Introduction

Hafnia-based ferroelectric random-access memory (FeRAM) has attracted tremendous interest as one of the future candidates for nonvolatile memory applications because of its thickness scalability and CMOS (complementary metal oxide semiconductor) process compatibility1,2,3,4,5,6,7. However, its endurance characteristic, which is one of the crucial reliability properties of FeRAM, is not comparable to that of commercialized memory devices like DRAM (Dynamic Random Access Memory). The endurance characteristics are significantly affected by interface traps, such as oxygen vacancies, in hafnia-based ferroelectric materials8. Therefore, optimizing the amount of oxygen vacancies in the ferroelectric layer is key to improving the endurance performance of ferroelectric-based devices. One of the crucial factors that governs the amount of oxygen vacancies in ferroelectric materials is the choice of material for the metal electrode. Tungsten has been widely used for ferroelectric devices because of its CMOS compatibility. Moreover, compared to TiN-based ferroelectric capacitors, ferroelectric capacitors with tungsten metal electrodes have shown better ferroelectricity due to their stronger tensile stress9. However, tungsten-based ferroelectric capacitors also exhibit poor endurance performance due to the strong oxygen scavenging effect of tungsten, which induces an excessive amount of oxygen vacancies in the ferroelectric layer10.

In this work, the device performance of HZO-based metal-ferroelectric-metal (MFM) capacitors with two different metal electrodes (i.e., metallic tungsten for W-based capacitors and oxidized tungsten for WO3-based capacitors) was compared and investigated for future FeRAM applications. O2 gas was added during the deposition of the tungsten metal electrode to form an oxygen reservoir and to prevent the oxygen scavenging effect of the tungsten metal electrode. The ferroelectric characteristics, including the endurance performance of W- and WO3-based MFM capacitors, were investigated. Quantitative analyses such as TEM, EDS, XPS, and GIXRD were conducted to analyze changes in the atomic bonding within the HZO layer and phase transitions in the W and WO3 metal electrode layers.

Materials and methods

MFM (metal-ferroelectric-metal) capacitors were fabricated, as shown in Fig. 1 First, standard cleaning processes (i.e., SPM, SC-1, and SC-2) and diluted HF (1:50) cleaning were performed on (100) p-type Si wafers with a resistivity of < 0.005 Ω∙cm. Then, the tungsten bottom electrode was deposited by DC sputtering. Note that two different gas flow conditions were used for the W deposition: (i) only Ar and (ii) Ar and O2. The power and process time were identical for all samples. Afterwards, a 10-nm-thick HZO (Zr-doped HfO2) layer was deposited by thermal ALD at a deposition temperature of 250 oC. Tetrakis (ethylmethylamino) hafnium (TEMAHf), tetrakis (ethylmethylamino) zirconium (TEMAZ), and H2O were used as source precursors for the ALD process to deposit the HZO film. Subsequently, tungsten was deposited by DC sputtering for the top electrode, followed by photolithography, etching, PR ashing, and PR stripping to define the contact pattern. The device area size is 80 × 80 µm2. The fabricated MFM capacitors were annealed at 400/500/600/700 oC for 30 s in an N2 atmosphere using rapid thermal annealing (RTA) for post-metallization annealing (PMA) to crystallize the HZO film.

Fig. 1
figure 1

(a) Microscopic image of ferroelectric capacitors with the area of 80 μm-by-80 μm. (b) Illustrated cross-sectional view and (c) fabrication process flow for the capacitor.

The device measurements were done with Keithley 4200 A-SCS parameter analyzer, to characterize the ferroelectric properties of those MFM capacitors. The current density versus voltage (I–V) and endurance characteristics were measured. A triangular waveform with the amplitude of 3 Vand 4 V was used to characterize the polarization-voltage (P-V) characteristics. A trapezoidal waveform with the amplitude of 3 V and 4 V was used for the endurance cycling. Note that the rise/fall time for both waveforms and the pulse width for trapezoidal waveform were set to 1 µs.

Results and discussions

Figure 2a and 2b show the Transmission Electron Microscopy (TEM) images and High Angle Annular Dark Field (HAADF) images from the Electron Dispersive Spectroscopy (EDS) of W- and WO3-based capacitors at PMA temperature of 600 oC. As expected, W appeared in the top and bottom of the capacitor, while Hf and Zr were present in the intermediate layer. The EDS and XPS depth profiles in Fig. 2c and 2d confirmed that the oxygen content in the WO₃-based electrodes was significantly higher than that in W-based electrodes. The oxygen-rich nature of the WO₃ electrodes suggests a suppression of oxygen vacancies in the HZO layer, which is critical for improving endurance performance. The EDS analysis showed clear evidence of higher oxygen concentration in the top and bottom electrode layers of WO₃-based capacitors. This corroborates the hypothesis that WO₃ acts as an oxygen reservoir, preventing the oxygen scavenging effect of tungsten and thus reducing oxygen vacancy formation in the HZO layer. This is further supported by the XPS analysis in Fig. 3a and 3c, which showed a significant reduction in the oxygen vacancy-related suboxide peaks in WO₃-based capacitors.

Fig. 2
figure 2

TEM images and EDS analysis of (a) W-based MFM capacitor and (b) WO3-based MFM capacitor. The XPS depth profile of (c) W-based MFM capacitor and (d) WO3-based MFM capacitor.

XPS spectra of Hf 4f and W 4f were analyzed to further investigate the presence of oxygen vacancies. Figure 3a and c show the X-ray photoelectron spectroscopy (XPS) analysis of Hf 4f spectra from the HZO layer in W- and WO3-based capacitors at PMA temperature of 600 oC. For W-based capacitors in Fig. 3a, the Hf 4f peaks appeared at higher binding energies (18.57 and 20.17 eV), indicating a higher concentration of oxygen vacancies in the HZO layer. The suboxide peak (HfO2-x) was prominent, making up approximately 46% of the total Hf 4f area, which signifies the existence of oxygen-deficient regions in the ferroelectric layer. The HfO2-x peak in the figure represents the suboxide peak, which is oxygen-deficient HfO2-xcaused by the redox reaction, indicating the presence of oxygen vacancies in the HZO layer11.

Fig. 3
figure 3

XPS spectra of (a) Hf 4f and (b) W 4f in W-based capacitor. The XPS spectra of (c) Hf 4f and (c) W 4f in WO3-based capacitor. GIXRD diffractogram for (e) W-based capacitor and (f) WO3-based capacitor.

In contrast, the WO₃-based capacitors (Fig. 3c) exhibited lower binding energy peaks for Hf 4f (18.45 and 20.05 eV), with the suboxide peak contributing only 31% to the overall area. This reduction in oxygen vacancies can be attributed to the oxygen introduced during the deposition of WO₃, which helps fill the vacancies in the HZO layer. The XPS spectra of W 4f in Fig. 3b and d further confirmed this effect. While the W-based capacitors displayed strong peaks corresponding to metallic W (W 4f5/2 and W 4f7/2), the WO₃-based capacitors showed oxidized W states (WO₃-x), reflecting the oxygen-rich environment in these capacitors.

The added oxygen atoms can diffuse into the HZO layer and fill up the oxygen vacancies, as shown in Fig. 4. Tungsten metal exhibits a strong oxygen scavenging effect, inducing excessive oxygen vacancies and deteriorating the endurance performance, as observed in the W-based capacitor12. The addition of O2 gas oxidizes the metallic tungsten to WO3, creating an oxygen reservoir and suppressing the oxygen scavenging effect of W, as evidenced by the XPS analysis.

Fig. 4
figure 4

Schematics of (a) atomic bonding of HfO2 with oxygen vacancies in MFM capacitor. (b) Oxygen vacancies filled up with oxygen diffused from tungsten metal electrode.

Figure 3e and 3f present the grazing-incidence X-ray diffraction (GIXRD) patterns in the range of 20 ~ 70° (2θ) for W- and WO3-based capacitors. In the W-based capacitors, peaks corresponding to metallic W were observed around 40° and 60°. However, the WO₃-based capacitors exhibited additional peaks associated with tungsten oxides (WO₃), confirming the successful oxidation of the tungsten electrode. These results corroborate the XPS analysis of W 4f in Fig. 3b and 3d.

For various PMA temperatures, the measured pristine (before endurance cycling test) current density–vs.–voltage (I–V) and polarization–vs.–voltage (P–V) characteristics of W- and WO3-based MFM capacitors are shown in Fig. 5. The P-V measurements were performed using a triangular pulse with an amplitude of 3–4 V. The endurance characteristics of each sample were measured, as shown in Fig. 6. A trapezoidal waveform with an amplitude of 3–4 V, a pulse width of 1 µs, and rising/falling times of 1 µs was used for the endurance cycling. The triangular pulse was used for the P-V measurements. Note that the endurance cycling pulse and P-V measurement pulse were set to the same value (i.e., 3 V/3V and 4 V/4V).

Fig. 5
figure 5

(a-d) Measured “pristine” current density versus voltage, and (e-h) measured “pristine” polarization versus voltage for four different post metal annealing (PMA) temperatures at voltage amplitude of 3 and 4 V: (a), (b), (e), and (f) W-based capacitors. (c), (d), (g), and (h) WO3-based capacitors.

Fig. 6
figure 6

Measured pristine 2Pr vs. the number of cycles of W-based MFM and WO3-based MFM capacitors at (a, b) 3 V, (c, d) 4 V. The number of endurance cycle is 100~ 108. The 10−1 indicates the pristine state. The annealing temperature was set to four different annealing temperatures (i.e., 400, 500, 600, and 700 ℃).

Table 1 summarizes the pristine 2Pr values and breakdown points in the endurance cycling of W- and WO3-based capacitors at various PMA temperatures and pulse amplitudes. The W-based capacitors exhibited excellent 2Pr characteristics, with a maximum value of 107.9 μm/cm2 at 700 ℃. The overall pristine 2Pr values of the WO3-based capacitors were lower compared to the W-based capacitors. This reduction can be attributed to the excessive oxygen in WO3, which causes the transformation of the o-phase to non-polar phases in the ferroelectric layer12. Although the pristine 2Pr of the WO3-based capacitors was lower than that of the W-based capacitors, the endurance performance was significantly enhanced by up to 3 orders of magnitude. This improvement is due to the reduced oxygen vacancies resulting from the oxygen-rich WO3electrode, as previously explained. Oxygen vacancies form undesirable conductive leakage paths in the HZO layer13,14. Consequently, the leakage paths were reduced with the decreased oxygen vacancies in the WO3-based capacitors, leading to enhanced endurance performance.

Table 1 Pristine 2Pr and breakdown point of endurance cycling test.

As shown in Fig. 7a, the performance of the capacitors fabricated in this work was benchmarked against previous studies (which were annealed at various annealing temperatures). The MFM capacitors annealed at 400 to 700 ℃ showed superior 2Pr compared to the previous works. The unusually high 2Pr of 107.9 µC/cm² in the W/HZO/W capacitors is noteworthy. We attribute this high remnant polarization to the combined effects of tensile stress and enhanced crystallization at high annealing temperatures (up to 700 °C). The W electrodes induce significant tensile strain on the HZO layer, which promotes the stabilization of the orthorhombic ferroelectric phase, leading to higher polarization. In Fig. 7b, the number of maximum endurances cycling before the breakdown of each ferroelectric-based capacitor are summarized and compared to each other. It turned out that the MFM capacitors in this work (vs. previous works8,10,15,16,17,18,19,20,21,22) shows comparable endurance performance even with the superior 2Pr value showed in Fig. 7a. Although the pristine 2Pr of WO₃-based capacitors was lower than that of W-based capacitors, the superior endurance performance makes WO₃ a more viable candidate for FeRAM applications, where long-term reliability is critical. The trade-off between polarization and endurance performance highlights the importance of selecting electrode materials that can balance both properties effectively.

Fig. 7
figure 7

The benchmarking plot of (a) 2Pr and (b) endurance characteristics of MFM capacitors in HfO2-based ferroelectric capacitors with various annealing temperatures.

Conclusion

The impact of W and WO3 electrodes on HZO-based MFM capacitors was investigated and compared. The W-based MFM capacitors exhibited superior remnant polarization compared to the WO3-based capacitors. However, the W-based capacitors showed deteriorated endurance performance. In contrast, the WO3-based capacitors demonstrated enhanced endurance performance due to the oxygen-rich reservoir effect. The added oxygen in WO3 prevented the oxygen scavenging effect of tungsten, thereby suppressing the excessive generation of oxygen vacancies in the HZO layer, and consequently improving endurance performance. This result indicates that adjusting the Ar/O2 gas ratio during the deposition of the tungsten metal can significantly affect the endurance performance of ferroelectric devices for future FeRAM applications.