Wavelength-dependent Optical Instability Mechanisms and Decay Kinetics in Amorphous Oxide Thin-Film Devices

We present a study on decay kinetics for a recovery process depending on the light wavelength selected in optical instability measurements against amorphous In-Ga-Zn-O (a-IGZO) thin-film devices. To quantitatively analyze optically-induced instability behaviors, a stretched exponential function (SEF) and its inverse Laplace transform are employed for a time- and energy-dependent analysis, respectively. The analyzed results indicate that a shorter wavelength light activates electrons largely from the valence band while metastable states are deionized with the respective photon energy (hv). In contrast, a longer wavelength illumination is mainly activating trapped electrons at metastable states, e.g. oxygen defects. In particular, at 500 nm wavelength (hv ~ 2.5 eV), it shows an early persistency with a much higher activation energy. This also implies that the majority of metastable states remain ionized, thus the deionization energy >2.5 eV. However, the decay trend at 600 nm wavelength (hv ~ 2 eV) is found to be less persistent and lower current level compared to the case at 500 nm wavelength, suggesting the ionization energy of metastable states >2 eV. Finally, it is deduced that majority of oxygen defects before the illumination reside within the energy range between 2 eV and 2.5 eV from the conduction band edge.


Results and Discussion
Experimental Observation of Optical Instabilities. Figure 1(a) shows the a-IGZO thin-film device with two electrodes fabricated with Mo (e.g. source and drain), i.e. a resistor-type. The detailed fabrication process used in this work is as follows. Note that, following a typical thin-film transistor process procedure 1-3 , starting with the glass wafer, a 50 nm-thick IGZO is formed to be used as a photo-absorption layer in the test device. Here, we used a RF-sputtering with an IGZO ceramic target. Here, we applied a low oxygen-gas partial pressure against Ar of 3%, followed by a thermal annealing at 250 °C. Here, a low oxygen-gas pressure helps to get a high electron density, thus the Ohmic conduction at the contact 1,3,16 . However, a space charge limited conduction may also happen depending on the process and illumination conditions 17,18 . Using this device, the drain-source current (I DS ) as a function of time was measured under illuminations for different wavelengths (λ), e.g. λ = 400, 450, 500, and 600 nm, while applying the drain-source voltage (V DS ) of 1 V, as described in Fig. 1(b). Here, an optical set-up with a light-emitting-diode is used being coupled with the KEITHLEY measuring the photo-current as a function of time, where allows a selection of 4 different wavelengths along with the same light intensity of 100 μW/cm 2 . And the peak value of the photocurrent measured at 300 sec is normalized. These optical measurements were conducted under illumination for 300 s followed by post-illumination recovery for 600 s. As the initial condition, the electron concentration (n 0 ) of the IGZO film before illumination can be estimated from the initial current I DS (t = 0) ~ 0.05 μA, using the definitions of conductivity (σ ~ qn 0 μ n ) and conductance (G = I DS /V DS = σ(S/L)), yielding the following relation, here, the electron mobility μ n is assumed to be ~20 cm 2 /V-s. With Eq. 1, n 0 is found to be about 6.25 × 10 16 /cm 3 for the given device geometries, such as channel length (L = 20 μm) and channel cross section (S = 5 × 10 −6 m 2 ). Here, S is defined as Wt s , where W is the channel width of 100 μm and t s is the channel thickness of 50 nm. As seen in Fig. 1(b), during the illumination, I DS is more quickly increased with a shorter wavelength light. More importantly, after the illumination, i.e. a recovery stage, their decays are relatively slow. In particular, a longer wavelength case shows a more persistent decay with a less portion of an initially-rapid decay. Comparing the current level at t = 0 s with that at t = 900 s, it is suggested that it should be related to the number of additional electrons (Δn) newly generated in the IGZO film during the illumination. For example, at 900 s, Δn for the illumination with λ = 500 nm, which is the most persistent case, is estimated based on Eq. 1 as follows, And Eq. 2 gives Δn ~ 5.38 × 10 17 /cm 3 . Here, the origin of Δn and this metastability, i.e. persistent photoconductivity (PPC), needs to be quantitatively investigated considering possible metastable states and related physical mechanisms.

Decay Kinetics and Mathematical Analysis.
It is known that information on metastable states is associated with the recovery period where a persistency of the film conductivity appears depending on the wavelength 14,15,19 , whereas the excitation stage shows a dramatic increase regardless of wavelengths, as depicted in Fig. 1(b). Thus, in order to mathematically quantify a decay behaviour, the SEF is applied to the recovery stage, i.e. decay period (see Fig. 2(a)), rather than the excitation period, as follows 17,18 , www.nature.com/scientificreports www.nature.com/scientificreports/ eff where τ eff is an effective time constant, and β is a stretched exponent. Note that an a-IGZO has many trap energy levels, so the SEF is required to analyze the decay characteristics 20,21 . It is mainly because, as seen in Fig. 2(a), the decay trend still shows non-linear characteristics in log-scale, implying that a single exponential is insufficient. And the amplitude of SEFs is normalizedfor a fair comparison across different cases. Here, the least square method is applied to extract optimum values of those parameters using MATLAB TM . Based on this computational procedure, parameters retrieved here are summarized in Table 1.
Based on those extracted parameters of the SEF for each case, the activation energy distribution (AED) of metastable states can also be mathematically estimated. For this, firstly, the SEF is converted to a function of frequency (f(S)) using the inverse Laplace Transform, i.e. L −1 {F(t)}, as follows 22 , j j eff where S is a frequency. Secondly, S in Eq. 4 can be replaced with an activation energy (E A ) using the following Arrhenius relation,

AE A
Here, ν AE is an attempt-to-escape frequency and kT is the thermal energy. And the solution of Eq. 4 can be represented as a function of E A while replacing S with Eq. 5, yielding, where γ is defined as β/(1 − β). Along with values of β and τ eff seen in Table 1, we can now find the AEDs for respective cases, as shown in Fig. 2(b). Here, ν AE is assumed to be 10 7 /sec for the a-IGZO 23 .
As shown in Fig. 2(b), it is found that a peak of the AED is usually shifted to a higher energy with increasing the wavelengthwithin the range from 400 nm to 500 nm. When we increase the wavelength further, however, the 600 nm-wavelength-light shows a narrow distribution with a peak at a lower energy in comparison even with the case of λ = 400 nm. It means that dominant metastable states are activated by illuminations at the wavelength range of 400 nm ~ 500 nm.

Instability Mechanisms and Capture Cross-sections.
To specify a physical mechanism for respective case (see Fig. 2(b)), a capture cross-section is drawn, as seen in Fig. 3. There are two main physical mechanisms. The basic process is the band-to-band excitation, i.e. electron emission directly from the valence band (VB), and another process is associated with metastable states, such as oxygen vacancies (V OX°) and oxygen interstitials  www.nature.com/scientificreports www.nature.com/scientificreports/ (I OX 2− ). In particular, those oxygen defects are ionized under the illumination having respective chemical reactions, i.e. V OX° → V OX 2+ + 2e − and I OX 2− → I OX 0 + 2e − . And this gives rise to a persistent photoconductivity (PPC), i.e. metastability 15,24 . The detailed explanation for respective cases is as follow.
As shown in Fig. 3(a) for λ = 400 nm, electrons are generated from the VB as well as oxygen defect states. The photon energy (hv) of this case is comparable with band-gap energy (E g ) of the a-IGZO and much bigger than ξ 1 (i.e. ionization energy). In addition, hv at λ = 400 nm (~3.1 eV) is much greater than ξ 2 (i.e. deionization energy), so that an instantaneous recombination of ionized metastable states can slightly happen with electrons supplied from the VB. As described in Fig. 3(b), the overall process at λ = 450 nm is similar to the case of λ = 400 nm. In this case, hv at λ = 450 nm is ~2.8 eV, which is also higher than both ξ 1 and ξ 2 . This suggests that the ionization of metastable states as well as instantaneous recombination (i.e. deionization) for some of ionized states can simultaneously happen as well. However, its photon energy is insufficient for the band-to-band excitation. Instead, the excitation process of VB electrons may get an assistance through valence band tail (VBT) states, thus an indirect excitation (see Fig. 3(b)). Similarly, for λ = 500 nm (i.e. hv ~ 2.5 eV), oxygen defects can still be ionized with hv bigger than ξ 1 , but an instantaneous recovery process is largely screened if hv < ξ 2 , as shown in Fig. 3(c). In this case, likewise, the excitation of VB electrons may also slightly happen with the assistance of VBT states. This implies that majority of ionized oxygen defects cannot be deionized, corresponding to the highest E A , as seen in Fig. 2(b). In other words, it becomes difficult for slow metastable states to be recombined instantaneously under that illumination condition where the wavelength is longer than 500 nm. Indeed, for the case of λ = 600 nm (see Figs 2(b) and 3(d)), since hv(~2.0 eV) is insufficient to overcome ξ 1 , so I OX 2− and majority of V OX 0 are hard to be ionized, but a small portion of the V OX 0 can be ionized. And this is consistent witha smaller increase of the measured current during the illumination with λ = 600 nm (see Fig. 1(b)). Therefore, the AED for λ = 600 nm shows a narrower distribution with a peak point which is lower than other cases. Figure 4 shows a summary of the capture cross-section discussed with Fig. 3. This finally suggests that the ionization of metastable states dominantly happens when hv > ξ 1 , and their deionization (i.e. recombination) is strongly screened when hν < ξ 2 . So, it is implied that the case with λ = 500 nm is satisfied with both conditions, so more persistent than any others in our experiments. From these results, the ranges of the ionization and deionization energies (ξ 1 and ξ 2 ) are estimated as follows, respectively, λ .

Conclusions
A quantitative analysis based on a stretched exponential function and its inverse Laplace transform is presented on optical instability mechanisms and respective decay kinetics in amorphous oxide thin-film devices. The presented analysis shows that decay behaviour is largely dominated by ionized metastable states which reside near the conduction band minima. Here, their instantaneous deionization with the valence band electrons are found to be more screened as increasing the wavelength up to 500 nm (hv ~ 2.5 eV). This results in a more persistency as a function of time, which explained with an insufficient photon energy to reach them located near the conduction band minima, suggesting the deionization energy >2.5 eV. With the 600 nm wavelength light (hv ~ 2.0 eV), however, the decay trend becomes slightly less persistent compared to the 500 nm case, suggesting the ionization energy of metastable states >2 eV. These results indicate that the majority metastable states before illumination are located in the energy range between 2 eV and 2.5 eV from the conduction band edge.