Field‐Induced Ferroelectric Phase Evolution During Polarization “Wake‐Up” in Hf0.5Zr0.5O2 Thin Film Capacitors

As an emerging nonvolatile memory technology, HfO2‐based ferroelectrics exhibit excellent compatibility with silicon CMOS process flows; however, the reliability of polarization switching in these materials remains a major challenge. During repeated field programming and erase of the polarization state of initially pristine HfO2‐based ferroelectric capacitors, the magnitude of the measured polarization increases, a phenomenon known as “wake‐up”. In this study, the authors attempt to understand what causes the wake‐up effect in Hf0.5Zr0.5O2 (HZO) capacitors using nondestructive methods that probe statistically significant sample volumes. Synchrotron X‐ray diffraction reveals a concerted shift in HZO Bragg peak position as a function of polarization switching cycle number in films prepared under conditions such that they exhibit extremely large (≈3000%) wake‐up. In contrast, a control sample with insignificant wake‐up shows no such peak shift. Capacitance – voltage measurements show evolution in the capacitance loop with switching cycle number for the wake‐up sample and no change for the control sample. Piezoresponse force microscopy measurements are utilized to visualize the domain switching with wake‐up. The combination of these observations clearly demonstrates that wake‐up is caused by a field‐driven phase transformation of the tetragonal phase to the metastable ferroelectric orthorhombic phase during polarization switching of HZO capacitors.


Introduction
Implementation of ferroelectric nonvolatile memory in semiconductor integrated circuits has been a decades-long goal. The widely-studied perovskite titanate ferroelectrics [1,2] have suffered from materials compatibility and dimensional scaling limitations that have inhibited their widespread application in semiconductor memories, although metal-ferroelectricmetal (MFM) capacitors in relatively low-density ferroelectric random access memories (FeRAM) have been demonstrated. [3] As a promising candidate for nonvolatile memory technologies, HfO 2based ferroelectric thin films have multiple advantages compared to perovskites including compatibility with silicon CMOS materials and process integration schemes, and ferroelectric properties that scale to nanometric film thickness, promoting dimensional scaling to advanced semiconductor technology nodes. [4][5][6] These films are integrated in MFM capacitors for FeRAMs and in transistor gate stacks for ferroelectric field-effect transistors (FeFETs), for applications such as memory computing. [7][8][9] However, HfO 2based ferroelectrics have their own challenges with respect to widespread adoption in semiconductor memories and related devices. One issue is the change in remnant polarization with field cycling. [10] Most of the ferroelectric HfO 2 -based films reported show a "wake-up" phenomenon, which refers to an increase in the remnant polarization during the initial switching cycles. To fully utilize the potential of HfO 2 -based ferroelectric memories, such as programming of multiple analog memory states, it is important to minimize the wakeup effect. To achieve this, it is imperative to understand what causes the polarization wake-up in HfO 2 -based films. There are different mechanisms reported that can contribute to the wake-up effect such as domain depinning from defect redistribution and field-induced structural transformation from the tetragonal (T-) phase to the ferroelectric orthorhombic (O-) phase. [10][11][12][13][14] Qi et al. used DFT calculations to investigate the energy barriers to polymorphic phase transitions in HfO 2 -based ferroelectric crystals, including the effect of applied electric As an emerging nonvolatile memory technology, HfO 2 -based ferroelectrics exhibit excellent compatibility with silicon CMOS process flows; however, the reliability of polarization switching in these materials remains a major challenge. During repeated field programming and erase of the polarization state of initially pristine HfO 2 -based ferroelectric capacitors, the magnitude of the measured polarization increases, a phenomenon known as "wake-up". In this study, the authors attempt to understand what causes the wake-up effect in Hf 0.5 Zr 0.5 O 2 (HZO) capacitors using nondestructive methods that probe statistically significant sample volumes. Synchrotron X-ray diffraction reveals a concerted shift in HZO Bragg peak position as a function of polarization switching cycle number in films prepared under conditions such that they exhibit extremely large (≈3000%) wake-up. In contrast, a control sample with insignificant wake-up shows no such peak shift. Capacitance -voltage measurements show evolution in the capacitance loop with switching cycle number for the wake-up sample and no change for the control sample. Piezoresponse force microscopy measurements are utilized to visualize the domain switching with wake-up. The combination of these observations clearly demonstrates that wake-up is caused by a field-driven phase transformation of the tetragonal phase to the metastable ferroelectric orthorhombic phase during polarization switching of HZO capacitors.
fields on polarization wake-up. [15] However, these calculations did not represent MFM stacks in which the electrostatic and mechanical influence of electrode interfaces may have a decisive effect on the magnitude of the electric field threshold value required for polar phase evolution. Zheng et al. performed insitu transmission electron microscopy (TEM) measurements to visualize the structural transformation to O-phase with an applied voltage cycle. [16] Because the field of view in TEM is restricted, and the preparation of ultrathin, electron-transparent cross-section specimens alters the electrostatic and mechanical boundary conditions on the thin films, it is not evident that the reported observations are representative of an actual ferroelectric device. Zhou et al. suggested that wake-up in Si-doped HfO 2 films originates due to the de-pinning of domains led by the reduction of the defect concentration at the bottom electrode interface with repeated voltage cycling. [17] Fields et al. used X-ray diffraction (XRD) to examine wake-up and fatigue effects in hafnium-rich hafnia-zirconia alloy film capacitors with Pt, W, and TaN electrodes and concluded that phase exchange occurs between T-/O-phases and the nonpolar monoclinic phase during fatigue. [18] Although it is proposed in reference 18 that a field-induced phase transformation from T-phase to O-phase results in the observed polarization wake-up, a peak shift consistent with the proposal was not resolvable by XRD. Moreover, other changes during wake-up (e.g., a possible decrease in XRD peak width) were within the error range of the measurements, as noted by the authors. While most of these prior reports examine a particular doped or alloyed hafnia-based ferroelectric composition, [13,[16][17][18][19][20][21] many of these studies do not benchmark their observations against a control sample.
In this work, the wake-up phenomenon is studied in Hf 0.5 Zr 0.5 O 2 (HZO) films using non-destructive measurement techniques. Two otherwise identically processed MFM capacitor arrays are studied with the HZO film grown at different atomic layer deposition (ALD) temperatures and have been examined exhaustively. [22] The two MFMs, ALD 200 °C sample, and ALD 250 °C sample, show drastically different polarization evolution as a function of the number of switching cycles and, therefore, serve as an ideal observation set. Synchrotron low incidence XRD and capacitance-voltage (C-V) measurements are used to test whether the observed evolution of switchable polarization with switching cycle number is caused by a field-induced HZO phase evolution of TiN/HZO/TiN capacitor stacks. Piezoresponse force microscopy (PFM) imaging of the HZO film surface as a function of voltage programming supplements the diffraction and electrical measurement results.

Synchrotron X-Ray Diffraction
The diffraction setup is shown schematically in Figure 1a. X-rays are incident on an array of 40 capacitors with 140 µm × 40 µm device size. Figure 1b,c shows the polarizationvoltage (P-V) hysteresis for the two samples. The ALD 200 °C sample shows a pinched loop in the pristine state, while the ALD 250 °C sample shows an ideal ferroelectric loop with high remnant polarization. This illustrates the impact of deposition temperature on the film properties of otherwise identically processed capacitors. A dissimilarity is observed in the diffraction pattern for the two samples in the pristine state, as presented in Figure 2a, which matches the previous report. [22] The main HZO film peak (labeled as the O/T peak) is observed at Q = 2.137 Å −1 for ALD 250 °C and at Q = 2.155 Å −1 for the ALD 200 °C, while the main Pt peak in Figure 2b is consistent for both ALD temperatures at Q = 2.79 Å −1 , eliminating the possibility of any alignment error. The ALD 250 °C sample is reported to have majority polar orthorhombic, O-phase (space group Pca2 1 ), with the O/T peak at lower Q values while the ALD 200 °C sample has the tetragonal, T-phase (space group P4 2 /nmc), with the O/T peak at higher Q positions. [22,23] This also explains the different P-V hysteresis data for the two samples. The switched polarization, as measured by the positive-upnegative-down (PUND) sequence, is plotted in Figure 2c www.advelectronicmat.de polarization, a 3000% increase after 10 7 switching cycles, whereas the ALD 250 °C sample has a minimal polarization increase with the number of switching cycles. The great differences in polarization endurance during voltage cycling of these samples are used to study the wake-up effect.
Diffraction measurements are performed on the two samples after different numbers of switching cycles to observe phase evolution with field cycling. The cycled conditions chosen for the XRD experiments are shown as black stars in Figure 2c. For the ALD 250 °C sample, diffraction patterns in the pristine state and after 10 4 switching cycles are collected. As shown in Figure 3a, no variation in the O/T HZO peak positions with voltage cycling is observed. This observation is in good agreement with the voltage cycling measurements of the ALD 250 °C sample, that do not show any significant polarization wake-up. Figure 3b shows diffraction data for the ALD 200 °C sample in the pristine state, and after 10 5 and 10 7 switching cycles. A peak shift is observed for the O/T peak to lower Q position with wake-up. The O/T peak shifts from Q = 2.155 Å −1 at pristine state to Q = 2.147 Å −1 after 10 7 switching cycles. However, the Pt peak position does not vary with field cycling, Figure 3c. This suggests that as the ALD 200 °C sample undergoes wake-up, a phase evolution in the HZO film occurs that can be observed as  www.advelectronicmat.de the O/T peak shift in the diffraction pattern. Since the shift is to the lower Q position, it is plausible that this sample undergoes a field-induced phase transformation from T-phase to O-phase, which is consistent with the observation from Figure 2a. The O/T peak width was also analyzed as a function of switching cycles for the two samples, see Supporting Information Table S1. There is no observed change in the O/T peak width for ALD 250 °C sample with cycling. There is a slight increase in the O/T peak width after 10 7 switching cycles, consistent with the O/T peak position shift.

Capacitance-Voltage Measurements
To further analyze the cycling behavior, C-V measurements on the MFM capacitors were performed. The C-V measurements were performed with a DC bias sweep between −3 V and 3 V and a 30 mV amplitude sinusoidal signal at 50 kHz frequency, Figure 4. The ferroelectric O-phase is reported to have a lower dielectric constant than the antiferroelectric T-phase. [21,24] Measurements conducted far from the peaks in the C-V curve are less influenced by domain wall contributions [25] to the capacitance and provide a more reliable measure of the dielectric constant. In the present case, the majority O-phase ALD 250 °C sample has a lower dielectric constant than the majority T-phase ALD 200 °C sample for all cycling states, Figure 4a,b. The dielectric constant for ALD 250 °C sample matches with the reported values for ferroelectric HZO capacitors with high switchable polarization. [26] Similarly, the ALD 200 °C sample has dielectric constant and C-V loops that match well with the reported anti-ferroelectric zirconia-rich HZO capacitors. [26] Across all cycling states, the ALD 250 °C sample has a pair of peaks, i.e., a butterfly loop which is a signature of ferroelectric capacitors, Figure 4a. [25][26][27] On the other hand, the ALD 200 °C sample shows two pairs of such peaks in the pristine state, Figure 4b,c, typical of majority anti-ferroelectric phase capacitors. [26,27] Figure 4c depicts a continuous evolution in the C-V loop shape for the lower deposition temperature sample during voltage cycling to switch the polarization switching until, after 10 7 switching cycles, it begins to resemble more closely a "butterfly" like loop with a single pair of peaks, suggesting a phase transition during voltage cycling.

Piezoresponse Force Microscopy
We also study the piezoelectric properties of these samples before and after cycling using PFM measurements on bare HZO surface without the presence of a top electrode. The MFM capacitors were cycled, and the top TiN electrode was removed using a wet chemical etch. The Pt/Ir coated conductive AFM tip (radius of < 25 nm) is used as an electrode to apply DC bias for poling the HZO film while the bottom electrode is grounded. Figure 5 shows PFM images that were measured in contact resonance frequency mode with a resonance frequency of ≈1 MHz and a drive amplitude of <600 mV. The amplitude and phase images show the polarization direction switched to the downward direction (yellow) in the region where the positive DC bias was applied. The topography images are shown in Supporting Information Figure S2. For the ALD 250 °C sample, the amplitude was saturated, and the polarization was fully switched in the area where a DC bias of 3 V magnitude or more is applied to the film in both the pristine state and after 10 4 switching cycles, Figure 5a,b. For the ALD 200 °C sample, no local domain switching was measured in the pristine state for any value of DC bias applied, Figure 5c. A fine pattern is observed in the phase image which may be correlated with surface features seen in the topography image, Figure S2c, Supporting Information. After cycling, however, the amplitude and phase images show contrast consistent with domain switching, the amplitude of which increases as the number of switching cycles increases from 10 5 to 10 7 cycles. This observed increase in domain switching can be linked to an increase in volume fraction of ferroelectric O-phase domains with increased field cycling. [28] Additionally, the amplitude of the PFM contrast for the ALD 250 °C sample is always higher than the amplitude for the ALD 200 °C sample, even after wake-up. This is consistent with the trend in polarization values measured during voltage cycling of the respective samples illustrated in Figure 2c.

www.advelectronicmat.de
We note that electrochemical reactions due to the applied DC bias can induce ferroelectric-like behavior in PFM measurements. [29][30][31] However, ferroelectric domains are observed outside the biased regions in Figure 5 which should not be affected by the electrochemical reactions. Non-ferroelectric HfO 2 can also give an electromechanical response from nonpiezoelectric effects. [29] The pristine state ALD 200 °C sample that undergoes the same bias treatment as the rest of the samples for PFM measurements and has majority of nonferroelectric T-phase does not exhibit any ferroelectric-like behavior, Figure 5c, indicating that electrochemical reactions are not the primary mechanism for the measured electromechanical response of these films. Therefore, the phase and amplitude changes observed as a function of DC bias should arise primarily from ferroelectric contributions. A quantitative analysis of the amplitude data is provided in Figure S3, Supporting Information. Additional measurements, beyond the scope of the present study, such as contact Kelvin probe force microscopy, [29,30] may provide insight into contributions to the PFM image contrast that arise from phenomena that are not simply related to ferroelectric polarization switching.

Conclusion
To summarize, we investigate the mechanism behind the "wake-up" observed in capacitors with HZO films grown at different ALD temperatures. Nondestructive techniques are used to avoid disrupting the local electrostatic and mechanical boundary conditions present in the thin film capacitors, to provide device-relevant interfaces, and to provide statistically meaningful analysis volumes to assess the structural changes that accompany field-induced polarization switching of the films. Synchrotron XRD analyses of otherwise-identical HZO film samples deposited under conditions that lead to very large polarization wake-up and control samples showing minimal wake-up demonstrate that polarization wake-up results from a field-induced phase change from an initially present tetragonal phase to the ferroelectric orthorhombic phase. Correlated C-V measurements and PFM imaging strongly support this conclusion. The potential for coexistence of multiple metastable phases of HfO 2 with similar formation energies [32] may facilitate such electric field-induced phase changes. To avoid the wake-up effect, it is important to employ process conditions for fabrication of pristine-state HZO thin film devices that maximize the O-phase fraction and minimize the T-phase fraction.

Experimental Section
Device Fabrication: For the capacitors, 10 nm thick HZO films were grown using plasma-enhanced ALD at 200 °C and 250 °C deposition temperatures and capped with 10 nm thick TiN top and bottom electrodes deposited by RF sputtering. A post-metal rapid thermal anneal (RTA) is performed at 500 °C in N 2 ambient for 30 seconds to crystallize the HZO film. Lithography is used to define the device area. Pt top contacts were deposited by e-beam evaporation and top TiN was selectively etched by SC1 solution to define the devices. The electrical behavior of these capacitors is measured using a semiconductor parameter analyzer (Keithley 4200-SCS) and the piezoelectric response was recorded using an atomic force microscope. (Asylum Research, MFP-3D). The XRD data were collected at beamline 2-1 at the Stanford Synchrotron Radiation Lightsource with an X-ray photon energy of 12 keV at an incidence angle of 1.2 degrees. To enhance the signal-to-noise ratio, rows of capacitors were defined and then electrically switched prior to XRD measurements to achieve defined cycled states. The HZO film surrounding the capacitors was etched so that only the capacitor stacks contribute to the recorded diffraction signal.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.