Reversible transition between the polar and antipolar phases and its implications for wake-up and fatigue in HfO2-based ferroelectric thin film

Atomic-resolution Cs-corrected scanning transmission electron microscopy revealed local shifting of two oxygen positions (OI and OII) within the unit cells of a ferroelectric (Hf0.5Zr0.5)O2 thin film. A reversible transition between the polar Pbc21 and antipolar Pbca phases, where the crystal structures of the 180° domain wall of the Pbc21 phase and the unit cell structure of the Pbca phase were identical, was induced by applying appropriate cycling voltages. The critical field strength that determined whether the film would be woken up or fatigued was ~0.8 MV/cm, above or below which wake-up or fatigue was observed, respectively. Repeated cycling with sufficiently high voltages led to development of the interfacial nonpolar P42/nmc phase, which induced fatigue through the depolarizing field effect. The fatigued film could be rejuvenated by applying a slightly higher voltage, indicating that these transitions were reversible. These mechanisms are radically different from those of conventional ferroelectrics.

nsuring reliable and repeatable electrical performance from ferroelectric (FE) thin films and devices present a significant challenge for conventional FE materials such as perovskitestructured Pb(Zr,Ti)O 3 (PZT) to enable their use in FE memory devices [1][2][3][4] . Among the potential problems, fatigue, which is induced by defect (mainly oxygen vacancy (V O )) generation 5 and the accompanying FE domain pinning 6,7 that occurs during repeated electrical stimulation, is one of the most critical concerns [1][2][3][4] . While the discovery of ferroelectricity in doped HfO 2 thin films by Böske et al. 8 in 2011 has again triggered enormous interest in FE memories [2][3][4] and several other novel applications 1,[9][10][11] , these films are not exempt from serious fatigue concerns 2,3 . However, the much higher binding energy of the Hf-O bond when compared with the Ti-O bond in PZT makes serious V O involvement in the fatigue in this material unlikely.
Another interesting but also undesirable effect in this new FE material is wake-up (a temporal increase in the switchable polarization during the first 10 4 -10 5 switching cycles) [12][13][14][15][16] . This has been ascribed to the decrease in the non-ferroelectric tetragonal interfacial layer (T-phase, P4 2 /nmc) caused by the reduced V O concentration in the layer 13,15 . However, a reactive TiN electrode would be likely to increase the V O concentration during repeated cycling 13,15 . Therefore, these two issues, i.e., severe fatigue and wake-up, as direct consequences of the V O effect are hardly understood.
In fluorite-structured FE materials, four of the eight oxygen ions within each unit cell were shifted along the c axis direction (O II ions), while the other four oxygen ions (O I ions) remained in centrosymmetric positions 17,18 . The energy of the polar phase (FE Pca2 1 , where all O II ions shift in the same direction) is only slightly higher than that of the antipolar phase (antiferroelectric (AFE) Pbca, where O II ions shift into antiparallel directions) bỹ 10 meV/f.u 17,19 . These specific crystallographic features of HfO 2 -based FE materials may render the transition between the FE Pca2 1 and AFE Pbca phases to be both fluent and reversible, depending on the bias voltage conditions, and maybe the origin of the fatigue and wake-up phenomena. However, this probability has scarcely been examined to date. This work used an advanced imaging technique, Cs-corrected scanning transmission electron microscopy in annular brightfield mode (STEM-ABF), which enables reliable imaging of light atomic columns (oxygen) in the presence of much heavier atomic columns (Hf and Zr) in fluorite-structured (Hf 0.5 Zr 0.5 )O 2 (HZO) films 20 . The pristine, woken-up, fatigued, and rejuvenated 10-to 15-nm-thick HZO films were prepared by electrical cycling of TiN/HZO/TiN FE capacitors. The atomistic structural evolutions occurring in each state were examined using the STEM-ABF technique. The involvement of V O effect was observed during fatigue cycling, which increased the interfacial T-phase, but the direct cause of the fatigue differed fundamentally from that in conventional FE materials. The wake-up, fatigue, and rejuvenation were mediated by the (reversible) transition between the FE Pca2 1 and AFE Pbca phases. This conclusion was further supported by theoretical calculations based on density functional theory (DFT). The theoretical findings on the switching path and the bias voltage dependence matched the experimental results very well.

Results
A TiN/HZO/TiN capacitor with HZO thickness of 15 nm was fabricated using the atomic layer deposition (ALD) process at a substrate temperature of 280°C (see "Methods" section). Because the FE orthorhombic phase (O FE -phase, Pca2 1 ) is not the thermodynamically stable bulk phase 21,22 , it could be achieved by kinetic means, supplemented by the surface energy/stress effects in the fine-grained (~10 nm) polycrystalline thin films 17,23 . This phase formation could be enabled by a specific kinetic phase transition from the amorphous as-deposited ALD material during post-deposition annealing (PDA) [24][25][26] . The typical thickness of the HZO film with the highest remanent polarization (P r ) was 10 nm, but a slightly greater thickness (15 nm) was used in this work to ease the STEM characterization. Figure 1a shows the cross-section of the pristine (before electrical cycling) TiN/HZO/TiN structure, which was observed by STEM in the high-angle annular dark-field (HAADF) mode. In the figure, one of the grains in the HZO film that penetrates through the entire film thickness was identified. The bright contrast region corresponds to the Hf and Zr ionic columns, which confirmed that the grain's crystal structure is O-phase. However, the different space groups (SGs) among the various O phases cannot be identified by recognition of heavy Hf/Zr ions alone. In particular, identification of the O FE -phase Pca2 1 (SG: 29) and the antipolar O AFE -phase Pbca (SG: 61) is an advantage of the STEM-ABF technique, which has a high sensitivity to light elements, i.e., oxygen ions. Zooming in to the dashed green square area in Fig. 1a in STEM-ABF mode allowed the oxygen ion positions to be determined clearly, as shown in Fig. 1b. Using the shifts in the oxygen ion positions from the centers of the four nearest Hf/Zr ion positions, the oxygen ions can be divided into O I -type (navy blue, at the centrosymmetric position) and O II -type (cyan, at the off-center position). Here, for ease of comparison of the two crystal structures, the polar O FE -phase was identified as  and ½00 1 directions (purple and green arrows, respectively), which can be identified more clearly in the enlarged insert images from Fig. 1b. The larger proportion with green arrows corresponds to the Pbc2 1 O FE domain with uniform polarization (P). The regions with alternating green-and purple-colored arrows might be regarded as the Pbc2 1 O FE -phase with a one-unit-cell-thick FE domain width, but an even more appropriate interpretation would be that it is the O AFE Pbca phase. Figure 1c exhibits the two kinds of atomic models of the Pbc2 1 and Pbca phases along [010] direction, in which the purple, navy blue, and cyan-colored solid balls correspond to the Hf/Zr, O I (center), and O II (off-center) ions, respectively. A detailed atomic structure comparison of the O FE Pbc2 1 and O AFE Pbca phases is shown in Supplementary Fig. S1, which contains selfconsistent experimental and theoretical ABF images. These results indicate that many of the O-phase single grains in the pristine state are composed of a mixture of the O FE Pbc2 1 and O AFE Pbca phases. This is expected because the free energies of these two phases are quite similar, and the crystallization stage during PDA processing thus involved a high probability of formation of these structures.
It is commonly believed that the hafnia-based ferroelectric capacitors undergo a wake-up-fatigue-breakdown trend under continuous switching cycles. Figure 2a shows the variations in 2P r and the leakage currents of a TiN/Hf 0.5 Zr 0.5 O 2 (15 nm)/TiN capacitor with increasing numbers of switching cycles (bipolar voltage pulses of ±5 V at f = 500 kHz). The 2P r was measured by ±3 V P-V loop, and the leakage current was measured at 2 V. During the first 10 4 cycles, wake-up occurs, and the leakage current remains almost constant. During subsequent cycling up to~10 7 cycles, the samples showed gradual fatigue (with 2P r decreasing from~20.7 to 13.7 μC/cm 2 ) and a simultaneous increase in the leakage current by more than 50 times. The sample broke down completely after 10 8 cycles. The fielddependent wake-up process was shown in Supplementary Fig. S2 (for the 15 nm Hf 0.5 Zr 0.5 O 2 film), where all 2P r values were measured using a ±3 V P-V loop after each decade of cycles. At low voltages (<3.2 V), no obvious wake-up was observed, but the 2P r value decreased monotonically with increasing numbers of switching cycles. However, at V a > 3.6 V, wake-up was clearly observed. Figure 2b shows the 2P r evolution of samples with 15nm-thick Hf 0.5 Zr 0.5 O 2 , 10-nm-thick Hf 0.6 Zr 0.4 O 2 , 10-nm-thick Hf 0.5 Zr 0.5 O 2 , 10-nm-thick Hf 0.4 Zr 0.6 O 2 , and the nanolaminated 10-nm-thick Hf 0.5 Zr 0.5 O 2 films, under bipolar V a pulses of ±1.8 MV/cm, which are lower than the critical wake-up field (Supplementary Fig. S3). The 2P r values remained relatively constant up to~10 4 cycles and decreased rapidly toward much smaller values at 10 7 cycles. The capacitor leakage currents were also measured at each decade of low-field cycling pulses, but no obvious increase in leakage current was observed, as shown in Fig. 2c. These findings indicated that the fatigue caused by low V a values is irrelevant to the generation of new defects. Figure 2d shows the low-V a fatigue and subsequent rejuvenation process. After the devices were fatigued using bipolar ±1.8 MV/cm, 10 7 pulse cycles, and then rejuvenated by subsequent application of ±3 MV/cm, 10 4 recovery pulse cycles. The rejuvenated capacitor could be fatigued and rejuvenated again, with ten cycles of repetitive fatigue/recovery process, as shown in Fig. 2e. This "temporary" low-V a fatigue and rejuvenation phenomenon is quite different from the conventional perception of fatigue, which is generally regarded as an irrecoverable process that causes permanent P r degradation and device failure 15,27,28 . These findings indicate that the changes in the internal (defect) structure of the HZO film during field cycling are different for the low and high V a values. In the low-V a cycling case, the main fatigue mechanism is probably the phase transition from the O FE Pbc2 1 to the O AFE Pbca phase, while the rejuvenation can be ascribed to the reverse phase transition without involving permanent structural change or V O generation. In contrast, the changes shown during high-V a cycling may involve a specific permanent structural change, e.g., V O generation and an accompanying phase transition. The subsequent STEM-ABF observations of the woken-up, fatigued, and rejuvenated samples confirmed the above hypotheses. Figure 3a shows a STEM-HAADF image of the woken-up sample, which was cycled 10 4 times at V a = ±4 V and f = 500 kHz to wake-up the sample effectively but minimize the damaging effects of defect generation. The image shows an O FE -or O AFEphase grain projected along the [010] zone axes. The corresponding STEM-ABF image is shown in Fig. 3b, in which the oxygen ion locations can be observed clearly. The image shows that most of the O II atomic columns shift along the [001] direction (green), with only two vertical O II columns with the opposite shift (purple) remaining; this indicates that almost the entire film becomes the O FE Pbc2 1 phase. The inset image in Fig. 3b shows a close-up that clearly reveals that all the O II ions shifted in the same direction. Therefore, the wake-up effect could be correlated positively with the structural transition from the O AFE -phase in the pristine sample to the O FE -phase after cycling. This is a distinctive interpretation of the wake-up effect in this material when compared with the previous interpretation, which described it as the result of V O redistribution from the interfacial nonpolar T-phase into the interior part of the O FE -phase film 15 . However, STEM observation of the interfacial region near the top electrode across the broad areas of both the pristine and wokenup samples (Supplementary Fig. S4a, b) showed that the interfacial T-phase thickness actually increased slightly, from 0.75-1.0 nm (pristine state) to 0.94-2.11 nm (woken-up state). This indicates that the interfacial nonpolar T-phase to O FE -phase transition may not be the major mechanism of the wake-up process. In contrast, the interpretation of the wake-up effect based on the O AFE -phase to O FE -phase transition in the bulk thin-film region does not involve such a difficulty. Additionally, it is consistent with the non-varying leakage current observed during the wake-up process (Supplementary Fig. S2).
The same reasoning implies that fatigue could also be induced by the reverse transition, i.e., from the O FE Pbc2 1 phase in the pristine state to the O AFE Pbca phase after the fatigue. To confirm this hypothesis, a pristine 15-nm-thick Hf 0.5 Zr 0.5 O 2 film was cycled 10 6 times at V a = ±2.5 V and f = 500 kHz, which corresponds to low-V a fatigue without involving an intermediate wakeup. This cycling operation reduced the 2P r value from~9.4 tõ 2.0 μC/cm 2 (read at ±2.5 V, data not shown here). Figure 4a, b shows STEM-HAADF and STEM-ABF images of such a low-V a fatigued sample, respectively. The images demonstrate that almost the entire area of the film was now in the O AFE -phase, which was further confirmed by the magnified view shown in Fig. 4c. To further verify the validity of the antipolar-polar transition model, the crystalline structure of other devices were examined at the low field fatigued and woken-up states, respectively. Supplementary Figs. S5-6 show the STEM-HAADF and STEM-ABF cross-section images of the 10-nm-thick nanolaminated Hf 0.5 Zr 0.5 O 2 film. They also show a clear Pbca phase structure after the low-field fatigue and Pbc2 1 phase-dominated structure after the wake-up. Similarly, Supplementary Figs. S7 and 8 show the reversible transition between the antipolar Pbca and polar Pbc2 1 phases as the origin of the low-field fatigue and rejuvenation in the 10-nm-thick Hf 0.5 Zr 0.5 O 2 film. As discussed below, the O AFE -phase free energy is lower than that of the O FE -phase by~10 meV/f.u., thus indicating that this transition is energetically favorably driven. It was noted that the low-V a fatigue only occurred with bipolar switching pulses. Stressing the device using unipolar voltages of the same amplitude did not cause fatigue (see Supplementary Fig. S9 and related discussion).
The question then becomes this: what is the critical field strength that determines whether fatigue or wake-up (and also rejuvenation) will occur? To answer this question, DFT was used to simulate the switching path from the Pbca global minimum energy phase to the Pbc2 1 local minimum energy phase. Details of the calculations are included in SM (Supplementary Fig. S10). The effective electric field (E eff ) distribution across the O FE region of the film, which must be smaller than the applied field (E app ) because of the presence of the non-FE layer, was also calculated by assuming the presence of the interfacial T-phase based on the STEM results ( Supplementary Fig. S11). The presence of nonpolar layers in series with the FE film induces the depolarization field, which reduces the effective field greatly during the switching cycles. The goal of these calculations is to estimate the critical field (E 0 ) that can make the polar O FE phase more stable than the O AFE phase by its electrostatic energy contribution.  Figure 5a shows the schematic diagrams of the energy landscape for Pbca (center global minimum) and Pbc2 1 (two local minima at finite P). Under the application of a negative E app (more precisely, E eff ), the −P s and +P s (saturation polarization) states of the Pbc2 1 phase become more unstable and stable, respectively, while the nonpolar phase energy remained unchanged. When E eff > E 0 , the Pbc2 1 phase becomes energetically more favorable than the Pbca phase; this is summarized in Fig. 5b. The calculations showed that E 0 is~0.8 MV/cm, i.e., when E eff is lower or higher than 0.8 MV/cm, the O AFE Pbca phase or the O FE Pbc2 1 phase becomes stable, respectively. The activation barrier (E a ) of the direct transition between Pbc2 1 and Pbca in HfO 2 was also calculated, which is directly related to the wake-up, fatigue, and rejuvenation. They are summarized with the calculated relative energy of Pbc2 1 to Pbca (ΔE), by~10 meV/f.u., which is consistent with previous reports 17,19 in Supplementary  Table S1. Although the E a depends on the transient path, i.e., the movement of O II (off-center) atoms, these calculation results are comparable with previous studies, including the recent report by Xu et al. 29 . Table 1 shows E eff across the O FE layer when the interfacial T-phase layer is present within the 0.75-1.0 nm range in the pristine state and in the 0.94-2.11 nm range in the woken-up state (the first and second rows). When V a is 3 V or higher, E eff was higher than E 0 for the pristine state, indicating that the film was prone to be woken-up, which is consistent with the results shown in Fig. 2c, except for the case where V a is 3 V. This exception must be related to the still high activation barrier between the two phases (reaching as high as~0.15 eV/f.u.), which prohibited fluent switching to the Pbc2 1 phase under this condition. In the woken-up sample case, E eff became smaller because of the involvement of the higher depolarizing effect. Therefore, a V a of at least~4 V was necessary to keep the Pbc2 1 phase more stable than the Pbca phase during cycling. These calculation results strongly support the hypothesis for the wake-up and fatigue processes discussed in Figs. 1-4.
Finally, the structural variations and accompanying electrical performances after fatigue cycling with a V a of ±4 V are discussed. Supplementary Fig. S12a shows a STEM-HAADF image of the TiN/HZO/TiN capacitor after high-V a fatigue testing (±4 V, 10 8   cycles), which reduced the 2P r value from 18.9 to 9.8 μC/cm 2 ( Supplementary Fig. S13a). The corresponding STEM-ABF image along the [010] direction in Supplementary Fig. S12b clearly shows the alternating distributions of the O II ion positions being shifted along the [001] (green) and ½00 1 (purple) directions. This finding demonstrates that high-V a fatigue was also induced by the transition from the O FE Pbc2 1 phase to the O AFE Pbca phase. Another notable finding from this sample is the development of a fairly thick interfacial T-phase, which can be seen in Supplementary Fig. S12a. The STEM-HAADF images of this high-field fatigued sample confirmed that the thickness of this layer was 3.25-4.25 nm. The last row of Table 1 indicates that E eff across the remaining~11.75-10.75 nm-thick O FE -phase is as low as 0.29-0.55 MV/cm for V a = 4 V under these circumstances because of the severe depolarizing field effect. These calculation results thus corroborated the finding that high-V a fatigue is induced by the transition from the Pbc2 1 phase to the Pbca phase and not by defect generation and the accompanying domain pinning.
Following the same line of reasoning used for rejuvenation of the low-V a fatigued sample shown in Fig. 2f, a similar rejuvenation was attempted using V a = 5.5 V that increased E eff tõ 0.84-1.09 MV/cm, as also shown in Table 1. Supplementary  Fig. S13b shows the evolution of the P-V curve with increasing rejuvenation cycling numbers up to 10 4 . As expected, the fatigued sample is partly recovered. This is further strong evidence that the high-V a fatigue was also induced by the phase transition mentioned above.
Similar electrical and structural characterizations were performed for the 5.6-nm-thick HZO film to prove the general applicability of the reversible phase transition model for the thinner films. The results are summarized in Supplementary Figs. S14-S21. Although the thinner film contained a higher portion of T-phase grains in the pristine state, in addition to the O-phase grains, it showed similar wake-up, fatigue, and rejuvenation processes. Therefore, it can be inferred that the reversible transition between the nonpolar Pbca phase and polar Pbc2 1 phase also occurs in the 5.6-nm-thick HZO film. The only difference is that the wake-up process in this thinner film is also contributed by the structural transition from the P4 2 /nmc T-phase to Pca2 1 O-phase.
The knowledge acquired based on the effects of the V O concentration on the phase stability indicates that the increase in the interfacial T-phase thickness is reasonable; it has been reported previously that the highest V O concentration stabilizes the T-phase over the O-and monoclinic (M)-phases 28,30 . It must then be reasonable to believe that the repeated electrical stimuli enhanced the chemical interactions between the HZO film and the reactive TiN electrode, which would also increase the V O concentration. It was also noted that these interfacial T-phases were mostly found around the top electrode interface. The bottom electrode interface was formed by depositing an HZO film on the TiN layer, which may produce a stable interface structure 31 . However, the top electrode interface was formed by depositing the TiN layer on the HZO film using a reactive sputtering process, which must have had a damaging effect on the HZO layer surface. Therefore, the top surface of the HZO layer must be vulnerable to oxygen concentration losses during electrical cycling tests.

Conclusion
The crystal structures of pristine, woken-up, fatigued, and rejuvenated O-phase grains in atomic layer-deposited ferroelectric HZO films were examined in detail using the Cs-corrected STEM-HAADF and STEM-ABF techniques. The results showed that the 180°domain boundary structure of the O FE Pbc2 1 phase is identical to the crystal structure of the O AFE Pbca phase, which is consistent with a previous theoretical expectation 18 . This structural relationship and the only slightly different energies of the two phases make the reversible phase change produced by the application of an electric field with a controlled magnitude feasible. The critical electric field for this transition was estimated to be~0.8 MV/cm using DFT calculations.
The wake-up phenomenon could be induced by applying a voltage that exerted a switching field across the O FE layer that was higher than the critical field. The interfacial T-phase grew as the number of switching cycles increased to the degree that was also dependent on the applied voltage. This interfacial T-phase generated a depolarizing field, which reduced the effective field applied to the remaining O FE phase film. As the interfacial T-phase evolves in tandem with the increasing numbers of cycles, the effective field then becomes lower than the critical field, even at the given applied voltage; this is accompanied by the transition from wake-up to fatigue. However, because the fatigue in this material was not induced by irreversible defect generation and domain pinning but instead was induced by a (reversible) phase transition between the polar and antipolar phases, the fatigued sample could then be rejuvenated by applying a voltage that increased the effective field to exceed the critical field. This degradation mechanism for the ferroelectric performance of the HZO film is critically different from that of conventional perovskite-based FE thin films. A similar model can be applied to other fluorite-structured FE thin films, including variously doped HfO 2 films.
While the oxygen loss caused by the repeated electrical stimuli was the fundamental origin of the performance degradation, it did not affect the phase stability and FE performance of the O FE phase film directly. Instead, the evolution of the interfacial T-phase caused the depolarization field effect over the remaining O FE phase-based portion, which then promoted the transition to the ground state O AFE phase. Therefore, preventing oxygen loss during repeated electrical operations remains the most crucial factor in achieving the ultimate performance levels when using these newly established FE films.

Methods
Device fabrication. The TiN/HZO/TiN capacitor fabrication processes were as follows. First, the TiN bottom electrodes were deposited on the SiO 2 /Si substrate by ion beam sputtering. Then, 10-15-nm-thick HZO thin films were deposited by ALD at 280°C. Hf[N(C 2 H 5 )CH 3 ] 4 , Zr[N(C 2 H 5 )CH 3 ] 4 , and H 2 O were used as the Hf precursor, the Zr precursor, and the oxygen source, respectively. The Hf:Zr ratio was controlled by alternating deposition of one cycle of HfO 2 and one cycle of ZrO 2 . Also, 10-nm-thick nanolaminated HfO 2 /ZrO 2 film, where each layer's thickness was 1 nm, was prepared. Subsequently, the top electrode (TiN) layer was sputtered and patterned into different sizes. Finally, the as-fabricated TiN/HZO/ TiN ferroelectric capacitors were annealed for 30 s at 500°C in a nitrogen atmosphere to crystallize the HZO.
Electrical measurements. The polarization-voltage (P-V) and leakage current characteristics of the metal-FE-metal devices were measured using a Radiant Workstation ferroelectric tester and an Agilent B1500 semiconductor parameter analyzer. During all electrical measurements, a bias voltage was applied to the bottom electrode, and the top electrode was grounded. The capacitor size was 100 µm × 100 µm.
Cs-STEM experiment. After the different electrical operations, the TiN/HZO/TiN capacitors were processed into cross-sectional samples using the focused ion beam (FIB) technique in the FEI Helios G4 system, including low-pressure polishing processes at 5 and 2 keV. The sample was treated in a Gatan 691 precision ion polishing system at 1-0.5 keV to remove any residual contamination and damage from the sample surface. The STEM experiments were then conducted on a JEM Grand ARM 300F microscope operating at 300 kV in the STEM mode with a probe aberration corrector. The HAADF and ABF images were acquired at a probe convergence semi-angle of~18 mrad. The STEM-HAADF images were acquired using an annular dark-field image detector with an inner semi-angle of more than 64 mrad. The STEM-ABF images were acquired using a bright-field image detector with its central section blocked using a beam stopper; the collection semi-angle was in the 12-24 mrad range.
DFT calculations. Ab initio calculations were performed using the Vienna Abinitio Simulation Package (VASP) 32,33 . The generalized gradient approximation of the Perdew-Burke-Ernzerhof (GGA-PBE) exchange-correlation functional 34 and Blöchl's projector augmented wave (PAW) approach 35,36 were used with an energy cut-off of 600 eV. The k-points were sampled and found to be 8 × 8 × 8 for the Pbc2 1 unit cell and 8 × 4 × 8 for the Pbca unit cell. The dielectric constant was obtained via density functional perturbation theory (DFPT) 37,38 , and the spontaneous polarization was calculated using the Berry phase method 39 . The activation barrier (E a ) for switching was estimated using the solid-state nudged elastic band (SS-NEB) method 40 .