Nanoscale etching of perovskite oxides for field effect transistor applications

Perovskite oxides have been under extensive research with their vast aspects of applications such as superconductivity,¹ dynamic random-access memory,² and negative capacitance field effect transistors (FETs).³ Recently, a wide bandgap (3.1 eV) perovskite material, BaSnO₃, has emerged as a potential semiconductor channel material for field effect transistors.^4–7 A high room-temperature mobility of up to μ = 180 cm²/V s with a high sheet density of n_s> 1 × 10¹⁴ cm⁻² has been demonstrated on molecular beam epitaxy (MBE) grown La-doped BaSnO₃ thin films,^6,7 as well as a high electron saturation velocity of 1.8 × 10⁷ cm/s.⁸ BaSnO₃-based heterostructure FETs (HFETs)⁹ and metal oxide semiconductor field effect transistors with Al₂O₃,¹⁰ HfO₂,¹¹ or LaInO₃¹² as gate dielectrics have been reported. One particularly interesting device proposed¹³ is to use a perovskite oxide that has a very large dielectric constant (such as BaTiO₃) as a gate dielectric on top of the BaSnO₃ channel to form a heterojunction so that the device can operate at high current and the high sheet charge density in the channel can be depleted at a low threshold voltage.

To develop such high performance perovskite oxide electronic devices, precise and anisotropic etching processes of perovskite oxides are required. It is difficult to etch perovskites in general by wet etchants. This is especially true for titanate perovskites. Wet etching processes have been only used for substrate surface treatment and preparation before epitaxial growth.^14–16 The wet etchants used are highly aggressive and can attack photoresists; therefore, the wet etching processes are not suitable for patterning etching for device fabrication. Therefore, the etching studies of perovskite oxides such as Pb(Zr_xTi_1 − x)O₃ (PZT),^17,18 Ba_xSr_1 − xTiO₃(BST),^19,20 and BaTiO₃²¹ so far reported have been focused on dry etching processes using either Cl- or F-based plasma chemistries. For Cl-based plasma chemistries, it has been found that the BCl₃/Cl₂/Ar plasma etching process has higher etch rates for etching of PZT than the Cl₂/Ar process because some oxycompound etching products such as B_xO_y and/or BCl-O are more volatile.¹⁸ For F-based plasma chemistries, it was found that the etch rates of BST were lower because of the stronger competition of plasma deposition and etching due to less volatility of etched products.²⁰ The etching study of BaTiO₃ by Li et al. using different F-based plasma chemistries suggested that the etch rate of SF₆ plasma was higher than that of C₄F₈ and CF₄.²¹ There are, however, little comparative studies reported on etching of perovskite oxides BaSnO₃ and BaTiO₃ using F-based and Cl-based plasma chemistries. Both of these two perovskite oxides are especially interesting for device engineering as mentioned above, because BSO is a semiconductor channel material due to its high mobility and BTO is a good candidate as a gate dielectric or barrier due to its very large dielectric constant.²² In this work, we report on the dry etching studies of BSO and BTO in an inductively coupled plasma reactive ion etching (ICP-RIE) system using both Cl-based and F-based plasma chemistries. Further, based on the etching processes developed in this study, BaTiO₃/BaSnO₃ HFETs with record high current density and transconductance are demonstrated.

The BSO samples used for etching studies have 40–50 nm thick (001) BSO on (001) SrTiO₃ (STO) or DyScO₃ (DSO) substrates. For etching of BTO thin films, the samples consist of BTO/BSO heterojunctions on STO substrates with a BTO thickness of 20 nm. Apart from the etching study of BTO, another reason we use these heterojunction samples is that field effect transistors can be directly fabricated as long as a base line etching process is established. All these samples were grown by MBE under active oxygen flow (RF plasma source). A postgrowth annealing process was performed to reduce oxygen vacancies. After the doped-BSO layer epitaxial growth, Hall measurement indicated that the doped-BSO layer had a charge density of n_2d = 5.5 × 10¹⁴ cm⁻² (N_d = 2.6 × 10²⁰ cm⁻³) and a mobility of μ = 52 cm²/V s. The details of the growth and annealing processed have been reported elsewhere.⁷ 

The dry etching studies were performed in a Plasma Therm SLR770 ICP-RIE system using photoresist (PR) as the etching mask. Patterns were created using Shipley S1813 PR by i-line stepper photolithography. After etching, the PR was stripped in hot N-methyl-2-pyrrolidone/acetone/2-propanol. The etching processes were performed with BCl₃/Ar and CF₄/Ar plasma chemistries for comparative studies.

For device fabrication, we used the BCl₃/Ar (40/10 sccm flow rates, 250 ICP power, 550 V plasma induced DC bias at 10 mT process chamber pressure) as the baseline etching process to etch BTO only for ohmic contact vias and BTO/BSO mesa etching for device isolation. Ti/Au (50 nm/100 nm) ohmic contacts were formed by electron beam evaporation and lift-off after BTO via etching. Pt/Au (50/100 nm) Schottky gates with a gate length of L_g = 1.5 μm and a gate width of W_g = 75 μm were formed by electron beam evaporation and lift-off on top of the BaTiO₃ barrier. The devices have a source-drain spacing of L_sd = 3.5 μm.

One of the key points for a good etching process is to consider how volatile the etched products are in favor of enhancing etch rates and achieving smooth etched surface. In general, the boiling point of an etched product is a good indication of its volatility. We, therefore, checked the boiling points of potential etching chemical products in this etch study from the literature²³ and they are summarized in Table I. As shown, the Cl-chemical products have significantly lower boiling points than their F-counterparts, indicating likely higher volatilities. To confirm this hypothesis, we performed experiment to look at the smoothness of etched surfaces and how the etch rate varies with etching durations.

In this experiment, we fixed all etching parameters (250 W ICP power, 50 W RF power, and 10 mTorr chamber pressure) and checked how the etched depths and smoothness changed with etching durations for CF₄/Ar and BCl₃/Ar (40/10 sccm flow rates) plasma chemistries. The etched depths of BTO and BSO as a function of etching durations are shown in Fig. 1. It is found that the depths of both BTO and BSO samples etched by BCl₃/Ar increase linearly with the time duration or constant etch rates of 3.2 and 2.8 nm/min are achieved, respectively. Contrarily, the etching rates of BTO and BSO by CF₄/Ar plasma are strongly dependent on the etching durations. As the time duration increases from 2 to 14 min, the average etch rate of BTO decreases from ∼1.8 to ∼0.7 nm/min and from 1.4 to 0.8 nm/min for BSO. This is likely the result of redeposition of F-etching chemical products. To further confirm this, we examined the smoothness of etched surfaces by atomic force microscopy (AFM). The samples were etched at the same conditions (BCl₃/Ar or CF₄/Ar with 40/10 sccm flow rates, 250 W ICP power, 50 W RF power, and 10 mTorr chamber pressure) but with different durations. As shown in Fig. 2, with 10 × 10 μm scans, the etched surface roughness for BCl₃/Ar samples is better than that of CF₄/Ar samples with root-mean-square (RMS) roughness changed from 1.9 nm to 4.9 nm for BTO and from 1.6 nm to 3.0 nm for BSO. It is noted that, even the samples were etched with deeper depths by BCl₃/Ar plasma, their RMS roughness values were found to be less than that of samples etched by CF₄/Ar plasma. Therefore, we concluded that the BCl₃/Ar plasma is more effective for etching of the perovskite oxide BaSnO₃ and BaTiO₃. This is mostly due to the fact that etching products of F-chemistries are less volatile than the case of Cl-based chemistries. Such an observation is also consistent with previous studies on etching characteristics of other perovskite oxides such as BST.²⁰ 

From the above, we conclude that BCl₃/Ar is a more viable plasma chemistry for FET device fabrication, as it gives an etch rate independent of etching duration. Next, we investigate how the etch rate of BCl₃/Ar etching processes is dependent on the plasma density and ion energy. A great advantage of an ICP-RIE system is that it can control the plasma density by varying the ICP power and ion energy by varying the RF power independently. In this experiment, we kept all process parameters (40/10 sccm flow rates for BCl₃/Ar, 10 mT chamber pressure) the same and varied only the ICP power or RF power. First, as shown in Fig. 3(a), at the fixed RF power of 100 W, the etch rate increases from 3.2 to 8.5 nm/min for BTO and from 3.1 to 6.5 nm/mm for BSO as ICP power increases from 150 to 300 W. This suggests that a high plasma density can promote a higher etch rate. It is interesting to note that the higher etch rate with the increase of the ICP power; hence, plasma density does not increase the roughness of the etched surface. As shown in Fig. 3(b), the RMS roughness of AFM images of etched surfaces is essentially independent of ICP power for both BSO (∼1.3 nm) and BTO (∼3.5 nm). Next, with the fixed ICP power of 300 W, the etch rates are more strongly dependent on the RF power or the ion energy. When the RF power was varied from 50 to 200 W, the plasma induced DC bias increased from 253 to 630 V. With these conditions, the etch rate increased from 3.2 to 10.9 nm/min for BTO and from 2.6 to 7.2 nm/min for BSO [Fig. 3(c)]. Also, as shown in Fig. 3(d), the RMS roughness of etched BSO surfaces increases only slightly when the ion energy is below 535 eV and increases from 1.32 to 2.83 nm when the ion energy is increased from 535 to 630 eV, while the roughness of the etched BTO surface is less dependent in the same ion energy range. Overall, the stronger dependence of the etch rate on ion energy is expected as the bombardment of energetic ions not only increases the physical sputtering rate but also assists the etching kinetics of reactive chlorine and boron radicals. In other words, it is suggested that the BCl₃/Ar etching process is influenced by both physical ions sputtering and chemical etching processes. The combination of both physical sputtering and chemical etching is in general desired for device fabrication as it can be easily tuned to the target etch rate and the etched profile with a smooth surface morphology.²⁴ 

After the baseline etching process was developed, we fabricated BTO/BSO HFETs using BCl₃/Ar plasma. The schematic cross-sectional view of the device is shown in Fig. 4(a). The layer structure consists of 20 nm BTO, 21 nm La-doped BSO, and 4 nm undoped BSO on the STO substrate. BCl₃/Ar plasma etching was performed to etch off BTO for ohmic contact vias and BTO/BSO for device mesa isolation. The etching condition is 40/10 sccm flow rates for BCl₃/Ar, 10 mT chamber pressure, 150 W RF power, and 250 W ICP power. Under this condition, the etch rate for BTO is 6.5 nm/min and that for BSO is 5.2 nm/min. The scanning electron micrograph as shown in Fig. 4(b) indicated the smooth side wall of the etched device mesa. The BaTiO₃/BaSnO₃ HFETs have a gate length of 1.5 μm and a source and drain spacing of 3.5 μm. A contact resistance of R_c = 0.34 Ω mm and a specific contact resistivity of ρ_C= 4.8 × 10⁻⁶ Ω cm² were extracted using the transmission line method. Capacitance-voltage (CV) measurements on fat FETs were performed from −5 to 0 V. The measured capacitance and loss are shown in Fig. 5(a) as a function of gate bias. From the integration of the CV sweep, a sheet carrier density of n_s = 5.0 × 10¹³ cm⁻² is obtained. Comparing the Hall measurement data with a high sheet carrier density of n_s = 5.5 × 10¹⁴ cm⁻², it is suggested that most of the 21 nm doped-BSO layer is depleted with a depletion width of 19.1 nm. The charge depletion is caused by the energy band alignments when the BTO cap layer is added and the Pt/Au Schottky barrier contacts are deposited. Therefore, the measured capacitance at 0 V is mostly dominated by the depletion capacitance as the 20 nm BTO dielectric oxide capacitance is much larger. To confirm that we calculated the dielectric constants of BTO and BSO (ɛ_BTO and ɛ_BSO) with measured capacitance values from −3 to 0 V shown in Fig. 5(a). As shown below, the total capacitance is calculated as the BTO capacitor (C_BTO) and the depletion capacitor (C_BSO) connected in series,

where V_bias is the bias voltage, V_bi is the built-in potential, W_d is the depletion thickness in BSO, and N_d is the doping density in BSO. As shown in the inset of Fig. 5(b), the extracted dielectric constant of BSO is 21 with no field strength dependence. The dielectric constant of BTO ranges from 135.8 to 100.7 with the electric field ranging from 0 to 300 kV/cm.

Figure 6(a) shows the family I-V characteristics of the fabricated HFET. The device exhibited a saturation current density of I_dmax = 287.6 mA/mm at a gate bias of 1 V. This is significantly higher than what have been reported on perovskite oxide FETs. Previously reported highlights on saturation current of perovskite oxide FETs include an I_max of 140 mA/mm demonstrated on LaAlO₃/SrTiO₃ HFETs with L_g = 60 nm,²⁵ I_max = 36 mA/mm on 3 μm SrSnO₃ MESFETs,²⁶ and I_max = 12.5 mA/mm on 100 μm fat HfO₂/SrTiO₃/BaSnO₃ HFETs,⁹ respectively. The device on-resistance R_on was determined to be 10.1 Ω mm from the output characteristics in the linear fit of Fig. 6(a) at V_g = 1 V.

The device transfer characteristics and transconductance as a function of the gate bias are shown in Fig. 6(b). The maximum transconductance of g_m = 91.3 mS/mm was measured at the gate bias of 0.4 V. To our knowledge, this is the highest transconductance so far reported on perovskite field effect transistors. Previously, a g_m value of 53 mS/mm was reported for highly scaled LaAlO₃/SrTiO₃ HFETs with a gate length of 60 nm.²⁵ Also, SrSnO₃ MESFETs with L_g = 3 μm were reported to have g_m = 17 mS/mm.²⁶ For long channel devices, HfO₂/SrTiO₃/BaSnO₃ HFETs with a gate length of 100 μm were reported to exhibit a g_m of 2 mS/mm.¹⁰ From the extrapolation of the linear regime of transfer characteristics to the intercept of the gate bias axis, we extracted the threshold voltage of V_th = −1.75 V. It should be mentioned that this threshold voltage is significantly lower than that of reported perovskite FETs with a high sheet charge density.^27,28 It should be emphasized that to be able to deplete 5 × 10¹³ cm⁻² electron charge at such a low threshold voltage is remarkable. The high transconductance and outstanding gate modulation efficiency as well as the low threshold voltage are attributed to the high dielectric constant of BaTiO₃ or the high gate capacitance (C_ox).

At current saturation, the drain saturation current can be expressed as

where μ_eff is the effective FET mobility. Therefore, the square root of on-state saturation current should have a linear dependence of gate bias V_g, as shown in Fig. 6(c). From the linear slope of the square root of I_dsat as a function of the V_g plot, we can extract the effective FET mobility using the following expression:

The extracted μ_FET has a value of 45.3 cm²/V s Here, we used the capacitance at 0 V gate bias shown in Fig. 5(a) as C_ox = 0.82 μF/cm². Obviously, both the high sheet charge density in the channel and high electron mobility constitute the high saturation current demonstrated in these devices.

According to Fig. 6(b), the device has a current on/off ratio of 114 with the off-state drain current being 2.5 mA/mm at V_g = −4 V and V_d = 4.5 V. This relatively low current on/off ratio is due to the gate leakage current. Figure 6(d) shows the gate to source and drain diode characteristics. At a gate bias of −4 V, the gate has a leakage current density of 2.7 mA/mm, which is basically the off-state drain current. The relatively high gate leakage current is mainly due to the high threading dislocation density of the epitaxially grown films. It has been reported previously that in BaSnO₃ films grown by MBE on SrTiO₃ substrates (lattice mismatch 5.4%), the dislocation density is ∼1.5 × 10¹² cm⁻² calculated from the mismatch dislocation spacing.⁷ Nevertheless, the record high maximum current density and transconductance as well as the capability to deplete 5.0 × 10¹³ cm⁻² sheet electron charge at a low threshold voltage of −1.75 V demonstrated on these BTO/BSO HFETs are remarkable. The high device performance is attributed to the high mobility and high sheet charge density heterostructure films grown by MBE, and the device fabrication engineering including the precise control of etching processes developed in this study.

In summary, we have performed the etching studies of MBE-grown perovskite oxide BTO and BSO thin films using BCl₃/Ar and CF₄/Ar plasmas in an ICP-RIE system. The results show that the etch rates of BCl₃/Ar based etching processes are independent of the etching duration, while the etch rates of CF₄/Ar etching processes decrease with the etching duration. Also, BCl₃/Ar plasma etching results in smoother etched surfaces than the case of CF₄/Ar plasma. All these effects were attributed to the low volatility of etching chemical products in F-based plasma. Using the BCl₃/Ar plasma etching process developed from this work, BTO/BSO HFETs with a gate length of 1.5 μm were fabricated. The devices demonstrated record a high current density of 287.6 mA/mm, a maximum transconductance of 91.3 mS/mm, and a high effective FET mobility of 45.3 cm²/V s. Remarkably, the device is capable of deleting 5 × 10¹³ cm⁻² electron charge at a low threshold voltage of −1.75 V due to the gate capacitance resulted from the high dielectric constant of BTO.

This work was supported by the DARPA DREaM program (ONR No. N00014-18-1-2034, Program Manager Young-Kai Chen, monitored by the Office of Naval Research, and Program Manager Paul Maki).