Introduction

Oxide semiconductors shows great promise as a high mobility channel layer in thin film transistors (TFT) fabricated at low or even room temperatures1,2. A classical oxide material is ZnO, which has mostly a poly-crystalline (pc) structure, i.e. pc-ZnO, suggesting the presence of potential barriers at the grain boundaries, which lowers the mobility (see Fig. 1a)3,4. To suppress these potential barriers, metal cations, such as Ga and In, were incorporated into the binary system, thereby stabilizing the amorphous phase over the crystalline phase. But the addition of Ga and In introduces compositional disorder and these cation states create potential fluctuations in the conduction band minima (EC)5,6. Although this tends to reduce the overall electron mobility as shown in Fig. 1b, the approach has been successful due to the low temperature processability and high mobility in the amorphous phase1,7.

Figure 1
figure 1

Schematic profiles of the conduction band minima (EC) in (a) poly-crystalline ZnO, (b) amorphous InGaZnO and (c) amorphous ZnON, respectively.

Here, ‘e’ denotes a free electron released into conduction band. In (a), φB is the potential barrier height at the grain boundary in poly-crystalline ZnO. In (b), φB is the potential barrier height due to compositional disorder in amorphous InGaZnO. (d) Schematic diagram to describe trap-limited conduction associated with the localized tail states. Here, nfree and ntrap denote free and trapped carrier densities at band tail states, respectively.

Full size image

As another approach for high mobility and optical stability in oxide semiconductors, anions, instead of cations, can be added8. For example, flourine (F) can be used for n-type doping of SnO2. However, F does not affect electron mobility as it is an anion, which puts disorder into the valence band not the conduction band9. As another example, when nitrogen (N) is incorporated into ZnO, this does not cause p-type doping but it forms an amorphous alloy, where the anion site disorder stabilizes the amorphous phase, making amorphous ZnON (a-ZnON). Again, this does not lower the electron mobility because the disorder is in valence band states and not the conduction band states. Another important effect is that the N 2p orbital lies higher than the O 2p orbital, so that the new N 2p states raise the valence band maxima (VBM) above its energy in ZnO9. It is known that the photo-induced instabilities of ZnO and IGZO are related to states due to O vacancies and interstitials that lie in the lower band gap region. These states are now covered up by the higher VBM states, so they can no longer give rise to such instabilities and persistent photoconductivity (PPC)10,11, which is a strong benefit for the N containing materials.

According to the recent literature11, the electron effective mass of a-ZnON (~0.19 m0) can be larger than that of crystalline ZnN (i.e. c-ZnN) and smaller compared to ZnO films. This may be explained with a disorder of a-ZnON, especially in amorphous phase. Indeed, the conduction band minima (CBM) of a-ZnON is also composed of both Zn 4 s and N 2p, similar to c-ZnN12,13. Here, N 2p is sensitive to bonding angle tilt in amorphous phase, thus high disorder in a-ZnON, although Zn 4 s is spherical and less sensitive to bonding angle disorder. This can yield a high density of localized tail states near the CBM in a-ZnON (see Fig. 1c). For a TFT with a-ZnON channel, the field-effect mobility is strongly affected by the presence of the localized tail states. This can be explained with trap-limited conduction theory (i.e. multiple trapping and thermal release events) (see Fig. 1c,d)5,14. One of the ways to reduce the localized tail state density is by thermal annealing, resulting in higher mobility (up to 110 cm2/V-s) reported in Ref. 8.

In this paper, we extract the density of localized tail states in a-ZnON TFTs, using current-voltage characteristics of the device. The extracted values of tail state density at the CBM (Ntc) and characteristic energy (kTt) are about 2 × 1020 cm−3eV−1 and 29 meV, respectively, thus kTt > kT (i.e. thermal energy) at T = 300K. This implies that trap-limited conduction is dominant at room temperature. In addition, it is found that the extracted field-effect mobility and its gate voltage-dependence are strongly dependent on the pre-factor 1/(α + 1), where α = 2(kTt/kT-1) for kTt > kT. Here, we derived a more accurate field-effect mobility expression, using the proposed self-consistent extraction method. Also, a weak persistent photoconductivity in a-ZnON TFTs with Mo electrode is observed under visible light illumination, suggesting ionization of defects, such as oxygen and/or nitrogen vacancies, located near the VBM. To check the effect of electrode on leakage current and PPC, we replaced the Mo with Cr for electrodes. It is found that the leakage current slightly increases, but the effect of illumination is maintained, suggesting the PPC is arising from a change in the intrinsic property of the channel layer.

Results and Discussion

Localized Tail States

Since the current-voltage (I-V) characteristics of the TFT are largely determined by the density of localized states, e.g. tail states in the channel layer, the density of localized states can be retrieved from the measured terminal characteristics15,16,17. As a first step, the free carrier density (nfree) is extracted from the measured I-V characteristics. Note that linear characteristics of the drain current vs. gate voltage (IDS-VGS) are required rather than saturation regime characteristics16,17,

Here, μband is the band mobility of the ZnON layer (note that this is a constant and the main unknown), εS the permittivity of a-ZnON (which is about 11ε0, where ε0 is vacuum permittivity), kT the thermal energy, W the channel width, L the channel length and VDS the drain voltage. Also, the carrier density of the free carriers (nfree) and trapped carriers (ntrap) at tail states can be derived with Poisson’s Equation along with Gauss’s law as follows,

where x is the distance from the front channel interface along the channel depth and φ the channel potential along x. Based on Equations (2) & (3), the carrier densities can be connected to gate voltage using a charge balance Equation: E(x = 0) = Cox(VGS − VTe). Here, Cox is the gate-insulator capacitance and VTe is an effective threshold voltage which can be represented as [1/VGS + 1/VT]−1 for VGS > 0 and VT > 0, where VT is a threshold voltage to be extracted with a linear extrapolation at a linear regime. This yields the following,

Squaring Equation (4) and taking its first derivative with respect to the surface potential φ(x = 0) = φS, ntrap can be obtained as,

Based on Equations (1), (4) and (5), the density of tail states (Ntail(E)) can be given as the first derivative of Equation (5),

Here, qφS = EF − EF0 where EF0 is the Fermi level at flat band (see the Supplementary Information for more detailed derivation procedure).

Based on the above extraction method, the density of localized tail states was retrieved using a-ZnON TFTs with W = 50 μm and L = 10 μm. Here, the linear IDS − VGS characteristics were measured at a small VDS (e.g. 0.01 V and 0.1 V), as seen in Fig. 2(a,b) shows output characteristics for different VGS (see also Figure S2 in the Supplementary Information). The basic parameter to be extracted from each I-V curve is VT. The VT values of these two cases (VDS = 0.01 V and 0.1 V) are 4.4 V and 4.3 V, respectively. The other unknown parameters, μband and Cox are 110 cm2/V-s and 19 nF/cm2, respectively8. Here, μband can be the maximum achievable mobility and the Cox value was calculated using a thickness of 300 nm and measured permittivity of ~6.5ε0 for the gate insulator Si3N4. Using these parameters and Equations (1, 2, 3, 4, 5, 6), the carrier densities (nfree and ntrap) and density of tail states (Ntail(E)) were extracted as seen in Fig. 3a,b, respectively. As indicated in Fig. 3b, the tail states (i.e. gap states near the EC) can be approximated as an exponential distribution14,15,16,

Here, Ntc is the tail state density at E = EC (i.e. conduction band minima) and kTt is the characteristic energy of the tail state. Applying Equation (7) into the plot shown in Fig. 3b, Ntc and kTt values were extracted as 2 × 1020 cm−3eV−1 and 29 meV, respectively.

Figure 2
figure 2

(a) Linear IDS-VGS characteristics measured at a small VDS (0.01 V and 0.1 V).

Inset: Schematic 3-D view of the fabricated a-ZnON TFTs along with this micro-photo. Here, the gate insulator is Si3N4, stacked layer of SiO2 and Si3N4 is used as Passivation and Mo and Cr are used for source, drain and gate terminals in two different devices, respectively. (b) Measured output characteristics, i.e. IDS-VDS, of the examined TFTs for different VGS.

Full size image
Figure 3
figure 3

(a) Extracted carrier densities (nfree and ntrap) and (b) density of tail states denoted as Ntail(E).

The extracted Ntail(E) for different VDS is similar to each other, suggesting that the extraction method is almost independent on VDS when VDS is small enough. This also implies that contact resistance effects are negligible at small VDS, e.g. 0.1 V and 0.01 V.

Full size image

Electron Mobility

To incorporate the effect of tail states into the field effect mobility (μFE), the trap-limited conduction theory is employed,

Here, nfree (with Boltzmann’s approximation for |EF − EC| > kT) and ntrap (for kTt > kT) are represented analytically as a function of Fermi level (EF)16,17, respectively, as follows,

where NC is an effective free carrier density [cm−3]. Note that the NC value of ZnON is about 2 × 1018 cm−3, which is calculated with an electron effective mass ~0.2 m010,11, where m0 is the electronic rest mass. With Equations (9) & (10), Equation (8) can be rewritten as a function of EF, assuming nfree < ntrap,

Here, EF − EC (=EF0 + qφS − EC) can be given as a solution of Equation (4), replacing VTe by VT just for the above-threshold regime,

Using Equations (11) & (12), Equation (11) can be represented as a function of VGS, following a power law,

where α = 2(kTt/kT − 1). Note that ξ is a prefactor independent on VGS. As seen in Equation (13), the field-effect mobility (μFE) is a function of VGS and linked with localized tail states in terms of the exponent (α) and constant (ξ) in the power law. Along with Equation (13), the current-voltage relation can be derived based on a drift transport equation and approximated with the condition, VGS − VT  ≫ VDS (i.e. linear approximation), as follows,

Due to the presence of the exponent (α) and gate-voltage dependence in the μFE expression seen in Equation (13), the first derivative of Equation (15) with respect to VGS (i.e. transconductance (gm)) is given as follows,

Note that the first term of Equation (16) cannot be zero since μFE is a function of VGS. So, the conventional way to get μFE seems to be insufficient and inconsistent. In Equation (16), the first derivative of μFE is shown and can be expanded with Equation (13), as follows,

With Equation (17), Equation (16) can be rewritten as follows,

From Equation (18), μFE is now given as follows,

As can be seen in Equation (19), due to the presence of α associated with localized tail states, the mobility can be reduced by the ratio of 1/(α + 1). Here, we defined the parameter ‘1/(α + 1)’ as the mobility scaling factor due to trapping events in the localized tail states. So, the conventional way yields an over-estimated value and hence an inconsistency. So, we believe that Equation (19) provides a more accurate μFE(VGS) while capturing the effects of localized tail states with the parameter α. Using Equation (19), μFE(VGS) was extracted using the retrieved value of kTt seen in Table 1. In addition, the following equation can also be defined to explain the portion of trapping (ψ) as,

Table 1 Summary of Extracted Parameters for a different VDS.
Full size table

This can be used as a measure of how trapping significantly affects the electron mobility. As shown in Fig. 4(a), μFE (at VGS = 9 V) is about 30.2 cm2/V-s for VDS = 0.01 V which is reduced by 21% (i.e. ψ = 0.21) compared to the conventional extraction route where μFE is considered as a constant (38.4 cm2/V-s without considering α), i.e. μgm. Note that we also tested the proposed mobility extraction method for VDS = 0.1 V. It is found that there is a discrepancy less than 1% compared to the case of VDS = 0.01 V, as summarized in Table 1. Here, we believe that the case of VDS = 0.01 V provides higher accuracy compared to VDS = 0.1 V since a smaller VDS is always better to satisfy the assumption that VGS − VT ≫ VDS, relating to Equation (15).

Figure 4
figure 4

(a) Field effect mobility (i.e. gate voltage dependent mobility) as a function of VGS for different VDS: 0.1 V, 0.01 V.

Here, μFE is shown from using Equation (19) in comparison with μgm calculated by the conventional way with transconductance (gm). (b) IDS-VGS characteristics measured after applying 20 V stress for each stress period (tS). Here, the gate voltage applied (VStress) is 20 V (Inset: retrieved ∆VT vs. tS).

Full size image

The band tail states not only affect the electron mobility but also affect bias instability. Indeed, it is known that the density of band tail states determines the rate of instability creation, e.g. threshold voltage shift (ΔVT). The ΔVT as a function of time (t) is defined by the following relation18,

Here, VStress is the bias stress, VT0 the pre-stress threshold voltage (~4.3 V) and t0 the characteristic time constant. In particular, β is a power-law exponent, which is proportional to the tail state density (Ntc, cm−3eV−1), So the following relation can be deduced,

where ξ is a constant in the dimension of cm3eV. To extract the values of β and t0, the I-V characteristics were measured after applying 20 V bias stress for each stress period, as seen in Fig. 4b. Based on this, the ΔVT as a function of time is retrieved, as seen in the inset of Fig. 4b, yielding β ~ 0.34 and t0 ~ 108 sec. Here, it is found that the value of β is smaller compared to a-Si TFTs ~ 0.418. As can be seen in Equation (22), this can be explained with a smaller Ntc of a-ZnON TFTs ~ 2 × 1020 cm−3/eV compared to a-Si TFTs ~ 1022 cm−3/eV19.

Persistent Photoconductivity

Additionally, we performed computations of the density of states (DOS) in a-ZnON. Figure 5 shows the computed total DOS and projected DOS (PDOS) onto Zn, O and N atoms for a-ZnON. Note that the fabricated a-ZnON film composition has been measured with the Rutherford backscattering spectrometry (RBS) (see the Supplementary Information). As shown in Fig. 5a, it is clear that the N forms the VBM and is located from −1.8 eV to 0 eV. This implies that oxygen defects (e.g. vacancies and interstitials), viewed as the origin of the PPC, are filled with N. Also, the band-gap is about 0.75 eV, as indicated in Fig. 5, suggesting that photon energies in the range from 0.75 eV ~ 2.55 eV may not give rise to the PPC in a-ZnON. To check this, we measured the drain current as a function of time (i.e. IDS-time plot) when VGS = 0 V and VDS = 0.01 V under illumination with 550 nm wavelength light (equivalent to 2.25 eV photon energy, hv), as seen in Fig. 6a. Here, the maximum optical power P0 is ~0.1 mW/cm2. As shown in Fig. 6b, the drain current (~2.3 × 10−10 A) under P0 is increased by 2 times more than that (~1.2 × 10−10 A) under 0.5P0. This suggests that the drain current under illumination is associated with excess carrier generation, which is linearly proportional to incident optical power, rather than electron trapping into gate insulator. After removal of illumination (after t = 80 s), the drain current is almost recovered, thus the examined a-ZnON is optically very stable. However, we find a small difference (i.e. IPPC ~ 3.4 × 10−11 A) between the initial (IDS ~ 9.1 × 10−11 A) and final (IDS ~ 1.25 × 10−10 A) stages. Thus, persistent photoconductivity (PPC) still exists, albeit mild as seen in Fig. 6b, implying that some of the excess electrons are generated from optically irreversible states. So, the PPC may be associated with some of the unfilled oxygen defects (OD). Besides oxygen-related defects, we also consider nitrogen-related defects (ND), such as nitrogen vacancies (NV), which exist especially at the vicinity of the valence band maxima20,21,22. Under illumination, they can also be ionized as NV0 → NVn+ + ne–, where n is an integer, e.g. 1 ~ 320,22. Since these constitute negative U-defects, the ionization process is irreversible even after removal of light, thus giving rise to the mild PPC22. These defects, collectively denoted as D0, can be ionized under illumination (D0 → Dn+ + ne−), relocating to the vicinity of the EC10,11,20,21,22,23,24,25. This maintains the increased Fermi level (EF) even after illumination, as described in Fig. 6c. Since we have a very mild PPC (see Fig. 6b), we should assume that the density of defects, including oxygen and nitrogen vacancies, is small, implying that the photocurrent under illumination mostly consists of electrons generated from the N 2p states in the valence band. In order to estimate the number of the ionized deficiency defects (Nidd, cm−2), we may use the following relation,

Figure 5
figure 5

(a) Computed total electronic density of states (DOS) and projected DOS (PDOS) for the Zn, O and N atoms in a-ZnON.

The VBM energy level is set to 0 eV and retrieved band-gap energy (Eg) is about 0.75 eV as indicated. (b) The model unit cell of a-ZnON obtained from the melt-quench simulations.

Full size image
Figure 6
figure 6

(a) Schematic 3-D view of the transistor examined with the light illumination and DC bias on VDS.

(b) Measured drain current (IDS) of the examined device with Mo electrode as a function of time. Here, light pulse is applied as described in the inset. (c) Possible density of states picture with respect to optical carrier generation mechanisms and PPC. Here, the generated electrons are denoted as ‘e’, EF and E’F denote Fermi levels before and after illumination, respectively and EV is the valence band maxima. (d) Schematic 3-D band diagram along the channel depth. Here, we show the effect of the positive gate bias on the recombination of ionized deficiency defects with induced electrons.

Full size image

Using the values of the parameters extracted in the previous sections (e.g. IPPC = 3.4 × 10−11 A, μFE = 30.2 cm2/V-s, W/L = 5, VDS = 0.01 V), the Nidd is retrieved as 1.43 × 108 cm−2. To electrically remove these ionized defects, we employed a positive gate-pulse scheme25, as seen in Fig. 6b. This yields a fully recovered current level (IDS ~ 9.2 × 10−11 A) which is similar to the current level before illumination. Hence, it is suggested that these ionized defects (Dn+) were eliminated with the forced recombination with the electrons (ne−) induced during the positive gate-pulse width (+ 10 V), i.e. Dn++ne− → D0, as described in Fig. 6d. Here, the number of induced electrons (i.e. Ne ≈ Cox × (10 − VT)/q ≈ 6.7 × 1011 cm−2) are much more than the Nidd ~ 1.43 × 108 cm−2, thus it is enough for a full recombination.

In addition, we replaced Mo with Cr for electrodes to check effects of metal on leakage current and PPC. Figure 7a shows the measured IDS vs. time for two devices with different metal electrodes, e.g. Mo and Cr, respectively. It is found that there is a small current difference before and after illumination for each case. And the difference (i.e. IPPC) is almost the same as 0.034 nA. This implies that the choice of electrode metal doesn’t affect the PPC. And the current difference between the Mo and Cr cases (ΔIDS) is shown on the right-hand-side y-axis. It is found that this current difference is always approximately 0.1 nA, suggesting that the leakage current is changed globally regardless of illumination and PPC. This can be explained with the reduced barrier height at source side (qϕb) due to a smaller work-function of Cr (~4.5 eV) compared to Mo (~4.6 eV), as shown in Fig. 7b,c. These results indicate that the choice of metal for electrodes does not affect the PPC.

Figure 7
figure 7

(a) Measured IDS vs. time for two devices with different metal electrodes, e.g. Mo and Cr, respectively.

Also, the current difference is shown on y-axis on the right hand side. Horizontal band diagrams for the cases of (b) Mo and (c) Cr electrodes, respectively.

Full size image

Conclusions

In conclusion, the density of localized tail states in ZnON thin film transistors (TFTs) has been extracted using current-voltage characteristics of the TFTs. The extracted values of Ntc and kTt are about 2 × 1020 cm−3eV−1 and 29 meV, respectively. Considering trap-limited conduction theory, the field-effect mobility expression has been derived and shown to be represented in terms of tail state parameters. In particular, the exponent (α) has been strongly connected to the mobility through kTt which is a key measure of the degree of the disorder of the channel layer. This suggests that a reduction of kTt is needed to achieve higher mobility. Additionally, it has been revealed that the examined ZnON is optically very stable showing only weak PPC which is thought to be arising from ionization of defects, such as oxygen and/or nitrogen vacancies, located in vicinity of the valence band maxima. This happens equivalently in both ZnON TFTs with Mo and Cu electrodes, suggesting the PPC is associated with a change of the intrinsic property of the channel.

Additional Information

How to cite this article: Lee, S. et al. Localized Tail States and Electron Mobility in Amorphous ZnON Thin Film Transistors. Sci. Rep. 5, 13467; doi: 10.1038/srep13467 (2015).