Introduction

The discovery of ferroelectricity in HfO2-based thin films inspired a renewed interest in field of ferroelectric (FE) materials and devices in this century1,2. The polarization in HfO2-based films is known to be originated from a metastable Pca21 orthorhombic (O) phase formed in the transformation from tetragonal (T) to monoclinic (M) phase though a polarized rhombohedral phase is also reported1,3,4,5,6,7. Different from perovskite type of conventional FE material, Pca21 O-phase in HfO2 are with fluorite-structure and mostly coexists with T- and M-phases, so are a new type of FE8. These films are with good CMOS compatibility and high scalability down to few nm9,10, overcoming the bottleneck issues of conventional FE materials and making FeRAM a competitive candidate for next-generation non-volatile memory devices7,11,12.

Polarization switching dynamics is one of the most important fundamentals for FE materials. It significantly affects the FE properties, such as coercive voltage (VC) and switching speed, which are critical in non-volatile memory applications. In HfO2-based FE thin films, however, it is considered to be significantly different from that in conventional perovskite type of FE materials13. The polarization in Pca21 O-phase in fluorite-structure of HfO2 is attributed to the four displaced oxygen anions instead of a single cation in the perovskite unit cell8. These anions are located in only one half of the Pca21 unit cell and the other half is similar to the T-phase structure, making locally independent dipole switching possible14. In addition, coercive field (EC) of HfO2-based FE films is much higher than that of conventional ones, indicating a different polarization switching dynamics15. It is worth noting that the high EC in HfO2-based films has been a key issue limiting the lowering of operation voltage and inhibiting their application in state-of-the-art memory devices nowadays.

To clarify the polarization switching dynamics in HfO2-based FE film, understanding of the intermediate state in polarization switching of Pca21 O-phase, from viewpoints of both structures and energetics, is the key. It determines the specific atom migration and kinetic switching barrier, which affects the electric characteristics significantly. So far, some theoretical studies based on density functional theory (DFT) have proposed several switching pathways through antipolar Pbca or 90° rotated polar Pca21 O-phase, through non-polar Pbcm or Pbcn O-phase, and through non-polar P42/nmc T-phase as the intermediate state16,17,18,19,20,21,22. Among them, the pathway through Pbca or 90° rotated polar Pca21 O-phase more likes the domain wall migration16,17,18,19. The pathway through Pbcm O-phase obeys the electrostatics physics and is described by the typical double-well energy function where Pbcm O-phase is energetically unstable16,17. However, the EC predicted by this pathway is one order higher than that of the experimental value.20 In contrast, a pathway through Pbcn O-phase based on the quasi-chiral characteristics of Pca21 O-phase shows lower EC21. And, the pathway through non-polar P42/nmc T-phase is also considerable because of its lower kinetic barrier16,17. Notably, intermediate T-phase can be energy favorable as a metastable state, so such a pathway is described by a triple-well energy landscape in some work16,22. In fact, such a polarization switching dynamics has been generally considered as the origin of the antiferroelectric (AFE) behavior of ZrO2 or HfO2-based film with higher doping concentration23,24.

Despite a lot of efforts demonstrated, a well-agreed model is still not clearly shown, let alone its guiding on the engineering of the related properties, particularly on the operation voltage lowering limited by the high EC. The key issue in clarifying the actual pathway among or beyond these proposals lies in the direct experimental verification consistent with theoretical prediction. It is quite challenging because the intermediate state is only a transient state during the polarization switching process, and it is difficult to capture. Therefore, a carefully designed special investigation methods is crucial. In this work, very interestingly, based on the qualitative analysis of experimental results using Landau-Ginzburg-Devonshire (LGD) model25,26,27, we predict a polarization switching pathway in FE-HfxZr1-xO2 (HZO) films through O–T–O phase transition with a triple-well energy function, namely, through T-phase as a metastable intermediate state. This is qualitatively similar to that predicted by a DFT in early stage but such a pathway appears in FE-HZO films, which are naturally evolved from AFE-HZO in pristine state in our analysis. More importantly, we provide the experimental evidence to directly verify the triple-well energy landscape in such FE-HZO films via its summation effect with electrical energy in dielectric (DE) capacitor (CAP) and the corresponding phase transition pathway via carefully designed in-situ grazing incidence X-ray diffraction (GIXRD) measurement under electric field. Moreover, we get a way to engineering EC and operation voltage of HZO films through tuning the energy barrier of O–T phase transition. As a result, a low EC along with low operation voltage is achieved for an applicable remnant polarization (Pr) together with low leakage current and high endurance performance in a HZO thin film.

Results and discussion

Theoretical prediction of triple-well energy function in FE-like HZO

We start by explaining how a triple-well energy function of FE-HZO is obtained. The evolution from AFE-like behavior to FE-like one in polarization switching cycling of HZO FE-CAP, reported as a kind of wake-up phenomenon28,29,30, is clearly observed in a Hf0.5Zr0.5O2 CAP as shown in Fig. 1a–c. AFE-like double hysteresis in polarization-voltage (P–V) characteristics merges into FE-like single one after 106 polarization switching cycling. Although Pr increases, namely wake-up continues, after hysteresis emerging after 106 cycling, we mainly focus on the transition from AFE-like to FE-like characteristic up to 106 cycling. The typical double hysteresis of AFE-HZO and single hysteresis of FE-HZO in polarization-electric field (P–E) characteristics have been usually represented qualitatively by LGD model in forms of triple- and double-well energy landscape, respectively10,16,23,31. Here, we fit the continual results of P–E characteristics in the evolution before 106 cycling using LGD equation, which represents the average characteristics of HZO film under a static state. The detail fitting method is discussed in section 1 of Supplementary material. Note that the first-order phase transition in LGD model is used because we have experimentally confirmed the first-order phase transition in HZO by investigating crystal phase and P–V characteristics in temperature change as discussed in section 2 of Supplementary material. Figure 1d exhibits that the relative energies of three wells corresponding to AFE characteristics continually change along with the coefficients in LGD model in the evolution. Namely, the two-sided wells become lower relative to the intermediate one. More interestingly, following the trend of change in relative energies of three wells, special states appear after the evolution, which are with triple-well energy landscapes but exhibit single P–E hysteresis, as shown in Fig. S1, which is also reported in Al2O3/HZO/Al2O3 stack32,33. Note that, although single P–E hysteresis are obtained macroscopically, like a typical FE material with double-well energy function, these states have triple-well energy functions and two stage of switching with double “S”-curve. This means a metastable state exists in the pathway of polarization switching even in FE-like HZO. Since such states are resulted from a change in the energy landscape itself rather than a change in the relative proportion of domains with AFE-like triple-well energy landscapes and FE-like double-well energy landscapes, as discussed in section 3 of Supplementary material, we define these characteristics as triple-well FE (TFE) ones in the following discussion.

Fig. 1: Prediction of triple-well energy landscape in HZO with single hysteresis of P–E characteristics.
figure 1

a Schematics and b TEM image of an HZO(Hf0.5Zr0.5O2)-CAP under tests. The thickness of HZO is 6 nm. c The evolution of P–V characteristics from AFE-like double hysteresis to FE-like single one observed in the HZO-CAP before 106 polarization switching cycling. d The corresponding evolution of energy landscape of the HZO-CAP before 106 polarization switching cycling, fitted from the I–V and P–V curves using LGD model. eP–E characteristic of the HZO-CAP just after 106 polarization switching cycling fitted by LGD model. f Corresponding energy landscape of the HZO-CAP just after 106 polarization switching cycling and its TFE switching pathway fitted by LGD model.

Full size image

Here, we consider that the HZO-CAP just after 106 cycling, which just transits from AFE-like behavior to the FE-like one, is in TFE state. In fact, the extracted Curie temperatures of this sample is shifting, supporting this prediction with first-order phase transition of LGD as shown in Fig. S6. Figure 1e, f shows the theoretical triple-well energy landscape and P–E characteristics by fitting the experimental results. Interestingly, the middle well corresponding to P = 0 is the energetically favorable but TFE-CAP is not in this state when V = 0. The middle well can only be reached during polarization switching when |V| ≥ VC as an intermediate state, but it is changed to be unstable at this moment. As a result, the middle well cannot be directly observed in single TFE-HZO CAP. To observe the middle well state with P = 0 and confirm a triple-well energy function in TFE-HZO, adding DE-CAP is an effective way as predicted in Fig. 2a. In this way, the energy of middle well relative to two-sided wells in the whole TFE + DE CAP is further lowered compared to that of single TFE CAP. This means the whole TFE + DE CAP will have an AFE-like energy function and the single P–E hysteresis in single TFE-CAP will split into double after connecting a DE-CAP in series.

Fig. 2: Splitting phenomenon of polarization switching peak in I–V curve of HZO-CAP induced by connection with DE-CAP in series.
figure 2

a Theoretical prediction of change in energy landscape and P–V loop when the TFE-CAP is connected to a DE-CAP in series. b Schematics of experiment setup to confirm the theoretical prediction. cI–V and P–V diagram of single HZO-CAP, after DE-CAP connection and after DE-CAP disconnection, respectively. The single polarization switching peak in I–V curve of single HZO-CAP splits into two peaks after connecting the DE-CAP. And, the I–V curves of HZO-CAP before connecting and after removing DE-CAP connection are basically the same.

Full size image

Experimental verification of triple-well energy landscape in FE-like HZO

To further verify the above theoretical prediction of a TFE-CAP, we connected a 200 pF DE-CAP in series with the HZO-CAP in Fig. 1a whose P–V characteristics is just change to single hysteresis and the capacitance is about 70 pF. Figure 2b shows the schematics of the experimental setup. Here we consider three current-voltage (I–V) and P–V tests of single HZO-CAP, after DE-CAP connection and after DE-CAP disconnection, respectively, as shown in Fig. 2c. The results clearly demonstrate that the single polarization switching current peak in I–V of single HZO-CAP splits into double after DE-CAP connection, and then recovers into single after DE-CAP disconnection. This phenomenon is exactly as predicted by the triple-well energy landscape in TFE-HZO as discussed above, and thereby can be considered as an experimental verification. It is worth noting that the I–V and P–V tests in three modes were performed continuously without any additional steps. And there is no obvious change in peak position and shape in the first and third mode, indicating HZO-CAP have not been changed by the DE-CAP connection. This is intrinsically different from the peak splitting phenomenon reported in other literature, which is considered induced by the local imprinting of domains or depolarization effects34. And the tiny peak at V ≈ 0 in the forward I–V curve of HZO-CAP before DE-CAP connection and after DE-CAP removal does not affect the peak splitting phenomenon, because it can be removed by optimizing the preparation process without influencing the peak splitting phenomenon as discussed in section 4 of Supplementary material.

In order to further confirm the experimental results, the impacts of the capacitance value of DE-CAP and polarization switching cycles applied on HZO-CAP before DE-CAP connection on the splitting phenomenon of polarization switching peaks in I–V curve were also investigated as in the section 5 of the Supplementary material. The switching peaks in I–V curve of single HZO-CAP splits wider with connection of smaller capacitance of DE-CAP. This is as expected from energy landscape change with matching of smaller capacitance of DE-CAP. Meanwhile, switching peak splitting phenomenon becomes weaker with more polarization switching cycles applied on HZO-CAP before connection of DE-CAP. This is also predicted as the side energy wells of the HZO-CAP become lower in the evolution of P–E characteristics, which requires smaller capacitance of DE-CAP to re-match the condition. In a word, these results further confirm the triple-well energy function of HZO evolved from AFE one.

Polarization switching pathway through O–T–O phase transition in TFE-HZO

Next, the polarization switching pathway correlated with the triple-well energy landscape in TFE-HZO is considered. Theoretically, since the TFE state evolves from AFE-like state naturally by polarization switching cycling, the material structures corresponding to three energy wells in TFE-HZO should be consistent with that of the pristine AFE-like HZO. Meanwhile, AFE-like behavior of HZO with T-phase has been generally considered to be originated from a reversible O–T–O phase transition induced by switching electric field22,23,24, and the electric field induced reversible O–T phase transition at microscopic scale has been indicated by a couple of transmission electron microscope (TEM) investigations35,36. If this is the case, the three wells in free energy function will correspond to energy of O↑-, T-, O↓-phase, respectively. Namely, the T-phase will be the metastable intermediate state in polarization switching of polar O-phase in TFE-HZO CAP, which have been also predicted by DFT as introduced above. And, it is understandable because structure of Pca21 O-phase is very similar to P42/nmc T-phase and both phases coexist in the film. However, the macroscope evidence is still absent even for an AFE-HZO film.

In order to confirm this macroscopically and systematically, a synchrotron-based GIXRD measurement was carefully performed on the pristine Hf0.5Zr0.5O2-CAP and typical AFE ZrO2 and Hf0.2Zr0.8O2 CAPs with in-situ electric field. Figure 3a shows the setup of the measurement where 15 nm HZO films was employed to enhance detection signal. The method to control the X-ray spot consistent with electrode area is described in the method section. Figure 3b shows the pulse sequence applied on the HZO-CAP, where ±4~±6.5 V waveform is higher than the switching voltage of HZO-CAP to make sure the polarization switching. In order to decrease the effect of direct current (DC) stress, the relatively low voltage of ±3 V is applied for in-situ measurement after the waveform, under which the polarized state is maintained. In analysis of the results, 2θ were transferred from λ1 of 1.24 Å into λ2 of 1.54 Å as the latter is generally used in literatures. Although the shapes of peaks from O- and T-phase cannot be directly distinguished, the relative change between both can be obtained from the position shift of the T (101)/O (111) peak, because the peak position of O (111) and T (101) has a difference of ~0.26° in 2θ.

Fig. 3: In-situ GIXRD measurement and results of 15 nm Hf0.2Zr0.8O2-CAP under electric field.
figure 3

a Schematics of in-situ GIXRD measurement under electric field. The X-ray spot size was about 2.5 times of the electrode size. b Voltage pulse sequence applied on 15 nm Hf0.2Zr0.8O2-CAP during in-situ GIXRD measurements. The orange triangle pulses were carried out to induce polarization switching, and GIXRD measurements were carried out under different DC field marked as segments (1)–(5). GIXRD spectrum of 15 nm Hf0.2Zr0.8O2-CAP under electric field corresponding to segments (1)–(5) in Fig. 3b after different c, d positive and e, f negative triangle pulses. More T–O phase transition occurs after triangle pulses with higher electric fields. g P–V curve, O (111)/T (101) peak position of 15 nm Hf0.2Zr0.8O2 sample, and the corresponding reversible T–O phase transition. h Schematic diagram of O–T–O polarization switching pathway in HZO film.

Full size image

Clearly, as shown in Fig. 3c, the GIXRD peak of AFE Hf0.2Zr0.8O2-CAP at ~30.4° contributed by T (101)/O (111) phase shows a negative shift from T-phase side to O-phase side under 3 V voltage after applying a triangle pulse with 4–6.5 V, and it shift back to T-phase side when the electric field is removed. The negative peak shift is along with the increase of the left shoulder and decrease of the right shoulder of T (101)/O (111) phase peak, corresponding to intensity increase of O (111) and decrease of T (101) phase peak. And the other way around. This phenomenon indicates the reversible T–O phase transition induced by applying and removing the electric field. Since it is not easy to get accurate decomposition between O (111) and T (101) phase peaks, the peak shift rather than intensity change is taken into account as the main reference for analysis. Note that the negative shift of the T (101)/O (111) phase peak is more obvious after applying a triangle pulse with higher electric field, which suggests that it is the triangle pulse itself instead of DC stress during GIXRD measurement that induces T–O phase transition. And the GIXRD peak shift is induced by the T–O phase transition rather than electric field induced strain as the GIXRD peaks position at ~34.3° are basically unchanged in Fig. 3d. The same phenomenon is observed when negative field is applied and removed, as shown in Fig. 3e. The negative peak shift of T (101)/O (111) phase under negative applied field (−5.5 V) is slighter compared with that under positive field with same strength (+5.5 V). This is because the polarization intensity change under the negative field is smaller as discussed in the section 6 of Supplementary material. Also, due to the earlier breakdown phenomenon under negative electric field, the maximum negative field is smaller (−5.5 V) than the positive one. Here, the magnitude of peak shift is not as large as the theoretical difference between O (111) and T (101) because only part of crystal experienced such a phase transition. The minimum size of X-ray spot is about 2.5 times of the electrode size, which will introduce additional XRD signal from the unaffected HZO film. And the polarization switching intensity is not very large (~4 μC/cm2) as shown in Fig. 3g, which indicates that even in the HZO film beneath the electrode, there are some unpolarized T-phase grains that have not been transformed into polarized O-phase. Based on the results in the Fig. 3c, e it can be expected that the T–O phase transition signal will be more obvious if a higher electric field is applied. To make the results more intuitively, we fitted the peak positions of O (111)/T (101) peaks and plotted them as a function of electric field together with polarization in Fig. 3g. We found that the peak shift became quicker when the applied field is approaching the transition field from non-polarized state to polarized state. This is consistent with the P–E curve and directly show the electric field induced reversible T–O phase transition occurring along with polarization state change.

More interestingly, we found that the intensity of peak at ~34.8° in GIXRD results contributed by O/T (002) phase increases under both positive/negative voltage and such a phenomenon disappear when the voltage is reset to be 0 V as well, as shown in Fig. 3d, f. Similarly, the peak intensity changes at ~34.8° also become quicker after applying a triangle pulse with higher electric field no matter it is positive or negative. Since the polarization direction of (002)-oriented O-phase is out-of-plane, (002)-oriented O-phase should be at least one of the main contributors to the polarization intensity in P–V loop. Therefore, we considered the consistent intensity change of O/T (002) peak with polarization change in P–V curves comes from the intensity change of O (002)-phase. And same GIXRD results are also observed in 15 nm AFE-ZrO2, as shown in section 6 of Supplementary material. This is another evidence of electric field induced reversible T–O phase transition. Here, the slight peak shift of O/T (002)-phase under electric field may be due to the peak position difference between O (002) and T (002)-phase in this film. Such a difference seems affected by HfO2 content, as the shift is less obvious in results of ZrO2 and the lattice parameter of O-phase is slightly change by increasing HfO2 content as reported in literatures2. Therefore, intensity change is taken into account mainly for analysis concerning T/O (002) peak. Notably, a reversible T–O phase transition with applying and removing electric field in such a film is also confirmed by the TEM measurements in-situ under the electric field as discussed in the section 7 of Supplementary material. All these results demonstrate the polarization of O-phase switches through T-phase as an intermediate state in AFE-HZO and ZrO2.

Similarly, the peak shift with applied voltage also occurs in 15 nm Hf0.5Zr0.5O2-CAP with AFE characteristics in pristine state, as shown in Fig. S20. This indicates the pristine AFE-like behavior in Hf0.5Zr0.5O2 has the same physical origin through reversible O–T–O phase transition with the typical AFE-HZO. But in this sample, the peak does notcompletely shift back to the initial position after one cycle of polarization switching. It is consistent with the reported results by TEM and suggests that a non-reversible T–O phase also occurs partly. This is understandable that the evolution from AFE to FE has occurred in the first cycle of polarization switching in some grains under the DC stress. Nevertheless, our results directly and clearly demonstrate the polarization switching pathway through T-phase as an intermediate state in pristine Hf0.5Zr0.5O2. And, naturally, such a pathway should continue in the films, which is evolved to TFE properties as schematically shown in Fig. 3g. Besides, other effects, such as ferroelastic switching might also play an important role28, which is discussed through the capacitance-voltage (C–V) and in-situ GIXRD tests in section 6 and 8 of Supplementary material.

Low E C achievement in TFE-HZO

The insight of the above polarization switching pathway through O–T–O phase transition in TFE-HZO CAP on its EC and operation voltage properties is finally considered. Since EC in FE material is generally associated with energy barrier between two directions of polarization state, for TFE HZO film, it should highly depend on the energy barrier between O- and intermediate T-phase as shown in Fig. 4a. Therefore, reducing the energy barrier of O–T–O phase transition should be an effective way to lower EC and operation voltage. Theoretically, it has been proposed that introduction of proper strain and more oxygen vacancies (VO) into HfO2-based films could significantly reduce the energy barrier between O- and T-phase16,17. Experimentally, we tried to increase VO concentration by using oxygen reactive TiN electrode along with increasing post-metallization annealing (PMA) temperature. Our results confirmed that EC can be significantly lowered through these ways. On the other hand, EC is further lowered along with enhancement in Pr by introducing a Al2O3 layer between top TiN electrode and HZO. It is understood that inserting of Al2O3 may modulate the strain or act as dopants in HZO, and the precursor may induce VO in HZO film during Al2O3 deposition37. The specific influence of these adjustments on EC is detailed in the section 9 of Supplementary material.

Fig. 4: Demonstration of optimized TFE-HZO film with low operation voltage and low EC.
figure 4

a Schematics of EC reduction through modifying the energy barrier of reversible O–T phase transition. b PUND characteristics of optimized HZO film under different operation voltage. 2Pr > 20 µC/cm2 is achieved under 0.65 V. c Benchmark of HZO film reported in this work as against some reported HZO films considering the Pr value at each operation voltage and electric field. The thickness of optimized TFE-HZO film is ~8 nm, which is measured by the TEM as shown in Fig. S30b. d Leakage current and e endurance performance of optimized HZO film. ~10−5 A/cm−2 leakage under 0.65 V and >1011 wake-up free endurance under 1.5 V is demonstrated on this device.

Full size image

As a result, we get an optimized TFE-HZO film whose fabrication process was shown in the section 10 of the Supplementary material. The film thickness is ~8 nm, which is measured by TEM as shown in Fig. S28b. The positive-up-negative-down (PUND) characteristics under different operation voltage of optimized HZO film whose P–E characteristics has just completed transition from double hysteresis to single one, is shown in Fig. 4b. With PUND measurement, the effects of possible leakage on Pr are basically eliminated. A high saturation 2Pr of ~54 μC/cm2 is obtained under ~1.7 V of operation voltage and an applicable 2Pr > 20 μC/cm2 is done under ~0.65 V. The corresponding I–V curves extracted by PUND test indicate the EC of optimized HZO film is ~0.6 MV/cm, as shown in Fig. S29. Figure 4c benchmarks the Pr under different electric fields in this work compared with other works7,11,12,38,39,40. Clearly, we achieved a low electric field along with low operation voltage for the same level of Pr. In addition, the leakage and endurance characteristics of the optimized films were also investigated. Figure 4d, e show a < 10−5 A/cm−2 leakage under 0.65 V and >1011 wake-up free endurance under 1.5 V. These results demonstrated the high comprehensive performance is achievable in TFE-HZO along with the low EC and low operation voltage.

In conclusion, a triple-well energy function along with single-hysteresis of P–E characteristics in HZO films evolved from pristine AFE ones has been theoretically predicted and experimentally confirmed. From this energy landscape, a polarization switching pathway, where the unpolarized tetragonal-phase acts as a metastable intermediate state during the polarization switching of polar orthorhombic-phase, has been also demonstrated. These results provide insights into EC engineering in FE-HZO by controlling the energy barrier of reversible O–T phase transition. As a result, a low EC ~ 0.6 MV/cm with low operation voltage <0.65 V for 2Pr > 20 μC/cm2 has been achieved together with high endurance performance and low leakage level in such a HZO films.

Methods

Device fabrication

The fabrication processes of HZO-CAPs in Figs. 1, 2, 4, S3–S6, S10–S12 and S26–S30 are basically the same. They started with RCA cleaning of p+ Si substrate, followed by 30 nm TiN films deposition by reactive magnetron sputtering at room temperature. The sputtering power is 250 W, the flow rate is Ar:N2 = 50:4 sccm, and the air pressure is 0.5 Pa. Then HZO films were deposited by atomic layer deposition (ALD) under 250 °C with precursors TEMAHf/TDMAHf, TEMAZr/TDMAZr and H2O. Optionally, ~2 nm Al2O3 top layers were deposited by 27 ALD cycles under 200 °C with precursors TMA and H2O. A bi-layer photoresist lithography process (LOR10A + AZ5214) was then applied for the sharp lift-off of 30 nm TiN followed by 30 nm W top electrodes by sputtering. The sputtering power for W is 80 W, the flow rate is Ar = 30 sccm, and the air pressure is 0.9 Pa. And the size of top electrodes is 40 × 40 μm2. Finally, PMA was carried out at 350–650 °C for 30 s ~ 3 h in flowing N2 ambient at 100 kPa. The detail process parameters are listed in Fig. S32 in section 11 of Supplementary material. The fabrication process of the device in Fig. 3, S18–S20, S23 and S24 are basically the same as above, while the HZO film thickness is 15 nm and the size of top electrodes is 200 × 4000 μm2, which is around 1/2.5 of the X-ray spot (~300 × 8000 μm2). The X-ray spot and electrode are aligned under the microscope using sticker development. The triangle pulses in Fig. 3b have applied voltages of ±4~±6.5 V and a frequency of 250 Hz. And the pulse 2 and 4 in Fig. 3b last 10 s for the CCD detector to image.

Material characterization

The thickness of the p+ Si/TiN/HZO/TiN/W films were determined by TEM. TEM lamella samples were prepared with TESCAN GAIA3. FEI TALOS F200X G2 operated at 200 kV was used for TEM imaging. The in-situ GIXRD measurements were carried out with incident X-ray source energy of 10 keV, corresponding to 1.24 Å wavelength. In order to enhance the effective signal strength corresponding to the electric field induced phase transition, the X-ray projection spot on the HZO sample should match the size of the electrode. In this work, the X-ray spot size was set to ~300 × 300 μm2, which was controlled by a slit near HZO sample. And the projected size of X-ray spot at 2° incidence angle on HZO sample is ~300 × 8600 μm2, around 2.5 times of electrode size (200 × 4000 μm2). The X-ray spot was positioned and marked under the microscope through the fluorescent sheet. And the electrode was aligned to the spot position through electric displacement stage. The GIXRD spectrums were collected by Dectris Eiger 500 K. To get enough XRD signal strength to capture the possible phase transition, the exposure time, along with the corresponding DC stress time, was set to 10 s. In order to eliminate the influence of long-term electric field stress on the HZO samples as much as possible, the DC stress was set to 4 V. And the samples were trained by a triangle pulse with different higher voltage (±3~±6.5 V) and 10−3 s rise time before the DC stress. Note that the max negative field of triangle pulse in Fig. 3e is limited to −5.5 V because the breakdown occurs during GIXRD signal capture after applying higher field. The 2θ of GIXRD results were calibrated by the standard LaB6 sample result under the same experimental conditions. The theoretical diffraction data of O-phase refers from P. Nukala’s work41, and that of T-phase refers from ICDD: 00-002-0733. In the analysis of GIXRD results, due to the possible presence of strain in HZO film or other system error, the XRD peaks of the actual samples shift from the theoretical values. Thus, the experimental peaks of O (111), T (101), O (020) and O/T (002) lie at around 30.23°, 30.5°, 34.3°, and 34.8°, respectively. The crystal structure drawings of O and T-phase in Fig. 3h were produced by VESTA42. In-situ electric field TEM imaging was performed using a Thermo Scientific Themis Z TEM operated at 300 keV. A PicoFemto in-situ STM-TEM electrical measurement system was employed to apply electric field across the HZO thin film during high-resolution imaging. The PicoFemto TEM holder enables precise electrical biasing with simultaneous atomic-scale structural characterization. Real-time HRTEM images were acquired at various voltage states, and Fast Fourier Transform analysis was conducted to determine the local crystal structure, allowing for phase identification between the tetragonal and orthorhombic phases and monitoring of the electric-field-induced phase transition dynamics.

Device characterization

Electrical characterization was carried out using a Keithley 4200A-SCS system with 4225-PMU and 4225-RPM module at room temperature and at atmosphere. The endurance tests on HZO-CAPs in Figs. 1, 2, S10–S13 and S16 were performed with 1 MHz of 3 V rectangle waveform, while the P–V and I–V characteristics were measured with 25 kHz of 3 V triangle waveform. The I–V tests in Fig. 3 at XRD samples were performed with 250 Hz of ±6 V triangle waveform. The endurance tests on optimized TFE-HZO in Fig. 4 were performed with 500 kHz of 1.5 V rectangle waveform, while the PUND characteristics were measured with 1 kHz of 0.6–3 V waveform. The influence of parasitic capacitance to ground in series circuit of HZO-CAP and DE-CAP was avoid by connecting DE-CAP to the side of PMU1 which provides waveform output. When DE-CAP is connected, the I–V tests were performed with ±6~±10 V instead of ±3 V of triangle waveform. The endurance tests on HZO-CAPs in Fig. S14 is similar to the above devices in Fig. 1 but the 3 V waveform was replaced by 4 V. The C–V tests were performed with the 100 kHz AC signal on Keysight E4990B, and the dielectric constant ε is extracted by ε = Cd/Sε0, where d is the thickness of HZO sample, S is the electrode size of HZO sample, and is ε0 vacuum permittivity. The I–V tests in Figs. S24, S25, S27–S29, and S31 were performed with 1 kHz of 2–3 V triangle waveform.