Light Effect on Amorphous Tin Oxide Thin‐Film Transistors

Amorphous tin oxide (a‐SnOx) is a potential transparent oxide semiconductor candidate for future large‐area electronic applications. The thin‐film transistor (TFT) mobilities reach ≈100 cm2 Vs−1, a mobility higher than other multiple cation‐based oxide semiconductor TFTs. Few optical properties have been reported so far and therefore both the effect of visible light and negative bias illumination stress (NBIS) on a‐SnOx TFT performances, known to dramatically impact oxide semiconductor‐based TFTs, have been investigated. The variation of density of states (DOS) due to NBIS by device simulation is analyzed, and a fourfold increase of the donor‐like states and a decrease in the band edge DOS from 2.3 to 2.0 × 1019 cm−3 eV−1 are showed. The evaluation of the effect of neutral, singly, and doubly ionized oxygen vacancies by density functional theory using 95 atoms reveals not only states in the gap of SnO2, but also variations in the electron density, and modifications in the crystal parameters compared to a structure without an oxygen vacancy. Material and device simulation analysis reveal that the oxygen vacancies have a dramatical impact on the DOS in the gap of SnO2 and can explain the NBIS phenomenon observed in a‐SnOx TFT.


Introduction
One of the purposes to study thin-film transistor (TFTs) is to implement them in large-area electronics, like active matrix organic light-emitting diodes (AMOLEDs). [1,2]Development of materials to improve the performances of TFTs has been one way.For that reason, amorphous indium gallium zinc oxide (a-IGZO) and other oxide semiconductors have been under the spotlight for more than 15 years.Made from cations with the (nÀ1)d 10 ns 0 (n ≥ 4) electronic structure, amorphous oxide semiconductors are characterized by high mobilities (over 10 cm 2 Vs À1 ) and high transmittance (>80% in the visible region).Polycrystalline single cation-based oxide semiconductors like indium oxide, zinc oxide, tin oxide, and multiple cation amorphous oxide semiconductors like zinc tin oxide, indium zinc oxide, and indium zinc tin oxide have also been investigated. [3,4][15][16][17] High-temperature-annealed (over 300 °C) and/or spin-coated devices may compete with vacuum-processed TFTs.The development of solution-processed TFTs has been accompanied by high-k dielectrics, as they provide low-voltage operation and high mobilities. [3,18][21] Cho et al. demonstrated amorphous IGZO TFTs with mobility of 48.3 cm 2 Vs À1 by cation composition investigation at a low annealing temperature of 250 °C. [22]ITZO TFTs with mobility of 27.7 cm 2 Vs À1 were also achieved. [23]Not only does ALD offer high performance devices, but also the process allows to build TFT in which the channel can be made of multiple layers, [24] like IGZO made of layers of In 2 O 3 /Ga 2 O 3 /ZnO and can lead to mobility over 100 cm 2 Vs À1 with careful plasma treatment on the Ga 2 O 3 layer. [25]High-k dielectrics, like Al 2 O 3 deposited by ALD, can lead to mobilities of 23.3 cm 2 Vs À1 . [26]lso, ALD can lead to explore materials which have not been implemented in TFT, like TiO 2 . [27]in oxide, SnO 2 , [4] was the first transparent conducting oxide material used as a channel region in a TFT, [13] and has recently achieved high mobilities of 147 cm 2 Vs À1 . [28]Other reports have confirmed the high mobility of the tin oxide TFTs, reaching 40, and even 90 cm 2 V À1 but the TFTs demonstrated low I on /I off ratios or high leakage currents. [29,30]SnO 2 TFTs have been investigated by ALD process.Demonstration of low temperature processed (max processing temperature of 130 °C) has led to TFTs with mobilities of 6.25 cm 2 Vs À1 when Sn(dmamp) 2 was used with O 2 plasma.The precursors are of paramount importance because at the low-temperature process of 60 °C, Sn(DMP) 4 with O 2 plasma can lead to mobilities of 12 cm 2 Vs À1 . [31]herefore, tin oxide as a polycrystalline material is a potential candidate for high-performance large-area electronics.On the other hand, we have recently reported the electrical, physical, and chemical properties of solution-processed amorphous tin oxide (a-SnO x ) in TFTs applications.The TFTs show high performance (mobility ≈100 cm 2 Vs À1 ) and high stability under various gate bias stresses. [32]Compared to other oxides, a-SnO x requires only one cation, which could lead to a lower cost of fabrication.To be used in AMOLEDs or other optoelectronic applications, interactions between light and a-SnO x TFTs need to be investigated.
[35] It is therefore necessary to understand the effect of light on other oxide-based devices to ensure possible use for photosensitive applications.On the other hand, one of the most problematic instabilities in oxide TFTs (namely, IGZO TFTs) is due to negative bias illumination stress (NBIS): a negative V GS is applied while the light is lit up on the TFT.The TFTs demonstrate a negative shift of their I-V transfer curves accompanied with an increase of the OFF current. [36,37][39] Let us note that the effect has been definitely identified, and has various intensities depending on the TFT structure.For example, using the regular SiO 2 gate insulator (GI) can lead to high NBIS effect, while using HfO 2 can reduce the effect. [40]Using an offset top gate contact, or a dual-gate structure can significantly reduce the NBIS on IGZO TFT. [41,42]n this report, we show the impact of light only, and NBIS on the a-SnO x TFT I-V characteristics.We then explain the behavior by the presence of gap states and their variation by technology computer-aided design (TCAD).Finally, we introduce density functional theory (DFT) calculations to understand the change in bandgap states and electrical properties with the presence of oxygen vacancies.

Light Effect on TFTs
We start with the effect of light on the TFTs. Figure 1a-d shows the variation of the TFT transfer curve with blue, green, yellow, and red light during 1 h, respectively.As one can see, the transfer curve does not vary significantly.The blue light does modify slightly more the transfer curve than the other lights; a small decrease in drain current and a small positive shift (≈0.1 V) are observed.Therefore, for high-energy photons, the current is decreased and the threshold voltage (V th ) is shifted positively, which signifies a decrease in the collected number of charge carriers.This trend is sensitively different from IGZO TFTs which show a significant change in the TFT I-V curves with light only. [43]et us first note that our channel layer is rather thin (9.2 nm). [32]It was reported that light effect on IGZO TFTs depends on the thickness of the channel.The thicker the IGZO layer is, the higher the impact of light is. [36,44]Recently, the effect of light was also investigated on indium gallium tin oxide TFTs. [45]The authors demonstrated that photosensitivity was dependent on the channel thickness, and as in IGZO TFTs, the thicker the layer, the larger V th shift was observed.
The composition is also important.For example, for indium oxide doped with tin and aluminum, the effect of light depended on the amount of tin. [46]At higher Sn contents, the TFT demonstrated a higher OFF current, but also a higher sensitivity to light.The authors claimed that the electrons are excited from gap states near the valence band.
Let us note that all these materials are multiple cation-based materials.Considering monocationic materials, ZnO demonstrated no influence under various lights, except the blue light, and demonstrated high photosensitivity with adding quantum dots on top of the TFT. [35]A similar trend was observed with In 2 O 3 , where a sensitive layer was required to enable photosensitivity. [47]o explain our results, we first now discuss the light penetration in the material.At first, we can consider that the penetration depth of light is ≈1/α, where α is the absorption coefficient.The absorption coefficient is shown in Figure 1e.For the light used, we obtain a depth penetration of 74 nm (blue) and 450 nm (red).Therefore, even high-energy photons are not completely absorbed in our 9.2 nm thin channel layer. [32]This can explain partially the results observed.We suggest that the scattering of the light-generated electrons and holes can explain the slight shift and current decrease in the experiment.Also, the recovery to the initial I-V curve occurs within 1 min for all lights, clearly demonstrating that the main phenomenon is due to charge recombination.
The leakage current is understood as the charge carrier that may exist in the channel while a negative V GS is applied.We previously reported the band diagram of a-SnO x /HfO 2 . [39]The valence band offset is smaller than 1 eV and therefore may provide a way for charge carriers (holes) to flow.The presence of holes is possible as we previously reported that our a-SnO x is made of a mixture of n-type SnO 2 and p-type SnO. [32]2.NBIS Next, we evaluate the NBIS on the a-SnO x TFT.To understand the behavior correctly, we first run NBS and illumination from the backlight unit (BLU) separately.The results are shown in Figure S1, Supporting Information.The NBS demonstrates not only an increase in ON-current, but also a decrease in OFF-current.The I GS current also increases when V GS > 0 V. But, within 1 h, the TFT recovers its initial I-V characteristics.We previously demonstrated that a similar trend in polycrystalline SnO 2 .We showed that this could be explained by the introduction of slow moving holes into the dielectric due to the small valence band offset between the dielectric and the tin oxide channel layer.[48] Under the light from the BLU, the TFT did not show considerable modification in the I-V characteristics.Considering the effect of the BLU only, let us remember that the V th shift is only small with only one wavelength.So, our result with the BLU is consistent with the results presented in Figure 1. Figure 2a-c shows the evolution and recovery of the a-SnO x TFT transfer and the output curves with time.In Figure 2a, we observe that with 1 h, the transfer curve shifts negatively, and there is more and more current available.The TFT V th shifts from À0.51 to À0.60, À0.72, and to À0.77 V after NBIS has been applied for 0, 100, 1000, and 3600 s, respectively.For the same NBIS stress times, the subthreshold slope (SS) varies from 76, 73, 70, and 72 mV dec À1 .The variation in SS is not significant during the stress.The higher current is also observed in the output curve in Figure 2c. Figure 2b,c shows the recovery of the transfer and output curves with time.With time, the current decreases, and the V th shifts positively.The I-V curve used at 0 h of recovery is the same I-V curve when NBIS has been applied for 1 h to the TFT.The V th varies from À0.77 to À0.66, À0.63, À0.55, and À0.52 V at a recovery time of 0, 1, 2, 16, and 39 h, respectively.This means that there are fewer and fewer electrons collected from the channel.At the same time points, the SS varies from 72, 74, 80, 73, to 78 mV dec À1 .We observe that the recovery is almost complete after ≈39 h.All values are gathered in Table 1.Also, we note that under NBIS the gate leakage (I GS ) increased, and during recovery, I GS decreased as shown in Figure 2a,b, respectively.The trend observed is in total agreement with the previously reported NBIS effect on IGZO TFTs.[43,49,50] We can also observe that there is an increase in the leakage current by almost one order of magnitude during NBIS, and a decrease of about one order of magnitude also during the recovery.A higher leakage current is associated with more charge carriers in the channel.These charge carriers would be due to the extra generated charge carriers and the ionized oxygen vacancies moving to the GI as will be discussed later. We nte that as the I GS decreases, so does the I DS in the negative V GS region, so the leakage current is I GS limited.Also, as stated above, the valence band offset is less than 1 eV so hole can easily flow into the dielectric.The long recovery is therefore demonstrating that slow mechanisms are required (such as the movement of ionized oxygen vacancies and their recovery to lower ionization states).
Ionization of oxygen vacancies (V O ) in IGZO TFTs could explain the shift of V th under NBIS: Vo-related subgap states make a transition from neutral states (Vo) to ionized positive charged state ( by donating free carriers within the metal-oxygen (M-O) network. [36,37]Removal of one/two electron(s) from Vo results in single/double-ionized vacancies as given below equations: where e denotes an electron.These free carriers (electrons in n-type semiconductors) would be responsible for the shift of the threshold voltage in the negative V GS direction.We note that single-ionized defects are not stable, hence they turn to neutral or double-ionized states. [36,37]To further understand the NBIS effect, we performed TCAD simulation on the I-V curves measured during and after NBIS.

Device Simulation
Now, to understand the TFT I-V evolution with time, we evaluate the a-SnO x gap states by TCAD.We show the measured data and the TCAD fitting using the density of states (DOS) model in Figure 3. Figure 3a-d shows the measured and modeled transfer characteristics, and Figure 3e-h shows the measured and modeled output characteristics of the fabricated a-SnO x TFT at initial, after 1 h of NBIS, after 1 h of recovery, and after 39 h of recovery, respectively.
We previously established that the DOS model consists of donor-and acceptor-like defects (N GD and N GA ) and of conduction band (CB) edge DOS (N TA ). [39,51]Figure 3i-l introduces the variation of the various subgap states in a-SnO x as calculated by TCAD.Under NBIS, N GD increased from 1.2 Â 10 17 to 4.5 Â 10 17 cm À3 eV À1 , and after 1 h of recovery N GD = 3.75 Â 10 17 cm À3 eV À1 , and after 39 h of recovery N GD = 1.35 Â 10 17 cm À3 eV À1 .As noted in our previous study, N GD can be understood as charged donors. [39]Therefore, the variation in N GD can describe in part the variation in ionized oxygen vacancies.
N TA , which represents M-O bond parameter distribution (angles, bond lengths, etc.) [39] is usually understood as a constant for a material.Yet, here, we considered the possibility of variation because of the possible role of ionized oxygen vacancies, which could modify locally the material structure.During NBIS, N TA decreased from 2.3 Â 10 19 to 2.1 Â 10 19 cm À3 eV À1 , remained constant even after 1 h of recovery, and reached back its initial value of 2.3 Â 10 19 cm À3 eV À1 after 39 h of recovery.More discussion on this point will be given later in the material section part.Finally, we note that the W TA remained constant throughout the experiment and had a value of 0.09 eV À1 .It is reported that structural disorders cause band tail states in  oxide semiconductor; however, overlapping s orbital from heavy metal in oxide semiconductor network provides low tail state peak (N TA ), steeper slope (W TA ), and higher field-effect mobility. [51,52]n oxide TFTs, the variation in the interface trap density (ITD) results in a parallel V th shift without affecting the subthreshold slope. [53,54]As shown in Figure 2 previously, the subthreshold swing does not change significantly throughout the NBIS and recovery experiments, but the final value after recovery (78 mV dec À1 ) is slightly higher than the initial value (76 mV dec À1 ).From the simulation, we observed that the ITD increased from 0.45 Â 10 11 to 3.2 Â 10 11 cm À2 during NBIS and decreased to 0.7 Â 10 11 cm À2 after the 39 h recovery process.We also considered the carrier concentration within the a-SnO x layer.The carrier concentration increases from 1.85 Â 10 18 to 2.1 Â 10 18 cm À3 during NBIS and decreases to its initial value.Table 1 gathers all the ITDs and carrier concentration values as a function of NBIS and recovery time.
The analysis of the DOS simulated by TCAD shows that no change in the acceptor-like state (N GA ) is required to explain NBIS-induced degradation.
We note that the high carrier concentration can lead to highperformance TFTs with HfO 2 GI.But in TFTs using SiO 2 , the carrier concentration needs to be reduced down to ≈10 16 cm À3 , or source/drain engineering could be necessary. [53,55]Also, the GI and the fabrication process have an impact on the stability of the TFTs under NBIS.For example, IGZO TFT using HfO 2 has demonstrated higher stability by controlling the GI deposition temperature. [40]Not only is NBIS affected by the choice of the GI, but also the performances are.Reliability can be improved by careful choice of thickness of Al 2 O 3 , [56] and can also improve memory effect [57] in TFTs using Hf ZrO 2 .Various studies on high-k dielectrics have shown the impact on the performances. [58]Therefore, it could be of interest to investigate the influence of NBIS on a-SnO x TFTs with other GIs.Other strategies could be doping the channel material, [59] or even using other TFT structure. [41]We note that we analyzed the so-called back channel-etched TFT structure as the structure is not only easy to fabricate, but also the most suited to study the effect of light. [43,60]57b,c] Yet, the ferroelectric properties of HfO 2 may appear after more than a few thousands of switching cycling, [57c] which we did not apply here, so the properties shown here should be a result of the active material a-SnO x .

Possible Defect Formation Mechanisms and NBIS Mechanism
The TCAD simulation demonstrates that two phenomena are present and can explain the NBIS instability and partial recovery.On the one hand, the variation in trap density suggests a trapping/detrapping mechanism, and on the other hand, the variations in N TA and N GD would suggest a variation in ionized oxygen vacancy density.The electrical degradation under NBIS in n-type oxide semiconductor TFT was previously explained as the charge trapping at GI/oxide semiconductor interface [61,62] and/or defect generation, mostly oxygen vacancies in the oxide semiconductor bulk as the degradation mechanism. [42]Considering multimetal cation-based oxide semiconductors (e.g., IGZO), oxygen defects are more probable than metal defects. [63]xygen vacancy defects are categorized as neutral vacancies (V O ), and single/double-ionized vacancies ( O defects are metastable and easy to transform into either V O or V 2þ O state without additionally supplied energy (by large gate field or optical excitations). [49]However, in the case of V 2þ O defects, thermal annealing after NBIS is required to achieve pristine/initial electrical performance recovery. [64,65]et us now discuss the possibility to form cation defects.A previous report demonstrated metallic vacancy (V MO ) and metal interstitials (V MI ) in oxide semiconductors. [66]A Zn vacancy in ZnO (V 2À Zn ) has an acceptor like behavior. [66]esides, the metal interstitial provides two electrons to the CB and forms a positively ionized charged state (V 2þ Zn ), which acts as a shallow donor in ZnO stoichiometric network. [66]Also, the p-type conductivity in SnO semiconductors is mainly because of Sn vacancy (V Sn ). [67]But, because of the recovery process and nonresponsive N GA from TCAD simulation, we understand that under NBIS, the formation of O is more favorable than metal vacancy/interstitials.The incomplete recovery after 39 h correlates to the fact that NBIS on a-SnO x TFT could result in the formation of both the As N GD has not reached its pristine value, we believe that some V 2þ O may remain in the material.Also, the number of interface traps still has not fully recovered after 39 h, so some extra electrons remain trapped.Yet, the carrier concentration has reached its pristine value.This could mean that some of the extra electrons given out by ionized oxygen vacancies are trapped near/at the insulator/semiconductor interface.Now, we discuss the possible mechanism related to oxygen vacancies involved during NBIS.Figure 4a shows the V O ionization process upon incident light irradiation with applied negative vertical gate field.Photoinduced ionized defects migrate to the GI/active interface due to a negative vertical gate field under NBIS and provide electrons to the CB.We note that the barrier height between the valence bands of HfO 2 and a-SnO x (À0.15 eV) suggests the possible penetration of holes from the semiconductor to the GI.Therefore, one possible mechanism involves the penetration of holes into HfO 2 .This is highly likely to happen because we observed an increase in gate leakage during NBIS.We note measurements of a-SnO x TFTs under NBS further confirmed the possible penetration of positively charged species in the GI. [32,39,48]We note that the midgap location of V O -related states is justified in the material simulation below.So, carrier concentration increased and the transfer characteristics shifted toward the negative V GS direction.The penetration of holes into HfO 2 may also contribute to the negative V th shift.Upon the removal of NBIS, and by keeping the TFTs at room temperature as shown in Figure 4b, NBIS-generated O defects may capture electrons and return to their original oxygen vacancy state ðV O Þ.However, the TFTs exhibit incomplete initial characteristics recovery, suggesting that all V 2þ O are not completely removed after recovery of 39 h.As mentioned above, V þ O are metastable species so they may be more likely to be present in the a-SnO x TFT because the recovery is almost complete, and the remaining unrecovered V th shift and current could be explained by the presence of V 2þ O : This is also confirmed by the slight increase in subthreshold swing from initial (76 mV dec À1 ) to final value (78 mV dec À1 ).

Material Simulation
Finally, we discuss the theoretical background of oxygen vacancies in SnO 2 and how they are related to the DOS of SnO 2 as calculated by DFT.Let us first keep in mind that the DFT analysis lies on the crystalline structure of SnO 2 and not on an amorphous structure.The results should provide a first step toward understanding the bandgap structure though.Indeed, for IGZO, it was demonstrated by DFT that the amorphous and crystalline phases had their bandgap modified when an oxygen vacancy was present.The oxygen vacancy leads to the presence of states in the middle of the bandgap of IGZO in both cases. [68,69]The relaxed structure of SnO 2 demonstrated a position of the Sn of 0,0,0 while u = 0.3068, close to other works (see Table S1, Supporting Information).From the basic SnO 2 structure, we then evaluated two 4 Â 2 Â 2 supercells.Both 4 Â 2 Â 2 structures had a length (a), width (b), and height (c) of 12.142964, 18.005593, and 36.011186Bohr, respectively.The structure without vacancies (structure A) had 96 atoms, while one oxygen atom was removed to evaluate the impact of oxygen vacancy on the properties of the material on structure B. Figure 5a shows the original SnO 2 structure.
Figure 5b shows the impact of oxygen vacancy in the structure of SnO 2 .The figure shows the difference in the positions of atoms due to the vacancy.In the vicinity of the oxygen vacancy, the Sn atoms tend to go near the vacancy, whereas other atoms tend to go away from the vacancy.We note that the change in bond length is in the pm range.Also, the variation in angle in both structures is within AE2 o .When the vacancy is ionized once (V þ O ) (Figure 5c), the Sn atoms tend to move away, and when the vacancy is doubly ionized (Figure 5d), the Sn atoms further go away from the vacancy.This trend is in total agreement with previous results reported for ZnO [63,66] and in In 2 O 3 . [70]The variation in angle and in bond length becomes larger.Some angles change by more than 10 o from structure A to the structure B with doubly ionized oxygen vacancy.Table 2 gathers bond angles and lengths around the oxygen vacancy position for all structures.The nomenclature used is shown in Figure 5e.
Structure A demonstrated a bandgap of ≈4.2 eV.The common values found experimentally are 3.9-4.2eV, and other calculations lead to values below 3 eV (see Table S1, Supporting Information).This is due to the Tran-Blaha method, which is known to enlarge the bandgap. [71]Figure 6a shows the total DOS of structure A. The presence of Sn 2s states forms the CB (Figure S2a, Supporting Information), while the valence band is mainly formed by O 2p states (Figure S2b, Supporting Information).
The total DOS of structure B is shown in Figure 6b.The figure demonstrates the presence of filled states by midgap.The filled states are mainly formed of Sn s and d orbitals (Figure S2c, Supporting Information) and oxygen p states (Figure S2d, Supporting Information).The oxygen vacancy, therefore, modifies the bandgap structure.We note that contrary to IGZO, deeplevel states are not near the valence band, but close to the middle of the bandgap, and are ≈2 eV below the CB, [37,50,72] but in accordance with the DFT results. [68,69]V þ O (see Figure 6c) leads to partially filled states and V 2þ O to empty states near the CB (Figure 6d).Also, we show the difference in electron density in the CB in both structures in Figure 7.In structure A, we can observe that the electron density of O is mainly associated with p-orbitals while for the Sn atom, the electron density can be associated with s-orbitals.The Sn atoms orbitals show antibonding characteristics.Also, the electron density is not 0 away from the atoms.These considerations suggest the delocalization of electrons in the CB.In structure B, on the other hand, we can observe that the density is highly concentrated in O and that there is a high density around the oxygen vacancy, and near the Sn atom.
On the other hand, the electron density around the Sn atom seems higher than in the pristine structure, and away from the atoms, the electron density also seems higher.As shown in 7(c) and (d), the ionization of the oxygen vacancy leads to the high density of electrons between Sn atoms.Therefore, the higher electron delocalization explained in this figure can explain the higher current measured in TFTs during NBIS.We introduced in the previous part the possibility of NTA being slightly changed during NBIS and recovery.NTA is related to the cation bonds, bond lengths, and angles. [63]As we demonstrated with the material simulation the angles and lengths of Sn-O are modified in the vicinity of the oxygen vacancy.Also, we showed that when the vacancy becomes singly ionized the volume around the vacancy becomes larger, and becomes even larger when the vacancy is doubly ionized.Therefore, the  assumption in the previous part was correct, and the bottom of the CB as simulated by device simulation should be modified to explain the impact of NBIS on a-SnO x TFT.Also, the presence of the oxygen vacancy clearly shows the enhancement in electron conduction in the CB.

Conclusion
We presented the influence of light on the a-SnO x TFT I-V characteristics.We discussed the origin and possible mechanisms of the variation of TFT I-V characteristics through device and material simulations.The experiments show that only blue light has a slight influence on the current and threshold voltage of the TFT.Under NBIS, the TFT shows a drastic V th shift and an increase in current.The investigation demonstrated the possible creation of defects in the bandgap, namely, donor-like states increasing from 1.2 to 4.5 Â 10 17 cm À3 eV À1 , while tail gap states could also have changed from 2.3 to 2.1 Â 10 19 cm À3 eV À1 .One of the mechanisms involves the trapping of charged carriers into/near the GI.We understood that the ionized oxygen vacancies that could migrate toward the channel/insulator interface could lead to an increase in current.As we could not observe a complete recovery after NBIS, we concluded that only a few oxygen vacancies could have been recovered to their original nonionized state.This was further supported by device simulation showing that the donor-like states did not recover their initial value (1.2 Â 10 17 cm À3 eV À1 ) and were 1.35 Â 10 17 cm À3 eV À1 after 39 h of recovery.The material simulation through DFT calculation further demonstrated the impact of oxygen vacancies in SnO 2 .The presence of oxygen vacancies modified the bond lengths and angles between Sn-O in the vicinity of the oxygen vacancy.A higher electron density is present between Sn atoms, and higher delocalization of electrons is demonstrated.We identified the presence of a band in the midgap of the SnO 2 bandgap due to the oxygen vacancy.These gap states are mainly formed of Sn s and d states and oxygen p states.The singly ionized vacancyrelated states are partially filled states in the midgap, whereas the doubly ionized vacancy-related states are localized ≈3 eV above the valence band.The presence of oxygen vacancy also allows a better electron delocalization and can explain higher current and mobility in the TFTs.Future works therefore include the use of other TFT structures (top-gate and dual-gate TFT) and materials (SiO 2 and contact electrodes) to reduce the effects of NBIS on a-SnO x TFTs.

Experimental Section
Solution Fabrication: The a-SnO x precursor solution was made by mixing 1 mmol of SnCl 2 (Sigma-Aldrich, 99.999%) into 6 mL of a mixture of acetonitrile and ethylene glycol (35 and 65 vol%, respectively).0.2 M HfO 2 precursor solution was prepared by mixing 1 mmol of HfCl 4 (Sigma-Aldrich, 99.999%) into 5 mL of the aforementioned mixture solvent.
TFT Fabrication: The fabrication process of bottom gate top contact a-SnO x TFTs was reported somewhere else. [32,39]In a nutshell, the TFTs were fabricated on a glass substrate with 40 nm of Mo as the gate, 95 nm of HfO 2 as the GI, and 9.2 nm of a-SnO x .We used 200 nm IZO as the source and drain electrodes.
Device Characterization: TFT electrical measurements: We measured the TFT I-V curves with an Agilent semiconductor parameter analyzer 4156C.The TFT width (W ) and length (L) were 50 and 10 μm, respectively.The threshold voltage (V th ) was defined as the gate bias (V GS ), giving a constant current of W=L Â 10pA.We extracted the SS as the minimum value of ð∂logI DS =∂logV GS Þ À1 from the linear region of the transfer characteristics.
Photosensitivity Characterization: We used red, green, orange, and blue LED, emitting a wavelength of 620, 545, 586, and 466 nm, respectively.We evaluated the impact of each wavelength at 10.000 Cd m À2 for 1 h.On the other hand, we used a typical white BLU (also set at 10.000 Cd m À2 ) used in an LCD monitor as a source of white light.The 1931 CIE x and y coordinates were 0.3054 and 0.3305, respectively.The CIE x and y coordinates and the wavelengths were extracted from a KONICA MINOLTA CS100A luminance meter.
NBIS was performed for 1 h by setting V GS = À2.5 V and the BLU emitting at 10.000 Cd m À2 .The I-V transfer curves were measured at 100, 1000, and 3600 s, while the BLU was set off.The recovery was measured by grounding all electrodes and measuring the I-V curves at various times (in the dark).
Device Simulation: Numerical simulation was performed by using a TCAD software to achieve the simulated TFT performances and to extract the bandgap states. [39,42]We create the TFT device structure file with the Athena module, and used the Atlas module to obtain the material properties and the DOS.The DOS mainly consists of exponentially distributed tail states for both the n-type/p-type semiconductors and Gaussian donor/ acceptor-like states. [39]a-SnO x being an n-type material, [32] we assume that the Fermi energy level is far away from the valance band tail state, and therefore, we mainly considered acceptor-like tail states, Gaussian acceptor-and donor-like states (states at the vicinity of the conductor band edge) to simulate the electrical properties. [73]We considered various measured quantities, i.e., the channel layer thicknesses, the bandgap, Hall carrier concentration, and HfO 2 dielectric constant as input of the simulation parameters to accurately investigate the a-SnO x TFTs with an HfO 2 dielectric layer. [39][76] These techniques have the limitation to either achieve the valance band or the CB DOS.Moreover, TCAD simulations provide results close to the one achieved by experiment. [60]he acceptor-like tail states near the CB edge can be derived as [73] DOS exp ðEÞ ¼ N TA ⋅ e where E is the energy, E C is the CB energy, N TA is the CB tail state DOS that intercepts E C , and W TA is the characteristic decay energy or slope of N TA .Similarly, the acceptor like Gaussian DOS is written below: [73] DOS GA ðEÞ ¼ N GA ⋅ e where N GA represents the Gaussian-distributed acceptor-like DOS at the midgap (E 0 ) and W GA is the width of the Gaussian acceptor-like state distribution.Also, the donor-like Gaussian DOS follows the equation: [73] DOS GD ðEÞ ¼ N GD ⋅ e where N GD represents the Gaussian-distributed donor-like DOS at the midgap (E 0 ) and W GD is the width of the donor-like Gaussian distribution.
Material Simulation: We used the Wien2k package for DFT calculations. [77]The method lies in the full potential linearized augmented plane wave method. [77,78]We started from the rutile structure of SnO 2 having a tetragonal unit cell (space group P4 2 /mnm).We used the lattice parameters of a = b = 4.738 Å and c = 3.188 Å, with Sn and O positions as (0, 0, 0) and (0.307, 0.307, 0). [79]We used muffin-tin radii R mt of 1.98 for Sn and 1.58 for O.The R mt K Max parameter (where R mt is the smallest muffin tin radius, and K max the magnitude of the largest K vector) was set as 6.5.The energy cutoff separating the core from the valence states was set at À7.0 Ry.
We first evaluated the SnO 2 structure, starting with the positions given above.We then relaxed the structure by finding the minimum energy of the structure by varying the c/a ratio, the volume, and the internal positions of the atoms.We made sure that the relaxed structure forces were below 1 mRy Bohr À1 , while the energy and the charge convergences were set to 0.0001 Ry and 0.001 e, respectively.From the relaxed structure, we then made 2 4 Â 2 Â 2 supercells.One structure had 96 atoms (structure A), from which an oxygen atom was removed to obtain structure B with 95 atoms.One or two electrons were then removed from the structure B to simulate a singly (V þ O ) and a doubly ionized oxygen vacancy (V 2þ O ).We used 1000 irreducible Brillouin-zone k-points for the basic cell and 10 k-points for the 4 Â 2 Â 2 supercells.On each relaxed structure, we then used the Tran-Blaha-modified Beck-Johnson exchange potential to have an opening of the gap. [71]We then analyzed the DOS and the electron density of structures A and B.

Figure 1 .
Figure 1.Effect of light on a-SnO x .a-d) The variation of the transfer curve before and after 1 h of 10.000 Cd m À2 of blue, green, yellow, and red light on an a-SnO x TFT, respectively.The light was turned on for 1 h over the TFT.The transfer curves were measured in the dark before and after the illumination.e) The absorption coefficient as a function of the photon energy.The TFTs were measured at V DS = 0.1 V.

Figure 2 .
Figure 2. Effect of NBIS on the TFT I-V characteristics and their recovery.a,b) The evolution of the transfer curve under NBIS and during recovery, respectively.c) The evolution of the output curve under NBIS and recovery.The transfer curves were measured at V DS = 0.1 V; the output curves were measured at V GS = 0.5 V.During recovery, the TFT was in dark, and the contacts were grounded.The output and transfer curves were then measured at the indicated times.

Figure 3 .
Figure 3. a-d) Transfer characteristics, e-h) output characteristics, and i-l) TCAD DOS model of the fabricated TFTs at initial, after 1 h of NBIS, 1 h of recovery, and after 39 h of recovery, respectively.In (a)-(h), the solid lines represent experimental results and the dashed lines are the TCAD fitting.The transfer curve and output curves were measured at V DS = 0.1 V and V GS = 0.5 V, respectively.

Figure 4 .
Figure 4. Illustration of a) NBIS and b) recovery mechanisms of a-SnO x TFT.Under NBIS, incident light transforms neutral oxygen vacancies (V O ) into single/double-ionized oxygen vacancies (V þ O =V 2þ O ) and provides electrons to the CB.Due to the large negative vertical gate field, these generated positively charged defects accumulate near the GI/a-SnO x interface.Therefore, carrier concentration increased and transfer characteristics shifted toward negative V GS directions.Upon the removal of NBIS, NBIS generated V þ O =V 2þ O defects capture electrons and they become neutral oxygen vacancy defects.However, our TFTs exhibit incomplete recovery of the initial characteristics, suggesting V 2þ O are remaining after the removal of NBIS.

Figure 5 .
Figure 5. Crystal structure of SnO 2 .a) The rutile structure of SnO 2 , b) the difference between structures A and B, c) the variation with a singly ionized oxygen vacancy (V þ O ), d) the variation with a doubly ionized oxygen vacancy (V 2þ O ), and e) the plane used for evaluating the bond and angle difference between the two structures.In (a)-(d), the gray (red) spheres depict Sn (O) atoms.In (b)-(d), the green arrows show the movement of atoms.In (e), the orange circle depicts the oxygen atom removed in structure A to obtain structure B.

Figure 7 .
Figure 7. Electron density contour in the CB of a) the pristine structure A, and b) structure B having an oxygen vacancy, c) structure B with a singly ionized oxygen vacancy (V þ O ), and d) structure B with a doubly ionized vacancy (V 2þ O ).The colour in each figure depicts the electron density.The atoms Sn and O are indicated in all structures for easiness of understanding.

Figure 6 .
Figure 6.The total DOS of a) the SnO 2 structure A with 96 atoms, the SnO 2 structure B with b) a neutral oxygen vacancy, c) a singly ionized vacancy (V þ O ), and d) a doubly ionized vacancy (V 2þ O ).

Table 1 .
Variation of TFT parameters under NBIS and recovery processes.

Table 2 .
Examples of bond lengths and angles in the plane as indicated in Figure5e.