Tiered deposition of sub-5 nm ferroelectric Hf_1-xZr_xO₂ films on metal and semiconductor substrates

Since the discovery of ferroelectricity in Si doped HfO₂,¹ numerous efforts have been made to explore a range of dopants which can induce ferroelectricity.² Of particular interest in the field of doped HfO₂ is the use of ZrO₂ due to the wide compositional window and low annealing temperature at which ferroelectric behavior is observed.³ Efforts have been made both experimentally and theoretically to characterize the compositional dependence of Hf_1-xZr_xO₂ (HZO) for the case of thin (<20 nm)^3–5 and thicker (>195 nm)⁶ films. Additionally, the effect of the bottom electrode has been investigated for the semiconductors Si and Ge⁷ and metals such as TaN,⁷ TiN/Ir,^8–12 and Pt.¹³ However, the effects of these parameters are understudied for sub-5 nm thick films (approaching the lower thickness window at which ferroelectricity has been achieved) in terms of ferroelectricity, wake-up, and endurance.

For metal-ferroelectric-metal (MFM) capacitors, either 20 nm of Ir was sputtered as a bottom electrode (Ir-BE) or 10 nm of TiN was deposited via atomic layer deposition (ALD) (TiN-BE) on a (100) Si wafer. Metal-ferroelectric-semiconductor (MFS) capacitors were deposited on a (100) oriented Ge substrate with a 6° tilt towards (111). Hf_1-xZr_xO₂ was grown via a Cambridge Nano Fiji 200 ALD system with Tetrakis(dimethylamido)hafnium(IV) (TDMAH) and Tetrakis(dimethylamido)zirconium(IV) (TDMAZ) as the precursors and H₂O as the oxygen source. A doping ratio for Hf:Zr of 3:1, 2:1, and 1:1 was used for the Hf_0.75Zr_0.25O₂, Hf_0.66Zr_0.33O₂, and Hf_0.5Zr_0.5O₂ compositions. The depositions were performed at 175 °C using an exposure mode type growth in which the Hf and Zr precursors were pulsed for 1.5 s, chamber flows stopped and held for 60 and 70 s, respectively, and finally the chamber purged for 90 and 100 s.

To systematically investigate a range of thicknesses, a tiered deposition approach was used to deposit the Hf_1-xZr_xO₂ thin films. After 36 ALD cycles were deposited, a subset of the films was removed from the chamber. Additional 4 ALD cycles were deposited before removing the next subset of films. This process was repeated up to 52 ALD cycles at which point all removed films were placed back in the chamber and a top 10 nm TiN electrode was deposited using Tetrakis(dimethylamido)titanium(IV) (TDMAT) and nitrogen plasma. Pt was sputter deposited and used as a hard mask to define squares ranging in area from 1200 to 14 400 μm². After deposition, a rapid thermal anneal was applied at 500 °C for 20 s in N₂. Electrical characterization was performed using an Agilent 33500B waveform generator and a Tektronix TDS5104B oscilloscope for polarization vs. voltage (P-V), endurance, and Positive-Up-Negative-Down (PUND) testing.¹⁴ High-resolution cross-sectional transmission electron microscopy (HR-XTEM) imaging of the structures was performed using an FEI Tecnai F20 S/TEM equipped with a Gatan UltraScan 1000P digital camera. Specimens for HR-XTEM imaging were prepared using an FEI Helios Nanolab 600 dual beam scanning electron/focused ion beam microscope; specimens were prepared such that the incident beam direction during HR-XTEM imaging was aligned with an in-plane ⟨110⟩ direction.

As determined from the HR-XTEM images in Figs. 1(a) and 1(c), the thicknesses of the 40 ALD cycle 1:1 HZO Ge-BE and Ir-BE films are ∼37 Å ± 1 Å. It can be seen in Fig. 1(c) that the HZO Ir-BE film is crystalline with well-defined lattice fringes, while in Fig. 1(a), the HZO Ge-BE film is amorphous with no clear lattice fringes. These results match the electrical data which demonstrate a measurable polarization from the hysteresis and PUND tests for the 40 cycle and up Ir-BE films and no measurable polarization for the Ge-BE films below 44 ALD cycles. From Figs. 1(b) and 1(d), it can be observed that the 52 ALD cycle Ge-BE and TiN-BE are both crystalline with well-defined lattice fringes and a thickness of ∼51 Å ± 1 Å.

Hysteresis curves for the different thickness films across the various substrates and doping ratios are shown in Figs. 2(a)–2(j). Hysteresis measurements were taken at 3.5 MV/cm at 1 and 2 kHz with a dynamic leakage current compensation performed.¹⁵ Wake-up initialization consisted of 10⁴ cycles of a bipolar pulse at 1 kHz. The first, and most apparent trend, is the drop in remanent polarization (Pr) with decreasing ALD cycles. For 1:1 HZO Ge-BE films [Figs. 2(c) and 2(h)], the Pr after wake-up [Fig. 2(h)] is 14.3 and 12.6 μC/cm² at 52 and 48 ALD cycles. The Pr drops dramatically to 4.2 μC/cm² at 44 ALD cycles and is nonexistent (<1 μC/cm²) at 40 and 36 ALD cycles. For 2:1 HZO Ge-BE films [Figs. 2(b) and 2(g)], the Pr after wakeup [Fig. 2(g)] is half of 1:1 HZO after wake-up with values of 7.0 and 4.6 μC/cm² at 52 and 48 ALD cycles and no measurable Pr below 48 ALD cycles. For 3:1 HZO Ge-BE films [Figs. 2(a) and 2(f)], the Pr after wakeup [Fig. 2(f)] is slightly higher than that for 2:1 HZO at 8.8 μC/cm² for 52 ALD cycles but abruptly drops to <2 μC/cm² for 48 ALD cycles and <1 μC/cm² for 44 ALD cycles and below. In terms of doping, it can be seen that 1:1 HZO Ge-BE films provide the highest Pr across all thicknesses and, furthermore, increasing the HfO₂ concentration seems to increase the minimum thickness needed to achieve a measurable Pr.

1:1 HZO TiN-BE and Ir-BE films also show a similar trend of decreasing Pr with decreasing ALD cycles but with some key differences compared to 1:1 HZO Ge-BE films. After wake-up, the 1:1 HZO TiN-BE [Fig. 2(j)] has a Pr of 4.1 μC/cm² at 52 ALD cycles and <1 μC/cm² for anything thinner. The 1:1 HZO Ir-BE films show a less abrupt trend after wakeup [Fig. 2(i)] with a Pr of 13.2, 10.8, 4.7, 2.2, and <1 μC/cm² for 52–36 ALD cycles. The small but distinct Pr at 40 ALD cycles supports the HR-XTEM results which show a crystalline structure [Fig. 1(c)]. This is in contrast to the 40 ALD cycle 1:1 HZO Ge-BE film which showed no crystallinity and no discernable Pr. At 52 ALD cycles, both the 1:1 HZO Ge-BE and TiN-BE films, which are crystalline as determined by HR-XTEM imaging, show hysteresis behavior. The results are consistent with the observation that thinner films may require higher temperatures to crystalize as shown experimentally for pure HfO₂ as well as in ferroelectric HfO₂ doped with silicon.^16,17 Additionally, the different bottom electrodes could result in variations in the surface energy which is known to have an impact on the crystallization temperature as well as the preferred crystal phase.^5,18

Another trend seen in Fig. 2 is the change in the virgin state Pr across the different bottom electrodes. For all Ge-BE films (3:1, 2:1, and 1:1) with a sufficient thickness, the virgin state shows ferroelectric behavior with a measurable remanent polarization. For 1:1 HZO Ir-BE films, the virgin state displays anti-ferroelectric behavior with no measurable Pr and anti-parallel dipolar alignment. For 1:1 HZO TiN-BE films, there is no ferroelectric or anti-ferroelectric behavior but instead initially a linear dielectric response for the thicknesses studied (3.7–5.1 nm). It is only after wake-up cycling that both the Ir-BE and TiN-BE films display ferroelectricity. The wake-up phenomenon has been studied extensively in HfO₂-based ferroelectrics and is generally attributed to a decrease in the internal bias within the bulk ferroelectric,^19,20 de-pinning of domains^21,22 (due to defect migration), and/or a field-induced phase transformation.^21,23–27 Using highly doped Ge as a bottom electrode has been observed before to exhibit a larger polarization and less “wake-up” than metal bottom electrode films.⁷ Ge-BE films may have a reduced number of trapped defects by virtue of the thinner oxide layer at the interface resulting in less internal bias or pinned domains. This is similar to the effect suggested by Florent et al. for the case of Al-doped HfO₂, Si insulator Si (SIS) devices.²⁸ 

While the hysteresis measurements in Fig. 2 provide a snapshot of the ferroelectric behavior, they do not fully capture the evolution of polarization as it would appear in a memory circuit due to the continuous triangle wave applied. To this end, a second evaluation was conducted using PUND tests with switched polarization (Psw) values of P-U reported. Tests were performed with a 1–10-1 μs pulse for the PUND pulses and a 1 kHz bipolar square wave for electric field cycling. An initial electric field of 3.5 MV/cm was applied and increased by 0.6 MV/cm until the device under test incurred dielectric breakdown. Psw (P-U) versus number of switching cycles is shown in Fig. 3 for Ge-BE, TiN-BE, and Ir-BE films for various thicknesses and electric fields. The Psw PUND results confirm the polarization trends seen from the hysteresis measurements. The 1:1 HZO Ge-BE films exhibit the highest Psw for 52 and 48 ALD cycles at 39.6 μC/cm² [Fig. 3(d)] and 34.0 μC/cm² [Fig. 3(c)] after 10⁴ switching cycles at 4.7 MV/cm but also display a rapid decrease in Psw at 44 ALD cycles {16.8 μC/cm² [Fig. 3(b)]} and no retention for the thinner films {Psw values >1 μC/cm² [Fig. 3(a)]}. The 1:1 HZO Ir-BE films continue to show a more graded drop in Psw with thickness with values of 27.1, 28.3, 21.0, and 13 μC/cm² [Figs. 3(d), 3(c), 3(b), and 3(a)] and are the only films to achieve retention at 40 ALD cycles with Psw values after 10⁴ switching cycles of 3, 7, and 13 μC/cm² [Fig. 3(a)] for fields of 4.7, 5.3, and 5.9 MV/cm. For 1:1 HZO TiN-BE films, only the 52 ALD cycle film displayed retention [Fig. 3(d)] and could be interrogated at an electric field of 6.5 MV/cm, 1.2 MV/cm, and 1.8 MV/cm higher than what the Ir-BE and Ge-BE films could withstand.

In terms of wake-up and fatigue, at 52 ALD cycles and 4.1 MV/cm, a wake-up effect is observed with the Psw for Ge-BE and Ir-BE films increasing from 7 and <1 μC/cm² in the virgin state to 35 and 17 μC/cm² after 10⁴ switching cycles [Fig. 3(d)]. Under the same conditions, the fatigue behavior is drastically different. The Psw of the Ge-BE film drops from 35 to 4 μC/cm² after 10⁶ switching cycles (89%), while the Psw of the Ir-BE film only drops from 17 to 13 μC/cm² (24%). The TiN-BE film at 52 ALD cycles shows an entirely different behavior with no fatigue apparent across all applied fields up to 10⁶ cycles. To further test the endurance properties of the 52 ALD cycle TiN-BE film, a second PUND test was performed up to 10¹⁰ cycles at a 100 kHz frequency (Fig. 4). 100 kHz allows for ease of testing at a greater number of switching cycles and also results in the wake-up effect being not as pronounced for the same number of switching cycles as in the 1 kHz case. This is attributed to wake-up being a function of time under bias as opposed to a number of switching cycles.^29,30 The 52 ALD cycle TiN-BE film at 4.7 MV/cm shows no sign of fatigue and is still in the process of wake-up throughout the test with the Psw value increasing from <1 μC/cm² at 10¹ switching cycles up to 21 μC/cm² at 10¹⁰ switching cycles.

The endurance properties of ferroelectric HfO₂ have been explored in numerous papers. Endurance as high as 10¹⁰ cycles for both Si-doped HfO₂ (Ref. 31) and La-doped HZO³² as well as 10⁹ switching cycles for HZO⁹ has been reported for 10 nm thick films. In those studies, a higher endurance was achieved by applying lower electric fields, as fields higher than 2.25–2.5 MV/cm would cause dielectric breakdown at a smaller number of cycles. The applied field of 4.7 MV/cm is higher than what has been reported and may be explained by the thin nature of the films. It is well-known in CMOS technology that the maximum breakdown strength of SiO₂ increases with the decreasing film thickness.^33,34 To further investigate the effect of the film thickness on endurance and dielectric breakdown, 1:1 HZO TiN-BE films of 100 and 240 ALD cycles (approximately 10 and 24 nm thick) were grown. For the thickest film of 240 ALD cycles, dielectric breakdown occurred at 3.5 MV/cm after 10⁶ switching cycles (Fig. 4). It was only at a lower field of 2.5 MV/cm that 10⁹ switching cycles were achieved. The thinner 100 ALD cycle film suffered from breakdown at a higher field of 4.7 MV/cm and achieved 10⁹ switching cycles at 3.5 MV/cm. The thinnest film of 52 ALD cycles did not suffer from breakdown at the highest applied field of 4.7 MV/cm and 10¹⁰ switching cycles. This behavior fits into the model of thinner films being more robust to dielectric breakdown than thicker films for the same applied field.

Of the three thicknesses tested, the 100 ALD cycle film exhibited the highest Pr with wake-up extending to 10⁸ cycles. The wake-up effect begins to taper off but still shows no signs of fatigue or dielectric breakdown at 10⁹ switching cycles and an applied field of 3.5 MV/cm. While the thickness is seen to have an effect on the maximum applied field, the 3.5 MV/cm field at 10⁹ switching cycles is still higher than that typically reported for a 10 nm HfO₂ MFM and may be due to the deposition conditions. The precursors utilized for HZO deposition were TDMA(H/Z) with H₂O as the oxidizer in place of the more commonly used tetrakis(ethylmethylamido) hafnium or zirconium (TEMA(H/Z) precursors and O₃ oxidizer. O₃ is typically preferred for higher wafer throughput since the required purge time is far shorter than for H₂O; however, one drawback is a higher carbon impurity concentration which has been shown to decrease the time to breakdown in HfO₂.³⁵ Liu et al. showed that increasing the deposition temperature lowers the carbon impurity level with O₃, which begins to reach values comparable to H₂O at a deposition temperature of 320 °C.³⁶ However, higher deposition temperatures also pose a potential issue with precursor decomposition (300 °C and 400 °C for TEMAZ and TEMAH).³⁷ Additionally, deposition conditions similar to the ones reported in this work have been shown to result in similarly high breakdown fields for Al and Si doped HZO.³⁸ 

In conclusion, the effects of the bottom electrode and doping ratios were explored for thin ferroelectric HZO films using a tiered deposition approach to systematically study the lower thickness limit with 1:1 HZO films grown on Ge, Ir, and TiN bottom electrodes and annealed at 500 °C. 1:1 HZO Ir-BE films exhibited a measurable hysteresis and PUND response at 40 ALD cycles or ∼37 Å. 1:1 HZO Ge-BE films displayed the highest Psw for 52 ALD cycles in both the virgin and woke-up states but were not ferroelectric at 40 ALD cycles. The 1:1 HZO TiN-BE films did not show ferroelectricity below 52 ALD cycles or ∼51 Å but displayed the best endurance of all the films studied with a wake-up effect continuing to 10¹⁰ switching cycles and no fatigue observed across all other thickness and tests.

This work was supported in part by NSF Grant No. ECCS 1610387. The authors acknowledge the use of the Nanoscale Research Facility (NRF) at the University of Florida and would like to thank Dr. Brent Gila and Andres Trucco for their helpful discussions and support. The Research Service Centers at the Herbert Wertheim College of Engineering at the University of Florida are acknowledged for use of the transmission electron microscope and dual beam scanning electron/focused ion beam microscope.