in the Supporting Information, Fig. S2b, d, f, h. Figure 1b,c are higher-magnification views of the red-marked regions in Fig. 1a,b, and Fig. 1d further enlarges the area in Fig. 1c. As seen in Fig. 1a, the TaN film delaminated from the substrate, attributed to volume expansion from oxygen incorporation during annealing and formation of TaO_xN_y; oxygen in the bottom TaN was confirmed by XPS depth profiling (Supporting Information Fig. S1). In Fig. 1b, the YNHO lattice is aligned from surface to TaN interface without dislocations; the undistorted region extends over 40 nm laterally, indicating grains exceeding ~ 40 nm × 40 nm (also see the Supporting Information Fig. S2e). Similar low dislocation density has been reported for La³⁺/Ta⁵⁺-codoped HfO₂, suggesting suppression of oxygen-vacancy formation²⁰.

Figure 1c is a processed high-magnification ABF-STEM image with nonperiodic noise removed (raw image: Supporting Information Fig. S2g). It indicates a noncentrosymmetric YNHO structure; for further analysis, Fig. 1d is used. In Fig. 1d, larger circles denote Hf, Y, and Nb, and smaller circles denote O. Atomic positions show vertical asymmetry in Hf(Y,Nb)–O distances, indicating absence of inversion symmetry along the vertical axis, while lateral positions are symmetric. Figure 1c also indicates the absence of a glide plane and screw axis. Among noncentrosymmetric space groups with mirror planes but without glide planes or screw axes, thirteen are possible: Pm, Cm, Pmm2, Cmm2, Amm2, Fmm2, Imm2, P4mm, I4mm, P3m1, P31m, R3m, and P6mm²⁴. Since it is unlikely for the fluorite structure to transform into trigonal or hexagonal phases, these can be excluded from the candidates. As will be discussed later, the grazing-incidence X-ray diffraction (GI-XRD) results indicate that the possibilities of monoclinic, rhombohedral, and orthorhombic symmetries are unlikely, narrowing the candidate space groups for YNHO to I4mm and P4mm. Further narrowing of the space group is carried out later using electron diffraction.

For the TaN/YNHO/TaN/n-Si structure, YNHO and TaN layers were 18 nm and 10 nm. Crystallization was achieved by annealing at 600°C for 1 and 10-min in N₂ at 1 atm, followed by natural cooling. Before TEM, this sample underwent endurance measurements to 10⁹ cycles using 6 V, 100 kHz square pulses (P–E and endurance are discussed later). Figure 1e,f shows high-magnification ABF-STEM images (Fig. 1f enlarges the red-marked region in Fig. 1e); additional ABF-STEM and HAADF-STEM images are in the Supporting Information Fig. S3a–h. In Fig. 1f, the red lines serve as auxiliary guides to verify that the oxygen atoms are arranged within the same plane. Because the pattern in Fig. 1f matches Fig. 1d, the same space group is preserved after annealing with top-electrode and electrical stress.

Figure 1g shows the electron diffraction pattern along [110] (fluorite unit-cell assumption) from the sample in Fig. 1e,f. Figure 1h presents simulated patterns for I4mm and P4mm; atomic positions and parameters are in the Supporting Information section 3. The beam-exposed region is shown in the Supporting Information Fig. S3a. Comparing experiment and simulation, I4mm does not reproduce the dark spots indicated by white arrows in Fig. 1g, whereas P4mm reproduces the spots marked by red circles. On this basis, YNHO is estimated to be P4mm.

Figure 1i shows EDX mapping of YNHO in TaN/YNHO/TaN/n-Si annealed at 600°C for 1-min. Panels (top-left to bottom-right) display the HAADF-STEM image and maps for Hf, O, Y, Nb, and Y + Nb. Y and Nb are uniformly distributed with no discernible pattern.

Based on these results, approximate atomic positions rendered in VESTA are shown in Fig. 1j²⁵. Ochre spheres represent Hf(Y,Nb) and red spheres represent O. Because YNHO is tetragonal, the unit cell differs from the conventional fluorite cell and corresponds to the red-outlined cell in the top panel of Fig. 1j; side and oblique views are shown below. A distinctive feature is that, unlike tetragonal HfO₂ (P4₂/nmc), oxygen atoms aligned horizontally lie on the same plane²⁶. As in Fig. 1d,f, the two oxygen planes are shifted in the same direction from the quarter-point along the c-axis, and the apparently body-centered Hf(Y,Nb) atoms are slightly displaced along c-axis. If Hf(Y,Nb) were exactly at the body center, their scattering would cancel that of the corner atoms and the dark spots in Fig. 1g would not appear. A bias in Y/Nb distribution could also cause contrast even for body-centered positions, but Fig. 1i shows uniform Y and Nb distribution, making this unlikely. These observations indicate that displacement of the oxygen planes and near-body-centered Hf(Y,Nb) shifts the centers of negative and positive charges, producing spontaneous polarization; its precise magnitude was not determined here and requires further study.

Next, GI-XRD measurement was performed on TaN/YNHO/TaN/n-Si and YNHO/TaN/n-Si to analyze the YNHO structure. YNHO and TaN thicknesses were 45 nm and 10 nm; preparation details are in the Experimental Section/Methods. Figure 2a shows the profile of a TaN/YNHO/TaN/n-Si sample annealed in 1 atm N₂, Fig. 2b that of YNHO/TaN/n-Si under the same conditions, and Fig. 2c that of YNHO/TaN/n-Si annealed in 1 atm Ar + O₂ (21%). Annealing temperatures, durations, and cooling conditions are listed to the right of Fig. 2c. No crystallization occurred at 400°C, whereas samples annealed at ≥ 500°C crystallized. In crystallized films, a peak appears near 30.5°, assignable to the (111) reflection of the orthorhombic/cubic phases or the (011) reflection of the tetragonal phase, while intensities near 28.5° and 31.5°—monoclinic (111) and (− 111)—are negligible. If YNHO were rhombohedral, d₁₁₁ and d_11 − 1 would differ in both length and multiplicity, and the (111) or (11 − 1) peaks around 30.5° should therefore split into two or appear as an asymmetric peak. However, since the peaks in Fig. 2a–c are all single and symmetric, the possibility of a rhombohedral structure is unlikely. Figure 2d shows the (113), (131), and (311) orthorhombic/cubic or (103), (013), (121), and (211) tetragonal reflections for a sample annealed at 600°C for 10-min in Ar + O₂ (21%) and slowly cooled at 6°C min^− 1. First, the TEM results indicate that YNHO is not cubic, thereby excluding the cubic phase from consideration. Next, if YNHO were orthorhombic, d₃₁₁, d₁₃₁, and d₁₁₃ would have the same multiplicity but different lengths, and the (311), (131), and (113) reflections should therefore split into three peaks of comparable intensity. In contrast, for the tetragonal case, the multiplicity of d₂₁₁ is twice that of d₁₀₃, so the (103), (013), (211), and (121) reflections should form two peaks with an intensity ratio of approximately 2:1. From the peak shape in Fig. 2d, the feature can be interpreted as two overlapping peaks with an intensity difference of about a factor of two, making it reasonable to conclude that YNHO is tetragonal rather than orthorhombic. Assuming two overlapping peaks, fitting gives positions at 59.95° and 60.34°. From these, the lattice parameters of the unit cell in Fig. 1j are a ≈ 3.59 Å and c ≈ 5.12 Å (fitting details in the Supporting Information section 4). Together with TEM and electron diffraction, these results indicate tetragonal P4mm formation under the employed annealing conditions, with minimal monoclinic content.

The profile for Ar + O₂ (21%) is noteworthy because oxygen-containing anneals in prior ferroelectric HfO₂ studies typically produce monoclinic or cubic phases (and, rarely, an antiferroelectric tetragonal phase)^27–31. Since YNHO nevertheless crystallizes into tetragonal P4mm after Ar + O₂ (21%) annealing, oxygen vacancies are not required to stabilize the ferroelectric phase in YNHO.

The topmost profile in Fig. 2c corresponds to the Fig. 1a–d sample annealed at 600°C and slowly cooled at 6°C min^− 1; monoclinic peaks are scarcely observed, suggesting the P4mm phase is not a quenched metastable state. Collectively, these data indicate that tetragonal (P4mm) YNHO is more stable than the metastable ferroelectric Pca2₁ phase reported for other ferroelectric HfO₂.

To further test this, GI-XRD was measured for YNHO/TaN/n-Si annealed in 1 atm air (N₂ + O₂) at 500–600°C for 10–100 h, followed by slow cooling at 0.5°C min^− 1 (Fig. 2e). For all samples, monoclinic peak intensities increased relative to 10-min anneals, yet the most intense peak remained near 30.5°, characteristic of the tetragonal phase, indicating a substantial tetragonal fraction even after 100 h. Annealing at 600°C for 100 h produces a larger monoclinic fraction than at 10 h; likewise, 100 h at 600°C yields more monoclinic content than at 500°C. Thus, while the monoclinic phase is thermodynamically stable, the tetragonal phase is sufficiently stable to remain dominant after prolonged annealing, i.e., nearly thermodynamically stable phase.

Next, we present ferroelectric property evaluations of YNHO films. Figure 3a–d shows P–E hysteresis, I–V curves, positive-up-negative-down (PUND) measurements, and endurance for a TaN/YNHO/TaN/n-Si structure annealed at 600°C for 1-min in N₂ and naturally cooled. Similarly, Fig. 3e–h presents the same measurements for a sample annealed at 600°C for 10-min in N₂ and naturally cooled. In both cases, the YNHO layer was 18 nm thick; the applied voltage was up to 6 V (3.3 MV cm^− 1). Measurement procedures and leakage-current data are provided in the Experimental Section/Methods, and Supporting Information section 5. The endurance in Fig. 3d,h was derived from remanent polarization (Pr) obtained by the PUND signals in Fig. 3c,g. The 10-min sample is the same as that used for TEM and electron diffraction in Fig. 1e–g.

Although overlapping and visually difficult to distinguish, Fig. 3a–c includes results from 1 to 10¹⁰ cycles. Measurement stopped at 10¹⁰ cycles due to system constraints; the sample remained operational. The P–E loops, I–V curves, and PUND responses are essentially unchanged from the first to the 10¹⁰th cycle. The 1-min sample shows Pr ≈ 11.4 µC cm^− 2 at 3.3 MV cm^− 1 (Fig. 3c), indicating stable endurance without wake-up or fatigue over 10¹⁰ cycles.

Similarly, Fig. 3e,f shows results from 1 to 10⁹ cycles. Measurement was stopped at 10⁹ cycles as a precaution because this sample was intended for TEM observation before breakdown. Compared with the 1-min sample, slight changes appear in P–E and I–V shapes with cycling, whereas the PUND signal remains essentially unchanged, suggesting that the variations arise outside the ferroelectric regions. The 10-min sample shows Pr ≈ 15.6 µC cm^− 2 (Fig. 3g) and endurance without wake-up or fatigue (Fig. 3h). These results indicate that an electric field of 3.3 MV cm^-1 does not induce ferroelectric–paraelectric phase transitions in YNHO.

To evaluate endurance stability, benchmarking was performed. Figure 3i compiles Max and Min Pr for polycrystalline HfO₂-based ferroelectric films reported to exceed 10¹⁰ cycles under fields ≥ 2.5 MV cm^− 1; the horizontal axis is Min Pr and the vertical axis is Max Pr. Data were estimated from endurance plots of La-doped Hf_0.5Zr_0.5O₂, V-doped HfO₂, SiO₂/Hf_0.7Zr_0.3O₂, Ga-doped HfO₂, HfO_0.61N_0.72/Hf_0.5Zr_0.5O₂, Hf_0.5Zr_0.5O₂, ZrO₂/Hf_0.5Zr_0.5O₂, and Ca-doped Hf_0.5Zr_0.5O₂^4,5,32–39. The black dotted line denotes Max Pr = Min Pr; red, blue, and green dotted lines mark Min Pr = 95%, 90%, and 80% of Max Pr. The YNHO values from Fig. 3d (Max Pr = 11.76 µC cm^− 2, Min Pr = 11.35 µC cm^− 2) lie to the right of the 95% line, indicating high stability.

The P–E and PUND results in Fig. 3a,c,e,g exhibit “slanted” or “sheared” hysteresis^40–42. A discussion of possible causes is provided in the Supporting Information section 6.

On the basis of the results reported thus far, although the ferroelectric phase of YNHO is not the thermodynamic ground state, it is sufficiently close in stability and can exist without oxygen vacancies. We next relate this observation to the improved endurance of YNHO.

Ferroelectricity in HfO₂ codoped with La³⁺ and Ta⁵⁺ (La_0.02Ta_0.02Hf_0.96O₂) has been reported in Ref. 20. In that study (and its Supplementary Information), it was noted that dislocation formation was suppressed due to a reduction in oxygen vacancies. However, the crystal structure remained orthorhombic, and both wake-up and fatigue were observed. These findings suggest that suppressing oxygen vacancies alone is insufficient to fundamentally resolve the issues of wake-up and fatigue.

In contrast, wake-up and fatigue are suppressed in YNHO, and the underlying reasons may be interpreted as follows. For YNHO, peaks at angles corresponding to the tetragonal phase appear irrespective of annealing atmosphere (Fig. 2b,c), indicating that small variations in oxygen-vacancy content have little effect on the crystal structure. Even if vacancies migrate under an applied voltage, such changes do not trigger transitions between the ferroelectric and paraelectric phases. These results indicate that, to achieve stable endurance, forming a ferroelectric phase that does not undergo vacancy-driven phase transitions is more important than merely suppressing vacancy formation or migration. The differences in properties between La_0.02Ta_0.02Hf_0.96O₂ and YNHO are presumed to arise from factors such as dopant concentration and the covalent radii of the dopant elements, although further investigation is needed to clarify the underlying mechanisms.

A remarkable and essential effect of valence-complementary codoping is that it does not merely stabilize a known metastable phase, but rather enables near-stabilization of a previously unreported ferroelectric phase in HfO₂-based systems. This distinction is not limited to the crystal structure. As noted in Refs. 43 and 44, in orthorhombic (Pca2₁) HfO₂, only half of the unit cells exhibit oxygen-ion displacement, resulting in a layered configuration composed of ferroelectric and non-ferroelectric subunits. This gives rise to a two-dimensional polar state characterized by alternating polar and non-polar sublayers (spacers). In contrast, YNHO adopts the P4mm structure, similar to BaTiO₃, and thus contains no such non-ferroelectric subunits. Moreover, while conventional ferroelectrics such as BaTiO₃ exhibit spontaneous polarization primarily through cation displacement, ferroelectricity in HfO₂-based systems arises mainly from oxygen displacement. As previously mentioned, YNHO exhibits spontaneous polarization primarily arising from oxygen displacement, similar to other ferroelectric HfO₂. These findings indicate that YNHO is distinct from both ferroelectric HfO₂ and conventional ferroelectrics, exhibiting characteristics that lie between the two. These results indicate that valence-complementary codoping stabilizes the ferroelectric phase through a mechanism fundamentally distinct from single-dopant strategies that stabilize the metastable orthorhombic phase.

Furthermore, valence-complementary codoping can be achieved solely by adjusting the composition ratio, without the need for specialized equipment or techniques. As such, it is readily compatible with existing structural designs and process technologies for ferroelectric random-access memory, ferroelectric tunnel junctions, and FeFETs, making it a strategy with a low barrier to implementation.

We demonstrate that valence-complementary codoping of Y³⁺ and Nb⁵⁺ in HfO₂ stabilizes a polar tetragonal P4mm phase without requiring oxygen vacancies. The resulting Y_0.06Nb_0.06Hf_0.88O₂ films exhibit robust ferroelectric switching and endurance up to 10¹⁰ cycles without wake-up or fatigue, with structural stability confirmed via TEM, GI-XRD, and ferroelectric properties measurement. This vacancy-independent route circumvents the reliability issues associated with field-driven vacancy migration in ferroelectric HfO₂. Our results provide a generalizable design strategy for stable ferroelectric HfO₂ via local charge inhomogeneity, compatible with ferroelectric memory applications.