Reliability Improvement from La2O3 Interfaces in Hf0.5Zr0.5O2‐Based Ferroelectric Capacitors

Further optimization of a typically reported ferroelectric capacitor comprised of a Hf0.5Zr0.5O2 ferroelectric thin film with TiN electrodes is explored by introducing an additional non‐ferroelectric La2O3 interfacial layer evaluated at different positions in the capacitor stack. The role of the interface to the ferroelectric layer is investigated and discussed, with the main focus directed toward the reliability of the device for non‐volatile memory applications. With this investigation, different degradation mechanisms determining electric field cycling and polarization retention are observed, and it is concluded that modifying the bottom interface between the electrode and the ferroelectric layer has the best potential to provide a benefit in device performance.


Introduction
Driven by the increasing data volumes that are produced in a given time as a result of the constant advancements in information technology, [1] the ability of electronic gadgets to manage and store these large amounts of data has become increasingly important. Accordingly, a novel, fast, low-power, and reliable memory technology is needed.
There are many novel non-volatile memory technologies under development that seek to satisfy the current memory demand. Each of these technologies exploits a particular material property in order to store data. For instance, ferroelectric memory technology utilizes the stable electric polarization states of a ferroelectric material to store data. Two prominent examples of this technology are ferroelectric random-access memory polar o-phase fraction and also leads to stronger depolarization fields. These two effects combined lead to a lower initial P r , a large wake-up effect during electric field cycling, and reduced data retention. [9,15] As a result, doping HZO thin films with La suggests potential improvements in their ferroelectric behavior but also disadvantages in data retention.
As a result, La 2 O 3 doping in HZO can be beneficial to improve device reliability but has a disadvantage in phase stabilization, wake-up behavior, and retention. Former publications started to characterize the impact of interlayers in HZO. [12,[16][17][18] As a main result, ultrathin Al 2 O 3 interface layers, having a thickness of ≈1 nm, at the top or bottom interface to the electrodes were shown to reduce leakage and enhance field cycling endurance, but detailed studies on retention and degradation mechanisms are missing. In an initial work on La 2 O 3 in comparison to Al 2 O 3 interlayers, La 2 O 3 was found to be more beneficial than the corresponding Al 2 O 3 layers for reliability improvement. [12] Based on this initial work, a more detailed characterization of different La 2 O 3 thicknesses at different positions in the HZO capacitor stack is reported in this manuscript. The effect of a La 2 O 3 interlayer on device reliability, such as retention in different stages of electrical field cycling, is also analyzed. In addition, a relationship between the degradation mechanisms affecting both reliability aspects is discussed.

Experimental Section
To avoid the impact on phase transition by La doping but keep the leakage current reduction capability of La, in this work, La 2 O 3 interfacial layers are introduced into an HZO-based thinfilm ferroelectric capacitor structure in three distinct configurations, as shown in Figure S1 (Supporting Information). The three configurations are referred to as the mid-interfacial layer (La 2 O 3 @MI), bottom interfacial layer (La 2 O 3 @BI), and top interfacial layer (La 2 O 3 @TI). The first configuration divides the HZO layer into two equally thinner films, and the second two separate the TiN electrode from the HZO layer. In addition, a standard capacitor with no La 2 O 3 interface layer was also fabricated for reference purposes.
For the fabrication of the capacitors, Hf 0.5 Zr 0.5 O 2 and La 2 O 3 thin films were deposited via atomic layer deposition (ALD) at 300 °C using O 2 plasma as an oxygen source. 3 and tris(isopropyl-cyclopentadienyl)lanthanum (La-(iPrCp) 3 ) were used as precursors for Hf, Zr, and La. The O 2 plasma was generated via an inductively coupled plasma source at 300 W and a 20 sccm O 2 gas flow rate. The combined dielectric film, comprised of 10 nm Hf 0.5 Zr 0.5 O 2 and additional La 2 O 3 , is deposited at the bottom, center, or top electrode interface. The La 2 O 3 thickness is varied between 6, 9 and 15 ALD cycles, which result in ≈2, 3, or 5 monolayers of the material. Based on a growth rate of 0.9 nm/ALD cycle, an interlayer thickness of ≈0.6, 0.8, and 1.4 nm can be assumed. TiN layers were used as bottom and top electrodes and were deposited via sputtering at room temperature using a Ti target and N 2 plasma.
A Brucker D8 Discover (Cu-Kα radiation with λ = 0.154 nm) was used for grazing incidence x-ray diffraction (GIXRD) (phase determination) and x-ray reflectivity (thickness) measurements. Electrical characterization was performed with an aixACCT TF Analyzer 3000 for dynamic hysteresis measurement (DHM), endurance, and relaxed polarization by applying the signal to the top electrode while keeping the bottom electrode grounded. The hysteresis measurements were carried out using triangular voltage signals with a maximum amplitude of 4 MV cm -1 and a frequency of 10 kHz. For preconditioning (wake-up) and endurance measurements, square voltage pulses with an amplitude of 4 MV cm -1 and 100 kHz frequency were applied. A Cascade Microtech probe station was used for retention measurements. The exact measurement procedure can be found in the literature. [19] At the same time, the imprint values are extracted from the retention data. Impedance spectroscopy was conducted with Zurich Instruments impedance analyzer MFIA in a frequency range from 1 mHz to 5 MHz.

Results and Discussion
In this study, the impact of La 2 O 3 interlayers at different positions within a TiN/ Hf 0.5 Zr 0.5 O 2 /TiN capacitor stack is evaluated. For this purpose, three types of capacitors are fabricated: one with 0.6 nm of La 2 O 3 at the bottom interface to the TiN electrode, another with a layer at the top interface, and a third with the same 0.6 nm La 2 O 3 in the center of the HZO film (0.6 nm corresponds to about two monolayers, which means that a closed layer is expected). In addition, La 2 O 3 layers with different film thicknesses from 0.6 to 1.4 nm are introduced at the bottom or top electrode interface. For initial structural characterization, GIXRD measurements are performed (Figure 1a,b). All diffraction patterns show a maximum at a 2theta value of ≈30.5°, which can be attributed to the o(111) or t(011) reflection of the orthorhombic Pca2 1 or the tetragonal P4 2 /nmc phase with only minor contributions of the monoclinic P2 1 /c phase at 2theta angles of 28.5 (m(-111)) and 31.5 (m(111)). Since only minimal differences are visible, deconvolution of the patterns is performed assuming constant stress/ strain for the HZO layer. Under this condition, the GIXRD patterns can be deconvoluted, as shown in the literature, [20] with a constant o(111) and t(011) peak at 30.4° and 30.8°, respectively (see SI, Supporting Information). The resulting phase contents are shown in Figure 1c,d. Placing only 0.6 nm La 2 O 3 at the top or bottom position in the film stack has only a minor impact on the phase content compared to the HZO film without an interlayer. The o-phase content remains between 48% -58%, with ≈20-30% t-phase and m-phase ( Figure 1b). In contrast, a clear reduction of the m-phase content to ≈0%, compensated by an increase of t-phase content to 50%, is visible for a La 2 O 3 interlayer in the middle of the HZO. The o-phase content is constant at 50%.
Having a further look at the impact of the La 2 O 3 interlayer thickness on the HZO phase content in Figure 1d, increasing La 2 O 3 thickness leads to a reduction of the o-phase from ≈60 to 40% parallel to an increase in t-phase from 20 to 35% with an almost constant m-phase content of ≈20%. Trends are similar for La 2 O 3 at the top and bottom interface. The peak width of the reflection at ≈30.6 is almost constant at ≈Δ2theta = 1° except of the La 2 O 3 in the middle interlayer case. Approximating the www.advmatinterfaces.de mean HZO crystallite size based on this peak width with the Debye-Scherrer equation, [21] the crystallite size for the middle interlayer is 6 nm which is roughly half of the crystallite size obtained for the rest of the cases. Results are similar to other literature studies with Al 2 O 3 interlayers in the center. [22,23] In 10 nm HZO films, crystallites typically extend from the bottom to the top interface, as opposed to a middle interlayer where crystallite growth is restricted, resulting in two HZO layers separated by the middle interlayer.
In the next step, field cycling endurance measurements are performed by applying a maximum field of 4 MV cm -1 with a cycling frequency of 100 kHz to the capacitors. For the sample set with 0.6 nm of La 2 O 3 at different positions of the interlayer, the change in P r with cycling, polarization versus voltage, and switching current versus voltage are plotted in Figure 2. All samples, except for the one with La 2 O 3 in the center of HZO, show a main ferroelectric switching peak at ≈1.2 MV cm -1 leading to a typical 2·P r value of 30-40 µC cm -2 , [24] which is slightly improved during wake-up cycling. Having a detailed look at the samples with 0.6 nm La 2 O 3 at different positions within the capacitor stack, the HZO film without interlayer has the highest 2·P r value of all samples during the complete cycling range. The ferroelectric switching peak has a small back-switching shoulder at zero field, which disappears almost entirely during field cycling ( Figure 2d). This change is typically attributed to several effects like field-induced phase changes or domain depinning, [25] leading to a slight 2*P r increase during wake-up cycling of <10%. The introduction of the La 2 O 3 layer at the top or bottom interface reduces the pristine 2*P r value for both cases down to 31-33 µC cm -2 , which can also be seen by www.advmatinterfaces.de the reduction of the ferroelectric switching peak with a minor impact on the back-switching. This trend of reduced ferroelectric content due to an interlayer is already visible in XRD measurements (Figure 1a,c), where the interlayers caused a slight reduction in o-phase content.
Only for the La 2 O 3 in the center position a clear ferroelectric switching and back-switching is visible, as typically seen for anti-ferroelectric-like HfO 2 /ZrO 2 -based capacitors. A higher field is required for a t-to o-phase transition leading to an enhanced switching field of 2.5 -3 MV cm -1 compared to pure ferroelectric switching at ≈1.2 MV cm -1 . Since back-switching is occurring at zero field, initially, only a remanent polarization of 20 µC cm -2 remains. Only after wake-up is the field-induced polarization converted to a clear ferroelectric polarization with 2*P r of 36 µC cm -2 . This matches well with the higher t-phase content, which was already determined by GIXRD measurements for this sample (Figure 1c).
All samples indicate fatigue after ≈10 5 cycles, but only the sample with La 2 O 3 at the top electrode showed enhanced leakage degradation (Figure 2d). Additionally, due to higher wake-up, lower P r, and stronger fatigue, samples with La 2 O 3 interlayers in the center of the HZO layer are not further evaluated in subsequent studies. One reason could be the quality of the La 2 O 3 interlayer. For La 2 O 3 deposition at the bottom TiN electrode, a subsequent HZO deposition leads to further oxidation of La 2 O 3 and possible removal of defects like oxygen vacancies. In contrast, this additional oxidation of La 2 O 3 is not occurring for the interlayer at the top interface. Here, TiN is instead pulling oxygen out during the crystallization anneal. [26,27] The more oxygen vacancy-rich top La 2 O 3 is expected to degrade earlier during field cycling. For a La 2 O 3 interlayer in the center of HZO, a smaller grain size is detected. This could lead to a higher grain surface, which is typically more defect rich.
To understand which interlayer thickness might be optimal at the top or bottom electrode, a sample set with a La 2 O 3 thickness split from 0 -1.4 nm La 2 O 3 was fabricated and characterized. Comparing field cycling endurance measurements, all samples with La 2 O 3 at the top electrode showed stronger leakage degradation and earlier breakdown (Figure 3b), similar to the 0.6 nm La 2 O 3 case (Figure 2a). As discussed before, the HZO sample without interlayers has stable remanent polarization until breakdown, but all samples with a La 2 O 3 interlayer at the top electrode indicate a leakage increase already after www.advmatinterfaces.de one-hundred field cycles leading to round-shaped hysteresis curves and drastic apparent P r increase. The effect is enhanced for larger interface thicknesses. In contrast, an introduction of La 2 O 3 at the bottom electrode is not showing such a strong degradation during field cycling (Figure 3a). Accordingly, the main focus remains on samples with La 2 O 3 at the bottom electrode.
The highest remanent polarization corresponds to the HZO capacitor without the La 2 O 3 interlayer. Introducing a 0.6 nm La 2 O 3 thick bottom interlayer reduces the 2*P r value to 33 µC cm -2 . Further La 2 O 3 thickness enhancement to 0.8 and 1.4 nm reduced the overall remanent polarization (Figure 4a). This correlates nicely with the o-phase content, as extracted from GIXRD. The main improvement by the interlayer is the improved field cycling endurance: since field cycling measurements were stopped after 10 7 cycles, no detailed evaluation can be done, but all samples with an interlayer have higher endurance values compared to the pure HZO layer.
Fatigue is worsened for larger La 2 O 3 thicknesses, as revealed by a reduced remanent polarization likely caused by domain pinning due to enhanced charge injection. For 1.4 nm La 2 O 3 , fatigue is occurring already after the first switching cycle (Figure 4a). Accordingly, the best endurance behavior with the lowest fatigue is achieved with the 0.6 nm La 2 O 3 interlayer sample.
The pristine switching peak position is shifting to higher fields with increasing La 2 O 3 thickness, indicating an increase in the internal bias field from 0 to 0.6 MV cm -1 (Figure 4b). This hysteresis shift is calculated from the average shift in switching field position for both polarization directions. The switching peak shift can be caused by the enhanced field needed for ferroelectric switching due to the potential drop in the interlayer and charges present in the layer. During field cycling, the value can be drastically reduced below 0.2 MV cm -1 , indicating charge redistribution or interface breakdown, as later discussed in the manuscript. In addition, polarization measurements are performed again after a 1 s delay to evaluate the impact of depolarization fields caused by the non-ferroelectric nature of the La 2 O 3 interlayer. [28] As seen in Figure 4b, the ratio between the relaxed polarization after 1 s and the non-relaxed polarization is enhanced from 83% to 93% for the pure HZO case with field cycling. With increasing interlayer film thickness, a stronger relaxation is visible, as expected, due to higher depolarization fields. For 1.4 nm La 2 O 3 , relaxed polarization is only 67% for the pristine sample and 73% after field cycling.

www.advmatinterfaces.de
For a better understanding of the degradation mechanisms during field cycling, impedance spectroscopy measurements were performed on an HZO capacitor without a La 2 O 3 interface ( Figure 5) and compared to capacitors with a 0.8 and 1.4 nm La 2 O 3 bottom interfacial layer ( Figure 6). As discussed in literature, during ALD deposition of the dielectric/ferroelectric layer on a TiN bottom electrode, the TiN layer is slightly oxidized, and a thin TiO x N y layer is formed. [27] Once a TiN top electrode is deposited, the layer stack is crystallized during a post-metallization anneal. Here, an additional TiO x N y layer at the top electrode interface is formed. Impedance spectroscopy can only distinguish dielectric layers with a different dielectric constant, as discussed later. Accordingly, similar to former publications, [13] a film stack as illustrated in Figure 5a is assumed. Nyquist and Bode plots are shown in both figures and fitted to a model presented in Figure 5a consisting of a resistive element for the electrodes, an resistor-capacitor in parallel (R-C) element for the bulk HZO film, and an R-CPE (CPE stands for constant phase element) for the TiO x N y interface regions as discussed in literature. [15] The reason for fitting the interface component with a CPE instead of a standard capacitive element, as done for the bulk, is due to the expected non-ideal capacitive behavior of the interface given its reduced thickness. Looking at the HZO capacitor without a La 2 O 3 interface layer ( Figure 5), the modeled data fits suitably with the experimental data, as indicated by the quality of the fit described by a chisquare of 0.0040 with a confidence level of 96%. With this, resistance values extracted from the measurement were: 150 Ω for the TiN electrodes, 505 MΩ for the HZO dielectric layer, and 780 Ω for the TiO x N y interface layers.
Comparing the results from Figure 5 to capacitors with an HZO layer with a La 2 O 3 bottom interfacial layer (Figure 6), a "two capacitors in series" model with distinct capacitances for HZO and La 2 O 3 , must be accounted for. To further confirm the presence of the interface layer, the imaginary part of the modulus and impedance is plotted as a function of frequency (Figure 5d). The imaginary part of the modulus has two maxima, highlighting the two different components in the dielectric layer. [29][30][31] Further details are discussed in the SI (Supporting Information).
Distinguishable properties of both materials caused by a different current flow path were expected [14] in the impedance spectroscopy analysis. That said, and as can be seen from Figure 6 in comparison to Figure 5, no significant difference was observed after introducing the interfacial layer, regardless of the thickness of the La 2 O 3 . The reason behind this is likely the similar dielectric constant of HZO and La 2 O 3 of ≈20-30, [32] which does not allow for a distinction between both layers. In addition, as for the HZO layer without a La 2 O 3 interface, a CPE element was required for an adequate data fit, indicating once more the presence of a TiO x N y interlayer. Similar to before, the resistance values extracted from the measurement were: 150 Ω for the TiN electrodes, 517 MΩ for the HZO dielectric layer, and 770-780 Ω for the TiO x N y interface layers.
In the following analysis, the film properties are evaluated after electrical field cycling with 10 4 and 10 7 cycles using a 3 MV cm -1 signal amplitude. Extracted values are plotted as a function of the number of field cycles in Figure 7. For all samples, the HZO bulk resistance remained practically constant in a range of 500 to 525 MΩ. A slight interface degradation is visible, as shown by the reduction in interface resistance from about 770 Ω to values of 730 Ω and below for all three samples. At the same time, the n-value of the CPE element, indicating the resistive to capacitive nature of the interface, is almost constant at ≈0.6, predicting a predominantly capacitive behavior of the interface. Overall, the HZO layer without the La 2 O 3 interface degrades much faster than the samples with La 2 O 3 , which correlates nicely with the field cycling endurance results. In addition, the cause of degradation can be explained by the deterioration of the TiO x N y interlayer, which is expected due to the high fields experienced in this thin film during polarization switching. [8] In effect, the La 2 O 3 layer may be invisible to impedance spectroscopy measurements due to its similar dielectric constant to HZO, but it clearly shows the improvement in field cycling endurance and leakage current. The remaining question would be whether a positive effect on polarization retention is visible.
To be classified as a non-volatile memory element, according to Joint Electron Device Engineering Council (JEDEC) industry standards and test specifications, the technology must show at least ten years of data retention at an elevated temperature www.advmatinterfaces.de (e.g., 85 °C) in the absence of any external power supply. [33] Following this criterion, the data retention test is performed for the different bottom interface configurations at different cycling stages. The details about the measured polarization states, that is, new same state (NSS), opposite state (OS), and same state (SS), and the corresponding measurement procedure can be found in literature. [19] Before presenting the experimental results of the retention test, it is important to mention that the data retention loss in HfO 2 -based ferroelectric materials is mainly caused by two independent physical mechanisms: [34] a built-in bias imprint effect (Figure 8a,b) and depolarization field induced back-switching. Thermal depolarization can also take place but is less significant under the tested conditions. In view of the previously shown results, 0.8 nm La 2 O 3 at the bottom interface is considered here as an example case of a device with a small non-polar t-phase portion and is therefore suitable to discuss retention loss due to imprint. The positive polarization charge at the metal-ferroelectric interface attracts negative charges either from charged oxygen movement or the trapping of electrons, while the opposite effect happens at the opposite metal-ferroelectric interface. [35,36] This accumulation of charge creates a bias field within the material that supports the downward polarization orientation while opposing the upward orientation. In an extreme case, this bias field can even become so strong that it permanently pins the domains in one direction and makes it unswitchable to the opposite one when a constant switching field is applied. [34] In Figure 8a,b, the difference in shifting of the respective I(E) curves for a cycled and pristine capacitor can be observed. The main difference is the broadness of the initial switching peak for the pristine case. That said, from Figure 8c it is evident that the cycling stage of the capacitor does not have a significant impact on the behavior of imprint in the initial phase with only slightly higher imprint after cycling. In contrast, from Figure 8d, increasing the interfacial layer thickness does seem to increase imprint drastically. This would indicate a difference in the degradation mechanism for retention compared to field cycling: unlike field cycling, where a progressive increase in cycles leads to a larger number of defects within the film and eventual breakdown of the device, imprint, and therefore Figure 5. a) Electrical equivalent circuit used for the impedance spectroscopy analysis: The different layers of the capacitor are assigned to the corresponding equivalent circuit elements. The bulk HZO is modeled using a parallel R-C element (marked in orange), the interface is modeled using a parallel R-CPE element (marked in blue), and the TiN electrodes are modeled using a resistor (marked in green). A b) Nyquist plot and c) Bode plot obtained from an impedance spectroscopy measurement of a standard HZO capacitor. The devices were probed from 1 Hz to 5 MHz with a 75 mV AC signal. The solid red lines in b) and c) describe the fit to the experimental data. Fitting was performed using the model shown in a). d) The imaginary part of the modulus and impedance plotted as a function of frequency for HZO.

www.advmatinterfaces.de
retention, is more dependent on the interface of the ferroelectric layer with the electrodes, where increasing an interfacial layer here leads to faster degradation as already indicated in the P r relaxation measurements discussed in Figure 4b. This is suggestive of a progressive charge accumulation in this region which generates the observed imprint. Furthermore, from Figure 8e, where the different retention states are compared, it is important to mention, first, that the 2•P r loss reaches ≈50% near ten years (>10 8 s), indicating that the polarization loss is not as significant as imprint and, second, that the state most influenced by imprint (OS) has the worst behavior. This highlights that the retention limit is not determined by the loss of polarization but by the shift of the hysteresis curve or, in other words, by imprint.
Overall, introducing an ultrathin interlayer into an HZObased capacitor might be beneficial for leakage and endurance properties, as formerly reported by several authors, [16][17][18] but other reliability concerns like retention and imprint need to be critically evaluated.

Conclusion
In summary, the use of La 2 O 3 interlayers in HZO-based capacitors was explored in this work regarding non-volatile memory applications. Starting with the most essential nonvolatile memory-related material property, the highest remanent polarization is present in HZO films without interlayers. Depending on the interface layer thickness, an increased reduction of the polarization is visible even though the interlayer was added to a constant HZO film thickness. The best position of the interlayer is at the bottom TiN electrode since the center-and top-position lead to numerous effects like P r reduction, wake-up increase, and leakage enhancement. The significant advantage of the interlayer is the field cycling endurance improvement likely caused by leakage reduction. That said, too large interlayer thicknesses lead to issues like P r reduction, charge accumulation at the interface, and retention degradation, as determined by P r relaxation and retention imprint measurements. La 2 O 3 seems to act as trap sites in addition to causing depolarization fields in HZO. Since the dielectric constant of La 2 O 3 is similar to HZO, the primary degradation during field cycling occurs in the lower k TiO x N y interfacial layer.   Figure 5a. Both Nyquist and Bode plots match very well with the experimental data, supported by a quality of the fit described by a chi-square of 0.0040 with a confidence level of 96%.
In conclusion, the ferroelectric-metal interface in a metalferroelectric-metal capacitor structure strongly impacts overall device performance concerning non-volatile memory application. Insertion of an interlayer causes a substantial change in the structural and electrical properties of the device, which then leads to a significant alteration of device performance as a non-volatile memory element. This work conducted a comparative study via interface engineering and concluded that covering the bottom electrode surface with a thin layer (<1 nm) of La 2 O 3 before depositing a 10 nm thick ferroelectric HZO on top improves the field cycling endurance performance of the ferroelectric capacitor as a memory element, but can be critical for other reliability concerns like retention. Accordingly, interlayers might be beneficial, but the interlayer thickness needs to be kept low. Further advances are necessary to improve imprint and retention.

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