Origins of Highly Stable Al-evaporated Solution-processed ZnO Thin Film Transistors: Insights from Low Frequency and Random Telegraph Signal Noise

One long-standing goal in the emerging field of flexible and transparent electronic devices is to meet the demand of key markets, such as enhanced output performance for metal oxide semiconductor thin film transistors (TFTs) prepared by a solution process. While solution-based fabrication techniques are cost-effective and ensure large-area coverage at low temperature, their utilization has the disadvantage of introducing large trap states into TFTs. Such states, the formation of which is induced by intrinsic defects initially produced during preparation, have a significant impact on electrical performance. Therefore, the ability to enhance the electrical characteristics of solution-processed TFTs, along with attaining a firm understanding of their physical nature, remains a key step towards extending their use. In this study, measurements of low-frequency noise and random telegraph signal noise are employed as generic alternative tools to examine the origins of enhanced output performance for solution-processed ZnO TFTs through the control of defect sites by Al evaporation.

Scientific RepoRts | 5:16123 | DOi: 10.1038/srep16123 To date, numerous reports have been published on the successful enhancement of output performance in solution-processed metal oxide semiconductor TFTs [11][12][13] . Our previous studies also addressed the concept of Al-evaporated ZnO TFTs exhibiting highlystable device performance in air [14][15][16] . Control of the carrier concentration and defect states by varying the time of Al evaporation onto a thin ZnO layer had a significant effect on device performance. However, the photoluminescence [PL] spectrum measurements (not shown in this figure) exhibited no large difference in the deep level statesafter various Al incorporation, except a slight peak shift in the near band edge spectrum depending on Al evaporation time. In particular, an enhancement in electrical features was likely to arise from a both a reduction in weak Zn-O bonding, whichis related to shallow donor trap states through Zn substitution by Al, and a decrease in adsorbed oxygen, which is related to the deep level trapping of electrons through the passivation role of Al-evaporated ZnO TFTs 14,15 . However, questions still remain regarding how carriers are transported between trap sites under different biases and what the main conduction features are after Al evaporation.
In an attempt to answer the aforementioned questions, low frequency noise (LFN) and random telegraph signal noise (RTN) analyses were utilized as alternative techniques to justify the role of Al evaporation on the enhanced electrical features discussed in previously published papers 14,15 . In general, the trapping of carriers within the bandgap of a semiconductor severely limits the performance of analog and digital devices. Therefore, noise characteristics that are governed by the trapping and de-trapping of carriers have garnered great interest in efforts to understand degradation and functionality in semiconductor devices [17][18][19][20] . In this work, observed LFN behaviors were attributed to traps within the bandgap of metal oxide semiconductors and two possible fluctuation sources: carrier number fluctuation (CF) and mobility fluctuation (MF). The number of CFs and MFs that play an important role in noise features were considered in order to verify the experimental findings at different sample and bias conditions [17][18][19][20][21][22] .
This letter addresses the origins of the LFN and RTN observed in various solution-processed ZnO TFTs with and without Al evaporation on the back channel of the ZnO active layer. The RTN responses were tested at each point of the drain currents, thereby providing two discrete switching events between two levels (high-and low-current states). Such events are induced by carrier capture and emission events, depending on the gate voltage. Plots of the drain current power spectral density versus the drain current of proper Al-evaporated ZnO TFTs confirmed a transition between CF and MF models in the vicinity of the threshold voltage. Based on the observed RTN features, we describe a possible mechanism for stable device performance after Al evaporation that is mainly associated with carrier trap and de-trap behaviors at trap centers in different gate voltage regimes. Figure 1 shows the transfer I-V curves and drain current normalized noise power spectral densities (S Id /I d 2 ) of pure ZnO [Sample A] and Al-evaporated ZnO TFTs [Samples B, C, and D] measured over three different voltage regimes; the channel width (W) and length (L) were 500 μ m and 50 μ m, respectively. For convenience, the three gate voltage (V g ) regimes are denoted as regime I (in the vicinity of the threshold voltage), regime II (above the threshold voltage), and regime III (far away from the threshold voltage; in this letter, V g = 30 V). The Al evaporation times for Samples B, C, and D of Fig. 1 were 10, 20, and 40 s, respectively. More detailed noise power spectral density (S Id ) versus frequency data for Samples A-D are given in Figure S1 of the Supporting Information. The representative transfer I-V curves for Samples A, B, C, and D are displayed in Fig. 1a-d, respectively; all samples exhibit typical n-channel behavior. It is widely believed that the hysteresis window feature in the I-V curves represents the deep level trapping of electrons with a long release time, whereas the mobility is primarily determined by shallow trap states with a short release time 23 . The I d values in all regimes increase with longer Al evaporation times. Sample A exhibited a large hysteresis feature, while the current performance of Sample C rapidly improved without hysteresis loss. The enhancement in current performance for Sample C can be attributed to a decrease in oxygen vacancy-related porous sites, which in turn leads to a reduction in the defect density as previously reported by other groups 15,16 . The observed I-V curves were separated into regions in the vicinity of the threshold regime (gray color) and above the threshold regime (blue color), depending on the drain current. Shown in Fig. 1e-h are the drain current normalized power spectral density (S Id /I d 2 ) plots used to observe LFN features for Samples A, B, C, and D, respectively. The S Id /I d 2 versus frequency (range = 4 Hz to 200 Hz) plots were recorded at a fixed drain-to-source voltage (V d = 10 V). Three points (black stars, red squares, and blue circles in the I-V curves of Fig 1a-d) were chosen for the S Id /I d 2 measurements. As evident in Fig. 1e-h, S Id /I d 2 decreases with increasing drain currents for all samples, reflecting typical LFN features observed for conventional metal oxide semiconductor field effect transistors (MOSFET) [17][18][19][20][21][22] . Thus, the LFN characteristics of our ZnO TFTs may be expressed by well-known LFN models frequently used for conventional MOSFET devices (this topic will be addressed later). In addition, S Id /I d 2 exhibited a frequency dependency with a slope of 1/f γ slope (γ = 1.2 ~ 1.7) for all samples. The similarity of the noise slopes implies that the nature of noise in all specimens is also similar, i.e., carriers are trapped and de-trapped at defect centers. When examining the results obtained for Samples A, B, and C, it can be seen that increasing the Al evaporation time served to increase γ . For example, the value of γ (1.6 ~ 1.7) for Sample C with stable I-V characteristics was larger than that of Sample A (1.0 ~ 1.3) with unstable I-V features. The larger γ value for Sample C may suggest that the fluctuation of carriers with long lifetimes has a more significant impact on the LFN behavior when compared to that of carriers with a relatively short lifetime. However, the value of γ (1.2 ~ 1.3) for Sample D was similar to that of Sample A, which means that the fluctuation of carriers with both long and short lifetimes contributes to the LFN characteristics over the entire frequency range. As such, more empirical results and comparisons are needed to establish a clearer empirical model or explanation for the origin of the various γ values observed in the samples.   Figure 2 shows the S Id /Id 2 curves corresponding to Samples A (black lines), B (blue lines), and C (red lines) in regimes I (Fig. 2a) and II (Fig. 2b); Sample D was excluded due to its high leakage response in regime I of Fig. 1d. As displayed in Fig. 2a,b, the noise level of Sample A was always higher than that of Samples B, C, and D in all regimes. Such results indicate an enhancement in noise characteristics upon Al evaporation onto the ZnO layer, possibly due to a reduction in carrier trap or scattering centers that act as the main LNF sources. In addition, a significant decrease, by a factor of 10 4 , in the magnitude of noise was observed for Sample D when compared to that of Sample A (see Fig. 2b). This also suggests a reduction in noise sources for the ZnO TFTs with longer Al evaporation times.

I-V response and typical low-frequency noise behaviors for ZnO-based TFTs.
To further clarify the above observations regarding the effect of Al evaporation, two specific noise models frequently implemented for field effect transistors (FET) were adopted. First, the CF model describes fluctuations in the carrier density due to electrons localized at trap sites near the channel-oxide interface or other deep level traps induced by weak structural bonding. Second, the MF model describes fluctuations in charge mobility due to fluctuations in the mean free path of electrons. According to the CF model 17 , S Id in regime I can be expressed as where q is the elementary charge, k is the Boltzmann constant, T is temperature, Not is the trap density per unit area, C ox is the gate capacitance per unit area, and γ is the slope parameter. Furthermore, the value of SId above the threshold region (regimes II and III) can be expressed as 17 where μ is the charge carrier mobility. According to the Equation 1 and 2, the S Id -I d dependence gives two different slopes, 2 or 1, for regimes I and II/III, respectively. According to the MF model based on Hooge's empirical law 17 , S Id can be expressed as where α H is the Hooge parameter that allows for a comparison of noise characteristics in different conductive materials. Equation 3 suggests a slope of 1 from the S Id -I d plot, which is similar to that derived from Equation 2.
To formulate a relationship between the two noise models and the observed results, plots of SId versus drain current (I d ) were generated for each sample over all regimes; the results are shown in Fig. 3. The SId plots as a function of drain current were recorded at f = 50 Hz and V d = 10 V. In regime I (gray color in Fig. 3), the slope was found to vary from 1.9 to 1.1 depending on the sample (Sample D was excluded due to its high leakage response in regime I). More detailed S Id -I d data recorded at different frequencies for Samples A-D are presented in Figure S2 of the Supporting Information. The slope of Sample A was about 2 times larger than that of Sample C in regime I, while the slopes of all samples in regime II and III (blue color in Fig. 3) were approximately equal to 1. Thus, the slopes of all specimens may be well-described by fitting the models to the SId plots over all regimes. For example, Sample A shows nearly quadratic variations as a function of the drain current Id with a slope close to 1.9 in regime I. This implies that the notion of the electrons drifting toward the drain side is appropriate for the CF model expected from Equation 1. That is, the LFN characteristics of Sample A are a result of carrier trapping and de-trapping by trap centers in regime I. The slopes obtained for all samples in regime I progressively vary from 2 to 1 with increasing Al evaporation time. Such a variation in slope indicates a rapid reduction in the density of traps inside the active region with Al evaporation. This in turn leads to change from CF in Equation 1 to MF in Equation 3. However, the current fluctuation of Sample B with a slope ~ 1.5 is described by almost equal proportions of both the CF and MF models. In regimes II and III, all SId values have a linear relationship (slope ~ 1) with Id. This relationship is in good agreement with the fitting results obtained from Equation 2 and 3. With the exception of a buried channel far away from the gate insulator, typical FETs allow for an effective channel region under the gate insulator interface that is associated with the trap density in the above threshold regime 24,25 . In our work, the ZnO TFTs permit the presence of an effective channel region placed close to the SiO 2 /ZnO interface due to the formation of thin ZnO active layer (< 10 nm). Therefore, the dominant LFN characteristics in regimes II and III can be described by the CF model. As expected from Equation 2 and the empirical results of Fig. 3, a reduction in the trap density upon Al evaporation corresponds to the LFN behaviors observed for Samples B, C and D.
To further illustrate the observed LFN behaviors, simple gate voltage-dependent band diagrams were constructed. Shown in Fig. 4a,b are band diagrams for Samples A and C, respectively, at two different gate bias regimes. The left and right sides of each figure represent the band diagrams when V g -V th < 0 and V g -V th > 0, respectively. (Where V th is the gate threshold voltage.) As displayed in Fig. 4a, Sample A contains a space charge region (SCR, denoted in blue) that is initially present during solution-processed growth. Large oxygen vacancies working as donors diffuse out toward the back channel surface (BCS) due to a surface potential energy generated by O 2− ions attracted from the exposed outside surface. This results in conduction band (E c ) and valence band (E v ) bending far from the Fermi level (EF), as depicted in Fig. 4a. The SCR includes weak Zn-O bonding as well as pore traps in the area close to depletion region, thereby establishing a narrow effective channel region under the SiO 2 gate layer due to the enhanced surface potential effect. Therefore, the LFN features of Sample A arise from weak Zn-O  Fig. 5. The RTN signals are commonly caused by specific carrier trapping and de-trapping processes that mainly occur at the channel oxide/SiO 2 interfaces, particularly in small-scale FET devices. The average carrier capture (referred to as a de-trapping process) or emission (referred to as a trapping process) fluctuations are well-described based on the Shockley-Read-Hall (SRH) theory 26 . The carrier emission and capture responsible for RTN take place through a tunneling process from or towards the channel region with specific time constants (τ c and τ e ) serving as typical RTN parameters. Here, τ c and τ e refer to the carrier capture and emission times, respectively. These times are determined by a separating algorithm from histograms extracted from time domain plots. According to the generally accepted SRH theory, the ratio between τ c and τ e is expressed as (a) Band diagrams of Sample A in the V g -V th < 0 (left) and V g -V th > 0 (right) regimes. In sample A, traps at the SCR (which is created by attracted O 2− ions) and SiO 2 /ZnO interface may contribute to the noise characteristics in all regimes. Thus, two trap energies (E t ) are placed at each trap center. (b) Band diagrams of Sample C in the V g -V th < 0 (left) and V g -V th > 0 (right) regimes. In Sample C, the noise behavior mainly arises from the SiO 2 /ZnO interface.
where E t is the trap energy and E F is the Fermi Level. Equation (4)  Shown in Fig. 5a,b are plots of the RTN responses for Samples A and C, respectively. Using a small active area (W = 10 μ m and L = 1 μ m), three gate voltages (V g -V th < 0, V g -V th ~ 0, and V g -V th < 0) in regime I were selected at V d = 10 V for the RTN measurements. In general, the value of I d in regime I was too high for observations of the RTN response. As evident in Fig. 5a,b, all time domain traces in the three V g regions show distinct high state (capture, τ c ) and low state (emission, τ e ) drain currents 26,27 . Displayed in Fig. 5c,d are the V g dependences of τ c (black squares) and τ e (red squares) extracted from histograms obtained from Fig. 5a,b, respectively. More detailed histograms for Samples A and C are given in Figure S3 of the Supporting Information. At first, the RTN of Sample C (Fig. 5d) can be described in two distinct gate voltage regions: the V g -V th < 0 regime, where τ c is a prime factor, and the V g -V th > 0 electron emission process in the V g -V th < 0 region, while Sample C displays two distinct gate voltage regions: a V g -V th < 0 region where electron capture is a prime factor, and a V g -V th > 0 region where electron emission is dominant. Sample C shows trends similar to those observed in typical MOSFETs.
Scientific RepoRts | 5:16123 | DOi: 10.1038/srep16123 region, where τ e is dominant. However, in Sample A (Fig. 5c) τ e was dominant in the V g -V th < 0 region and both τ c and τ e exhibited almost equal contributions in the V g -V th > 0 regime. That is, Sample C exhibits RTN behaviors that are typical of MOSFETs, while Sample A does not follow such trends. Therefore, it is believed that additional contributions to noise are present in Sample A (as described earlier). The ratios of τ c to τ e for Samples A and C as a function of gate voltage are shown in Fig. 5e,f, respectively. Positive and negative slopes were obtained for Samples A and C, respectively. Furthermore, x t for Sample C was about 5.9 nm from the SiO 2 /ZnO interface when the contribution to noise from the oxide/SiO 2 interface only is considered. However, x t for Sample A was about − 0.96 nm with respect to the SiO 2 /ZnO interface. According to Equation 5, such a result does not represent a typical trap depth. That is, a simple analysis of Sample A yields information on the presence of a relatively large number of trap centers in the SCR. These trap centers mainly contribute to the RTN features, as indicated from the LFN characteristics in Figs 2 and 3. In addition, the negative slope of Sample C supports the usual trend of noise sources created by interfaces only, while the presence of a positive slope for Sample A provides experimental evidence for additional noise contributions arising from the SCR. Therefore, we believe that the above experimental findings clearly demonstrate the role of Al evaporation in reducing trap centers that are initially present during the preparation of solution-processed ZnO TFTs.
In summary, the origins of LFN and time-domain RTN features were explored in solution-processed ZnO TFTs with or without Al evaporation on the back channel of the ZnO active layer. The noise associated with Sample A (no Al evaporation) arises from typical trap centers due to unstable bonding at the SiO 2 /ZnO interface as well as from weak Zn-O bonding and pore traps in a space charge region as a result of O 2− ion absorption into the back channel surface of the ZnO layer. The LFN of Sample A is mainly determined by carrier fluctuations in all regimes, and the RTN response shows more dominant carrier emission (trap) time counting than capture (de-trap) time counting in the threshold voltage regime. In contrast, the enhanced noise features of Sample C may be ascribed to a reduction in weak Zn-O bonding through Zn substitution by Al and a decrease in adsorbed oxygen via passivation in the Al-evaporated ZnO TFTs. The LFN of Sample C is a mobility noise in regime I, and the RTN characteristics exhibit behavior typically observed in conventional Si-MOSFETs, where only the interface noise contribution is considered. Thus, we anticipate that the ability to control and improve the noise features of TFTs by proper Al evaporation will lead to practical applications for solution-processed oxide semiconductors.

Measurement.
Transistor current-voltage characteristics were measured using a HP4156 semiconductor parameter analyzer. Forward and reverse gate biases of the hysteresis were analyzed between − 20 and 30 V (at V d = 10 V) in order to generate typical transfer curves. LFN and RTN tests were conducted in a shielded probe-station with the aid of an HP4156 semiconductor parameter analyzer, a BTA 9812B low noise amplifier (for LFN), a SR570 low noise amplifier (for RTN), and an HP 35670A dynamic signal analyzer.