Resistive switching in an amorphous ZnO dielectric film prepared on a Ga-doped ZnO transparent electrode

Abstract

The resistive switching behaviour of amorphous ZnO (a-ZnO) sandwiched between Ga-doped ZnO (GZO) transparent conductive oxide and Al electrode is reported. Transparent GZO films were deposited on polymer substrates as bottom electrodes using pulsed DC magnetron sputtering at 100 °C, on which a-ZnO films were deposited by RF magnetron sputtering at room temperature. The layered structure prepared in this manner was semi-transparent to visible light and its current–voltage hysteresis was representative of a bipolar resistive switching behaviour. The observation of such a resistive switching behaviour was attributed to the employment of a-ZnO as a dielectric layer and the use of Al and GZO as electrodes, which enabled the formation of Schottky barrier only at the a-ZnO/GZO interface. The conduction through the dielectric layer during the high resistance state was due to the Schottky emission as deduced from the consideration of band structures and the fitting of the current–voltage relations to the various conduction models. Switching to the low resistance state was attributed to the filament formation due to the migration of oxygen vacancies during the set process. In control experiments where crystalline ZnO was used as the dielectric layer, resistive switching behaviour was not observed.

---

Introduction

The material ZnO has many desirable properties suitable for electronic and optoelectronic applications due to its wide bandgap, adjustable electrical properties by doping with wide range of dopants, chemical stability and nontoxicity, morphological variety ranging from thin films to nanostructures combined with simplicity of processing, availability and low price of raw materials, etc.^1–3 While ZnO shares application fields with highly successful GaN materials, there are some applications where ZnO has advantages such as transparent thin film transistors, transparent oxide electrodes, photovoltaic cells, etc.^4,5 In addition to the excellent optoelectronic properties, it has been demonstrated that both doped and undoped ZnO films show resistive switching characteristics when they are sandwiched between metallic electrodes such as Pt, Au, Ag, etc. to form metal–insulator–metal (MIM) stacks.^5–9 The resistive switching behaviour of the MIM structure originates from the switching of the electrical resistances of the insulating (dielectric) ZnO layer between two distinctive states under specific bias conditions.

Meanwhile, recent advances in electronic device have prompted works on transparent (and optionally flexible) device structure which has potential application to wearable and portable computing devices.^1,10,11 Transparent devices require transparent circuitry including transparent memory cells and it would be desirable if the memory cells have simple layout and structure even if they are prepared from transparent electronic materials for higher transparency. ZnO would be suitable for such an application once the metallic electrodes in the MIM structure can be replaced with transparent electrodes. Further, it would be convenient from the processing viewpoint if the transparent electrodes and the dielectric layers are based on a single material system. There are ZnO-based transparent electrodes that have been suggested as ITO substitutes, viz. Ga-doped and Al-doped ZnO (GZO and AZO, respectively).^12,13 However, GZO and AZO films showing metallic conductivity have Fermi levels located above the conduction band edge of ZnO and therefore they cannot form Schottky junctions when brought into contact with crystalline ZnO (c-ZnO). This limitation is expected to be overcome if amorphous ZnO (a-ZnO) is used instead because a-ZnO has markedly different band compared with c-ZnO.^14,15

While a recent study demonstrated the possibility of the resistive switching behaviour of a-ZnO as prepared on the Pt/Ti/SiO₂/Si substrate,¹⁶ this study reports resistive switching behaviour of the a-ZnO dielectric film sandwiched between Al and GZO electrodes. The a-ZnO film was deposited on a GZO transparent conductive oxide film grown on a polymer substrate of polyethylene naphthalate (PEN). The bipolar switching was observed only when the dielectric layer was prepared from a-ZnO, which could be explained by the differences in the band structures of the a-ZnO and c-ZnO.

Experimental

Initially, 100 nm-thick GZO films were deposited on a transparent polymer substrate of polyethylene naphthalate film (Teonex® PEN, product of DuPont, 220 μm in thickness) as bottom electrodes using pulsed DC magnetron sputtering at 100 °C. A square-shape ZnO ceramic target (400 mm × 400 mm) containing 5.7 wt% of Ga₂O₃ was used as the source. Sputtering power was fixed at 2.4 kW with the pulse frequency of 50 kHz and the duty cycle of 72%, which gave the pulse factor of 14.4 μs. Before the deposition of the GZO films, the sputtering chamber was evacuated to the base pressure of about 3.0 × 10⁻⁴ Pa. Subsequently, Ar (5 N) gas was supplied into the chamber for the film deposition at the working pressure of 0.66 Pa. These GZO films showed the sheet resistance of 210 Ω sq⁻¹—with the carrier concentration of 1.60 × 10²⁰ cm⁻³ and the Hall mobility of 21.5 cm² V⁻¹ s⁻¹—as measured by van der Pauw Hall technique (Bridge Technology, Ecopia HMS-3000 Hall Measurement System). The optical transmittance was about 90% in the visible range as measured by UV-Vis spectrometry (JASCO, V-570) while the optical bandgap was estimated to be 4.2 eV from a Tauc plot. On the GZO side of the GZO/PEN substrate, a-ZnO dielectric layers were deposited by RF magnetron sputtering at room temperature using a 3-inch ZnO ceramic target as a source with the sputtering power of 180 W. Initial vacuum level of the sputtering chamber was 6.7 × 10⁻⁴ Pa while the working pressure was maintained at 0.67 Pa via the supply of Ar/O₂ (5 N) mixture having the oxygen content of 6% (Ar/O₂ = 9.4/0.6). This ZnO layer, it being amorphous confirmed by X-ray diffraction (XRD, X'Pert PRO MPD, PANalytical), was highly resistive with the sheet resistance of 60 kΩ sq⁻¹. Finally, the device structure was completed by adding 50 nm-thick Al electrodes in the forms of disk-like dots (200 μm in diameter) using electron-beam evaporation though a patterned shadow mask. The current–voltage (I–V) characteristics of the device were evaluated using a semiconductor parameter analyzer (Agilent, E5270B).

Results and discussion

Fig. 1(a) shows a schematic of the MIM structure composed of Al/a-ZnO/GZO. Also shown is the XRD pattern for the ZnO dielectric layer, which indicates that it is amorphous. Transparent GZO films were deposited on PEN substrates as bottom electrodes using pulsed DC magnetron sputtering at 100 °C, on which a-ZnO films were deposited by RF magnetron sputtering at room temperature. Al dots were deposited as top electrodes on a-ZnO/GZO. The MIM structure prepared in this manner was semi-transparent to the visible light as demonstrated in Fig. 1(b). Fig. 1(c) shows the I–V hysteresis of this structure when it is biased through the sequence as indicated in the figure with the arrows and sequential numbers. The positive bias was defined such that the Al top electrode was connected to the positive terminal. Initially, the forward bias (along the direction 1) caused small linear increase in the current up to about 2 V. This was followed by non-linear increase in the current up to the bias voltage of 4 V. At this point, an abrupt increase in the current to the compliance limit of 0.5 mA was recorded. Such an I–V behaviour indicates that the dielectric was initially in the high-resistance state (HRS) and then switched to the low-resistance state (LRS) at 4 V. The LRS was maintained with further increasing the voltage to 5 V. When the bias was reversed (along the direction 2), the current remained at the compliance limit of 0.5 mA until the voltage decreased to 1 V. From this point on, the current decreased linearly with reverse bias to −1 V (along the direction 3) where the current compliance limit was reached in the reverse direction. This compliance limit was maintained until the reverse bias reached −5 V. The linear I–V relation between 1 and −1 V indicates that the LRS is maintained under the reverse bias after the initial set process. When the bias was changed again to the forward direction (along the direction 4), the current began to decrease rather rapidly and non-linearly, without any sudden drop of the current, from the bias of −2 V to 0 V. During this sequence, the device returned to the HRS. The asymmetry of the I–V hysteresis can be better seen when it is drawn in semi-log plot as shown in Fig. 1(d). This I–V hysteresis is representative of a typical bipolar resistive switching behaviour with the set and reset voltages of 4 and −2 V, respectively.

Fig. 1 (a) Schematic showing the layout of the ZnO-based MIM stack. The XRD pattern in the inset indicates that the ZnO dielectric layer is amorphous. (b) Photograph demonstrating the optical transparency of the device. (c) Linear and (d) semi-log plots of the I–V hysteresis of the Al/a-ZnO/GZO device.

For the analysis of the I–V hysteresis, it was reconstructed in the double logarithmic plot for the initial forward bias (sequence 1) and the bias reversal (sequence 2) in Fig. 2(a). When the device was in the HRS, the log(I)–log(V) relation had the slope of 1.2 up to the bias voltage of 2 V implying ohmic-dominant conduction, which is typically observed in many MIM structures at low applied electric field regimes.¹⁷ From this point on, the slope of the log(I)–log(V) relation increased to 2.9 with further increasing the bias voltage. At the bias of 4 V, the slope was almost vertical corresponding to the switching to the LRS. Once the LRS was reached, the log(I)–log(V) relation exhibited the slope of 1.0 as the voltage decreased from 1 to −1 V implying that the conduction is completely ohmic. Contrary to the set process, the reset process was characterized by the gradual decrease in the slope of the log(I)–log(V) relation from 1.6 to 1.1 as the bias was once again reversed (sequence 4) by varying the voltage from −2 to 0 V, as shown in Fig. 2(b). This gradual change in the slope of the double logarithmic I–V relation, combined with small slopes close to 1, suggests that the final sequence of the I–V hysteresis is related to predominantly ohmic conduction.

Fig. 2 Double logarithmic plots of the I–V characteristic of the Al/a-ZnO/GZO device during (a) set and (b) reset processes.

The abrupt increase in the slope to 2.9 at 2 V, which is maintained to 4 V (Fig. 2(a)), indicates a change in conduction mechanism from ohmic to something else with increased electric field across the dielectric. For the MIM structures, several conduction mechanisms have been suggested. In order to find what model best describes the observed conduction behaviour for the applied bias section from 2 to 4 V, the I–V relations for this bias range have been fitted to various conduction models in the forms of proportionalities and the results are plotted in Fig. 3. In MIM structures, conduction can be controlled by the metal/insulator interface (electrode-limited conduction). Conduction models in this class include Schottky emission (Fig. 3(a)), Fowler–Nordheim (F–N) tunneling (Fig. 3(b)) and thermionic field emission (Fig. 3(c)).¹⁸ In other cases, conduction can be controlled by the dielectric layer (bulk-limited conduction). Poole–Frenkel (PF) emission (Fig. 3(d)), hopping (Fig. 3(e)), space-charge-limited current (SCLC) (Fig. 3(f)) and ohmic conduction belong to this class.¹⁸ If the electron traps in the insulator layer controls the conduction while the electronic mean free path is smaller than the layer thickness, the trap-limited type conduction mechanism would prevail.¹⁹ In such a case, the Schottky emission is modified such that the I–V behaviour resembles that of the PF emission (Fig. 3(d)).

Fig. 3 Fitting of the high slope region (slope = 2.9) of Fig. 2(a) to the various conduction models (represented as proportionalities appropriate for each model) for MIM structures: (a) Schottky emission, (b) Fowler–Nordheim tunneling, (c) thermionic field emission, (d) Poole–Frenkel emission (and also Schottky emission with trap-limited conduction), (e) hopping and (f) space-charge-limited current.

Since the I–V behaviours are all represented in proportionality forms, one can first ‘guess’ possible conduction mechanisms by comparing the coefficients of determination, R², of the linear fitting results. From the R² values in Fig. 3, Schottky emission, trap-limited Schottky emission, PF emission and hopping are chosen as potential candidates. It is now possible to further narrow down the candidates by inspecting the slopes of the proportionalities. For these possible four conduction models, the slopes and the estimated dielectric constant, ε_r, or mean hopping distance, a, are summarized in Table 1 for the fixed temperature (298 K) and insulator thickness (100 nm), together with the appropriate proportionalities and slope formulae. If the conduction by Schottky emission is assumed, the dielectric constant ε_r of the insulator layer is estimated to be 1.88 from the slope of the fitting to the ln(I) vs. relation. This value seems to be rather smaller than reported values of about 4 to 11 for c-ZnO.²⁰ However, it is consistent with the measured ε_r of 1.86 via capacitance–voltage relations shown in Fig. 4. It is believed that such a low values is due to the differences in the crystalline quality of the a-ZnO and c-ZnO, i.e. the amorphous has open structure with many structural defects and broken bonds. If the trap-limited Schottky emission is chosen instead, the ε_r is estimated to be 4.38 from the fitting to the ln(I/V) vs. relation which falls within the reported values range,²⁰ but it is different from the measured value. On the other hand, if the PF emission is considered, which is often suggested as the main conduction mechanism in the MIM stacks simply from the fitting I–V behaviours to the ln(I/V) vs. relation without considering the magnitude of slopes,²¹ the ε_r is estimated to be 17, an unrealistically high value even for c-ZnO. Finally, the mean hopping distance is calculated to be 2.55 nm from the slope of the ln(I) vs. V fitting corresponding to hopping conduction. This hopping distance, which represents the mean spacing between electron traps, is much smaller than the thickness of the insulator layer (100 nm) but is about twice larger than those reported for MIS or MIM structures having Pr₂O₃ and MgO as insulating layers elsewhere (in the range between 1 and 1.5 nm),^22,23 implying difficulty of conduction by this mechanism.

Table 1 Estimated physical properties of the a-ZnO dielectric layer from the fitting of the I–V behaviours of the MIM device to various conduction models. In the formulae for the slopes, e is the electron charge, k is the Boltzmann constant, T is the temperature, d is the thickness of the insulator, ε_r is the dielectric constant of the insulator, ε₀ is the permittivity of vacuum and a is the mean hopping distance. The slope was taken from the fitting results in Fig. 3 and the ε_r or a was estimated from the formulae for the slopes

Conduction model and proportionality	Slope	Estimated εr or a
Schottky emission:		εr = 1.88
Trap-limited Schottky emission:		εr = 4.38
Poole–Frenkel emission:		εr = 17.5
Hopping: ln(I) ∝ V		a = 2.55 nm

---

Fig. 4 Capacitance–voltage relations of the a-ZnO dielectric layer measured under the alternating current with the frequency of 1 MHz.

Through the estimation of the physical parameters using the slopes of the fitting line to the various conduction models, the possible conduction mechanisms of the MIM structure just before the switching to the LRS are now reduced to Schottky emission. If the Schottky emission is to operate in the MIM stack, there must be a Schottky barrier at the metal/insulator interface. For the formation of such a Schottky barrier, the work function Φ of the dielectric ZnO should be smaller than that of a metallic electrode because the ZnO dielectric layer showed n-type conductivity. At a glance, this condition cannot be met since the c-ZnO has larger work function than Al (Φ_Al = 4.3 eV) and GZO (Φ_GZO = 3.5 eV as estimated from the previously observed bandgap of the GZO electrodes).²⁴ At the same time, the electron affinity of c-ZnO, χ_c-ZnO, is the same as the Φ_Al and larger than Φ_GZO, as schematically illustrated in Fig. 5(a), eliminating the possibility of forming the potential barriers at the interfaces. The situation changes once the dielectric layer is changed to a-ZnO.¹⁴ First, the bandgap energy of a-ZnO is widened to 6.8 eV, much wider than its crystalline counterpart, while the locations of the valence band edge and Fermi level, E_F, are not much different from those of c-ZnO.²⁵ In addition, in a-ZnO, the bandgap (mobility gap) is not empty but filled with defect-related bands. There is a 2 eV-wide localized band related to the Zn3d⁹(4s4p)¹ states beginning from 1.6 eV above the valence band edge. The Zn3d⁹(4s4p)¹ states are deficient of electrons, i.e. they are positively charged, which are taken by the lattice oxygens to form a charge transfer band related to the O2p⁵ states (negatively charged). This O2p charge transfer band stretches between the Zn4s4p band and the conduction band with some overlap with the Zn4s4p band.

Fig. 5 Comparison of the possible energy band structures of the model resistive switching devices having (a) c-ZnO and (b) a-ZnO dielectric layers.

Now, the band structure of a-ZnO is considered together with those of the Al and GZO electrodes connected in series as shown in Fig. 5(b). In the HRS, the electrons near the E_F of the GZO electrode would face repulsive force from the negatively charged traps in the O2p charge transfer band of the a-ZnO under the forward bias condition. Further, there is a potential barrier to the bottom of the conduction band of a-ZnO (Φ_GZO − χ_a-ZnO = 2.9 eV). Therefore, electrons in the GZO can be injected into the a-ZnO layer for conduction only when there is sufficient electric field to reduce the height and thickness of the potential barrier, i.e. the conduction type would be Schottky emission. During the operation of the Schottky emission in the HRS, charged defects such as oxygen vacancies (V_O²⁺), which typically exist in ZnO as native defects,⁴ would migrate toward the GZO electrode (connected to the negative terminal). Eventually accumulated oxygen vacancies would form the conductive filament which bridges the Al and GZO electrodes to reach the LRS.^6,26,27

Once the bias is reversed in the LRS (Al electrode is connected to the negative terminal), electron conduction is initially possible through the conductive filament. However, as the Al electrode serve as oxygen reservoir, diffusion of oxygen from the Al electrode into the a-ZnO layer would gradually annihilate the oxygen vacancies near the Al/a-ZnO interface, resetting the device to the HRS.^6,26,28 Even after the conduction filament begins to rupture, however, electrons in the Al electrode can be injected into the electron-deficient (positively charged) Zn4s4p localized band with no barrier due to the location of the E_F of Al (Fig. 5(b)). When these electrons reach the a-ZnO/GZO interface, they face a weak potential barrier with the height only of 0.2 eV, which can be overcome easily under the reverse bias (GZO connected to the positive terminal). Because of this relative ease with which the electrons can flow under the reverse bias, there was no sudden drop of the current level while the conduction itself was dominantly ohmic type. In summary, the asymmetries in the barrier structures at the Al/a-ZnO and GZO/a-ZnO interfaces would result in the observed asymmetric nature of the I–V behaviour.^6,27,28

If the assertion so far were indeed the case, one may reason that the resistive switching behaviour should no longer be observed once the dielectric layer is replaced with c-ZnO, which typically shows n-type conductivity even without doping.⁴ Since then the contact on either side of the c-ZnO dielectric layer with metallic electrodes would become ohmic as deduced from the band structure shown in Fig. 5(a). In the present study, this possibility was verified by introducing two different ZnO dielectric layers with gradually improved crystalline quality, which were achieved by increasing the O₂ concentration in the sputtering atmosphere from 6% to 18%, in the MIM device structure. The resulting I–V hystereses are shown in Fig. 6(a) and (b) where the crystallinity of the dielectric layers is shown together as XRD patterns in the insets. When the dielectric layer was only partially crystalline (Fig. 6(a)), the resistive switching behaviour was still observed, but the set and reset voltages could not be clearly defined. As the crystallinity of the dielectric layer improved further (Fig. 6(b)), no switching behaviour was observed while the I–V behaviour became completely ohmic (slope of the log(I)–log(V) relation in the inset is 1.0) although this film still retained some portion of the amorphous phase. This ohmic conduction is understood recalling that χ_c-ZnO is the same as the Φ_Al and larger than Φ_GZO (Fig. 5(a)) which allows easy two-way transfer of carrier electrons to the dielectric layer from both the electrodes.

Fig. 6 (a and b) Resistive switching behaviours of the devices with varying crystallinity of the dielectric layers. The XRD patterns in the insets show the evolution of the crystallinity of the dielectric layers from amorphous to crystalline with varying Ar/O₂ flow ratios. (c) Comparison of the optical absorption spectra of the ZnO dielectric films with varying Ar/O flow ratios and (d) accompanying Tauc plots for the estimation of mobility gaps (or bandgaps).

The changes in the band structure with improved crystallinity could be inferred from the observed changes in the optical absorption characteristics of the dielectric films. Fig. 6(c) and (d) show the absorption spectra of the dielectric films as functions of the photon energy of the incident light, and the corresponding Tauc plots, respectively. There are absorption edges near 5 eV and all the spectra are characterized by large tails. These absorption tails originate from the impurity states in the bandgap,²⁹ which would correspond to the Zn- and O-related defect bands in a-ZnO. In addition, the films showing crystalline characteristics in the XRD patterns have secondary absorption edges at about 3.3 eV that are more pronounced with improving crystallinity. This would be related to the band to band transition of electrons in the c-ZnO phases. The mobility gap estimated from the intercepts of the tangent lines with the photon energy axis in the Tauc plot is 5.3 eV for the a-ZnO film deposited with Ar/O₂ ratio of 9.4/0.6. The gap narrowed to 5.1 and then 4.9 eV when Ar/O₂ ratios decreased to 8.8/1.2 and 8.2/1.8, respectively, reflecting increasing portion of the crystalline phase.¹⁵ Further, the additional band gaps derived from the secondary absorption edges are estimated to be 3.28 eV corresponding to that of c-ZnO.

Conclusions

It was possible to observe a bipolar resistive switching behaviour in the MIM structure where one of the electrodes was prepared from GZO transparent conductive oxide at 100 °C and the insulating layer was introduced by depositing a-ZnO at room temperature. Fitting of the observed current–voltage relations to various conduction models and consideration of the band structures of the component materials suggested that the conduction through the a-ZnO dielectric layer before the set process is related to the Schottky emission. The use of Al as one of the electrodes resulted in the asymmetry in the barrier structures at the electrode/dielectric interfaces and this was suggested as the reason for the observed asymmetry in the conduction behaviour under forward and reverse biases. Switching to the low resistance state from the high resistance state was attributed to the filament formation due to the migration of oxygen vacancies by the electric field during the set process. The conclusions were supported by the gradual disappearance of the resistive switching behaviour with improved crystalline quality of the dielectric layers achieved by increased oxygen concentrations in the working atmosphere during their preparations. At this stage, it was not possible to observe stable endurance and retention capabilities of the MIM device, but once this issue is cleared, transparent resistive switching device based on ZnO would become a possibility.

Acknowledgements

This research was financially supported by Changwon National University in 2015–2016.

Notes and references

1. J. F. Wager, D. A. Keszler and R. E. Presley, in Transparent Electronics, Springer, New York, 2008.
2. A. Janotti and C. G. Van de Walle, Rep. Prog. Phys., 2009, 72, 126501.
3. T. C. Collins and R. J. Hausenstein, Fundamental Properties of ZnO, in ZnO Materials for Electronic and Optoelectronic Device Applications, ed. C. W. Litton, D. C. Reynolds and T. C. Collins, Wiley, New York, 2011.
4. Ü. Özgür, D. Hofstetter and H. Morkoç, Proc. IEEE, 2010, 98, 1255.
5. W.-Y. Chang, Y.-C. Lai, T.-B. Wu, S.-F. Wang, F. Chen and M.-J. Tsai, Appl. Phys. Lett., 2008, 92, 022110.
6. N. Xu, L. Liu, X. Sun, X. Liu, D. Han, Y. Wang, R. Han, J. Kang and B. Yu, Appl. Phys. Lett., 2008, 92, 232112.
7. X. L. Wang, Q. Shao, C. W. Leung, R. Lortz and A. Ruotolo, Appl. Phys. Lett., 2014, 104, 062409.
8. F. M. Simanjuntak, O. K. Prasad, D. Panda, C.-A. Lin, T.-L. Tsai, K.-H. Wei and T.-Y. Tseng, Appl. Phys. Lett., 2016, 108, 183506.
9. Y. C. Yang, F. Pan, Q. Liu, M. Liu and F. Zeng, Nano Lett., 2009, 9, 1636.
10. J. F. Wager, Science, 2003, 300, 1245.
11. D. J. Rogers and F. Hosseini Teherani, ZnO: From Transparent Conducting Oxide to Transparent Electronics, in Encyclopedia of Materials: Science and Technology, ed. K. H. Jürgen Buschow, R. W. Cahn, Me. C. Flemings and B. Ilschner, Elsevier, Amsterdam, 2nd edn, 2010.
12. L. Ellmer and A. Klein, ZnO and Its Applications, in Transparent Conductive Zinc Oxide, ed. K. Ellmer, A. Klein and B. Rech, Springer, New York, 2008.
13. K. Ellmer, Nat. Photonics, 2012, 6, 809.
14. D. Schmeißer, J. Haeberle, P. Barquinha, D. Gaspar, L. Pereira, R. Martins and E. Fortunato, Phys. Status Solidi C, 2014, 11, 1476.
15. W. H. Huang, S. J. Sun, J. W. Chiou, H. Chou, T. S. Chan, H.-J. Lin, K. Kumar and J.-H. Guo, J. Appl. Phys., 2011, 110, 103705.
16. Y. Huang, Z. Shen, Y. Wu, X. Wang, S. Zhang, X. Shi and H. Zeng, RSC Adv., 2016, 6, 17867.
17. F.-C. Chiu, P.-W. Li and W.-Y. Chang, Nanoscale Res. Lett., 2012, 7, 178.
18. C. Hamann, H. Burghardt and T. Frauenheim, in Electrical Conduction Mechanisms in Solids, VCH Verlag, Berlin, 1990.
19. J. G. Simmons, Phys. Rev. Lett., 1965, 15, 967.
20. O. Madelung, U. Rössler and M. Schulz, in II–VI and I–VII Compounds; Semimagnetic Compounds, Springer, Berlin, 1999.
21. H. Schroeder, J. Appl. Phys., 2015, 117, 215103.
22. F.-C. Chiu, C.-Y. Lee and T.-M. Pan, J. Appl. Phys., 2009, 105, 074103.
23. F.-C. Chiu, W.-C. Shih and J.-J. Feng, J. Appl. Phys., 2012, 111, 094104.
24. K.-J. Ahn, S. Lee, W.-J. Kim, G. Y. Yeom and W. Lee, Mater. Sci. Semicond. Process., 2013, 16, 1957.
25. T. C. Yeha, Q. Zhu, D. B. Buchholz, A. B. Martinson, R. P. H. Chang and T. O. Mason, Appl. Surf. Sci., 2015, 330, 405.
26. T. Gu, J. Appl. Phys., 2014, 115, 203707.
27. L. Shi, D. Shang, J. Sun and B. Shen, Appl. Phys. Express, 2009, 2, 101602.
28. W.-Y. Chang, C.-S. Peng, C.-H. Lin, J.-M. Tsai, F.-C. Chiu and Y.-L. Chueh, J. Electrochem. Soc., 2011, 158, H872.
29. J. I. Pankove, in Optical Processes in Semiconductors, Dover Publications, New York, 1975.

---