Thermal retention of atomic layer deposited Hf_0.5Zr_0.52 films using H₂O and O₂–H₂ plasma oxidation methods

The increasing demand of mobile communications and data transfer necessitates faster, lower power, and denser memory. Ferroelectric hafnium zirconium oxide (HZO) has become an intriguing candidate for the next generation of memory devices owing to the material system's inherent nonvolatility, ferroelectric stability below 10 nm, CMOS compatibility, and low power operation. Despite the promising possibilities, ferroelectric HZO faces reliability challenges toward the path to device integration. Cycling lifetime, or endurance, has been demonstrated up to 4 × 10¹⁰ switching cycles, but this still lags behind the 1 × 10¹⁵ switching cycles of commercial lead zirconate titanate (PZT).^1,2 Another pressing challenge is thermal retention, which is the ability of a ferroelectric device to maintain its polarization state when subjected to elevated temperatures. The two mechanisms responsible for loss in retention are thermal depolarization and imprint. Thermal depolarization is the temporary loss of polarization as the temperature of the ferroelectric film approaches the phase transition temperature.³ Thermal imprint is the change in coercive fields that cause the device to favor a particular data state.³ Studies have been performed to examine and improve the thermal retention of HZO devices,^4–10 but more developments are needed to match the 10 year retention at 125 °C achieved with PZT.¹ 

In this effort, the effect of the oxidizing source is explored on the thermal retention characteristics of HZO films. The first set of films utilized H₂O as the oxidizing source, while the second films used the recently reported treatment of sequential O₂ and H₂-plasmas, which was shown to improve the virgin state switched polarization and reduce the wake-up effect compared to H₂O oxidation.¹¹ Devices were fabricated in a TiN–HZO–TiN stack at 200 °C via atomic layer deposition (ALD). TiN layers were deposited with tetrakis(dimethylamido)titanium(IV) (TDMAT) and nitrogen plasma, while for the Hf_0.5Zr_0.5O₂ layer, tetrakis(dimethylamido)hafnium(IV) (TDMAH) and tetrakis(dimethylamido)zirconium(IV) (TDMAZ) pulses were alternated with two different oxidation methods: O₂-plasma oxidized followed by H₂-plasma treated (O₂–H₂ plasma films) and H₂O (H₂O films). 50 nm thick top contacts of platinum were sputtered and used as a hard mask to etch the exposed TiN in 65 °C H₂O₂, creating devices with areas of 1600 to 14 400 μm². Rapid thermal anneals (RTA) were then carried out at 300, 350, 400, 500, 600, and 700 °C for 20 s in N₂. Same state and opposite state thermal retention tests were collected at room temperature (∼25 °C) after baking up to 100 h at 125 °C. Switched polarization values were measured through positive-up-negative-down (PUND) pulses with rise/hold/fall times of 1–10–1 μs, taken with an Agilent 33500B waveform generator and a Tektronix TDS5104B oscilloscope. (Figure S1 shows the pulse sequence for PUND measurements and provides further information on the testing procedure.) Hysteresis measurements were also taken using the same equipment at 1 and 2 kHz.

The evolution of switched polarization as a function of annealing temperature was systematically examined for the O₂–H₂ plasma and H₂O oxidized HZO films. The switched polarization is plotted vs RTA anneal temperature in Fig. 1 for both oxidizing sources. For lower annealing temperatures (350, 400, 500, and 600 °C), the plasma process results in higher virgin state switched polarization values of 3.4, 24.5, 26.6, and 30.4 μC/cm², compared to 0, 6.5, 9.0, and 20.4 μC/cm² for the water oxidized films. It is of importance to note that the O₂–H₂ plasma film achieves a switched polarization of 3.4 μC/cm² when annealed at 350 °C, while the H₂O film is paraelectric with no switched polarization. This is consistent with other plasma based HZO films where ferroelectricity and a measurable switched polarization were observed for HZO films annealed at 300 °C and 350 °C.^12,13 When the annealing temperature is raised to 700 °C, however, the switched polarization of the plasma film decreases down to 11.4 μC/cm², while the switched polarization of H₂O film continues increasing up to 30.8 μC/cm². Ultra-thin HZO films are typically polycrystalline with a mix of orthorhombic (polar), tetragonal (nonpolar), monoclinic (nonpolar), and cubic (nonpolar) phases. The composition of the various phases comprising the films is dictated by the various formation energies. Dopant type and concentration,^14–16 annealing temperature and duration,^17–19 electrode choice,²⁰ and deposition type (ALD or PVD)²¹ can vastly influence the resulting ferroelectricity and switched polarization of the film. For ALD of HZO films, conditions such as deposition temperature²² and precursor/oxidation type,²³ or pulse and purge time^24–26 can also impact the ferroelectricity of the films. In this work, the O₂–H₂ plasma process is theorized to create oxygen vacancies in the film layer by layer during the deposition, resulting in an enhancement of the orthorhombic phase and virgin state switched polarization.¹¹ Post deposition annealing can also introduce oxygen vacancies into the film and enhance the orthorhombic phase;²⁷ thus, annealing the H₂O films up to 700 °C may explain the similar virgin state switched polarization as seen in the O₂–H₂ plasma film annealed at 500 °C. Annealing the O₂–H₂ plasma films at 700 °C may create an abundance of oxygen vacancies, which begin to have an opposite effect and degrade the virgin state switched polarization.²⁸ Since the O₂–H₂ plasma films annealed at 500 °C and the H₂O films annealed at 700 °C exhibit similar switched polarizations, the cycling endurance, leakage current, thermal retention, and imprint are investigated.

The cycling endurance and the leakage current of the O₂–H₂ plasma film annealed at 500 °C and the H₂O film annealed at 700 °C were compared. Both the O₂–H₂ plasma films and the H₂O films have similar virgin state switched polarizations of 27.4 μC/cm² and 30.8 μC/cm². Figure 2(a) reveals further similarities with the switched polarization of the two films remaining within 10% of each other through 10⁵ and after 10⁸ switching cycles and within 20% of each other between 10⁶ and 10⁷ switching cycles. Figure 2(b) shows that the two films also have similar breakdown characteristics when measured on virgin state devices. The H₂O films, however, have on average half an order of magnitude higher leakage current and break down half a volt earlier. Before subjecting to thermal bakes, the devices of both films were initialized with 10k switching cycles at room temperature utilizing a 1 kHz bipolar square wave. An average of five fresh devices were initialized and averaged for all bake times.

Same state switched polarization measurements capture the thermal depolarization of the films, as they are collected at room temperature (∼25 °C) immediately after the applied elevated temperature, when the polarization has been temporarily reduced. Since thermal depolarization must be captured by the first read pulse following the bake, single P, U, N, or D reads were applied over a minimum of four devices per bake time. (Two devices baked in a positive polarization, for N and U reads, and two devices baked in a negative polarization, for P and D reads.) As seen in Fig. 3, both the O₂–H₂ plasma and H₂O films show robust thermal retention at 125 °C. The O₂–H₂ plasma films start with a P–U switched polarization of 39 μC/cm² and end with a switched polarization of 26 μC/cm² after 100 h at 125 °C. The |N–D| switched polarization shows a more stable trend, dropping from 40 μC/cm² to 34 μC/cm² after 1 h but only dropping 3 μC/cm² down to 31 μC/cm² after 100 h at 125 °C. The H₂O films, however, fare even better with both P–U and |N–D| switched polarizations remaining above 40 μC/cm² for 100 h. Grazing incidence x-ray diffraction (GIXRD) scans (Fig. S4) collected on both films show similar peak locations near 30°, 35°, 43°, 51°, and 60° 2θ, typical of polycrystalline HZO films. The highest intensity peak around 30° is measured to be 30.59° for the O₂–H₂ plasma film and 30.57° for the H₂O film. While the slightly lower 2θ position for the H₂O film could indicate a shift toward the orthorhombic phase, the difference is minimal. A larger difference between the peaks is seen in Fig. S4(d), indicating a broader peak for the H₂O film with a full width at half maximum (FWHM) of 1.1° compared to 0.92° for the O₂–H₂ plasma film. A broader peak can indicate an overlap of separate peaks or a mix of crystal phases. Interestingly, the H₂O and O₂–H₂ peaks have similar bounds at the higher 2θ, but the broadness of the H₂O peak extends further than the O₂–H₂ plasma film down to the lower 2θ. This can also indicate a shift toward the orthorhombic phase. While both processes are theorized to increase oxygen vacancies that help stabilize the orthorhombic phase (O₂–H₂ through oxide reduction during the H₂-plasma treatment or the H₂O films due to oxygen scavenging by the TiN electrodes during the higher temperature annealing), the broader GIXRD peak and better thermal retention of the H₂O film may indicate higher stability of the orthorhombic phase, as compared to the O₂–H₂ plasma film.

While the thermal depolarization of both films shows promise, same state retention bake measurements can often convolute thermal depolarization with the imprint effect (the tendency for a particular polarization state to become favored with exposure to extended elevated temperatures). One way to decouple these effects is to perform a second read (restore pulse) after the bake step by re-writing and re-reading the data state stored at room temperature. Also shown in Fig. 3 are the subsequent restore pulses. As can be seen, there is minimal difference between the bake and restore pulses, with the largest gap of 3 μC/cm² observed for the O₂–H₂ plasma film after 100 h at 125 °C. Such negligible changes between bake and restore pulses imply minimal thermal depolarization. The P–U switched polarization of the O₂–H₂ plasma film for both bake and restore pulses, however, drops as a function of bake time, with the largest drop of 13 μC/cm² observed after 100 h at 125 °C. This effect is most likely attributed not to thermal depolarization then but imprint. One method to examine the effect of imprint is to perform opposite state retention measurements. Opposite state switched polarizations were measured by placing the devices in a polarization state opposite to the state that they were baked in and then applying an appropriate read pulse. For example, devices baked in an N state would then be written into a P state followed by an N or U read, while devices baked in a P state would be written into an N state followed by a P or D read. Figure 3 shows the opposite state retention in terms of P–U and |N–D| switched polarizations. The H₂O films again show robust retention characteristics with switched polarizations of 32 μC/cm² and 27 μC/cm² after 100 h at 125 °C for P–U and |N–D|, respectively. The O₂–H₂ plasma films, however, show a complete loss of retention after 100 h at 125 °C, with P–U and |N–D| switched polarizations of 4 and 5 μC/cm² after 50 h.

To further examine the imprint effect, hysteresis loops were collected at room temperature for all devices after same and opposite state measurements. The overlaid hysteresis loops can be seen in Fig. 4 below. For devices baked in a P state, both O₂–H₂ plasma and H₂O films show a large initial shift in the coercive fields from time 0 to 1 h at 125 °C and then minimal to no shift in coercive fields from 1 h to 100 h at 125 °C. Investigating further, the H₂O films show an expected symmetric, equal initial shift in both positive and negative coercive fields of 0.5 MV/cm (1.8 to 1.3 MV/cm and −0.9 to −1.4 MV/cm from time 0 to 1 h at 125 °C). The O₂–H₂ plasma films, however, oddly show an asymmetric initial shift in coercive fields with the positive coercive field shifting by 0.6 MV/cm (1.8 to 1.2 MV/cm) but the negative coercive field only shifting 0.2 MV/cm (−1.2 to −1.4 MV/cm). For devices baked in an N state, the H₂O films show a minimal symmetric shift in the positive and negative coercive fields of 0.2 and 0.14 MV/cm from time 0 to 100 h at 125 °C. The O₂–H₂ plasma films again show an asymmetric shift, but in this case, the negative coercive field continues to shift starting at −1.2 MV/cm at time 0 and shifting all the way down to −0.4 MV/cm after 100 h at 125 °C. This continued shift in the negative coercive field for the O₂–H₂ plasma film helps to elucidate the observed loss in opposite state retention. For a device baked in an N state, a small negative coercive field would not significantly impact same state retention measurements, P and D reads would still yield the expected results. When measuring opposite state retention, however, the devices would need to be placed in a P state before N and U reads are applied. With a small negative coercive field, it is possible for the P state polarization to be lost to reverse switching into the more favorable N state polarization. Subsequent N reads would result in a low integrated polarization and U reads into a high integrated polarization. (Essentially appearing as D and P reads, respectively.) This can be confirmed by examining the individual P, U, N, D polarizations from Fig. 3 (Figs. S2 and S3). For the O₂–H₂ plasma films, after 100 h at 125 °C, a same state N read polarization has an average integrated value of 36 μC/cm², while an opposite state N read polarization has an average integrated value of 12 μC/cm², a decrease in 24 μC/cm². Similarly, after 100 h at 125 °C, a same state U read polarization has an average integrated value of 0.1 μC/cm², while the opposite state U read polarization has an average integrated value of 26 μC/cm², an increase in 26 μC/cm². Since the opposite state polarization is reported as P–U and |N–D| values, a dramatic decrease in N integrated polarization and increase in U polarization results in the observed loss of retention.

With the observed electrical behavior of the O₂–H₂ plasma film consistent between opposite state PUND and hysteresis measurements, it is natural to wonder why the loss of opposite state retention and asymmetric shifting of the hysteresis loops is not observed for the H₂O films, which show robust opposite state retention and symmetric hysteresis shifts. One explanation could be due to the asymmetry in the deposition for each of the processes. For the O₂–H₂ plasma films, the alternating O₂ and H₂ pulses are applied during the deposition of the HZO layer. The O₂-plasma pulse could be oxidizing the bottom electrode to form a TiO_xN_y interfacial layer. Although it has been shown that treating TiN after deposition with in situ ALD H₂-plasma can improve the metallic quality of ultra-thin (11 nm thick) TiN by reducing oxidation,²⁹ the relative time of the O₂-plasma (20 s) vs H₂-plasma (5 s) treatments suggests that the formation of a TiO_xN_y interfacial layer is more likely. Theoretically, the formation of an interfacial layer at the bottom interface may explain the asymmetric shifts in the hysteresis in the O₂–H₂ plasma film, as proposed by Zhou et al.³⁰ 

In conclusion, deposition of ferroelectric HZO films was explored for two different oxidation methods, O₂–H₂ plasma and H₂O. It was found that similar polarization values of 30.4 μC/cm² and 30.8 μC/cm² were achieved for the O₂–H₂ plasma films annealed at 500 °C and H₂O films annealed at 700 °C. Thus, these films were selected to explore the thermal retention characteristics. Both O₂–H₂ plasma and H₂O films displayed strong same state thermal retention with P–U and |N–D| switched polarization values above 25 μC/cm² after 100 h at 125 °C. O₂–H₂ plasma films, however, showed asymmetric shifts in the coercive fields and subsequent loss of opposite state retention after 100 h at 125 °C, while the H₂O films exhibited symmetric shifts in the coercive fields, and P–U and |N–D| opposite state thermal retention above 25 μC/cm² after 100 h at 125 °C. Exploring and improving the thermal retention of ferroelectric HZO films is a critical step in the near future for realization of commercial devices.

See the supplementary material for the pulse sequence and testing procedure for PUND measurements and the GIXRD scans.

This work was supported, in part, by NSF Grant No. ECCS 1610387. The authors acknowledge the use of the Research Service Centers at the Herbert Wertheim College of Engineering at the University of Florida and would like to thank Dr. Brent Gila and Andres Trucco for their invaluable insights and support. The authors would also like to extend a special thanks to Kristy Schepker for her appreciated support on GIXRD characterization.