Ta₂O₅/SiO₂ Multicomponent Dielectrics for Amorphous Oxide TFTs

1. Introduction

Amorphous oxide (AO) thin films have greatly progressed in a relatively short time, having found market application in the display industry where materials such as indium-gallium-zinc oxide (IGZO) appear as an advantageous alternative to Si technologies [1,2]. Besides conventional electronics, their characteristics make them suitable for concepts such as transparent and flexible electronics [3,4,5] or even paper electronics [6,7,8,9], allowing for interesting applications in fields such as medical, security and item tracking [10,11], crucial under the scope of the Internet of Things (IoT). One of the main advantages of AO is their good properties even when fabricated at low temperatures, with temperatures below 200 °C being imposed when considering flexible substrates or even paper substrates. Lower annealing temperatures unavoidably result in poorer device performance and stability. When considering these lower annealing temperatures the IGZO properties are known to be strongly related to its processing conditions [12] and adjustment of the cation ratio in the material thus plays a major role in its optimization [13]. In addition, employing dielectrics with high dielectric permittivity, ε_r, (high-κ dielectrics) can compensate for poorer performances by reducing driving voltages (e.g., as required in power-efficient applications within IoT) and improving gate voltage swing due to higher gate capacitances [14,15]. For low temperature deposition of dielectrics, physical techniques such as pulsed laser deposition (PLD) [16,17], thermal evaporation [18,19] and sputtering can be used, the latter allowing the deposition of most materials without any intentional substrate heating [20], at a large scale and with low contamination [21]. Several high-κ materials have been employed for gate dielectrics in low-temperature IGZO TFTs (or other ZnO-based TFTs) including: Al₂O₃ [22,23,24], HfO₂ [25], Ta₂O₅ [26,27], Y₂O₃ [28,29,30] and ZrO₂ [31]. Nevertheless, high-κ materials present some disadvantages, aggravated by low thermal budgets, when compared to conventional dielectrics such as SiO₂. While having a relatively low permittivity (ε_r = 3.9), SiO₂ is stable, has a very high band gap (E_g) of 9 eV and has a low defect density making it a good insulator with a high breakdown voltage. Additionally, it is amorphous and has a good interface with IGZO. On the other hand, ionic bonds in high-κ dielectrics result in high defect concentrations with oxygen vacancies (V_O) being the primary source of traps. These can be a source of fixed charges or act as electron traps, scattering carriers in the channel (decreasing mobility), changing the threshold voltage (V_th) and assisting oxide breakdown and gate leakage mechanisms [32], decreasing device performance, stability and reliability. Furthermore, high-κ materials are often polycrystalline (even at low temperatures) with grain boundaries contributing both to degraded surface properties and acting as preferential paths for leakage current and impurity diffusion [33,34]. When choosing the dielectric material, band alignment should be considered as at least 1 eV of conduction/valence band offset is desirable for blocking electron/hole injection. ε_r is normally inversely proportional to E_g (Figure 1), and the band alignment of several dielectrics with IGZO can be found in the work of Hays et al. [35].

Incorporation of higher E_g materials with high-κ dielectrics effectively increases E_G and can result in amorphous materials to much higher temperatures due to their increased disorder [36]. Regarding low temperature (T < 200 °C) ZnO-based TFTs with multicomponent dielectrics, sputtered Bi_1.5Zn_1.0Nb_1.5O₇ showed ε_r ≈ 51 but was polycrystalline at room-temperature (RT) [37] while sputtered Ba_0.5Sr_0.5TiO₃ (ε_r ≈ 28) was amorphous at RT but presented significant leakage [38]. The insertion of MgO into the latter dielectric improved insulation at the expense of ε_r (to close to 18) and it was applied to IGZO TFTs with good performance on plastic substrates [39]. In previous studies conducted by our group, sputtered HfO₂ was combined with SiO₂ or AlO_x effectively preventing the crystallization of the material at RT [20,40]. Similarly, Ta₂O₅ was combined with SiO₂ or AlO_x resulting in TFTs with good insulation and good performance at T ≤ 150 °C [14,27]. Another approach to reduce leakage is the use of multilayered gate dielectric stacks in which SiO₂ layers (or other low defective materials) are employed at the dielectric/semiconductor interface. In general, these layers result in lower trap densities at the interface improving device performance and stability while imposing a higher barrier for carrier injection. Sputtered HfO_xN_y/HfO₂/HfO_xN_y [41] and HfO₂/SiO₂ [21,42] stacks showed improved interface quality and insulation properties when compared to the respective single layer high-κ dielectrics. Hsu et al. showed excellent flexible IGZO TFTs fabricated at RT with electron beam evaporated SiO₂/TiO₂/SiO₂ [43]. Solution based processes can also be used for depositing high-κ dielectrics such as HfO₂, ZrO₂, and Ta₂O₅, permitting devices with very good performance [44,45] and multilayered stacks with these processes were shown to be promising even when considering low temperatures (T < 150 °C), as shown by Carlos et al. [46], while others have shown that this approach can even improve mechanical flexibility [47]. While masking some of the high-κ dielectrics disadvantages, these approaches can significantly decrease the effective oxide ε_r. Moreover, while the multicomponent and multilayer concepts in dielectrics were already demonstrated, the dielectric layer composition and the gate dielectric stack architecture need to be carefully considered to obtain the best combination of performance and reliability. This is particularly relevant when imposing low thermal budgets (T ≈ 150 °C), given that in such cases the usual benefits of high temperature annealing to improve film quality cannot be considered. These low temperatures are extremely relevant in the current scenario of flexible electronics, where aspects such as hybrid integration with temperature sensitive technologies as organics or usage of unconventional substrates as paper are considered [7,11,48]. Within this context, this work presents a study of the effect of composition in multicomponent dielectric layers composed both by sputtered Ta₂O₅ (a high-κ dielectric with ε_r = 25) and sputtered SiO₂. Besides having a high dielectric permittivity and an amorphous structure, Ta₂O₅ can be deposited by sputtering with a good growth rate without requiring the application of very high power [14]. A low thermal budget (T ≤ 150 °C) was considering in this work, for compatibility with flexible substrates. Multilayered stacks comprising these multicomponent layers and SiO₂ layers were also studied and IGZO TFTs employing these dielectrics were assessed in terms of performance, stability, and reliability.

2. Materials and Methods

2.1. Device Fabrication

IGZO TFTs were fabricated with a staggered bottom gate structure on Corning glass by using standard photolithography patterning procedures, with UV patterning on a Suss MA6 aligner (SUSS MicroTec, Garching, Germany). All layers were produced by radio frequency (RF) magnetron sputtering in an AJA ATC-1300F system (AJA International Inc., North Scituate, MA, USA) without intentional substrate heating. The gate electrodes were sputtered from a Mo target in an oxygen free atmosphere with an RF power density of 3.8 W/cm² resulting in a final thickness of 60 nm. The multicomponent dielectric films were produced by co-sputtering from 2 inch ceramic targets of Ta₂O₅ and SiO₂ under an argon + oxygen atmosphere and the power applied to the Ta₂O₅ target was varied between 50 and 150 W, while power of 150 W to the SiO₂ target was kept fixed, resulting in films with different Ta₂O₅:SiO₂ contents. A substrate bias of 84 V during the dielectric depositions was used as it is known to result in denser and smoother films [21]. A dielectric film of only Ta₂O₅ was also produced for investigating the role of the SiO₂ incorporation in the high-κ dielectric stacks. The 40 nm semiconductor film was sputtered from a 2 inch multicomponent ceramic target of IGZO 2:1:2 (In₂O₃:Ga₂O₃:ZnO mol) with an RF power density of 4.9 W/cm² in an argon + oxygen atmosphere, resulting in an amorphous film with a 4:2:1 (In:Ga:Zn) atomic ratio [49]. The source and drain electrodes were sputtered in the same way as the gate dielectric. All layers were patterned by lift-off technique with the exception of the dielectric which was patterned by plasma etching in SF₆ atmosphere. Resulting TFTs had a width-to-length ratio (W/L) of 320/20 (μm/μm). The devices were annealed on a hot plate at 150 °C for 1 h. A schematic of the device cross-section is shown in Figure 2a. For capacitance analysis, metal-insulator-semiconductor (MIS) devices with the same dielectric layers were produced using p-type Si wafers as substrates and Mo top contacts with areas of 1.88 × 10⁻³ cm². The multicomponent dielectrics were also deposited in p-type Si wafers for Rutherford backscattering spectrometry (RBS) for compositional analysis and for spectroscopic ellipsometry (SE).

Additionally, devices employing gate dielectrics consisting of a stack of a multicomponent layers and a SiO₂ layer where also fabricated. In these stacks the multicomponent layer was produced with a power of 100 W in the Ta₂O₅ target and a power of 150 W in the SiO₂ target, with expected thickness of 200 nm for this layer. The SiO₂ layer in these stacks is employed either at the gate/dielectric interface or at the dielectric/semiconductor interface by sputtering with a power of 150 W before or after the multicomponent layer, respectively, and without breaking vacuum during the entire dielectric stack deposition, with a nominal thickness of 15 nm for the SiO₂ layer. These dielectric stacks and the multicomponent single layer dielectrics are schematized in Figure 2a.

2.2. Films and Devices Characterization

The stoichiometry of the dielectrics was assessed by Rutherford backscattering spectrometry (RBS) using a 2 MeV He beam delivered by a 2.5 MV van de Graaf accelerator. Two solid state detectors placed at 140° and 165° were used to collect the backscattered particles. The RBS spectra were analyzed with IBA DataFurnace NDF software [50]. A HORIBA-Jobin Yvon spectroscopic ellipsometry system was used with an incident angle of 70° in a spectral range between 1.5 and 6.5 eV. The acquired data was analyzed with DeltaPsi 2 software (v2.6.6.212, Horiba, Bensheim, Germany) and fitted using the Tauc–Lorentz dispersion formula [51]. TFTs and MISs were characterized using a Keysight B1500A (Keysight Technologies, Santa Rosa, CA, USA) semiconductor parameter analyzer in a EPS 150 probe station (Cascade Microtech, Beaverton, OR, USA). The MIS devices were characterized through capacitance–voltage (C-V) analysis at 100 kHz.

3. Results and Discussion

3.1. Multicomponent Dielectric Properties

RBS analysis was performed on the multicomponent films sputtered with different powers applied to the Ta₂O₅ target. The RBS spectra are presented in Figure 2b, in which the barrier for each element is indicated by vertical arrows. It is noticeable that with the increase of the power in the Ta₂O₅ target the Ta barrier height increases whereas the Si barrier decreases. Analysis of this data allowed assessing of the stoichiometry of the films, confirming the different Ta₂O₅ and SiO₂ contents within the dielectrics as summarized in

Table 1. Normalized Ta₂O₅ and SiO₂ content in the sputtered T_xS_{100 − x} films, obtained by RBS, and corresponding dielectric permittivities.

Name	Power in Ta2O5 Target (W)	Ta2O5 Content (mol.%)	SiO2 Content (mol.%)	εr
T39S61	50	38.8	61.2	10.0
T60S40	75	59.9	40.1	13.2
T69S31	100	68.6	31.4	15.7
T74S26	125	73.9	26.1	16.4
T75S25	150	75.0	25.0	17.8
Ta2O5	100	100.0	0.0	22.7

. RBS revealed some incorporation of Ar (values presented in Table S1) across all film compositions, as is common for films sputtered in Ar rich atmospheres [52]. Figure S1 presents the Ar content in the dielectric for different Ta₂O₅ contents, showing clearly that the Ar incorporation is more pronounced for higher Ta₂O₅ contents. Nevertheless, the Ar content across all compositions (4.3 ± 0.7%) does not change significantly (from 3.4% to 5.5%) and the Ta₂O₅ and SiO₂ contents presented in Table 1 are thus normalized to 100%, for simplicity. According to the normalized Ta₂O₅ and SiO₂ compositions the multicomponent dielectric layers are denominated as “T_xS_{100 − x}” where $x$ is the approximate Ta₂O₅ percentage of the material, and $100-x$ is thus the approximate SiO₂ percentage. “T” and “S” thus correspond to Ta₂O₅ and SiO₂, respectively, under this nomenclature. For clarity, when the dielectric is based on a single cation, the usual chemical formulas are employed, namely, Ta₂O₅ or SiO₂. For each of these layers its thickness was extracted from the analysis of the SE data, with values in the 200−250 nm range [53]. As for the multilayered gate dielectric stacks, their multicomponent layers have “T₆₉S₃₁^” composition and are then denominated “SiO₂/T₆₉S₃₁” and “T₆₉S₃₁/SiO₂”, according to the position of the SiO₂ layer in the gate dielectric stack, as schematized in Figure 2a.

Knowing the areal density extracted from RBS and the thicknesses from SE (Table S1), the atomic density for each composition can be calculated. It was shown in a previous work that at least for the range of power used here, the Ta₂O₅ density is independent of the power density in the Ta₂O₅ target [53] and since the SiO₂ target power density is kept for all compositions, the atomic densities for each composition can be assumed to depend linearly on the content of each material as per (1)

$$\rho ={\rho }_{T{a}_2{O}_5}.x+{\rho }_{Si{O}_2}\left(100-x\right)={(\rho }_{T{a}_2{O}_5}-{\rho }_{Si{O}_2}).x+{\rho }_{Si{O}_2}$$

(1)

where the approximate percentages of Ta₂O₅ and SiO₂ are $x$ and $100-x$, respectively. Figure 2c shows the atomic density for the different dielectric compositions. From the linear fitting of the data, the oxides’ atomic densities are estimated as ${\rho }_{Si{O}_2}=\left(6.40 \pm 0.10\right)\times {10}^{22}$ cm⁻³ and ${\rho }_{T{a}_2{O}_5}= \left(7.74 \pm 0.17\right) \times {10}^{22}$ cm⁻³ which are below only 3.3% and 1.1% of the theoretical values, respectively, assuming mass densities of 2.2 g·cm⁻³ for SiO₂ [54] and 8.2 g·cm⁻³ for Ta₂O₅ [55]_.

Regarding surface roughness, SiO₂ films are known to be smooth and even with the increase of Ta₂O₅ content in the multicomponent material, ellipsometry and AFM showed that these films are still smooth with T₆₉Si₃₁ layers and Ta₂O₅ films presenting roughness below 0.5 nm and 1.1 nm, respectively [53,56]. Furthermore, XRD characterization showed that these films are amorphous up to 900 °C [56].

From the C-V characterization of the MIS structures (Figure S2), the dielectric permittivity (ε_r) was calculated for the different multicomponent dielectrics, as presented in Figure 2d and in Table 1. As expected, by incorporating Ta₂O₅ the dielectric constant can be greatly increased when compared to that of pure SiO₂. Assuming that ε_r depends linearly on the content of each material as per (2)

$${\epsilon }_r={\epsilon }_{r,T{a}_2{O}_5}.x+{\epsilon }_{r,Si{O}_2}\left(100-x\right)=({\epsilon }_{r,T{a}_2{O}_5}-{\epsilon }_{r,Si{O}_2}).x+{\epsilon }_{r,Si{O}_2}$$

(2)

where $x$ and $100-x$ are the approximate percentages of Ta₂O₅ and SiO₂, respectively, each oxides’ permittivity can be estimated as ${\epsilon }_{r,Si{O}_2}=1.29\pm 1.06$ and ${\epsilon }_{r,T{a}_2{O}_5}=22.48\pm 1.82$ from the linear fitting of the data. These values are relatively lower than the theoretical permittivities (3.9 and 25, respectively) which can be attributed to the defective structure of low-temperature sputtered oxide dielectrics. This is especially noticeable for the SiO₂′s dielectric constant, highlighting the significance of the Ta₂O₅ incorporation as a method for achieving reasonable dielectric permittivities. Regarding the multilayered stacks, while employing SiO₂ layers, they in fact have a capacitance between that of the T₆₉S₃₁ and the T₇₅S₂₅ gate dielectric stacks and result in effective dielectric permittivities close to that of the T₆₉S₃₁ gate dielectric stack.

3.2. Device Characterization

3.2.1. TFT Performance

The performance of TFTs employing the different dielectric compositions was evaluated and a transfer curve for the T₆₀S₄₀ composition is presented in Figure 3a as an example. Figure 3b summarizes the field-effect mobilities (μ_FE) and subthreshold slopes (SS) obtained with the different T_xS_{100 − x} dielectrics. With the addition of Ta₂O₅ content, a slight degradation of the mobility from ≈ 16.3 to 14.8 cm²·V⁻¹·s⁻¹ is observed. This can be justified by the decrease of quality of the T_xS_{100 − x}/IGZO interface with the addition of Ta₂O₅, where defects such as fixed charges can cause the scattering of carriers, decreasing their mobility. The trap density at the interface (D_it) was extracted from (3) in which k, T, and e have their usual physical meanings and C is the dielectric capacitance.

$$SS\approx ln\left(10\right)\frac{k.T}{e}\left(1+e.\frac{D_{it}}{C}\right)$$

(3)

As presented in Figure 3c, $D_{it}$ tends to increase with the addition of Ta₂O₅ content, showing that the high-κ oxide results in poorer interface quality with IGZO when compared to SiO₂. Nevertheless, for the studied range of Ta₂O₅ incorporation, all the devices show good performance with mobilities above 14 cm²·V⁻¹·s⁻¹, SS lower than 0.3 V/dec, on/off ratios close to 1 × 10⁷ and gate leakage currents (I_G) close to 1 pA. Regarding the devices employing the multilayered gate-stacks, for the SiO₂/T₆₉S₃₁, dielectric properties close to that of the T₆₉S₃₁ were found, which should be related to the similar T₆₉S₃₁/IGZO interface. For the TFTs employing a T₆₉S₃₁/SiO₂ layer, mobility was slightly higher than most T_xS_{100 − x} compositions, and a great improvement is shown in SS (0.15 V/dec) and consequently D_it, showcasing the significantly better quality of the SiO₂/IGZO interface.

3.2.2. Insulation Reliability

Arising from higher defect densities and lower band gaps, employing high-κ dielectrics can often result in high I_G. In fact, TFTs employing a Ta₂O₅ dielectric layer revealed poor insulation which compromised the extraction of their transfer characteristics. Regarding the T_xS_{100 − x} dielectrics, while I_G was below 1 pA for TFTs considered to be working properly (see Figure 3a), in some devices an abrupt increase of I_G is seen with the increase of gate voltage. To quantify the dielectric reliability for each T_xS_{100 − x} composition (and for the multilayered structures), a leakage probability (P_Leak) was determined as the frequency of TFTs presenting gate leakage out of 18 similar devices. In practice, I_G was either in the noise level (for properly working TFTs) or higher than several nA (considered as leakage). Figure 4a presents the leakage probability for TFTs with channel widths and lengths of 320 μm and 20 μm, respectively.

Whereas a good reliability is seen for Ta₂O₅ contents of 60% and lower, it decreases abruptly for contents of 69% and higher. This can be attributed either to an increase of the conductivity of the dielectric layer or to a decrease of the conduction band offset (∆E_C) between IGZO and the dielectric layer, allowing the injection of carriers into the dielectric trough conduction mechanisms such as thermionic emission and field emission [57]. The band gaps of the T_xS_{100 − x} dielectrics, obtained from ellipsometry measurements [53] are shown in Figure 4b. For all compositions, the band gaps are much closer to that of Ta₂O₅ (≥4 eV [58,59]) than that of SiO₂ (8.9 eV), a trend that has been seen before for other high-κ/SiO₂ mixtures [14,20,35]. Nevertheless, the increase of band gap with the incorporation of SiO₂ is still relevant: to block charge injection, a conduction band offset above 1 eV is desirable and while the valence band offset to IGZO is unknown, the measured T_xS_{100 − x} band gaps are very close to 1 eV above the IGZO’s band gap (as represented in Figure 4b). This means that the slight band gap increase by the incorporation of SiO₂ may play a critical role in blocking the injection of charge. Leakage probabilities for the multilayered dielectrics are also presented in Figure 4a, and these are compared to the T₆₉S₃₁ single layer. The SiO₂ layer at the gate/dielectric interface (SiO₂/T₆₉S₃₁) significantly reduces the leakage probability, demonstrating the good insulation properties of this stack. Nevertheless, with the SiO₂ layer at the dielectric/IGZO interface (T₆₉S₃₁/SiO₂), this is even further enhanced, with no leaking devices being found. Compared to the SiO₂/T₆₉S₃₁ layer, this improvement can be attributed to a decrease of charge injection from the IGZO to the gate dielectric stack (e.g., hot electron injection) due to either the higher ∆E_C, better interface quality, or both. This shows that multilayered gate dielectric stacks are a viable approach to improve device performance without sacrificing reliability.

3.2.3. Stability

It is interesting to notice that the transfer curves for these devices present counterclockwise hysteresis. When employing SiO₂ dielectrics the hysteresis is known to be clockwise resulting from electron trapping in the dielectric/semiconductor interface trap-sites. Nevertheless, this counterclockwise hysteresis has been previously reported at times for some high-κ dielectrics, as will be discussed later. This device hysteresis was shown to increase with decreasing V_GS step (which increases measurement time) as shown in Figure 5a for the T₆₀S₄₀ composition. Considering V_On the V_GS for I_DS ≈ 100 pA, the measured hysteresis (∆V_On) for different V_GS steps for the different T_xS_{100 − x} compositions is presented in Figure 5b. Counterintuitively, the hysteresis is higher (in magnitude) for higher SiO₂ content and this will be addressed later. As for the multilayered gate dielectric stacks, the hysteresis was shown to be close to that of the corresponding single layer (T₆₉S₃₁), even in the case of a SiO₂/IGZO interface.

The hysteresis’ direction is usually tied with the direction of the V_th shift seen during positive gate bias stress (PGBS), unless different mechanisms play a role in these [60]. TFTs with the different dielectrics where submitted to PGBS with a gate voltage of 10 V for 1 h, followed by 1 h of recovery at V_GS = 0 V. During both, a drain voltage of 0.1 V was applied allowing addressing of the value of V_th by (4), which was fairly consistent with the V_th extracted from transfer curves before and after stress and recovery.

$$I_{DS}={\mu }_{FE}{C}_{ox}\frac{W}{L}\left(\left(V_{GS}-V_{th}\right)V_{DS}-\frac{V_{DS}^2}{2}\right)$$

(4)

The threshold voltage shift (∆V_th) for the different compositions is shown in Figure 6a. Similarly to the devices’ hysteresis, V_th also shifts towards more negative values during PGBS, as opposed to the typically reported positive V_th shift associated with electron trapping at the interface. V_th shifts with higher magnitude are once again observed for the SiO₂-richer compositions, as summarized in Figure 6b. Interestingly, in the same time frame, V_th shows only a partial recovery (for V_GS = 0 V), unlike in the faster relaxations typical of cases dominated by electron (de)trapping, whereas applying a negative gate bias (V_GS = −10 V) resulted in a fast recovery of the threshold voltage (Figure S3a).

In Figure 6b, the V_th shift for the T₆₉S₃₁/SiO₂ layer is also shown. While its shift magnitude is significantly lower than that of the T_xS_{100 − x}/IGZO interfaces, it is still negative, suggesting that this anomalous shift is not an interface phenomena and it is in competition with the electron-trapping at the SiO₂/IGZO interface during the PGBS [60].

Considering conventional applications (for which V_th shifts are undesirable), Ta₂O₅-richer compositions present as better alternatives both by resulting in higher dielectric permittivities and for lower V_th shifts. While these compositions can be unreliable in terms of gate leakage, it was demonstrated that thin SiO₂ layers at the dielectric/semiconductor interface successfully prevent gate leakage even for thinner overall gate dielectric stacks (with effectively higher capacitances). This shows that the multicomponent and multilayered approach can be a feasible method for achieving high device performance by incorporation of high-κ dielectrics without sacrificing device reliability. In fact, while the presented V_th shift magnitudes are considerably high, further addition of SiO₂ layers can be used to counterbalance this anomalous shift. This approach permits achieving devices with lower and positive V_th shift magnitude, while still maintaining desirable values for effective ε_r of 10–15. This is shown in Figure 7, for devices previously reported by our group [60] which employed similar multilayered gate dielectric stacks with an increased number of SiO₂ layers as shown in the inset of Figure 7b, resulting in ε_r ≈ 13. While positive V_th shifts in transfer curves measured after discrete periods of gate bias stress suggest the charge trapping mechanism only (Figure 7a), closer inspection along the duration of the stress by monitoring the drain current (Figure 7b) allows observing that the anomalous shift of V_th occurs during the first few minutes of stress. This clearly shows that the two mechanisms are in competition with charge trapping dominating, eventually resulting in an overall V_th shift of 1.3 V after 1 h of bias stress. Similar devices were successfully applied for a flexible radiation sensing system, both for timing signals generation for addressing the sensors in the irradiated matrix [61] and for a high-gain transimpedance amplifier for amplification and voltage transduction of the sensors current signal [10], with both implementations requiring robust device operation.

3.3. Mechanism of the Anomalous V_th Shift

A discussion of the mechanisms involved in the observed anomalous V_th shifts is now presented. While these have been reported for TFTs employing high-κ gate dielectrics (such as Ta₂O₅ [62,63,64], ZrO₂ [9] and HfO₂ [65]), other reports often show normal shift directions. High-κ dielectrics are known to be prone to have high defect density (probably playing a role in the anomalous shift) and it should be expected that different processing methodologies can result in distinct material qualities, leading to different device behaviors. The mechanisms that are normally used to explain anomalous V_th shifts [34] are: charge (de)trapping from the gate dielectric [62,65] ionic migration within the dielectric [63] (understandable as a slow polarization of the dielectric material [64]) and defect creation [66]. Regarding defect creation, it can lead to an increase in the carrier concentration, resulting in the decrease of the V_th, and it was shown before for poor quality semiconductors [66]. In this case, this can be disregarded as no change in the transfer curves’ SS was observed, demonstrating the stability of the semiconductor (even considering the low thermal budget employed here). Furthermore, it is expected that the involved mechanism is dielectric related as this anomalous shift is not observed in our IGZO TFTs when employing other dielectrics. Regarding the charge trapping mechanisms, a positive net charge change is required for a negative V_th shift. Hole trapping at the dielectric/semiconductor interface can be disregarded as the voltage polarity is opposite to what would be required, and wide band gap n-type oxides require optical excitation for holes to be generated in the first place. Then, electron detrapping from the dielectric to the gate must be considered (e.g., from negatively charged oxygen interstitials: I_O^– or I_O^2– [65]). However, detrapping at this interface cannot change the electron concentration at the semiconductor, and thus cannot explain the anomalous shift if charge migration is not assumed [34,64]. To investigate the mechanism, the time dependency of the threshold voltage shift during the gate bias stress was studied and a power-law dependence $\left(\mathsf{\Delta }{V}_{th}\propto t^n\right)$ was found, as shown in Figure 8a for the T₃₉S₆₁ and T₇₅S₂₅ compositions. While exponential dependencies are seen when charge (de)trapping is the relevant process, power dependencies are often related to reaction–diffusion mechanisms. For the compositions presented in Figure 8a, the exponent n changed from ≈0.5 to ≈0.6 after approximately 1 min. For the intermediate T_xS_{100 − x} compositions, the exponent changed from ≈0.33 to ≈0.5 after approximately 1–5 min (Figure S4a,b). Similar dependencies were noted for the multilayered stacks (Figure S4c), further suggesting that the mechanism is not dominated by either of the interfaces. Power law time dependency of ${\mathsf{\Delta }\mathrm{V}}_{\mathrm{th}}$ is known for the diffusion of H species in MOSFETs during negative bias temperature stress, where exponents of 1/3 and 1/2 are associated to trap-generation and diffusion of the charged species H⁺ and H₂⁺, respectively [67]. Aleksandrov et al. also associated n = 1 to first-order chemical-reaction kinetics and n closer to 0.5 to diffusion and drift kinetics of H⁺ ions [68]. While further investigation is needed to understand the involved species in our devices, it is interesting to notice that Ta₂O₅ dielectrics are known for their ionic conductivity, with H⁺ and V_O migration in its bulk being known [69] and partly responsible for their application as resistive switching layers [70]. Regarding the V_th recovery (for V_GS = 0 V in Figure 6a), an exponential dependency with time is apparent: while recovering quickly initially, it seems that, at least in the presented time frame, V_th is slowly tending to values significantly larger than the initial ones. Charge migration is often ruled out when fast recoveries are seen (as without driving force the species cannot quickly return to their initial positions), but the relatively small recovery seen across all compositions seems to imply that migration has to be considered. For further evaluation, the full recovery for the dielectric composition presenting the higher V_th shift (T₃₉S₆₁) was studied and is shown in Figure 8b. The recovery took place in a much larger time frame (>1 month) than that of the stress process (1 h) and followed an exponential behavior with a time constant $\mathsf{\tau }\approx 8\times {10}^5$ s which suggests a different recovery process than that seen up to 1 h immediately after the stress. As stated before, applying a negative V_GS resulted in a much faster recovery of V_th (Figure S3a), of just a few minutes. This is in agreement with negative bias stress measurements made in as-fabricated devices (Figure S3b), where a fast increase of V_th is observed for the first minutes of stress, with the reverse ∆V_th direction in relation to the applied gate bias. Afterwards, V_th is fairly stable along the stress time, given the absence of photoinduced holes (measurements under dark conditions) to sustain the commonly observed negative V_th shifts under negative bias illumination stress (NBIS) [71].

These results point to the anomalous shift being related to the migration of charge within the dielectric, but the nature of the migrating species or type of defects involved is not understood yet.

While Ta₂O₅ is known to have a considerable defect density, tending to form suboxides, the V_th instability was found to be more pronounced for SiO₂-richer T_xS_{100 − x} compositions, implying these are more defective. Oxygen displacement in the network due to Si’s higher oxygen affinity than Ta is a possible mechanism. In fact, SiO₂/high-κ dielectrics interfaces are known to form defects, such as dipoles (V_O-I_O pairs) caused by oxygen displacement (due to deferring oxygen areal densities in these materials [72,73], or oxygen vacancies [74,75]. Note, that the oxygen vacancy formation energy in Ta₂O₅ is low compared to other high-k dielectrics [76]. Detrapping from either these defects to the gate, their migration inside the dielectric layer, activated by gate bias, or both, can result in the observed anomalous V_th shifts. Another possible mechanism is proton migration though the bulk of Ta₂O₅ [77,78,79], but in this case it would be expected that the anomalous V_th shift would have to increase with increased Ta₂O₅ content, contrary to our findings. Further investigation should be conducted to determine the nature of the charged species involved in the anomalous V_th shift.

4. Conclusions

Multicomponent gate dielectric stacks comprised of both SiO₂ and high-κ Ta₂O₅ were explored for oxide TFTs with low thermal budgets, compatible with flexible electronics (T = 150 °C). Co-sputtering of both materials as well as the addition of single SiO₂ layers in the gate dielectric stack were considered. While the incorporation of Ta₂O₅ effectively increases the relative permittivity, the resultant gate dielectric stacks present poor insulation for Ta₂O₅ contents ≥69%, due to either their poorer insulation capability or poorer band misalignment with the semiconductor. The multilayered approach of including a thin SiO₂ layer at the gate/dielectric interface improved the device reliability by decreasing the probability of devices having significant leakage current (P_Leak) from 56% to 17% when comparing devices with similar multicomponent layers (T₆₉S₃₁). Nevertheless, including a thin SiO₂ layer at the IGZO/dielectric interface not only blocks the charge injection into the dielectric, resulting in very reliable gate insulation (P_Leak = 0%), but also results in devices with better mobility (16 cm²·V⁻¹·s⁻¹) and SS (0.15 V·dec⁻¹), due to the superior interface between IGZO and SiO₂ than between IGZO and T_xS_{100 − x} mixtures (SS > 0.2 V·dec⁻¹), with Ta₂O₅ richer compositions presenting the poorer interface properties (μ_FE = 14.8 cm²·V⁻¹·s⁻¹ and SS ≈ 0.25 V·dec⁻¹ for T₇₅S₂₅). Counterclockwise hysteresis of the transfer curves and anomalous V_th shifts seen during PGBS were, unexpectedly, higher in magnitude for SiO₂ richer compositions, with ΔV_th ≈ −102 V and −30 V, after 1 h stress, for the T₃₉S₆₁ and the T₇₅S₂₅ compositions, respectively. Similar behavior of devices employing IGZO/SiO₂ interfaces discarded the interface dominance of the processes while the power law time dependency of V_th during gate bias stress, with exponents ≈ 0.5, points towards reaction-diffusion processes. Additionally, the recovery process was very slow, presenting a very high time constant, >8 × 10⁵ s. These results imply that the anomalous shift is caused by the migration of charged species inside the dielectric. The increase of V_th shift magnitude with the SiO₂ content may suggest that it promotes oxygen vacancies in the mixture by displacing oxygen from Ta₂O₅ due to its higher oxygen affinity. Regardless, the anomalous shift can be effectively counterbalanced by the inclusion of additional SiO₂ layers in the gate dielectric stack: ΔV_th (at 1 h of PGBS) for the T₆₉S₃₁ and T₆₉S₃₁/SiO₂ stacks are ≈ −41 V and −24 V, respectively, while for the stack with increased number of SiO₂ layers a positive shift of only 1.3 V is obtained. This multilayer approach enables devices that are both reliable and present good performance (mobility = 16 cm²·V⁻¹·s⁻¹, ε_r ≈ 13) and that can be successfully implemented into novel flexible electronic applications.