Atomic Layer Deposition of AlGaN on GaN and Current Transport Mechanism in AlGaN/GaN Schottky Diodes

Abstract

The current transport mechanism of AlGaN/GaN Schottky diodes prepared by atomic layer deposition (ALD) was explored using current–voltage (I–V) and capacitance–voltage (C–V) measurements. The Schottky barrier height decreased and the ideality factor increased with increasing the temperature. Poole-Frenkel mechanism was found to rule the forward current conduction, involving the dislocation-related trap states in AlGaN layer. The activation energy of traps was estimated to be about 0.6 eV under high reverse bias, related with trap-assisted tunneling. Frequency dispersion in the C–V data was not significant. C–V hysteresis measurements with the sequential scans with increasing the maximum voltage in accumulation showed the increase in the flatband voltage shift, which was associated with the charge trapping occurring in the interfacial oxide layer near the AlGaN/GaN interface. This work suggests that the suppression of Ga–O formation during the initial ALD process is a critical factor to improve the device performance.

Fig. 6 (a) Schematic diagram of the conduction band offset in the AlGaN/GaN structure with the PF emission barrier at forward bias and (b) schematic band diagram at reverse bias.

1. Introduction

A high-performance AlGaN-based ultraviolet (UV) light-emitting diodes (LEDs) has also been successfully fabricated.^1,2) In addition, Al(Ga)N/GaN junctions have been considered significantly for electronic devices such as high electron mobility transistors (HEMTs) due to the formation of a two dimensional electron gas (2DEG).^3–5) In such devices, performance and reliability can be improved by using AlGaN with a large Schottky barrier height (SBH), which can decrease the leakage current and increase the breakdown voltage.⁶⁾ For high speed operation in scaled down devices, high-Al-composition barrier is necessary to obtain high power levels.⁷⁾ However, such devices suffer from high gate leakage current due to a strong polarization electric field and high surface state density, hindering the device performance and reliability of GaN-based HEMTs.⁸⁾

Much research has been performed to explain the current transport mechanisms for gate leakage current. It was shown that the reverse leakage in AlGaN/GaN can be analyzed using a Poole–Frenkel (PF) emission mechanism with a trap-assisted process.⁹⁾ The investigation of CF₄ plasma-treated AlGaN/GaN Schottky barrier diodes has shown that the reverse leakage current increased exponentially with increasing temperature, indicating that a thermally activated transport mechanism is involved.¹⁰⁾ The temperature-dependent current transport mechanism in Al₂O₃/Al_0.55Ga_0.45/GaN structures was shown that the reverse PF emission current in low-field, mid-field, and high-field region were related to trap states with activation energy of 0.41, 0.49, and 0.71 eV, respectively, and the forward PF emission current was related to traps with activation energy of 0.65 eV.¹¹⁾ Dependence of gate leakage current on Al mole fraction of AlGaN/GaN HEMTs has shown that the reverse leakage current is mostly dominated by PF emission at higher mole fractions, due to higher electric field across the barrier.¹²⁾

The atomic layer deposition (ALD) growth of AlN has been investigated intensively.^13–16) In contrast, the growth of Al_xGa_1−xN by ALD has rarely been investigated.¹⁷⁾ From both technological and scientific points of view, material properties of Al_xGa_1−xN alloys by ALD should be understood thoroughly, which is fundamentally important to the design of practical devices based on these materials. In this respect, we prepared atomic layer deposited AlGaN on GaN substrate and investigated the current transport mechanism of AlGaN/GaN Schottky diodes.

2. Experimental

AlGaN thin films were deposited on c-plane (0001) GaN substrate (thickness: 325 µm) using a thermal ALD system (Atomic classic, CN-1, Korea) after cleaning process in a HCl:H₂O (1:1) solution. The temperature was ramped up to 335°C to deposit about 5 nm thick AlGaN layers using trimethylaluminum (TMA), triethylgallium (TEG), and NH₃ as precursors. The sequence of ALD reaction for AlN (GaN) film consisted of TMA (TEG) feeding - N₂ purge - NH₃ feeding - N₂ purge. Note that the growth of solely GaN film was not possible by thermal ALD due to the limited reaction. Hence, incorporation of GaN for AlGaN film was enabled by combining ALD reactions of GaN and AlN film at a ratio of 1:1. The analysis using Auger electron spectroscopy revealed that the composition ratio between Al and Ga was about 34:6, corresponding to Al_0.85Ga_0.15N. AlGaN/GaN Schottky diodes were fabricated with a Pt front contact (diameter: 500 µm, thickness: 50 nm) and an Al back contact (thickness: 100 nm) to investigate the electrical properties. Temperature dependent current–voltage (I–V–T) measurements were carried out with a Keithley 238 current source after placing samples on a hot chuck connected with a temperature controller and capacitance–voltage (C–V) measurements were performed using a HP 4284A LCR meter with an AC voltage of 30 mV.

3. Results and Discussion

Figure 1 shows the atomic force microscopy (AFM) surface images for the GaN substrate and AlGaN layer. The root-mean-square roughness (rms) values measured over 5 µm × 5 µm scan areas were measured to be 3.32 and 1.22 nm, respectively, for the GaN substrate and AlGaN layer. As shown in Fig. 1(c), pits exceeding 10 nm in depth were observed for the GaN substrate, which could affect the uniformity of overgrown AlGaN layer. Although the rms value was decreased for the AlGaN layer, still about 8 nm deep pits were observed. Hence, these features might create a nonuiform interfacial layer after metal contact formation, affecting the current transport. Figure 2 shows the I–V curves measured at various temperatures. The increase of current values with increasing the temperature is observed. Based on the thermionic emission (TE) model,¹⁸⁾ the temperature dependent Schottky barrier heights (SBHs) and ideality factors were obtained, which are shown in Fig. 3. With increasing the temperature, the barrier height was almost the same and then increased. The ideality factor increased up to 375 K and then decreased with the temperature. When the lateral barrier inhomogeneity is associated,^18,19) the barrier height increases and the ideality factor decreases with increasing the temperature, which is the case above 375 K in Fig. 3. The increase of barrier height with temperature was also explained by the following way:²⁰⁾ With increasing the temperature, some of the thermally activated metal electrons are nullified by the induced polarization positive charges in AlGaN and others are transported into 2DEG. More AlGaN donor atoms are thus ionized at higher temperature and the free electrons drift into the 2DEG. Consequently, the depletion layer width increases and the barrier height increases as well. According to Maeda et al.,²¹⁾ the decrease in the barrier height in Ni/n-GaN Schottky diodes was explained by the shrinkage of the bandgap (lowering of the conduction band edge) in GaN with increasing the temperature. Meanwhile, Zhang et al. observed the similar trend in the Al_xGa_1−xN diodes.²²⁾ From their explanation, the thermally activated metal electrons might tunnel into the semiconductor depletion region at higher temperature, reduced the depletion widths and lowered the effective barrier height. An increase in ideality factor at high temperature implies that the diode quality decreased and other transport mechanisms such as trap-assisted tunneling, thermionic field emission, and field emission contributed to the current flow. The temperature dependence in Fig. 2 indicates that simple tunneling mechanism cannot be used to explain the experimental data.

Fig. 1

Atomic force microscopy (AFM) surface images for (a) GaN substrate (b) AlGaN layer. Both (c) and (d) shows the topographic heights along the solid lines from A to B in (a) and (b), respectively.

Fig. 2

Semilogarithmic current–voltage (I–V) curves measured under different temperatures. The inset shows a schematic layer structure of AlGaN/GaN devices.

Fig. 3

Temperature dependences of barrier height and ideality factor.

As shown in Fig. 2, a sharp increase (about one order of magnitude) of reverse current as increasing the bias voltage occurred when the voltage reached a certain value, especially at lower temperatures. With increasing the bias voltage, the electrical field strengthens the inverse piezoelectric effect, leading to the expansion of AlGaN layer. Then the strain could exceed a critical value, which in turn caused the strain relaxation occurring through defect formation such as dislocations.^23,24) On the other hand, we did not observe such a sharp increase in reverse current for AlGaN(20 nm)/GaN Schottky diodes (not shown). For relatively thick AlGaN, strain relaxation might occur already during the ALD deposition and the increase of reverse voltage did not affect the inverse piezoelectric effect across the AlGaN layer. However, further investigation is required to understand the exact mechanism.

To clarify the dominant transport mechanism, Poole-Frenkel (PF) emission model was applied to the forward bias current, given by^25,26)   

\begin{equation} \ln (J/E) = m(T)E^{1/2} + b(T) \end{equation} (1)

with   

\begin{equation} m(T) = \frac{q}{kT}\sqrt{\frac{q}{\pi \varepsilon_{0}\varepsilon}},\quad b(T) = - \frac{q\phi_{t}}{kT} + \ln C \end{equation} (2)

where ϕ_t is the electron emission barrier height from the trap states, ε is the relative dielectric permittivity of the gate insulator at high frequency, ε₀ is the permittivity of free space and C is a constant. From the lineal fits to the PF plots of ln(J/E) vs. E^1/2 shown in Fig. 4(a), both m(T) and b(T) values were obtained at each temperature. The dielectric constants of AlN and GaN at high frequency are 4.77 and 5.35, respectively.²⁵⁾ Then, it can be assumed that the dielectric constant of AlGaN is an intermediate value between these two values. From the m(T) values in Fig. 4(b), ε was found to be 5.82, which is comparable to the theoretical value. This indicates that the PF emission can describe the experimental I–V data properly. From the b(T) values in Fig. 4(c), ϕ_t was obtained as 0.69 eV, meaning that the defect states localized at 0.69 eV below the AlGaN conduction band contributed to the PF emission. Zhu et al. obtained a value of 0.65 eV from Al₂O₃/Al_0.55Ga_0.45N/GaN HEMTs, which was associated with the PF emission assisted by oxygen vacancies in oxide layer.²⁷⁾ In this work, however, we did not deposit high-k oxide layer. Using the effective SBH of 1.06 eV at room temperature, the difference between SBH and ϕ_t is found to be 0.37 eV. The commonly observed dislocation-related traps are known to be located at 0.3 eV.²⁵⁾ Hence, the PF emission can be regarded to involve the dislocation-related trap states in AlGaN layer.

Fig. 4

Poole-Frenkel (PF) emission plots of ln(J/E) vs. E^1/2 with the linear fit to the experimental data, (b) calculated m values vs. temperature, and (c) calculated b values vs. temperature.

Similarly the PF emission was applied to the reverse leakage current above −2 V (not shown). However, the obtained dielectric constant was 25.0, which is much higher than the theoretical value. Hence, the PF emission is not appropriate to explain the current transport. Instead, trap assisted tunneling (TAT) would contribute to the current transport. Li et al. reported that the leakage current mechanism of AlGaN/GaN on Si substrate was dominated by the PF emission at low and medium reverse bias region, and TAT in the higher reverse bias region.²⁸⁾ Based on the analysis using Fowler-Nordheim emission, Zhang et al. obtained the smaller value of effective mass for AlGaN compared to GaN and they suggested that the defect-related tunneling transport mechanisms are more prevalent in AlGaN than in GaN.²⁹⁾ The off-state gate current in AlGaN/GaN HEMTs is shown to arise from two parallel gate to substrate tunneling paths and a model to explain this current has shown that TAT dominates below 500 K, and direct tunneling dominates at higher temperatures.³⁰⁾ Kotani et al. observed low dislocation density in AllnN HEMTs on GaN substrate and they suggested that the leakage current in the low bias region is governed by a dislocation-related PF emission, and the leakage current in the high reverse bias region originates from field emission due to the large internal electric field in the AlInN barrier layer. They also showed that band modulation by impurity doping and insertion of insulating layers beneath the gate electrodes would be necessary for the reduction of gate leakage current.³¹⁾ At high reverse bias (above −6 V), the current density in Fig. 2 seems to be almost saturated with reverse bias but increases with increasing the temperature, indicating a thermally activated process with an exp(−E_A/kT) dependence where E_A is an activation energy.^10,28) From the linear fits indicated by the solid lines in Fig. 5, the activation energy at each bias voltage was estimated to be about 0.60 eV. Similar E_A value was obtained by previous work.^8,10)

Fig. 5

Reverse leakage current density as a function of temperature at different reverse biases.

According to the argument by Pelá et al.,³²⁾ the bandgap of Al_0.85Ga_0.15N was calculated to be about 5.65 eV. The conduction-band offset (ΔE_C) between AlGaN and GaN was assumed to be 0.75ΔE_g, where ΔE_g is the bandgap difference between AlGaN and GaN.³³⁾ Assuming the bandgap of GaN to be 3.42 eV, ΔE_C was calculated to be 1.67 eV. Based on these values, the schematic diagram of the conduction band offset in the AlGaN/GaN structure under forward bias is drawn as shown in Fig. 6(a). The schematic diagram under high reverse bias (above −6 V) is shown in Fig. 6(b). In this condition, electrons are thermally emitted with an activation energy of 0.6 eV and then tunnel through the thin AlGaN layer.

Fig. 6

(a) Schematic diagram of the conduction band offset in the AlGaN/GaN structure with the PF emission barrier at forward bias and (b) schematic band diagram at reverse bias.

Figure 7(a) shows the C–V curves measured at various frequencies. The frequency dispersion in both the depletion and accumulation regions is not significant. The theoretical dielectric constant of the AlGaN layer is ε = 9.5 − 0.5x ≈ 9.08,³⁴⁾ for x = 0.85. The effective dielectric constant of AlGaN was determined to be ∼11.3 from the equation ε = C_AlGaNd/ε₀A, where C_AlGaN is the measured accumulation capacitance at 500 kHz, d the thickness of AlGaN layer, and A is the effective contact area. The obtained value is similar to the theoretical value. Figure 7(b) shows the C–V hysteresis plots measured at 100 kHz. All the hysteresis responses were measured starting from inversion to the accumulation region, and subsequently sweeping back towards inversion and there was no hold time in accumulation. During this measurement, the sequential scans were performed from depletion (−3 V) to accumulation (V_g,max). Here, V_g,max varied from 2 to 2.6 V. The inset in Fig. 7(b) shows both the flatband voltage shift (ΔV_FB) and the trapped charge density (Q_T = (C_OXΔV_FB)/qE_g) along the GaN bandgap (E_g). It is seen that the flatband voltage shift increased with increasing V_g,max value. This may be due to the increase in the slow-charge trapping states to the AlGaN layer at higher electric fields.³⁵⁾ Regarding the similar observation in metal/high-k/In_0.53Ga_0.47As MOS capacitors, Lin et al. suggested that the charge trapping taking place primarily in the interfacial oxide transition layer between the In_0.53Ga_0.47As and the ALD deposited oxide layer was responsible for the C–V hysteresis.³⁶⁾ To clarify the interface trap density further, we employed the parallel conductance method.

Fig. 7

(a) Capacitance–voltage (C–V) curves at different frequencies and (b) C–V hysteresis plots measured at 100 kHz with the sequential scans performed from depletion (−3 V) to accumulation (V_g,max). The inset in (b) shows the flatband voltage shift and the trapped charge density.

In the conductance method, the parallel conductance (G_p/ω) vs. the radial frequency (ω = 2πf) plots contain the information on the trap density (D_T) and the trap response time (τ_T) through the following equation,³⁷⁾   

\begin{equation} \frac{G_{p}}{\omega} = \frac{q\omega \tau_{T}D_{T}}{1 + (\omega \tau_{T})^{2}} \end{equation} (3)

Based on the experimental and fitting data shown in Fig. 8(a), the energy level of traps below the GaN conduction band, denoted as E_C–E_t were determined from τ_T according to the Shockley-Read-Hall statistics³⁷⁾   

\begin{equation} \tau_{T} = \frac{1}{v_{\textit{th}}\sigma_{n}N_{C}}\exp \left(\frac{E_{C} - E_{t}}{kT} \right) \end{equation} (4)

where v_th is the thermal velocity, σ_n is the electron capture cross section, and N_C is the effective density of states in the GaN conduction band. The calculated trap densities and time constants are shown in Fig. 8(b). Very narrow energy distribution at 0.26∼0.29 eV with the low time constants below 0.1 µs implies the fast border traps. The narrow energy distribution at 0.36∼0.39 eV with the corresponding time constants between 2 and 5 µs and another energy distribution at ∼0.48 eV with the slow time constants between 5 and 230 µs are also observed. The obtained trap density is higher than the previous works.^38,39) In those works, however, AlGaN/GaN layers were grown by metalorganic chemical vapor deposition. It was shown that border traps near the gate-dielectric/GaN interface were to be suppressed by the AlN interfacial layer in the Al₂O₃/GaN structure, associated with the suppression of detrimental Ga–O bonds.⁴⁰⁾ Hence, further investigation should be focused on reducing the Ga–O bonds during the initial ALD deposition process in order to obtain high quality of AlGaN layer with higher Ga mole fraction.

Fig. 8

(a) Frequency-dependent parallel conductance-voltage (G_p/ω–V) curves, (b) trap density vs. energy level, and (c) time constant vs. voltage.

4. Conclusion

In conclusion, we used ALD to deposit AlGaN on n-GaN substrate and explored the electrical properties of Pt/AlGaN/GaN Schottky diodes by using I–V and C–V measurements. From I–V data, it was found that the barrier height decreased and the ideality factor increased with increasing the temperature. The dominant current transport mechanism udder forward bias was found to be PF emission involved with the dislocation-related traps. Under high reverse bias, the trap-assisted tunneling contributed mainly to the current flow. C–V hysteresis measurements with the sequential scans showed the increase in both the flatband voltage shift and the trap charge density, associated with the charge trapping taking place in the interfacial oxide layer near the AlGaN/GaN interface.

Acknowledgements

This work was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2017R1D1A1B03030400).

REFERENCES

• 1)   Y.  Liao,  C.  Thomidis,  C.  Kao and  T.  Moustakasa: Appl. Phys. Lett. 98 (2011) 081110.
• 2)   T.  Wang,  Y.  Liu,  Y.  Lee,  Y.  Izumi,  J.  Ao,  J.  Bai,  H.  Li and  S.  Sakai: J. Crys. Growth 235 (2002) 177–182.
• 3)   D.  Marti,  S.  Tirelli,  A.  Alt,  J.  Roberts and  C.  Bolognesi: IEEE Electron Device Lett. 33 (2012) 1372–1374.
• 4)   S.  Turuvekere,  A.  DasGupta and  N.  DasGupta: IEEE Trans. Electron Devices 62 (2015) 3449–3452.
• 5)   R.  Swain,  K.  Jena and  T.  Lenka: IEEE Trans. Electron Devices 63 (2016) 2346–2352.
• 6)   Y.  Lv,  Z.  Lin,  T.  Corrigan,  J.  Zhao,  Z.  Cao,  L.  Meng,  C.  Luan,  Z.  Wang and  H.  Chen: J. Appl. Phys. 109 (2011) 074512.
• 7)   M.  Higashiwaki,  T.  Matsui and  T.  Mimura: IEEE Electron Device Lett. 27 (2006) 16–18.
• 8)   L.  Xia,  A.  Hanson,  T.  Boles and  D.  Jin: Appl. Phys. Lett. 102 (2013) 113510.
• 9)   D.  Yan,  H.  Lu,  D.  Cao,  D.  Chen,  R.  Zhang and  Y.  Zheng: Appl. Phys. Lett. 97 (2010) 153503.
• 10)   W.  Ha,  S.  Chhajed,  S.  Oh,  S.  Hwang,  J.  Kim,  J.  Lee and  K.  Kim: Appl. Phys. Lett. 100 (2012) 132104.
• 11)   J.  Zhu,  X.  Ma,  B.  Hou,  W.  Chen and  Y.  Hao: Appl. Phys. Lett. 104 (2014) 153510.
• 12)   S.  Turuvekere,  D.  Rawal,  A.  DasGupta and  N.  DasGupta: IEEE Trans. Electron Devices 61 (2014) 4291–4294.
• 13)   M.  Alevli,  C.  Ozgit,  I.  Donmez and  N.  Biyikli: J. Vac. Sci. Technol. A 30 (2012) 021506.
• 14)   H.  Shih,  W.  Lee,  W.  Kao,  Y.  Chuang,  R.  Lin,  H.  Lin,  M.  Shiojiri and  M.  Chen: Sci. Rep. 7 (2017) 39717.
• 15)   H.  Kim,  H.  Yoon and  B.  Choi: Nanoscale Res. Lett. 13 (2018) 232–242.
• 16)   Z.  Bougrioua,  J.  Farvacque,  I.  Moerman and  F.  Carosella: Phys. Status Solidi B 228 (2001) 625–628.
• 17)   C.  Ozgit-Akgun,  E.  Goldenberg,  A.  Okyay and  N.  Biyikli: J. Mater. Chem. C 2 (2014) 2123–2136.
• 18)   R.  Tung: Mater. Sci. Eng. Rep. 35 (2001) 1–138.
• 19)   I.  Jyothi,  H.  Yang,  K.  Shim,  V.  Janardhanam,  S.  Kang,  H.  Hong and  C.  Choi: Mater. Trans. 54 (2013) 1655.
• 20)   F.  Li,  X.  Zhang,  Y.  Duan,  X.  Xie and  C.  Lü: Chin. Phys. B 18 (2009) 5029.
• 21)   T.  Maeda,  M.  Okada,  M.  Ueno,  Y.  Yamamoto,  T.  Kimoto,  M.  Horita and  J.  Suda: Appl. Phys. Express 10 (2017) 051002.
• 22)   X.  Zhang,  F.  Li,  C.  Lü,  X.  Xie,  Y.  Li and  S.  Mohammad: J. Semicond. 30 (2009) 034001.
• 23)   J.  del Alamo and  J.  Joh: Microelectron. Reliab. 49 (2009) 1200–1206.
• 24)  J. Joh and J. del Alamo: IEDM Tech. Dig., (2006) pp. 415–418.
• 25)   H.  Zhang,  E.J.  Miller and  E.T.  Yu: J. Appl. Phys. 99 (2006) 023703.
• 26)   P.  Fiorenza,  G.  Greco,  F.  Giannazzo,  R.  Nigro and  F.  Roccaforte: Appl. Phys. Lett. 101 (2012) 172901.
• 27)   J.  Zhu,  X.  Ma,  B.  Hou,  W.  Chen and  Y.  Hao: Appl. Phys. Lett. 104 (2014) 153510.
• 28)   Y.  Li,  G.  Ng,  S.  Arulkumaran,  G.  Ye,  Z.  Liu,  K.  Ranjan and  K.  Ang: J. Appl. Phys. 121 (2017) 044504.
• 29)   H.  Zhang,  E.  Miller and  E.  Yu: J. Appl. Phys. 99 (2006) 023703.
• 30)   S.  Karmalkar,  D.  Sathaiy and  M.  Shur: Appl. Phys. Lett. 82 (2003) 3976.
• 31)   J.  Kotani,  A.  Yamada,  T.  Ishiguro,  S.  Tomabechi and  N.  Nakamura: Appl. Phys. Lett. 108 (2016) 152109.
• 32)   R.  Pelá,  C.  Caetano,  M.  Marques,  L.  Ferreira,  J.  Furthmüller and  L.  Teles: Appl. Phys. Lett. 98 (2011) 151907.
• 33)   I.  Smorchkova,  C.  Elsass,  J.  Ibbetson,  R.  Vetury,  B.  Heying,  P.  Fini,  E.  Haus,  S.  DenBaars,  J.  Speck and  U.  Mishra: J. Appl. Phys. 86 (1999) 4520–4526.
• 34)   O.  Ambacher,  J.  Smart,  J.  Shealy,  N.  Weimann,  K.  Chu,  M.  Murphy,  W.  Schaff,  L.  Eastman,  R.  Dimitrov,  L.  Wittmer,  M.  Stutzmann,  W.  Rieger and  J.  Hilsenbeck: J. Appl. Phys. 85 (1999) 3222–3233.
• 35)   B.  Ren,  M.  Sumiya,  M.  Liao,  Y.  Koide,  X.  Liu,  Y.  Shen and  L.  Sang: J. Alloy. Compd. 767 (2018) 600–605.
• 36)   J.  Lin,  Y.  Gomeniuk,  S.  Monaghan,  I.  Povey,  K.  Cherkaoui,  E.  O’Connor,  M.  Power and  P.  Hurley: J. Appl. Phys. 114 (2013) 144105.
• 37)  D. Schroder: Semiconductor Material and Device Characterization, (Wiley, New York, 2005).
• 38)   D.  Gregušová,  R.  Stoklas,  Ch.  Mizue,  Y.  Hori,  J.  Novák,  T.  Hashizume and  P.  Kordoš: J. Appl. Phys. 107 (2010) 106104.
• 39)   Q.  Wang,  X.  Cheng,  L.  Zheng,  L.  Shen,  J.  Li,  D.  Zhang,  R.  Qiana and  Y.  Yu: RSC Advances 7 (2017) 11745.
• 40)   S.  Liu,  S.  Yang,  Z.  Tang,  Q.  Jiang,  C.  Liu,  M.  Wang,  B.  Shen and  K.  Chen: Appl. Phys. Lett. 106 (2015) 051605.