Background

HfO2-based ferroelectric thin films are considered as promising gate-stack materials for ferroelectric field effect transistors (FeFETs) because of their complementary metal-oxide-semiconductor (CMOS) compatibility and scalability. Among several kinds of gate stack structures that can be used in FeFETs, a metal/ferroelectric/insulator/semiconductor (MFIS) represents a more practical configuration because it follows the current MOS device architectures and matches well with the modern high-k metal-gate (HKMG) processes. Therefore, great efforts have been made to design and fabricate FeFETs with MFIS gate stack structures for applications in embedded nonvolatile memories, negative capacitance field effect transistors, artificial neurons, synapses, and logic-in-memory devices [1,2,3,4,5,6,7,8].

Up to now, high-density and fast write speed FeFETs with MFIS gate stack structures have been successfully fabricated using HKMG processes [9, 10]. In addition to the high integration density and fast write speed, a large memory window (MW) and a high reliability with respect to retention and endurance are also critical for employing FeFETs for nonvolatile memory applications [11,12,13,14]. Owing to a large band offset to silicon, a high coercive field and a moderate dielectric constant of the HfO2-based ferroelectric thin films, HfO2-based FeFETs with MFIS gate stack structures exhibit reliable retention properties (10-year extrapolation) [15,16,17]. However, although the HfO2-based thin films demonstrate moderate endurance over 1 × 109 switching cycles [14, 18], HfO2-based FeFETs with MFIS gate stack structures have a rather limited endurance ranging from 1 × 104 to 1 × 107 switching cycles [17, 19,20,21,22,23]. Theoretically, employing of high-k insulator layers is expected to reduce the electric field in the MFIS gate stack, which would alleviate the band bending, thereby improving the endurance properties and the MWs of the HfO2-based FeFETs [12, 14]. Experimentally, Ali et al. verified that increasing the k value of the ultrathin insulator layer (i.e., using SiON instead of SiO2) can effectively improve the endurance properties as well as the MW of the HfO2-based FeFETs [13]. In our previous research [24], we reported that the insertion of a crystalline ZrO2 high-k layer in the MFIS gate stacks could improve the crystalline quality and suppress the formation of monoclinic phase in Hf0.5Zr0.5O2 (HZO) thin films, which leads to a large MW of 2.8 V characterized by DC voltage sweep method.

In this work, we report on the characterization of the MWs, retention, and endurance of the HZO-based FeFETs with and without crystalline ZrO2 seed layers by using fast positive and negative voltage pulse measurements. Moreover, the advantage of employing crystalline ZrO2 seed layers on the MW and endurance properties is discussed.

Methods

The n-channel FeFETs with and without ZrO2 seed layers were fabricated using a gate last process, as described in [24]. The ZrO2 seed layer and the HZO layer were both grown at a growth temperature of 300 oC by atomic layer deposition (ALD). The schematic of the fabricated FeFETs is shown in Fig. 1a, whose channel width (W) and length (L) were 80 and 7 μm, respectively. Meanwhile, TaN/HZO/TaN and TaN/HZO/ZrO2/TaN capacitors were also fabricated to evaluate the ferroelectric properties of the HZO thin films. The polarization–voltage (P–V) hysteresis loops of the capacitors were measured using a Radiant Technologies RT66A ferroelectric test system, while the device characteristics of FeFETs were measured by an Agilent B1500A semiconductor device analyzer with a pulse generator unit (B1525A) [20]. Two main test sequences used for MW and endurance measurements are shown in Fig. 1b and c. For MW and retention measurements, program/erase (P/E) pulses were first applied to the gates of FeFETs, and read operations were performed at different time intervals using ID–VG sweep (VD = 0.1 V) to sense VTH. Generally, VTH is determined as a gate voltage corresponding to a drain current of 10−7 A∙W/L [25], and the MW is defined as the difference of VTH values between programmed and erased states. For endurance measurements, the MW was measured after a certain number of alternating P/E pulses.

Fig. 1
figure 1

a Schematic of the fabricated FeFETs. The additional crystalline ZrO2 seed layer is marked by black gridlines. b, c Test sequences used for MW and endurance measurements

Full size image

Results and Discussion

Figure 2a shows the P–V hysteresis loops of the TaN/HZO/TaN and TaN/HZO/ZrO2/TaN capacitors. Remarkably, the TaN/HZO/ZrO2/TaN capacitor possesses even better ferroelectric properties than the TaN/HZO/TaN capacitor, which is consistent with the reported results [26], indicating that the crystalline ZrO2 seed layer could indeed improve the crystalline quality and suppress the formation of monoclinic phase in HZO thin films [24]. Figure 2b shows the ID–VG curves of the HZO-based FeFETs with and without additional crystalline ZrO2 seed layers after P/E pulses. The red symbol lines represent the ID–VG curves after applying a program pulse of 7 V/100 ns, while the blue symbol lines represent the ID–VG curves after applying an erase pulse of − 7 V/100 ns. One can see that the ID–VG curves of both FeFETs show counterclockwise switching characteristics, suggesting that the MWs of the present FeFETs are originated from the polarization switching of the HZO layers, rather than the charge trapping and injection. Nevertheless, the HZO-based FeFET with the additional crystalline ZrO2 seed layer exhibits an improved MW of 1.4 V, approximately 1.8 times larger than that (0.8 V) of the HZO-based FeFET without the additional crystalline ZrO2 seed layer. Moreover, the obtained MW of 1.4 V is comparable to the best results reported to date [9, 11, 14, 17, 21,22,23, 27].

Fig. 2
figure 2

a P–V hysteresis loops of TaN/HZO/TaN and TaN/HZO/ZrO2/TaN MFM structures measured at 4 V and a frequency of 5 kHz. b ID–VG curves of HZO-based FeFETs with (w) and without (w/o) ZrO2 seed layers after a program pulse (+ 7 V/100 ns) and an erase pulse (− 7 V/100 ns)

Full size image

Reliability with respect to the retention of the HZO-based FeFETs with and without additional crystalline ZrO2 seed layers was also evaluated. Figure 3 shows the VTH retention characteristics after applying a program pulse of 7  V/100 ns and an erase pulse of – 7 V/100 ns at room temperature. It is clear that the VTH values are approximately linear with the logarithmic time scale. The extrapolated MW after 10 years for the HZO-based FeFET with the additional crystalline ZrO2 seed layer is 0.9 V, larger than that (0.6 V) for the HZO-based FeFET without the additional crystalline ZrO2 seed layer. Since the thick capacitance equivalent thickness (CET) of the ZrO2 (1.5 nm)/SiO2 (2.6 nm) gate insulator layers would lead to an enhanced depolarization field in the gate stack [13, 15], further improvement in retention properties could be expected if the thickness of the SiO2 layer is reduced.

Fig. 3
figure 3

Retention characteristics of HZO-based FeFETs with and without ZrO2 seed layers

Full size image

Figure 4 shows the evolution of ID–VG curves after ± 7 V/100 ns alternating P/E cycles. For the FeFET without the additional crystalline ZrO2 seed layer, both significant shift and slope degradation in the ID–VG curves are observed from the early stages of P/E cycling, and the ID–VG curves in the erased states exhibit more slope degradation compared with the program states. For the FeFET with the additional crystalline ZrO2 seed layer, although the ID–VG curves in erased states exhibit an obvious positive shift during the early stages of P/E cycling that is attributed to the “wake up” effect [13, 28,29,30,31,32], no obvious shift of ID–VG curves in the program states is observed up to 1 × 103 cycles. Moreover, for the FeFET with the additional crystalline ZrO2 seed layer, the ID–VG curves in both erased and program states exhibit only a slight slope degradation up to 1 × 103 cycles.

Fig. 4
figure 4

Evolution of ID–VG curves of HZO-based FeFETs a without and b with ZrO2 seed layers with P/E cycling

Full size image

According to previous reports [12, 28, 33], the parallel shift in ID–VG curves is attributed to the gradual accumulation of trapped charges in the gate stack, while the slope degradation in ID–VG curves is the result of interface trap generation. Since trapped charges can be de-trapped by electrical means, but generation of interface traps is irreversible, minimizing interface trap generation is extremely important for improving the endurance properties [28]. The interface traps generated by P/E cycling (ΔNit) can be described using Eq. (1) [34, 35]:

$$ \Delta \mathrm{SS}=\frac{\Delta {N}_{it} kT\ln 10}{C_{FI}{\varnothing}_F} $$
(1)

where ΔSS is the change of the subthreshold swing, k is the Boltzmann constant, T is the absolute temperature, CFI is the total capacitance of gate stack, and ∅F is the Fermi potential. The ΔNit as a function of the P/E cycle for the HZO-based FeFETs with and without additional crystalline ZrO2 seed layers is shown in Fig. 5. Clearly, for the FeFET without the additional crystalline ZrO2 seed layer, the ΔNit increases obviously from the early stages of the P/E cycling, and ΔNit in the erased states is much larger than that in the program states. However, the ΔNit for the FeFET with the additional crystalline ZrO2 seed layer almost does not change up to 1 × 103 cycles, and it is always smaller than that for the FeFET without the additional crystalline ZrO2 seed layer. Because inserting the additional ZrO2 seed layer reduces the electric field in the gate stack and thus the band bending is weaker, the interface trap generation is alleviated [12, 14].

Fig. 5
figure 5

Evolution of ∆Nit with P/E cycling

Full size image

Figure 6 shows the evolution of gate leakage current characteristics (IG–VG curves) of HZO-based FeFETs with and without ZrO2 seed layers with P/E cycling. For the FeFET without the additional crystalline ZrO2 seed layer, the gate leakage current increases dramatically from the early stages of the P/E cycling. However, the gate leakage current for the FeFET with the additional crystalline ZrO2 seed layer almost does not change up to 5 × 102 cycles, and it is always smaller than that for the FeFET without the additional crystalline ZrO2 seed layer. It is reported that the increase in the gate leakage current might be related to the generated interface traps [28]. The reduction in the gate leakage current with cycling for the FeFET with the additional crystalline ZrO2 seed layer would be attributed to the suppression of interface trap generation.

Fig. 6
figure 6

Evolution of gate leakage current characteristics (IG–VG curves) of HZO-based FeFETs a without and b with ZrO2 seed layers with P/E cycling

Full size image

The VTH values for program and erase states extracted from the ID–VG curves of the HZO-based FeFETs with and without additional crystalline ZrO2 seed layers are shown in Fig. 7. The HZO-based FeFET with the additional crystalline ZrO2 seed layer always exhibits a larger MW than the HZO-based FeFET without the additional crystalline ZrO2 seed layer. Moreover, the MW of the HZO-based FeFET without the additional crystalline ZrO2 seed layer decreases obviously from the early stages of P/E cycling, while the MW of the HZO-based FeFET with the additional crystalline ZrO2 seed layer decreases slightly up to 1 × 103 cycles. As the P/E cycling number is further increased, the HZO-based FeFET with the additional crystalline ZrO2 seed layer also shows obvious degradation in the slope of the ID–VG curves and the MW, due to the enhanced generation of interface traps. However, the MW of the HZO-based FeFET with the additional crystalline ZrO2 seed layer is still larger than 0.9 V up to 1 × 104 cycles, which is approximately 2.3 times larger than that (0.4 V) of the HZO-based FeFET without the additional crystalline ZrO2 seed layer. As discussed previously, the decrease of the required electric field for obtaining more saturated polarization states are probably responsible for the improved endurance properties.

Fig. 7
figure 7

Evolution of VTH with P/E cycling

Full size image

Conclusions

The MWs as well as the reliability with respect to retention and endurance of the HZO-based FeFETs with the TaN/HZO/SiO2/Si and TaN/HZO/ZrO2/SiO2/Si MFIS gate stacks were characterized by fast voltage pulse measurements. The results show that the HZO-based FeFET with the additional crystalline ZrO2 seed layer exhibits a large initial memory window of 1.4 V and an extrapolated 10-year retention of 0.9 V, larger than the initial memory window (0.8 V) of the HZO-based FeFET without the additional crystalline ZrO2 seed layer. Moreover, the reliability with respect to the endurance of the HZO-based FeFET can be improved by inserting the crystalline ZrO2 seed layer in between the HZO layer and the SiO2/Si substrate. The MW and endurance improvement of HZO-based FeFETs with ZrO2 seed layers are primarily related to the improved crystalline quality of the HZO layer and the suppressed generation of interface traps due to the decrease of the required electric field for obtaining more saturated polarization states. On the basis of this work, it is expected that employing of a direct crystalline high-k/Si gate stack would further improve the MWs and reliability of the HfO2-based FeFETs, and thus warrant further study and development.