Origin of light instability in amorphous IGZO thin-film transistors and its suppression

Radiating amorphous In–Ga–Zn–O (a-IGZO) thin-film transistors (TFTs) with deep ultraviolet light (λ = 175 nm) is found to induce rigid negative threshold-voltage shift, as well as a subthreshold hump and an increase in subthreshold-voltage slope. These changes are attributed to the photo creation and ionization of oxygen vacancy states (VO), which are confined mainly to the top surface of the a-IGZO film (backchannel). Photoionization of these states generates free electrons and the transition from the neutral to the ionized VO is accompanied by lattice relaxation, which raises the energy of the ionized VO. This and the possibility of atomic exchange with weakly bonded hydrogen leads to metastability of the ionized VO, consistent with the rigid threshold-voltage shift and increase in subthreshold-voltage slope. The hump is thus a manifestation of the highly conductive backchannel and its formation can be suppressed by reduction of the a-IGZO film thickness or application of a back bias after radiation. These results support photo creation and ionization of VO as the main cause of light instability in a-IGZO TFTs and provide some insights on how to minimize the effect.


Results and discussion
Effect of light radiation. Figure 1b shows the effects of monochromatic light on the performance of the a-IGZO TFTs. The wavelength of the light (λ) is varied from 800 to 172 nm, which corresponds to photon energy of 1.6 to 7.2 in electron volts (eV), and the transfer characteristics are measured after light irradiation for 500 s. Although small, the negative ΔV TH after irradiation with photon energies less than 3.0 eV suggests photoexcitation from sub-gap states 22 . A slight jump in the ΔV TH is apparent for photon energies in the range of 3.0-3.4 eV and indicates the onset of band-to-band excitation. The photon energy of the deep UV light (~ 7.2 eV) is large enough to induce substantial negative ΔV TH (Fig. 1b). For a clear understanding of the effects of light in the a-IGZO TFTs, we therefore use the deep UV light (λ = 172 nm) in all investigations that follow. The negative ΔV TH induced by deep UV light is accompanied by an increase in SS and the formation of a 'hump' in the transfer characteristics, which increases with exposure time (Fig. 1c). Also, quite noticeable is that the µ FE doubles after radiating the TFTs with deep UV for 3,600 s, consistent with a photo-induced increase in free carrier concentration.
Flat band parameters extracted from the combined analysis of the I D -V GS (Fig. 2a) and C-V (Fig. 2b) characteristics are listed in Table 1. Detailed extraction methods can be found in Ref. 17. From the plot of surface potential (Ψ S ) vs. V GS (Fig. 2c), the flat band voltage (V FB ), taken as the V GS corresponding to Ψ S = 0 V, decreases from 0.36 to − 5.89 V after light radiation. At the same time, the Fermi level (E F ) moves closer to the conduction band (E C ) by approximately 0.023 eV and flat band carrier concentration (n FB ) increases from 1.4 × 10 16 to 3.4 × 10 16 cm −3 . Interface trap density (N int , the algebraic sum of negative and positive interface charge density) also increases from − 7.7 × 10 10 to 3.7 × 10 11 cm −2 eV −1 , while gap trap density per unit energy (dN gap /dE, the sum of interface and bulk density of states) increases from 1.3 × 10 12 to 3.6 × 10 12 cm −2 eV −1 . Note that a negative value of N int indicates that negative charge exceed positive charge by that amount and vice-versa, given that N int is the algebraic sum of positive and negative interface charge density.
These photoinduced changes in µ FE , SS, n FB , V FB , E F -E C , N it , and dN gap /dE are all consistent with accumulation/trapping of net positive charge at the gate-insulator/a-IGZO interface and creation of defects in the a-IGZO layer. The increase in SS indicates formation of the defects and given that µ FE increased and E F moved closer to E C , some of these defects must be in the bulk of the a-IGZO and donor-like. The defects must also be positively  www.nature.com/scientificreports/ charged (ionized), as the value of N it changed from a negative to a positive number, making the increase in n FB a sum of band-to-band and sub-gap photoexcitation.
Recovery after UV light radiation. The photoinduced negative ΔV TH is quite repeatable over many samples with varying channel dimensions. In particular, the ΔV TH is independent of L ( Supplementary Fig. S1) and rigid such that recovery in the dark is almost negligible at temperatures ≤ 100 °C (Fig. 3). At 150 °C, the transfer characteristics recover slowly and fail to return to the initial state even if the TFTs stay in the dark for a few days (Fig. 3c). Complete recovery is only achievable after annealing at 250 °C in vacuum for two or more hours. Therefore, the ionized donor-like defects generated by light radiation must be metastable states 23,24 . However, the exceedingly small but rapid initial recovery (within 100 s) at low temperature could be due to delayed electron-hole recombination via non-metastable gap states. As shown in the Fig. 3d, the time dependent ΔV TH during the recovery period is well described by the stretched exponential equation given by where V 0 is the ΔV TH at infinite time. The equation describes the superposition of processes with a spread of time constants (quantified by the stretch parameter β < 1) around an average value τ. A plot of the time constant τ as a function of temperature can be fitted to a straight line (Fig. 3e) described by the equation where T is the temperature in kelvins and k B is the Boltzmann constant. This equation is typical of thermally activated processes with an activation energy E τ . Here, the E τ required to reverse the formation of the defects created during the deep UV light radiation is estimated to be approximately 0.99 eV. This value is close to the activation energy previously obtained for the creation of a V O related double donors during NBIS 25 .
Location of defects. The nature of these defects is investigated by XPS depth profile analysis of thin-film stacks (glass/SiO 2 /a-IGZO/SiO 2 ) before and after UV exposure (Fig. 4). The O 1 s spectrum is deconvoluted into four energy peaks: O-M, V O , O-H, and Si-O. Peak O-M, which is centered at 529.9 eV, is attributed to O 2− ions binding with In, Ga, and Zn atoms-and thus represents the quantity of the oxygen atoms in a fully oxidized stoichiometric environment. Peak O-H is related to metal-OH (hydroxyl) bonds and centered at 532.1, whereas peak Si-O is related to Si-O bonds and centered at 532.4. The V O peak at 531.2 eV stems from the deficiently bonded oxygen in the a-IGZO layer containing nonstoichiometric oxide species, such as In 2 O 3−x , Ga 2 O 3−x , and ZnO 1−x , which are associated with V O 26, 27 . Area percentages of each peak before and after radiation are listed in Table 2.   (Fig. 4b), a significant amount is present at the top (Fig. 4a) and bottom (Fig. 4c) surfaces of the pristine a-IGZO film as expected. The amount of V O is also larger at the surfaces compared to the bulk. Similar situations have been reported before, identifying V O as a surface feature, confined within 0.5 nm of the top surface of the film 16,28 . While deep UV radiation results in the creation of very few V O defects in the bulk (Fig. 4e) and at the bottom surface ( Fig. 4f) of the a-IGZO, a substantial amount is created at the top surface ( Fig. 4d), where the area percentage increases from 10 to 24% ( Table 2). The area percentage of the Si-O peak also increases from 15 to 28% at the top surface, indicating the diffusion of Si atoms into the a-IGZO layer, possibly from the broken weak Si-O bonds at the top interface. It should be pointed out that a large overlap exists between the energy peaks O-H (532.1 eV) and Si-O (532.4 eV), making it difficult to separate the two. To ensure a reasonable peak ratio, we fixed the full width half-maximum of all components to 1.
The use of optical electron paramagnetic resonance (EPR) experiments has shed light on the nature of V O in ZnO single crystals 29,30 . V O bind two electrons in their electrically neutral state (V O 0 ) but can also exist in a singly ionized (V O + ) and doubly ionized (V O 2+ ) state. The increase in the concentration of V O states at the top surface of the a-IGZO film after light radiation indicates photo-creation of the V O . However, this increase is too small to account for ΔV TH > 10 V after light radiation (Fig. 1c). A plausible explanation would be the photoionization of existing and newly created V O 0 to V O 2+ states-a process which donates two electrons to E C . The presence of two electrons in V O 0 triggers the surrounding Zn 2+ ions to move inwardly, reducing the physical size of the vacancy. Consequently, the binding energy increases as the overlapping between the Zn 2+ wave functions intensifies, such that ε(+ /0) < ε(2 + /0) < ε(2 + / +), where ε(+ /0) is the transition energy from the V O + to V O and similarly for the other two cases [31][32][33]  If that occurs, free electrons occupying lower lying states in the conduction band minimum will need to be activated, thermally or otherwise, to these higher V O 2+ state levels for the reverse reaction, making the V O 2+ a metastable donor defect. Being ionized states that are degenerate with the E C , the V O 2+ will not be detected by XPS 28 . However, the negative ΔV TH increase in SS after light radiation is consistent with the photo-creation of shallow states. Moreover, the binding energies of the In3d 5/2 , Ga2p 3/2 , and www.nature.com/scientificreports/ Zn2p 3/2 peaks all show a shifting towards higher binding energy, indicating a decrease in the electron density around the metals ( Supplementary Fig. S2). This is consistent with ionization. Hydrogen also plays a role in the instability of the a-IGZO TFTs. In fact, a high concentration of hydrogen (in the order of 10 20 cm −3 ) has been observed in a-IGZO thin films, without intentional exposure of the films to hydrogen during their deposition 37,38 . Incorporated hydrogen can fill V O sites in a-IGZO films, forming stable + 1 charge states as donors 39 . If these states are also degenerate with the E C , they will be hard to detect by XPS. However, a significant increase in the amount of hydrogen after UV light radiation is detected by ToF-SIMS depth profiling of the SiO 2 /IGZO/SiO 2 structures used in this study. The increase is mainly confined to the top surface of the a-IGZO layer as can be seen in Supplementary Fig. S3d. Additionally, ToF-SIMS also detected a smaller amount of gallium, zinc, and oxygen at the top interface compared to the bulk ( Supplementary Fig. S3), which is consistent with a larger population of defects at the top interface.
Hydrogen impurities generate two types of defects, depending on whether they are bonded to an oxygen (OH) or to a metal (M-H) site. The OH defects have negative formation energy, implying a spontaneous formation whenever hydrogen is present. These defects do not generate states in the gap but act as donors until a high electron concentration is achieved, after which hydrogen starts to bind itself to metal sites, forming acceptors which compensate for the surplus of electrons in the a-IGZO. As the M-H bond requires an electron to form, it thus helps to limit the carrier concentration in a-IGZO, and forms states just above the E V 40 . ToF-SIMS results

M-O (%) V O (%) O-H (%) Si-O (%)
Before UV After UV Before UV After UV Before UV After UV Before UV After UV  41 . Furthermore, the positive shift in binding energy of the In3d 5/2 , Ga2p 3/2 , and Zn2p 3/2 peaks is mainly confined to the top half of the a-IGZO film ( Supplementary Fig. S2) and a linear plot of the transfer characteristics in Fig. 1c shows two different slopes for negative and positive V GS ( Supplementary  Fig. S4), which clearly indicates the presence of two logical channels with two different resistances. Taking all this into consideration, the formation of the subthreshold hump in the transfer characteristics after deep UV radiation can thus be explained by the creation of shallow donors at the top surface of the a-IGZO film. This will create a parasitic channel (backchannel) with a conductance that is higher than that of the a-IGZO bulk (frontchannel) 42 . Figure 5 shows how this conductive backchannel is manifested as a hump in the transfer characteristics of the a-IGZO TFTs post radiation. In Fig. 5a, the transfer characteristics are divided into three regions, labeled I, II, and III. In region I, the a-IGZO layer is depleted of electrons due to the strong negative V GS and only holes remain in the channel. In a-IGZO, it is difficult to induce holes with negative V GS , owing to the large concentration of V O located less than 1 eV above E V 43 . So, these holes must be the photogenerated holes that have not yet recombined. The holes drift towards the source (S) under the influence of the positive V DS , resulting in a small hole current and lowering of the source barrier for electron injection (Fig. 5b) 23,44 . In region II, electron accumulation begins at the top surface of the a-IGZO, and the backchannel becomes conducting, while the frontchannel is still not (Fig. 5c). In region III, both the front and backchannel are conducting, and I D is the sum of the currents flowing in the two channels (Fig. 5d).
In region III, the current due to the backchannel is lower than that of the bulk (front) (Fig. 5a) because the latter is thicker than the former. TCAD simulation of the TFT transfer characteristics with (Fig. 6a) and without (Fig. 6b) a highly conductive top surface of the a-IGZO film, also yielded the same results (Fig. 6c), verifying the hump formation mechanism. Here, a 2 nm-thick a-IGZO layer with a donor concentration (n gd ) of 7 × 10 18 cm −3 is used to represent the highly conductive a-IGZO top surface (Fig. 6b). n gd of the bulk is 1 × 10 17 cm −3 . Other density of states parameters have the same values as those reported in 45 . A depth profile analysis of the electron concentration (Fig. 6d) indicates that a highly conductive backchannel is responsible for the negative ΔV TH . Consistent with the mechanism for the "hump", only the backchannel is conductive when the V GS is negative (e.g., V GS = − 10 V) as shown in Fig. 6e. When the V GS is positive (e.g., V GS = 10 V), conduction occurs in both the backchannel and frontchannel (Fig. 6f).
It is important to note that the hump starts at the same point (at approximately V GS = 0 V), regardless of the UV light exposure time, although it progressively stretches into the negative V GS direction with increasing exposure time (Supplementary Fig. S4). This is consistent with the increase in the number of photoionized V O at the top surface of the a-IGZO film and the consequent increase in the backchannel current with UV exposure time. In other words, the negative ΔV TH is mostly due to defects generated at the top surface of the a-IGZO film. Application of negative V GS during light illumination pushes the E F towards E V and increases the concentration of holes, favoring the formation of these defects, which is why the effect is larger under NBIS. www.nature.com/scientificreports/ It is expected for one to suspect that the results presented herein show defect generation only at the top surface because absorption of light with high photon energy (~ 7.2 eV) is limited to the top surface. However, the penetration depth for λ = 172 in a-IGZO has been reported to be 30 nm for intensity 8 times smaller than the one used herein 46 . In fact, by considering the a-IGZO film thickness (d) of 20 nm and incident light intensity (I 0 ) of 400 mW/cm 2 , the transmitted light intensity (I t ) can be estimated from to be approximately ~ 54 mW/cm 2 , which is too high to eliminate absorption at the bottom surface of the a-IGZO. Here, α is the absorption coefficient, which is assumed to be approximately 10 6 cm −1 for λ = 172 nm based on the extrapolation of previously published results 47 . Therefore, the origin of the light instability in a-IGZO TFTs is indeed due to the ionization of V O states that are intrinsic to the top surface of the a-IGZO films.  . The above-mentioned mechanism is similar to the Staebler Wronski effect in a-Si:H 49 , where hydrogen plays a major role in a mechanism known as bond switching 50 . In this case, a weak Si-Si bond is also broken by using the energy released from the recombination of photo-generated electron-hole pairs, and two stable dangling bonds can be formed if a hydrogen atom from a neighboring Si-H bond is exchanged for one of the created dangling bonds. Therefore, atomic exchange after photo-creation of defects in a-IGZO with weak bonds present in the random lattice could play a role in the photo-creation of defects also in a-IGZO. Note that metal vacancies or interstitials are also present in the amorphous network of the IGZO 35 . Additionally, as hydrogen in the excess of ~ 10 20 cm -3 has been detected in a-IGZO 37 , a significant amount of OH and V O -related species are apparent at the top surface of the a-IGZO films presented herein.
After photo-creation of V O , the oxygen can be accommodated interstitially or through the formation of peroxides, which involve the covalent bonding of two oxygen atoms (O-O). While interstitial oxygens are stoichiometric defects which act as acceptors 34 , peroxides are donors that spontaneously form when a large concentration of holes is available in the E V 41 . The mechanism responsible for the deep UV light instability in a-IGZO is thus summarized diagrammatically in Fig. 7.

Suppression of the light instability.
Having identified the backchannel as the source of most of the light instability, owing to the defects that occupy the top surface of the a-IGZO, the effect of applying a back bias to reduce the instability is investigated. Double gate TFT structures with transparent top gates that fully (Fig. 8a) or partially (Fig. 8e)   www.nature.com/scientificreports/ Extracted TFT parameters can be found in Supplementary Fig. S5. The DG TFT results are interesting and support the light instability mechanism described above. First, the existence of a full TG bias suppresses the negative ΔV TH , for all bias conditions (see Fig. 8b-d and Supplementary Fig. S5a). However, the off-state leakage current (I OFF ) significantly increases after UV light radiation ( Fig. 8 and Supplementary Fig. S5b). This indicates that I OFF is the result of photogenerated holes drifting from the D to the S. While a grounded or negatively biased TG suppresses electrons from the conductive backchannel, which is the reason for the negligible ΔV TH , it promotes hole induction, consistent with the high I OFF . I OFF is also high during the DG sweep (grounded TG) because the hole layer is simply shifted towards the frontchannel, which is negatively biased. Grounding or negatively biasing the TG does not make a big difference with regards to the size of the off-state current because hole induction by V GS is difficult in a-IGZO, owing to the large concentration of V O located less than 1 eV above E V 43 . The holes responsible are the photogenerated holes that are yet to recombine. Second, negative ΔV TH and increase in SS occur after UV light radiation for 3600 s when a partial TG is implemented ( Fig. 8f and Supplementary Fig. S5c). However, the negative ΔV TH is not as large as that of a single gate TFT (Fig. 1c), indicating partial suppression of the conductive backchannel electrons by the partial TG. Additionally, the SS increases without the formation of a hump during the TG sweep (Fig. 8f) because there is only one channel (the backchannel) involved, supporting the mechanism for the "hump". However, suppression of the effect of the conductive backchannel electrons can be seen during the BG (Fig. 8g) or DG (Fig. 8h) sweep with a partial TG but evidence of it just starting to appear is apparent. The increase in the on-state current (I ON ) after UV light radiation during the TG sweep (Fig. 8f) is consistent with a photoinduced increase in the conductance of the offset regions. Note that the increase in I ON after UV light radiation is almost negligible when a full TG is implemented (Supplementary Fig. S5b). Consequently, the change in field-effect mobility after UV light radiation is exceedingly small when a full TG is used (Supplementary Fig. S5d).
These results provide further evidence showing that defects occupying the top surface of the a-IGZO are the source of the light instability in a-IGZO TFTs and that their effect can be suppressed by applying a back bias. This is especially important in applications such as transparent displays where the use of light shields is not possible. Although back biasing does not prevent the off-state currents from increasing, this is not expected to be a problem as they do not exceed 100 pA and they quickly recover at room temperature when the holes recombine. An alternative to applying a back bias, for instance in single gate inverted staggered TFTs (Fig. 1a), would be to reduce the thickness of the a-IGZO film. By doing this, the bottom (front) gate will not only have control of the backchannel, but the total number of defects is also reduced 9 . As shown in Fig. 9, using an a-IGZO film thickness of 7 nm results in negligible ΔV TH .  www.nature.com/scientificreports/ Fabrication process optimization to ensure a clean back interface is thus another way to minimize the light instability. The use of etch-stopper layers 51 , damage-free source/drain metal etchants 19 , high quality encapsulation layers 16 , high temperature or long-time annealing 12 , high pressure or water vapor-assisted annealing 17,18 and treatments 52,53 , are some of the optimization techniques that can be implemented.

Conclusion
By radiating a-IGZO TFTs with deep UV light (λ = 175 nm), we have shown that the origin of light instability in a-IGZO TFTs is V O and hydrogen-related defects that are located at the top surface of the a-IGZO film. Under light illumination, metastable donor states of these defects are created, making the backchannel conductive. This conductive backchannel results in rigid negative ΔV TH that is accompanied by an increase in SS and a 'hump' in the subthreshold region of the transfer characteristics. Most importantly, we have also shown that this instability to light illumination can be suppressed by applying a back bias or decreasing the thickness of the a-IGZO film in applications, such as transparent displays in Heads Up Displays (HUDs) or smart glasses, where the use of light shields is not permissible. Fabrication process techniques that ensure a clean top surface of the a-IGZO film, such as plasma treatments, water vapor-assisted annealing, and high-quality encapsulation layers are also necessary to minimize the light instability.

Methods
Device fabrication. We fabricated the a-IGZO TFTs with the inverted staggered structure (Fig. 1a) using the backchannel etch (BCE) process described in detail elsewhere 19 . In this process, molybdenum (Mo) layers deposited by sputtering at 200 °C are used for the gate (60 nm) and source/drain (150 nm) electrodes. Silicon-dioxide (SiO 2 ) layers deposited through plasma-enhanced chemical vapor deposition at 360 °C and 200 °C are used for the gate-insulator (200 nm) and the passivation layer (200 nm), respectively. The 20 nm-thick a-IGZO layer is deposited at 200 °C by reactive sputtering using a polycrystalline IGZO target (In 2 O 3 :Ga 2 O 3 :ZnO = 1:1:1 mol%). The gate-insulator (G.I.) and the a-IGZO layer are deposited consecutively in a cluster deposition tool, without breaking vacuum, to achieve a clean G.I./a-IGZO interface. A hydrogen peroxide-based etchant (pH = ~ 5) is used to pattern the source/drain electrodes to minimize backchannel acid corrosion 19 , and to ensure a reproducible unstressed state, the devices are annealed at 250 °C in vacuum for 2 h before measuring.
Characterization. We used the Agilent 4156C precision semiconductor parameter analyzer and the Agilent E4980A Precision LCR meter to measure the current-voltage (I-V) and capacitance-voltage (C-V) characteristics, respectively. For the C-V measurement, we superimposed the DC gate-voltage (V GS ) on a small AC signal (0.1 V) of frequency 1 kHz, while keeping the source and drain shorted. We derived the field-effect mobility (µ FE ) in the linear regime (with drain voltage (V DS ) = 0.1 V) from the transconductance (gm = ∂I D /∂V GS ) by using µ FE = (gm*L)/(W*C OX *V DS ). Here, I D , C OX , L, and W are the drain terminal current, the G.I. capacitance per unit area, the channel length, and the channel width, respectively. We took the SS as the minimum value of (∂log(I D )/∂V GS ) −1 and the threshold voltage (V TH ) as the V GS corresponding to I D of 1 nA. We carried out the light illumination experiments at room temperature with the TFT electrodes in the floating state. To further characterize the effects of deep UV light radiation, we extracted flat band parameters by a combined analysis of the TFT transfer (I D -V GS ) and C-V characteristics according to methods previously described in 20 . We characterized thin film stacks of SiO 2 (100 nm)/a-IGZO (20 nm)/SiO 2 (100 nm) by using Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS) and X-ray photoelectron spectroscopy (XPS) depth profiles. For ToF-SIMS, an ION-TOF (Münster, Germany) instrument (TOF-SIMS V) equipped with a Bi1 + (30 keV, 1 pA) and Cs + (3 keV, 30 nA) gun is used and raster areas for sputter and analysis are 200 μm × 200 μm and 50 μm × 50 μm, respectively.