Ferroelectric Hafnia-Zirconia Superlattices
- Ferroelectric hafnia-zirconia superlattices are engineered nanolayered structures that use HfO2 and ZrO2 sublayers to stabilize polar phases under precise epitaxial conditions.
- Advanced deposition methods like pulsed laser deposition enable atomic-scale control and sharp interfaces, which tailor the phase competition and ferroelectric performance.
- Exploiting interfacial energetics and strain management, these superlattices achieve enhanced polarization, endurance, and negative-capacitance effects, crucial for CMOS integration.
Ferroelectric hafnia-zirconia superlattices are artificially layered fluorite-oxide heterostructures in which , , or sublayers are stacked with nanometer- or unit-cell-scale periodicity so that interface energetics, epitaxial constraint, and thickness confinement reshape the competition among monoclinic, tetragonal, orthorhombic, antipolar, and rhombohedral polymorphs. Their central purpose is to stabilize a polar phase in a materials family whose bulk equilibrium reference is typically nonpolar monoclinic hafnia, while preserving the silicon compatibility that makes hafnia-based ferroelectrics relevant to CMOS-derived device platforms (Li et al., 1 Jul 2025, Mukherjee et al., 2024).
1. Superlattice concept and structural families
Two experimental superlattice lineages dominate the recent literature. One uses ordered epitaxial superlattices, with examined at total thickness around $20$ nm and extended over $4,6,10,20,40,60,$ and $100$ nm. In that notation, each repeat period contains unit cells of 0 followed by 1 unit cells of 2, and the experimentally optimized period is 3 (Li et al., 1 Jul 2025). The second uses 4 superlattices, where 5, 6, 7 is the number of repetitions, and 8 is the total thickness in nm; the studied 9 compositions are 0, 1, and 2, usually with a 3 sublayer-thickness ratio (Gent et al., 7 Jul 2025).
These architectures are embedded in a broader fluorite-derived phase landscape. In pure hafnia, the nonpolar monoclinic 4 phase is the bulk ground state, whereas the polar orthorhombic 5 phase is the phase most commonly associated with experimental ferroelectricity. First-principles and symmetry analyses additionally identify tetragonal 6, orthorhombic 7, rhombohedral 8 and 9, and antiferroelectric or antipolar orthorhombic parents such as 0 or 1, depending on the specific structural genealogy and boundary conditions (Raeliarijaona et al., 2023).
| Superlattice family | Reported polar assignment | Representative result |
|---|---|---|
| 2 | Predominant polar orthorhombic order | Optimized 3 period (Li et al., 1 Jul 2025) |
| 4 | Polar rhombohedral 5 in all constituent sublayers if the LSMO–6 interface is preserved | 7 at 8 total 9 (Gent et al., 7 Jul 2025) |
| Ideal Zr/Hf 2/2 and 4/4 superlattices | Fully polar 0 orthorhombic superlattices | Polarization 1–2 but not thermodynamic ground states (Mukherjee et al., 2024) |
This diversity is not merely terminological. It reflects the fact that phase selection depends on orientation, epitaxial symmetry, layer thickness, interfacial sharpness, and whether the relevant comparison is an ideal relaxed superlattice, an oxide-epitaxy model system, or a nanometer-scale capacitor stack.
2. Epitaxial realization and structural fingerprints
The 3 superlattices were grown epitaxially by pulsed laser deposition on 4-oriented 5 buffered with a 6 nm 7 bottom electrode. In that process, LSMO was deposited at 8 using 9 Hz and $20$0, whereas the HZO comparison films and the superlattices were deposited at $20$1 with $20$2, followed by cooling at $20$3 in $20$4 Pa oxygen. The resulting heterostructures showed a strong XRD peak near $20$5 assigned to orthorhombic $20$6, clear Kiessig fringes in X-ray reflectivity, and a root-mean-square roughness of about $20$7 pm for the $20$8 superlattice. Pole figures displayed $20$9 diffraction spots around 0, consistent with four orthorhombic crystallographic domains rotated by 1 and an effective out-of-plane [111] orientation (Li et al., 1 Jul 2025).
The same study used O 2-edge X-ray absorption spectroscopy and X-ray linear dichroism to distinguish the superlattice from a random-alloy HZO reference. In the 3 superlattice, the anisotropy between 4 and 5 showed positive XLD at the 6 manifold and negative XLD at the 7 manifold, matching prior fingerprints of polar hafnia phases, whereas the HZO alloy film showed reversed dichroism and 8 splitting characteristic of a substantially different, more monoclinic-like electronic structure. HAADF-STEM and EDXS further resolved alternating Hf- and Zr-rich layers with only about one atomic layer of mutual diffusion, and high-resolution images of both constituents revealed the zig-zag cation arrangement expected for orthorhombic 9 viewed along the [101] zone axis (Li et al., 1 Jul 2025).
In the $4,6,10,20,40,60,$0 family, the films were also grown by PLD on $4,6,10,20,40,60,$1, but with different growth conditions: $4,6,10,20,40,60,$2 layers at $4,6,10,20,40,60,$3, pure $4,6,10,20,40,60,$4 layers at $4,6,10,20,40,60,$5, and cooling at $4,6,10,20,40,60,$6 in $4,6,10,20,40,60,$7 mbar oxygen. STEM-HAADF showed atomically sharp interfaces, low roughness, negligible ion interdiffusion, and explicit $4,6,10,20,40,60,$8 nm thickness control. Because out-of-plane diffraction near $4,6,10,20,40,60,$9 could not by itself separate rhombohedral, orthorhombic OIII, and tetragonal-related reflections, pole figures and in-plane diffraction were used; the reported $100$0-fold symmetry around the $100$1 pole and the $100$2 offset between out-of-plane and in-plane $100$3 reflections were taken as evidence for a rhombohedral structure with four domains, each of threefold symmetry (Gent et al., 7 Jul 2025).
3. Mechanisms of polar-phase stabilization
The mechanistic picture that emerges is multiscale. In the $100$4 superlattices, first-principles calculations attribute the enhanced ferroelectric stability to kinetic suppression of nonpolar phase formation and to increased interfacial formation energy for unfavorable phase combinations. When $100$5 is constrained by neighboring orthorhombic $100$6, the transition barrier $100$7 in $100$8 rises by approximately $100$9, while the 0 barrier remains nearly unchanged. In the same framework, the calculated 1-2 interface energy increases from about 3 in an HZO reference system to about 4 in the actual 5-6 superlattice, with corresponding critical thicknesses of about 7 and 8 nm. The proposed consequence is that sharp elemental discontinuity suppresses monoclinic invasion over a wider thickness range (Li et al., 1 Jul 2025).
A second mechanism is epitaxial symmetry selection. High-throughput DFT phase diagrams for epitaxial hafnia show that orthogonal in-plane lattice conditions and the associated shear-strain penalty imposed on the monoclinic 9 phase can make 00 thermodynamically preferred under experimentally accessible conditions, especially for 01- and 02-oriented films on YSZ- and STO-derived templates. For isotropic substrates, the reported favorable windows are 03 Å for 04 and 05–06 Å for 07; the key point is that the decisive selector is not scalar compressive-versus-tensile strain alone, but the full epitaxial boundary condition, including in-plane anisotropy and angle (Zhu et al., 2023).
Pure-hafnia theory contributes a third element: the origin of switchability inside the polar phase. One symmetry analysis identifies the experimentally relevant 08 phase as a proper ferroelectric descended from a centrosymmetric 09 parent through a single unstable zone-center 10 mode, with a shallow double-well depth of 11. In that picture, strain is not required for switching once the polar phase exists, even though strain and epitaxy can be decisive in accessing or stabilizing the relevant precursor manifolds (Raeliarijaona et al., 2023).
By contrast, ideal fully relaxed Zr/Hf superlattices present a thermodynamic limit. In DFT studies of 2/2 and 4/4 ZrO12/HfO13 superlattices stacked along the three pseudo-cubic directions, all structures relaxed from the 14 starting point remain fully polar, with polarization 15–16. Their energy penalties, however, are 17–18, so they are competitive rather than favorable relative to monoclinic superlattice states. The stated explanation is that both HfO19 and ZrO20 share a monoclinic ground state and a relatively low-energy polar orthorhombic local minimum, so layering the two largely inherits bulk phase competition rather than eliminating it (Mukherjee et al., 2024).
4. Ferroelectric response, thickness scaling, and reliability
The best electrical performance reported for 21 occurs at 22. For the 23 nm series, the 24 superlattice shows 25, 26, and 27–28, together with pronounced switching peaks in the 29-30 curves and butterfly-shaped 31-32 loops. Relative to a 33 nm random-alloy HZO film, the same 34 structure exhibits increases of about 35 in 36, 37 in 38, and 39 in dielectric constant. Thickness scaling is especially unusual: optimized 40 superlattices remain ferroelectric from 41 to 42 nm, 43 decreases with thickness and reaches about 44 at the thick end, and a 45 nm sample maintains 46 with less than 47 variation over 48 switching cycles, whereas the HZO control breaks down after 49 cycles (Li et al., 1 Jul 2025).
The same work reports weak frequency and temperature dispersion. At 50 nm thickness, hysteresis loops measured from 51 Hz to 52 kHz change very little, and robust ferroelectricity is retained from 53 K to 54 K. The frequency insensitivity is interpreted in a nucleation-limited switching framework, in which the energetic barrier for domain nucleation dominates over facile domain-wall motion (Li et al., 1 Jul 2025).
The 55 family emphasizes a different optimization axis: deliberate use of thin 56 layers as a polarization booster and as an endurance-enhancing interface network. The best 57 superlattice, which corresponds to alternating 58 nm 59 and 60 nm 61 sublayers, shows 62 and 63, more than twice the 64 of a 65 nm 66 monolayer of equal average composition. The highest polarization is reported for the 67 family, where the best member exhibits 68 and 69 at 70 total 71 content (Gent et al., 7 Jul 2025).
Reliability trends in that family are equally central. Endurance is defined as the maximum number of cycles for which 72 is maintained. High-Zr solid-solution monolayers can break down after only 73 cycles, and a 74 solid solution reaches 75 but breaks down after 76 cycles. By comparison, the highlighted 77 superlattice can be cycled 78 times while maintaining 79 under the lower-field endurance condition. The proposed reason is that added interfaces redistribute oxygen-vacancy accumulation away from electrode-localized conductive filaments (Gent et al., 7 Jul 2025).
5. Electronic structure, switching pathways, and finite-size physics
Electronic-structure calculations show that ordered hafnia-zirconia superlattices are not electronically equivalent to random HZO alloys at the same average composition. In the orthorhombic 80 phase, the reported GGA-1/2 band gaps are 81 eV for o-82, 83 eV for o-84, and 85 eV for o-86. For ordered 87 ferroelectric superlattices stacked along the polar 88-axis, the gap decreases as 89 increases: about 90 eV for H1Z1-SL, 91 eV for H2Z2-SL, 92 eV for H3Z3-SL, 93 eV for H4Z4-SL, 94 eV for H6Z6-SL, 95 eV for H12Z12-SL, and 96 eV for H24Z24-SL. The stated mechanism is an internal electric field generated by asymmetric ferroelectric interfaces: because Hf is more ionic than Zr, interfacial III-coordinated oxygens acquire unequal charge states at the two interfaces, the planar-averaged Hartree potential tilts across the layers, and some Zr-derived conduction states are shifted downward in energy (Huang et al., 2023).
Switching and phase competition are also strongly size dependent. A Landau-Ginzburg-Devonshire treatment developed for HfO97 thin films, nano-islands, multilayers, and heterostructures predicts a ferroelectric window 98. In that framework, the upper critical size is controlled by the effective mismatch strain, including misfit dislocations and lateral relaxation, whereas the lower critical size is controlled by the depolarization field and correlation effects. For ideal screening, the FE phase is stable under compressive strains below roughly 99; for 00 Å, the minimum required compressive strain strengthens to about 01, and ferroelectricity disappears entirely above 02 Å at 03. The same theory gives a switching barrier of about 04 and activation fields of about 05–06 (Morozovska et al., 9 Jan 2026).
Mesoscale domain physics introduces an additional layer of complexity. A 3D phase-field model for ultrathin mixed-phase hafnia/zirconia-based films with orthorhombic/tetragonal coexistence finds that smaller ferroelectric grains and a larger angle of the polar axis with respect to the out-of-plane direction enhance negative-capacitance stabilization, with the strongest enhancement near 07–08. The same simulations report that a negative lateral gradient coefficient along even one axis drives unit-cell-scale antipolar modulation, hardens walls, and prevents negative-capacitance stabilization. This suggests that superlattice optimization cannot be reduced to phase fraction alone; domain-wall mobility and anisotropic wall coupling remain decisive (Kumar et al., 2024).
An additional interfacial mechanism emerges from ultrathin epitaxial La-doped hafnia on YSZ(111). There, the as-grown state is reported to be largely orthorhombic antiferroelectric 09, and mechanical poling converts it into ferroelectric 10, while orthorhombic order becomes more stable as thickness decreases to one unit cell because the o-AFE state carries a vanishing depolarization field and the interface is nearly isomorphic. This suggests that FE/AFE competition, rather than direct FE stabilization alone, may be relevant to some nanoscale hafnia-zirconia heterostructures as well (Li et al., 2024).
6. Phase assignments, design rules, and unresolved issues
A persistent issue in the field is that different hafnia-zirconia superlattice architectures stabilize different polar symmetries. Ordered 11 superlattices on STO/LSMO are described as predominantly polar orthorhombic, with 12 reflections, [111] texture, and zig-zag 13-like cation arrangements in both constituents. By contrast, the 14 superlattices on STO/LSMO are interpreted as rhombohedral 15, with pole-figure signatures taken to exclude a purely orthorhombic assignment. Ultrafine fluorite-matched systems on YSZ introduce yet another motif, namely an orthorhombic antiferroelectric precursor that can be poled into a ferroelectric state (Li et al., 1 Jul 2025, Gent et al., 7 Jul 2025, Li et al., 2024).
These differences are not easily reduced to a single universal phase map. The available data indicate that period, orientation, interface sequence, sublayer thickness, and the identity of the first layer at the electrode all matter. In the 16 series, the optimized period is 17, and reducing periodicity suppresses monoclinic peaks near 18 and 19. In the 20 family, the explicitly stated design rules are to increase Zr content in 21, keep both 22 and 23 layers thin, keep 24 below 25 nm, use more interfaces for better endurance, and start the stack with 26, not 27, so that the LSMO–Hf-containing interface is preserved. The best structures are described as asymmetric superlattice constructions consisting of thin layers where the strain state in each sublayer is maintained such that 28 is more elongated (Li et al., 1 Jul 2025, Gent et al., 7 Jul 2025).
At the same time, the thermodynamic limit remains unsettled. Ideal relaxed Zr/Hf superlattices are predicted to remain polar but not become true ground states, which implies that real-device performance still depends on epitaxial strain, interface chemistry, defect kinetics, electrostatics, and finite-size effects beyond the zero-temperature perfect-interface limit. This suggests that “ferroelectric hafnia-zirconia superlattice” is best understood not as a single structure type, but as a design space in which polar orthorhombic, rhombohedral, mixed polar/antipolar, and even negative-capacitance-enabled states can be selected by controlling the relative strength of monoclinic suppression, elastic matching, depolarization management, and interfacial sharpness (Mukherjee et al., 2024, Zhu et al., 2023).
