Oxygen Scavenging in HfZrOx‐Based n/p‐FeFETs for Switching Voltage Scaling and Endurance/Retention Improvement

The authors demonstrate improved switching voltage, retention, and endurance properties in HfZrOx (HZO)‐based n/p‐ferroelectric field‐effect transistors (FeFETs) via oxygen scavenging. Oxygen scavenging using titanium (Ti) in the gate stack successfully reduce the thickness of interfacial oxide between HZO and Si and the oxygen vacancy at the bottom interface of the HZO film. The n/p‐FeFETs with scavenging exhibit an immediate read‐after‐write with stable retention property and improved endurance property. In particular, n‐FeFET with scavenging exhibits excellent endurance property that does not show breakdown up to 1010 cycles. The charge trapping model in the n/p‐FeFETs is presented to explain why the effect of oxygen scavenging is more pronounced in n‐FeFET than in p‐FeFET. Finally, further switching voltage scaling potential is estimated by scavenging and HZO thickness scaling. It is believed that this work contributes to the development of low‐power FeFET and the understanding of FeFET operation.


Introduction
In conventional computing based on the von Neumann architecture, the issues of power consumption and signal delay due to data transfer between the two units cannot be avoided since the memory unit and the logic unit are spatially separated. These issues become more serious as devices are scaled down and as large-capacity information processing such as which is a lower thermal budget than conventional rapid thermal annealing, was carried out as postmetallization annealing (PMA), but IL growth inhibition was not effective and the P r was less than 10 µC cm −2 . Meanwhile, oxygen scavenging, which has been examined to reduce or eliminate the IL thickness in high-k/metal gate (HK/MG) technology, [16,17] has recently been investigated in ferroelectric HZO-based devices. Oxygen scavenging is a method to inhibit the growth of IL of SiO x by inserting a highly reactive metal, for example, Ti, into the top metal stack. Thus, oxygen scavenging can decrease the IL thickness without increasing the process temperature and also reduce the trap density. Although oxygen scavenging has been reported in capacitor structure, [18,19] relevant studies have not been much investigated in FeFET.
Herein, we successfully demonstrated oxygen scavenging in HZO-based n/p-FeFETs using Au/Ti/TiN/HZO/Si gate stack as an approach to achieve IL scaling and interfacial trap density reduction, which was manifested by transmission electron microscopy (TEM) and electron energy-loss spectroscopy (EELS), respectively. We systematically investigated oxygen scavenging to improve the important performances of FeFETs such as V sw , endurance, and retention properties. In particular, n-FeFET with scavenging was not only capable of immediate read-after-write at ±3 V/1 µs pulse but also exhibited excellent endurance properties of higher than 10 10 cycles. The effect of oxygen scavenging was greater in the n-FeFET than in the p-FeFET, and we presented a band structure model to explain it. Finally, based on our observation, we suggested potential V sw scaling in HZO-based n-FeFETs. We believe that oxygen scavenging has great potential for improving the performance of FeFETs. Figure 1a shows the advantages and disadvantages of charge trapping caused by IL in the MFIS structure. IL degradation by charge trapping inhibits endurance properties, and read delay, which occurs mainly in the program (PGM) operation of n-FeFETs, becomes an obstacle to immediate read-after-write capability and retention properties. However, charge trapping also enhances polarization, called charge-assisted polarization.

Fabrication of HZO-Based FeFETs via Oxygen Scavenging
The role of IL in the ferroelectric behavior of the MFIS structure is to induce additional voltage drop and depolarization field as well as charge trapping. Voltage drop and depolarization field across the IL cause an increase in switching voltage and inhibition of retention properties. Since the detrimental role of IL in ferroelectric behavior can mostly be improved by reducing the IL thickness, IL thickness scaling is one of the promising methods of IL engineering in FeFETs. Figure 1b is a schematic illustration of oxygen scavenging used for IL thickness scaling. Even if IL is removed in the substrate preparation step, IL is formed during gate stack growth. Afterward, in the PMA process, IL is regrown and scavenged in the gate metal stack without scavenging (Au/TiN) and with scavenging (Au/Ti/TiN), respectively. Considering the Gibbs free energy, since Ti in a gate metal stack with scavenging is thermodynamically favored to scavenge oxygen by remotely decomposing SiO x , [20,21] the insertion of Ti in the gate metal stack can effectively reduce the IL thickness. The detailed fabrication flow of FeFETs is shown in the Experimental section and Figure S1a, Supporting Information. Figure 2a-c shows the cross-sectional TEM images of the asdeposited Au/Ti/TiN stack, Au/TiN stack after PMA, and Au/Ti/ TiN stack after PMA. Although IL was removed with buffered oxide etchant during substrate preparation, the formation of IL during gate stack growth is shown in Figure 2a. In addition, as expected from the schematic illustration above, the TEM images in Figure 2b,c clearly show that IL in the gate metal stack without scavenging and with scavenging was regrown and scavenged after PMA. The energy dispersive spectrometer (EDS) maps of the gate stacks shown in Figure 2d-f displayed that Ti in the gate metal stack was oxidized after PMA, which indicates Ti plays a decisive role in oxygen scavenging. Additionally, diffusion of the Au layer into the Ti layer and slight oxidation of TiN were observed.
In order to examine whether oxygen scavenging by Ti affects the crystallinity and phase of HZO films, grazing incidence X-ray diffraction (GIXRD) and X-ray photoelectron spectroscopy (XPS) analysis were performed and shown in Figure  3. As shown in Figure 3a, no peaks appeared in as-deposited HZO film while HZO without scavenging and with scavenging exhibited peaks at 30.8 and 35.5°, [22] indicating that the amorphous as-deposited HZO transformed to a ferroelectric polycrystalline structure, a mixture of orthorhombic and tetragonal phase, after PMA in both HZO without scavenging and with Adv. Electron. Mater. 2023, 9,2201257  scavenging. In Figure 3b, the high-resolution narrow scan XPS spectra of Hf 4f and Zr 3d showed that the peaks of Hf 4f 7/2 and Zr 3d 5/2 appeared at 17.7 and 183.1 eV [23] for as-deposited HZO, HZO without scavenging, and HZO with scavenging. There were no differences in chemical bonding states and stoichiometry between the samples. Interestingly, XRD and XPS analysis revealed no structural differences between HZO without scavenging and with scavenging, thereby, the difference in electrical properties between the devices, which will be discussed in the following section, would be due to the IL difference.

Effects of Oxygen Scavenging on HZO-Based FeFETs
Firstly, we investigated the DC transfer curves of the FeFETs as in Figure 4a. At ±4 V sweep, n-FeFETs and p-FeFETs exhibited counter-clockwise and clockwise hysteresis, respectively, indicating ferroelectricity of the gate stack. As previously reported, the MW of n-FeFETs was larger than that of p-FeFETs. [24] In addition, MW of n-FeFETs and p-FeFETs increased from 1.41 to 2.32 V and from 1.17 to 1.28 V, respectively, by oxygen scavenging. The MW increase by oxygen scavenging was more pronounced in n-FeFETs than in p-FeFETs.
The detailed ferroelectric properties of the FeFETs were investigated based on the pulse measurement scheme shown in Figure S1b-e, Supporting Information. The transient I-V curves and P-V hysteresis curves in Figure 4c-f were obtained by quasi-static C-V (QSCV) measurement using bipolar triangular pulse and double-pulsed QSCV (DQSCV) measurement using the positive-up-negative-down (PUND) method. [3] With scavenging, V sw decreased from 3.19 V to 2.35 V (26% reduction) in n-FeFETs and from 3.34 to 2.75 V (18% reduction) in p-FeFETs. As V sw was lowered, P r increased significantly at the same pulse amplitude of 3 V. Figure 4e-f show the P r after removing the leakage component according to the inset formula. In n-FeFETs, the P r increased from 9.3 to 28.2 µC/cm 2 (203% increase), and in p-FeFETs, P r increased from 9.1 to 25.2 µC/cm 2 (177% increase). The significant enhancement of P r indicates that the oxygen scavenging technique is very effective for IL engineering in FeFETs. Additionally, transient I-V curves and P-V hysteresis curves according to the applied pulse can be found in detail in our previous work. [18] Since a large amount of electron trapping is known to degrade the retention properties in FeFETs of the MFIS structure, we investigated MW as a function of retention time. MW after 1 µs, 10 ms, 1 s, and 1000 s of program/erase (PGM/ ERS) at pulse amplitudes from 3 to 5 V and widths from 1 µs to 1 ms are mapped in Figure 5. Figure 6 displays the retention properties up to 1000 s at specific pulses of ±3 V/1 µs and ±4 V/1 ms. In the n-FeFET without scavenging, when operating with a low and short pulse (near the lower left corner of the map in Figure 5a), the write operation could not be performed ( Figure 6a) and a read delay was observed ( Figure 6b) because charge trapping was dominant over ferroelectric polarization at a short retention time (<10 ms), resulting in read delay, and then reversed at longer retention time (>1 s). In contrast, the device with scavenging was not only capable of immediate read-after-PGM with a low and short pulse, which is typically very difficult in n-FeFET, [25] but also the MW behaviors with a time were more stable than the device without scavenging (Figures 5b and  6a-c). The increase in MW according to the retention time occurs as trapped electrons by the positive pulse applied during PGM operation are gradually de-trapped. [3,25] The ΔV th values at the retention time of 1000 s after the PGM operation were 0.81 and 0.26 V (Figure 6c) in the devices without and with scavenging, respectively, corresponding to de-trapped electrons of 6.6 × 10 12 and 2.7 × 10 12 cm −2 . It suggests that electron trapping itself was alleviated by oxygen scavenging. Similarly, the p-FeFET without scavenging could not perform the write operation and underwent read delay in low and short pulses, and p-FeFET with scavenging allowed immediate read (Figure 5c,d  and 6d,e). Additionally, the p-FeFETs showed more stable V th after PGM operation and smaller MW compared to the n-FeFETs. In the p-FeFETs without and with scavenging, the ΔV th values at the retention time of 1000 s after the PGM operation were 0.30 and 0.09 V (Figure 6f), respectively, which were less ΔV th than in the n-FeFETs. This corresponds to detrapped electrons of 2.4 × 10 12 and 9.4 × 10 11 cm −2 . The stable V th value in p-FeFETs according to the retention time after PGM operation is because there are fewer electrons to be de-trapped after PGM operation and less charge-assisted polarization.
The charge trapping behavior is clearly demonstrated by retention measurement with standby voltage (V stby ) from +1 V to −1 V in Figure S2e, Supporting Information. In retention measurement after PGM operation, positive V stby prevents electron de-trapping and consequently slows down V th decrease while negative V stby promotes electron de-trapping, accelerating the V th decrease. The effect of V stby was more pronounced in the n/p-FeFETs with scavenging so that at V stby of −1 V, V th started to rebound from 1 ms in n-FeFETs with scavenging, implying domain switching, while n-FeFETs without scavenging showed rebound from 10 ms. As the p-FeFETs show a more stable V th , V th rebound by V stby was not noticeable and increased or decreased according to V stby . On the other hand, in retention measurement after ERS operation, the effect of V stby on V th change was not remarkable. These results strongly suggest that electron trapping/de-trapping would be the dominant mechanism for retention properties rather than the hole-associated mechanism.
To investigate the impact of the scavenging on the endurance properties, we applied cyclic bipolar square pulses to the devices and measured the MW (Figure 7). Generally, at lower and shorter pulses, the endurance properties of ferroelectric HZO film were higher. [18] In FeFETs without scavenging and with scavenging, the net electric field at the same V g is different due to the difference in IL thickness. Therefore, the comparison of endurance properties at the same V g is not fair because FeFETs with scavenging are under more severe electrical stress than FeFETs without scavenging. For a fair comparison, endurance properties of FeFETs without scavenging and with scavenging were compared at the pulses exhibiting the similar initial MW. First, in the n-FeFETs, FeFETs without scavenging and with scavenging showed MW of 0.60 and 0.61 V at ±3 V/100 µs and ±3 V/1 µs, respectively, and MW of 1.58 and 1.56 V at ±4 V/100 µs and ±4 V/1 µs, respectively. At ±4 V/100 µs and ±4 V/1 µs pulses, the breakdown occurred at 10 7 and 10 8 cycles in n-FeFETs without scavenging and with scavenging, respectively. At the pulses of ±3 V/100 µs and ±3 V/1 µs, n-FeFET without scavenging showed breakdown at 10 10 cycles while n-FeFET with scavenging did not show breakdown until 10 10 cycles. Meanwhile, in the p-FeFETs, FeFETs without scavenging and with scavenging exhibited similar MW at the pulses of ±3 V/1 ms and ±3 V/1 µs and at the pulses of ±4 V/10 µs and ±4 V/1 µs, respectively. Similar to n-FeFETs, p-FeFETs with scavenging showed 1 to 2 orders of magnitude higher endurance properties at the pulses with similar initial MW than p-FeFETs without scavenging. The superior endurance properties of  FeFETs with scavenging than that of FeFETs without scavenging would be attributed to the reduced electron trapping/ de-trapping at the IL by oxygen scavenging. In addition, the V th in the PGM state of FeFETs with scavenging was more stable during the endurance test than that of FeFETs without scavenging, suggesting that charge trapping by positive pulses was reduced. [26] On the other hand, the breakdown of p-FeFETs tended to occur in about 1 order of cycle earlier than that of n-FeFETs, which may be attributed to hole trapping as seen also in our previous works without scavenging process. [3] Based on the assumption that the impact of the oxygen scavenging would be originated from the amount of the trap site   Figure 8b,c) are sensitive to the crystallinity of the material and oxygen vacancy. [27] It has been reported that annealing at higher temperature tends to induce more crystallization and sharpen the peaks. [28,29] However, our HZO films without scavenging and with scavenging were grown and annealed under identical thermal processing conditions, and no difference in crystallinity was confirmed by XRD and XPS analysis. Thus, it can be inferred that the difference in EELS spectra of HZO without scavenging and with scavenging is attributed to oxygen vacancy. At the O-K edge of the HZO without scavenging, notable doublet peaks (Figure 8b) appeared at the middle and the top whereas a peak split was not seen at the bottom, implying oxygen deficiency at the bottom. [27] On the other hand, the HZO with scavenging showed doublet peaks even at the bottom (Figure 8c), indicating that oxygen vacancy at the bottom interface was reduced by oxygen scavenging. [27] A schematic illustration of oxygen vacancy distribution is shown in Figure 8d. Less oxygen vacancy at the bottom highly correlates with the superior retention and endurance properties of FeFETs with scavenging compared to that of FeFETs without scavenging.
Finally, based on the above understanding, we provide the band diagram and the charge trapping model of n/p-FeFETs in Figure 9. As shown in Figure 9a, the band diagram of the n-FeFET and p-FeFET at the initial state are almost identical. In HZO, oxygen vacancy creates defect levels near the conduction band that can trap electrons. [30] In addition, the large valence band offset between Si and SiO x induces less hole trapping. Accordingly, electron trapping has a more significant effect than hole trapping on the behavior of HZObased FeFETs. Figure 9b,c describe electron trapping caused by positive pulse during PGM operation in FeFETs without scavenging and with scavenging, respectively. (Additionally, as noted in Figure 1a, as the IL became thinner, the voltage drop across the IL decreased and the voltage drop across the HZO increased.) In n-FeFET with scavenging, less electron trapping occurs than in n-FeFET without scavenging, as confirmed by retention measurements and EELS analysis. However, in p-FeFETs, since electron trapping occurred less even with the device without scavenging, the reduction of trap density by scavenging was relatively small. The trap density decreased from 6.6 × 10 12 and 2.7 × 10 12 cm −2 (reduced by 3.9 × 10 12 cm −2 ) in n-FeFETs and from 2.4 × 10 12 to 9.4 × 10 11 cm −2 (reduced by 1.5 × 10 12 cm −2 ) in p-FeFETs as calculated above. Figure 9d shows the band diagram for retention with V stby after the PGM operation. As shown in Figure S2ab, Supporting Information, positive V stby suppressed electron de-trapping and slowed the V th lower shift in the retention measurement of n-FeFETs. In p-FeFETs, since there are fewer trapped electrons, an additional PGM operation occurred ( Figure S2c,d). On the other hand, negative V stby enhanced electron de-trapping and accelerated the V th lower shift in n-FeFETs. Therefore, it was observed that V th was rebound and ERS operation was performed by negative V stby . In the Adv. Electron. Mater. 2023, 9, 2201257 case of p-FeFETs, because there were fewer trapped electrons, the rebound of V th did not appear and only ERS operation was performed. The band diagram can evident the more impact of the scavenging in n-FeFET than in p-FeFET and the major contribution of electron trapping on retention properties in FeFET rather than hole trapping. Furthermore, in Figure S3, we also estimated potential V sw scaling by fabricating HZO-based n-FeFETs with HZO  thicknesses of 7, 10, 14, and 20 nm. According to the V sw trend with HZO thickness, it is expected that the V sw of FeFET without scavenging can be lowered to 2.1 V as HZO becomes thinner. By applying oxygen scavenging, it will be possible to further lower the V sw of the FeFET to 1.6 V. Moreover, by modeling the HZO and IL as series capacitors, we extracted the applied voltage across the HZO and predicted the V sw trend when the IL is completely eliminated. The equation in the inset shows that the V g required for driving decreases as the IL thickness decreases. Ultimately, we forecasted that n-FeFET will be feasible of switching below 1 V by the optimization of the scavenging process and HZO thickness scaling.
Finally, a benchmark [3,25,[31][32][33][34] in Table S1, Supporting Information highlights the low-voltage operation, good retention, and high endurance properties of our FeFETs realized by scavenging. In particular, the n-FeFET with scavenging showed MW of 0.63 V and endurance of >10 10 cycles at low and short pulses (3 V/1 µs) where PGM/ERS operation was impossible Adv. Electron. Mater. 2023, 9, 2201257 without scavenging. Oxygen scavenging which is induced at a low temperature of 500 °C has the potential to improve the performance of HZO-based FeFETs.

Conclusion
We have systematically investigated the effect of oxygen scavenging in n/p-FeFETs and developed a charge-trapping model. The decrease in IL thickness induced by oxygen scavenging using Ti metal was manifested by TEM cross-section analysis, and XRD and XPS analysis confirmed that there was no structural change in HZO film by oxygen scavenging. By oxygen scavenging, MW in the transfer curve of the same V g sweep increased in both n/p-FeFETs, and a decrease in V sw and an increase in P r were shown by QSCV and DQSCV measurements. In addition, in the retention measurement, n/p-FeFETs with scavenging enabled immediate read-after-write at ±3 V/1 µs pulse, which was not possible with n/p-FeFETs without scavenging. The V th change according to the retention time was more stable in n/p-FeFETs with scavenging than n/p-FeFETs without scavenging, indicating less charge trapping. Our FeFETs with scavenging exhibited improved endurance property of >10 10 cycles. Reduction of oxygen vacancy at the bottom interface of HZO film by oxygen scavenging which was confirmed by EELS analysis contributes to improved retention and endurance properties. The excellent properties of the FeFETs with scavenging show great potential for low-power FeFETbased applications.

Experimental Section
Fabrication: To investigate the effect of oxygen scavenging, FeFETs were fabricated by a gate last process as shown in Figure S1a, Supporting Information. First, S/D regions were formed and native SiO x on Si substrates was removed by buffered oxide etchant. Then, 10-nm-thick HZO was deposited by atomic layer deposition. Tetrakis (ethylmethylamino) hafnium and tetrakis (ethylmethylamino) zircomium were used as precursors and ozone was fed as the oxidant. Next, gate metal stacks of Au (80 nm)/TiN (20 nm) and Au (80 nm)/Ti (20 nm)/ TiN (20 nm) were formed for FeFETs without scavenging and with scavenging, respectively. Postmetallization annealing (PMA) was conducted at 500 °C for 1 min for crystallization and oxygen scavenging. Finally, S/D electrodes were formed of Si-doped Al, followed by PMA at 300 °C for 5 min.
Material Characterization: Cross-sectional view images, energy dispersive spectrometer (EDS) maps, and electron energy loss spectroscopy (EELS) spectra of gate stacks were obtained by transmission electron microscopy (TEM; JEOL, JEM-2100F). The structural characterization and chemical bonding state analysis of the HZO films were performed by grazing incidence X-ray diffraction (GIXRD; Rigaku, SmartLab) and X-ray photoelectron spectroscopy (XPS; ThermoFisher Scientific, K-Alpha+).
Electrical Characterization: For electrical characterization of FeFETs without and with scavenging, a parameter analyzer (Keithley, 4200A-SCS) with a pulse measurement unit (Keithley, 4225-PMU) and remote preamplifier/switch modules (Keithley, 42250RPM) was used. The pulse scheme for device measurement is shown in Figure S1b-e, Supporting Information. Quasi-static CV (QSCV) and double-pulsed QSCV (DQSCV) measurements were used for V sw extraction (Figure S1b, Supporting Information) and P r measurement ( Figure S1c, Supporting Information). Similar to a positive-up-negative-down (PUND) method, P r was derived from DQSCV results by subtracting the current measured in U (and D) pulses from the current measured in P (and N) pulses. Endurance and retention measurements were performed as shown in Figures S1d,e, Supporting Information.

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