Fully transparent field-effect transistor with high drain current and on-off ratio

Transparent oxide thin-film transistors (TFTs) have been studied extensively for over a decade for applications that include displays, wearable electronics, and smart windows.^1–3 Due to the relatively high mobilities of the transparent oxide semiconductors used, when scaled to small dimensions, transparent oxide TFTs are expected to have high drain current (I_D, drain-to-source current per channel width) and low power consumption, which are important for energy-sensitive applications. Tin-based oxide systems are promising materials for transparent oxide TFTs due to the superior oxygen stability and high mobility of these materials at room temperature.^4,5 Among stannates, the transparent perovskite BaSnO₃ is attracting worldwide attention following the recent discovery that La-doped BaSnO₃ has both high mobility at room temperature and excellent oxygen stability. Mobilities as high as 320 cm² V⁻¹ s⁻¹ and 183 cm² V⁻¹ s⁻¹ have been achieved in La-doped BaSnO₃ single crystals⁴ and epitaxial thin films,⁶ respectively, at room temperature. The high mobility of BaSnO₃ is attributed to the small effective mass originating from the large dispersion of the conduction band from the Sn 5s orbital,⁷ high dielectric constant (K ≈ 20)⁸ which reduces dopant scattering,⁴ and small phonon scattering.⁹ 

The performance of TFTs based on BaSnO₃ has been steadily improving in recent years, with the highest I_D achieved being 0.021 mA/μm.¹⁰ This performance is limited by the large size of the BaSnO₃-based TFTs, which have channel dimensions of 100–200 μm. These large devices have been patterned using metal shadow masks during growth, i.e., without photolithography. Attempts to use photolithography to define smaller devices have been plagued by the creation of oxygen vacancies in the BaSnO₃ film, including the undoped BaSnO₃ buffer layer, during the ion-milling process. These vacancies make the undoped BaSnO₃ buffer layer conductive, shunt the device, and degrade performance.¹¹ This problem is well known in other oxides as well.¹² 

Here, we report growth of high mobility La-doped BaSnO₃ thin films on MgO substrates using molecular-beam epitaxy and use reactive ion etching (RIE) to pattern BaSnO₃–based TFTs. With appropriate conditions, this etching technique enables BaSnO₃ to be etched with a controllable etch rate while preserving surface roughness and electrical properties including conductivity and mobility. The resulting micrometer-scale photolithography defined BaSnO₃–based transparent TFT has a peak field-effect mobility of 17.2 cm²/V⁻¹ s⁻¹, an on-off ratio over 1.5 × 10⁸, I_D over 0.468 mA/μm, and a peak transconductance (g_m) of 30.5 mS/mm. As we describe below, this is one of the highest performance fully transparent oxide TFTs ever reported.

Epitaxial La-doped BaSnO₃ films were grown on (001) MgO substrates by adsorption-controlled growth using molecular-beam epitaxy with barium, lanthanum, and SnO₂ sources. The details of the growth process are given elsewhere.⁶ Epitaxial growth of single-phase La-doped BaSnO₃ on MgO substrates was confirmed by x-ray diffraction (XRD) measurements; the full width at half maximum of the BaSnO₃ 002 peak is 0.31° (see Fig. S1 of the supplementary material). MgO substrates, with a bandgap of about 7.8 eV,¹³ are used to avoid the potential complications of persistent photoconductivity from the substrate arising from light exposure during the photolithography process.¹⁴ On the MgO substrate, an undoped BaSnO₃ layer followed by a La-doped BaSnO₃ layer is grown to form the channel of the TFT as is typical for BaSnO₃-based TFTs.^10,15–19 The purpose of the undoped BaSnO₃ layer is to reduce the density of threading dislocations that propagate into the overlying doped (and conducting) BaSnO₃. Even with such a buffer layer, the threading dislocation density of BaSnO₃ films grown on MgO as well as other commonly used substrates is of order 10¹¹ cm⁻²;^6,15,20 an ideal substrate for BaSnO₃ has yet to be demonstrated, but several lattice matched candidates are being developed.^21–23 

In order to make micrometer scale TFTs, it is critical to be able to lithographically pattern the high mobility La-doped BaSnO₃ channel layer without degrading its electrical performance. A few methods have been used in the past to pattern a La-doped BaSnO₃ channel layer. Early BaSnO₃ TFTs used shadow masks to define the channel structure during deposition.^10,15–19 This method can pattern the La-doped BaSnO₃ layer without degrading its surface roughness and electrical properties, but the channel length is limited to about 100 μm. More recently ion-milling has been employed to pattern the La-doped BaSnO₃ layer. Although ion-milling can etch La-doped BaSnO₃ to micrometer scale dimensions with controllable etch rate, the size of the channel length reported was about 100 μm.¹¹ Importantly, the resulting BaSnO₃ TFT with an undoped BaSnO₃ layer could not be depleted at room temperature and its on-off ratio was only 2. The reason for the inability to deplete the TFT at room temperature was attributed to significant charge trapping,¹¹ but our own experiments reveal that undoped BaSnO₃ becomes conductive when subjected to ion-milling. The conductivity of an undoped BaSnO₃ layer increased more than 10⁸ times following ion-milling even when we use the smallest available acceleration voltage of our equipment, 400 V. Such an enhancement in conductivity of the undoped BaSnO₃ layer is consistent with it is being caused by the creation of oxygen vacancies from the high energy bombardment of Ar ions during the ion-milling process, as has been reported for SrTiO₃.¹² 

The undesired conductance enhancement in unintentionally doped BaSnO₃ resulting from ion-milling can be reduced by annealing the sample in flowing oxygen at temperatures of about 800 °C, but even after such aggressive annealing, the conductance does not fully return to its original state. The effect of annealing on ion-milled undoped BaSnO₃ in flowing O₂ at temperatures from 300 °C to 800 °C is given in the supplementary material (see Fig. S7 of the supplementary material). Therefore, we believe that the conductance of the underlying undoped BaSnO₃ buffer layer caused by the ion-milling process is partially responsible for the poor room-temperature performance of the prior BaSnO₃-based TFT device subjected to ion-milling.¹¹ Ion-milling can also alter film stoichiometry and lead to charge trapping from defects along the side walls. For this reason, we have developed an alternate etching process to pattern the BaSnO₃ layers into a TFT.

In order to overcome the limitations of previous techniques used to define BaSnO₃-based TFTs, we employ reactive-ion etching (RIE) using chlorine and argon gases to etch the BaSnO₃ epitaxial stack. By using a reactive gas such as chlorine, the RIE technique can etch a BaSnO₃ film with a chemical process rather than a purely physical process, in contrast to ion-milling. Under appropriate conditions, the undoped BaSnO₃ film conductivity issue that we ascribe to the creation of oxygen vacancies can be resolved. Furthermore, with RIE, we are able to achieve a controllable etch rate that is not dependent on sample quality. In short, our RIE etching process for BaSnO₃ provides a controllable etch rate while preserving surface roughness and electrical properties including conductivity and mobility—qualities critical to the fabrication of high-performance TFTs. BCl₃ gas can also be used for RIE etching of BaSnO₃ while preserving surface roughness and electrical properties including conductivity and mobility, but the etch rate is lower than when chlorine gas is used. Therefore, using chlorine gas without adding BCl₃ gas is preferable. A higher etch rate is desirable to reduce etch time, but there is a trade-off between the etch rate and damage to the BaSnO₃ film. In making the TFTs, the etch rate has been maximized without damaging the properties of the BaSnO₃ film by adjusting RIE etching parameters. The RF power of the RIE is around 80–120 W with a DC voltage in the 200–250 V range. The total gas pressure is around 1–2 mTorr, and a gas ratio between 1:5 and 5:1 is used for the Cl₂ and Ar gases.²⁸ 

The device patterning and fabrication process was completed in the cleanroom environment of Cornell’s NanoScale Science and Technology Facility (CNF), except for the atomic layer deposition (ALD) process. For light exposure during photolithography, an Autostep 200 stepper was used. The channel layer was etched using a PT720/740 RIE etching tool. For device isolation, the channel layer was etched down about 18 nm, considerably beyond the 10 nm thickness of the La-doped BaSnO₃ top layer and well into the undoped BaSnO₃ buffer layer. After the etching process, indium tin oxide (ITO) was deposited as source and drain contacts using a PVD75 sputtering tool. The source and drain contacts were patterned by lift-off. A blanket HfO₂ gate dielectric layer was then deposited over the full structure by atomic-layer deposition (ALD), at 300 °C with ozone as an oxidant, which leads to have a deposition rate of 0.8 Å/cycle. The thickness of the HfO₂ layer was 20 nm. To complete the TFT, ITO was deposited again on the gate dielectric to form the gate electrode using the same sputtering tool as used earlier in the process. The gate contact was also patterned by lift-off. Electrical measurements were performed using a Cascade probe station with a Keithley 4200 SCS system. All device measurements presented were measured at room temperature in ambient air.

Figure 1 shows a schematic of the BaSnO₃–based TFT measured in Figs. 2 and 3, together with a scanning electron microscope image of the fully patterned device. The mobility and carrier concentration of the as-grown La-doped BaSnO₃ channel layer in this sample are μ = 65.2 cm² V⁻¹ s⁻¹ and n = 1.81 × 10¹⁹ cm⁻³, respectively, from Hall measurements at room temperature. The channel length and width of the TFT are L = 0.3 μm and W = 0.93 μm, respectively. This is the smallest BaSnO₃–based TFT yet reported.^{10,11,15–19} Further scaling is limited by the resolution of the light exposing tool.

The characteristic curves of the BaSnO₃ TFT are shown in Fig. 2. Figure 2(a) shows the transfer characteristic of the device: I_D vs V_GS (gate-to-source voltage) and μ_FE (field-effect mobility) vs V_GS. As shown in Fig. 2(a), the TFT can be depleted completely at room temperature—a first for a photolithographically defined BaSnO₃-based TFT.¹¹ The field-effect mobility of the device is calculated from the relation $μFE=LCoxWVDS∂IDS∂VGS$⁠.^30, I_DS, L, C_ox, and W are the drain-to-source current, the channel length, the capacitance of the gate dielectric per unit area, and the channel width, respectively. The capacitance of the HfO₂ gate dielectric was measured as shown in Fig. S9 of the supplementary material with a metal-oxide semiconductor (MOS) structure with dimensions 50 μm × 50 μm. This MOS capacitor has the same La-doped BaSnO₃ channel layer and the same ITO gate electrode as the TFT, i.e., the layers measured in the MOS capacitor were deposited on the MgO wafer and patterned into the MOS capacitor test structure and into the TFT. For calculating μ_FE, the maximum C_ox at the given voltage range has been used so as to not overestimate μ_FE. The calculated μ_FE is 17.2 cm² V⁻¹ s⁻¹. The large difference between the field-effect mobility and the Hall mobility can be attributed to the contact resistance. The contact resistance includes the sheet resistance of the ITO contact layer and interface resistance between the ITO contact layer and the La-doped BaSnO₃ channel layer. In the supplementary material (see Fig. S10), we utilize the measured value of the contact resistance to calculate the field-effect mobility; the result is 72 cm² V⁻¹ s⁻¹, which is comparable to the Hall mobility of the as-grown channel layer. From the details of this calculation, it is clear that the high contact resistance is significantly limiting the TFT performance; a lower resistance contact will increase the field-effect mobility and thus the drain current. There is dispersion in the capacitance at measurement frequencies higher than 100 kHz, and it is believed to be related to the trapping of charge by defects or impurities within the HfO₂ film and at the interface between the HfO₂ and La-doped BaSnO₃ channel layer since the cutoff frequency is 51.6 MHz. The cutoff frequency is calculated from the relation $f=12πRC$ and is much higher than 100 kHz. The on-off ratio is over 1.5 × 10⁸. This is the highest on-off ratio among BaSnO₃ TFTs when an undoped BaSnO₃ buffer layer is used.^{10,11,15–18} The subthreshold swing has been calculated from the equation $S=∂ log IDS∂VGS−1$⁠, and S is 0.15 V dec⁻¹. Figure 2(b) shows the transconductance of the device at V_DS = 1 V as a function of V_GS. The transconductance is calculated from the relation $gm=∂IDS∂VGS$⁠. The maximum transconductance is 30.5 mS/mm at V_DS = 1 V. This is much higher than the previous record transconductance in a BaSnO₃ based TFT, which is only 2 mS/mm.¹¹  Figure 2(c) shows the output characteristic of the device, which is I_D vs V_DS (drain-to-source voltage) by varying V_GS from 2 to −7 V. As shown in Fig. 2(c), the drain current reaches over 0.468 mA/μm. The slow increase in the drain current with respect to V_DS is attributed to the high contact resistance. The resistance of the ITO/La-doped BaSnO₃/ITO channel structure vs channel length is plotted in Fig. S10 of the supplementary material. The source/drain contact resistance between the ITO contact layer and the La-doped BaSnO₃ channel layer, which is 5.1 kΩ μm, is calculated by extrapolating a linear fit to the plot. The high contact resistance affects the slow response of I_D vs V_DS in the measured output characteristics as shown in Fig. S11 of the supplementary material; thus, the drain current can be improved by lowering the contact resistance.

The drain current vs channel length at V_GS = 2 V and V_DS = 5 V is plotted in Fig. 3. The drain current is inversely proportional to the overall length of the channel except at the shortest channel length of 0.3 μm, showing little degradation with respect to device scaling. The close agreement between the calculated field-effect mobility (with the effect of contact resistance included) and Hall mobility indicates that the deviation from linear behavior at the short channel length of 0.3 μm in the inset of Fig. 3 is likely due to the contact resistance and not short channel effects such as velocity saturation. At a channel length of 0.3 μm, the channel resistance is comparable to the contact resistance and the effective voltage applied to the channel is greatly diminished, lowering drain current. A lower resistance contact is expected to reduce the deviation evident in Fig. S11 (a) between the ideal square-law transistor model and actual TFT performance.

A comparison of the drain current vs on-off ratio among transparent oxide channel TFTs is shown in Fig. 4.^{3,5,10,15–19,32–57} The high I_D of the highest performance TFT with a transparent oxide channel was achieved on cleaved and transferred flakes of β-Ga₂O₃ onto an opaque substrate;⁴⁸ this is not a scalable technology. These data are indicated as gray marks. Our BaSnO₃-based TFT exhibits one of the best performances of all transparent oxide TFTs and is comparable to the best TFTs made with transparent oxide channel materials (in device structures in which other elements are not transparent). Its drain current is second best among all fully transparent scalable oxide channel TFTs.⁵⁷ 

In summary, a fully transparent submicron TFT based on BaSnO₃ has been fabricated with a high drain current and on/off current ratio. This breakthrough is made possible by (1) high mobility bare films in combination with (2) the development of a micrometer-scale etching method that preserves the surface roughness, conductivity, and mobility of BaSnO₃ films. These results demonstrate the tremendous potential of BaSnO₃ for the future of transparent electronics.