Introduction

With the scaling down of transistor, the integration level of integrated circuit (IC) is continuous growing. An accompanying power dissipation problem is urgent to be solved. In order to circumvent this problem, the operation voltage of the transistor should be reduced [1]. The subthreshold swing (SS) of MOSFET cannot be below 60 mV/decade at room temperature, which restricts the reduction of threshold voltage VTH and supply voltage VDD [2]. Many efforts have been devoted to the research and the development of devices with novel transport and switching mechanisms to beat the Boltzmann limit, including negative capacitance field-effect transistor (NCEFT) [3, 4], resistive gate FET [5], nano-electro mechanical FET (NEMFET) [6, 7], impact ionization metal-oxide-semiconductor (I-MOS) [8, 9], and tunneling FET [10, 11]. Among them, NCFET has aroused much attention because it can achieve a steep SS without losing the drive current [12,13,14,15]. Doped HfO2 (e.g., HfZrOx (HZO) and HfSiOx) has been widely used in NCFETs [4, 16, 17]; it is compatible with the CMOS process [18]. A theoretical study has shown that the undesired hysteresis occurs due to unmatched ferroelectric capacitance CFE to underlying MOS capacitance CMOS in NCFET [19]. However, the effect of matching between CFE and CMOS on the electrical characteristics of NCFETs is still a concern in the experiments.

In this work, the electrical characteristics of NC Ge FETs with different MOS capacitances are studied based on the different matching between CFE and CMOS. Although SS less than 60 mV/decade does not appear, the hysteresis-free transfer characteristics and better electrical properties are obtained. Apparent peaks of CFE versus VFE curves demonstrate NC effect of HZO based NCFETs. The better matching of CFE and CMOS contributes to steeper SS and higher on current, which is beneficial to the logic applications.

Methods

The key fabrication process of Ge NCFETs is shown in Fig. 1a. Four-inch n-Ge(001) wafers with a resistivity of 0.088–0.14 Ω·cm were used as the starting substrates. After pre-gate cleaning, Ge wafers were loaded into an ultra-high vacuum chamber for surface passivation using Si2H6. Two passivation durations of 40 and 60 min were used. Then, TaN/HZO/TaN/HfO2 stack was deposited. The thicknesses of the HfO2 dielectric layer and HZO FE layer are 4.35 and 4.5 nm, respectively. After gate patterning and etching, source/drain (S/D) regions were implanted using boron ions (B+) at an energy of 30 keV and a dose of 1 × 1015 cm−2. S/D metal Nickel was formed using a lift-off process. Finally, rapid thermal annealing at 450 °C for 30 s was carried out. Control MOSFET with TaN/HfO2 stack was also fabricated. Figures 1b and c show the schematics of fabricated NCFET and control MOSFET, respectively. The internal metal gate in the fabricated NCFET counterbalances the potential at the channel surface, which is called the MFMIS structure.

Fig. 1
figure 1

a Key process steps of fabricated NC devices. The schematics of the fabricated b NCFET and c control MOSFET

Results and Discussion

Figure 2a plots the measured IDS-VGS curves of a pair of NCFET and control MOSFET with 40 min surface passivation. Both devices have a gate length LG of 3.5 μm. The NC device with 40 min passivation has a significantly improved IDS than the control MOSFET. The transfer curves of NCFET exhibit a non-hysteretic feature. Point SS versus IDS curves in Fig. 2b show that the NC transistor has improved SS over the control device, although SS of sub-60 mV/decade does not appear. Figure 2c shows that NC transistor obtains a significantly boosted linear transconductance Gm over the control device at VDS of − 0.05 V. Figure 3 compares the electrical performances of NCFET and control MOSFET with surface passivation for 60 min. Similarly, the IDS, point SS and Gm of NCFET are superior to that of control MOSFET.

Fig. 2
figure 2

a The measured IDS-VGS curves of the NCFET and control MOSFET with 40 min passivation. Comparison of b point SS versus IDS and c Gm characteristics between NC FET and control MOSFET

Fig. 3
figure 3

a The measured IDS-VGS curves of the NCFET and control MOSFET with 60 min passivation. Comparison of b point SS versus IDS and c Gm characteristics between NCFET and control MOSFET

Figure 4a shows the statistical results of the drive current of NCFETs and control MOSFETs at VDS of − 0.05 V and VGS-VTH = − 1.0 V. NCFETs demonstrate 18.7% and 35.6% improvement in IDS for the 60 min and 40 min surface passivation, respectively, in comparison with the control devices. It is speculated that the NCFETs passivated for 40 min have a better matching between CMOS and CFE over the NC devices with 60 min. Figure 4b shows that NCFETs obtain 26.4% and 51.3% improvement in maximum transconductance Gm,max for 60 min and 40 min surface passivation, respectively, in comparison with the control devices. It is seen that the control MOSFETs with surface passivation for 40 min have a higher IDS and Gm,max than the devices passivated for 60 min, which is due to the larger CMOS induced by the smaller equivalent oxide thickness (EOT). The internal metal gate provides an equipotential plane; the device can be equivalently modeled as a capacitive voltage divider. The total capacitance CG is a series of CFE and CMOS. The internal gate voltage is amplified owing to the NC effect. The internal voltage amplification coefficient β =  ∣ CFE ∣ / ∣ CFE ∣  − CMOS gets the maximum when |CMOS| = |CFE| [20, 21]. Achieving the optimized matching of CFE and CMOS is the prerequisite of the improvement of on current.

Fig. 4
figure 4

The statistical a IDS and b Gm results of NCFETs and control MOSFETs with 40 and 60 min passivation durations

The extracted Vint versus gate voltage VGS curves are shown in Fig. 5a. Vint of NC transistor can be extracted on account of the hypothesis that IDS-Vint curve of NC transistor is exactly identical with IDS-VGS curve of the control device. The internal voltage amplification coefficient dVint/dVGS is shown in Fig. 5b. dVint/dVGS > 1 is achieved in the wide sweeping range of VGS for the NCFET with 40 min surface passivation, contributing to a steeper SS than the control device during the measuring process, which is due to the local polarization switching [22]. It is consistent with the aforementioned results in Fig. 2b. For the NCFET with 60 min passivation, the internal voltage amplification coefficient dVint/dVGS > 1 is achieved during the range of VGS < 0 V for the double sweeping of VGS, which is in agreement with the elevated SS in Fig. 3b.

Fig. 5
figure 5

a Extracted Vint as a function of VGS curves. b The internal voltage amplification coefficient versus VGS curves

Figure 6a shows the extracted CMOS versus VGS curves for NC transistor, which is relying on the Vint-VGS in Fig. 5a and the CG-VGS curves of control MOSFETs. The extracted CMOS is in good agreement with the measured CG. Hence, the validity of the calculation method is demonstrated. The CFE and CMOS versus VFE curves are depicted in Fig. 6b. From the initiation of NC effect, the absolute value of negative CFE of the transistor exceeds CMOS for double sweeping of VGS all the time in Fig. 6b. |CFE| > CMOS and CFE < 0 can cause hysteresis-free characteristics, and the matching of CMOS and CFE is beneficial to the logic applications [23, 24]. Hysteresis-free characteristics in Figs. 2a and 3a are observed attributed to all the domain matching and inhibited charge trapping [25]. The stable polarization switching is responsible for the non-hysteretic characteristics [26]. Furthermore, the large internal gate gain dVint/dVG > 1 is ascribed to the slight discrepancy between |CFE| and CMOS in the subthreshold region, resulting in the steep SS of NC device. Meanwhile, there is a better matching between CFE and CMOS for the NCFET with 40 min passivation than the NCFET with 60 min passivation. Thus, this provides direct evidence to indicate that the NCFET with 40 min passivation possesses a better electrical performance than the NCFET with 60 min passivation. The FE polarization changes the VFE; hence the charge of FE varies. The total charge multiplies, which is attributed to the FE polarization besides the increment of VGS. In other words, for the given VGS, the charge in the channel increases so the IDS improves. As a consequence, the steep SS of transfer characteristic appears in the experiments.

Fig. 6
figure 6

a Measured CG and extracted CMOS as a function of VGS. b CFE and CMOS versus VFE curves

Conclusions

The hysteresis-free transfer characteristics are obtained for the NCFETs with 40 and 60 min passivation. NC Ge pFETs with 40 min passivation have better electrical characteristics than the NC device with 60 min passivation in experiments. We also demonstrate the NC effect of HZO based NCFETs. For NCFETs, the steep SS and dVint/dVGS > 1 are obtained. The NCFET with 40 min passivation has achieved a good matching between CFE and CMOS, which contributes to the non-hysteretic characteristics. The different NC behaviors are considered to be related to the microscopic domain wall switching in the FE thin films.