Gallium nitride (GaN) has been developed as a basis semiconductor for InGaN and AlGaN ternary compounds for such applications as green, blue, up to deep ultra-violet light emitters and high power electronic devices  . However, the full potential of GaN based devices has been restricted due to the lack of suitable substrate  . The sapphire substrates are generally used owing to low cost and high temperature stability  . But they introduce threading dislocations in a typical range of 109 - 1011 cm−2 due to lattice and thermal mismatch between epitaxial layer and substrate    . High density of these structural defects forms below-gap states in group III-V semiconductors (such as GaAs, InP and GaN) which act as non-radiative recombination (NRR) centers in the crystal and degrade the device efficiency and lifetime    . The insertion of buffer layer between substrates and epilayers has generated a lot of research interest for decreasing defect density in GaN based optoelectronics and microelectronics devices   . It has been reported that the insertion of thin AlN buffer layer between GaN epilayer and sapphire substrate can reduce tensile growth stress and dislocation density which in turn improve crystalline quality compared to that of the LT-GaN buffer layer     . Recently, the high temperature AlGaN MSFET with AlN buffer layer and better surface morphology and crystalline quality of thick AlGaN have been realized for the growth on the AlN buffer layer   . However, for further improvement of GaN based device performance, it is still insufficient to understand the formation mechanism of defect states and structural optimization for eliminating them during the growth process. The GaN epilayers grown on LT-GaN buffer and AlN buffer layers has been characterized by photoluminescence (PL), scanning electron microscopy (SEM) and atomic force microscopy (AFM) studies     , but these methods give little information about NRR centers. Deep Level Transient Spectroscopy (DLTS) has been also used to study the deep levels in GaN epilayers  but its applications are restricted due to the necessity of preparing suitable sample for the measurements. On the other hand, our two-wavelength excited photoluminescence (TWEPL) is a versatile non-contacting and non-destructive scheme; no need to arrange any special kind of sample preparation. A comparative study of these types of samples has not been reported yet by this method.
In this work, TWEPL has been used for the detection and characterization of NRR centers in n-type GaN layers on a LT-GaN buffer layer and aAlN buffer layer grown on sapphire substrates. The change in PL peak intensity due to the addition of the BGE light over that of the AGE is observed as a function of AGE density, BGE density, and temperature. The NRR parameters have also been evaluated by systematically solving the rate equations and fitting the results with experimental data.
2.1. Sample Structure
Two n-type GaN layers with Si concentration of 3 × 1016 cm−3 were grown on LT-GaN (sample A) and AlN buffer layer (Sample B), respectively, by metal organic chemical vapor deposition (MOCVD) method. The detailed structure of the samples is shown in Figure 1. The n-GaN (1.7 μm) layer was grown at 1050˚C after the sequence of LT-GaN (30 nm) or AlN (1.0 μm) buffer and i-GaN (3.0 mm) layer on c-plane (0001) sapphire substrate. All the layers except the buffer layer are same for both samples.
The experimental setup for characterizing the n-GaN layers by TWEPL is shown in Figure 2. The sample was mounted in a temperature-controlled cryostat and
(a) Sample A (b) Sample B
Figure 1. Structure of n-type GaN samples; (a) sample A with LT-GaN buffer layer and (b) sample B with AlN buffer layer.
Figure 2. Experimental setup of the TWEPL measurement.
irradiated by diode-pumped solid-state (DPSS) laser of 4.66 eV (266 nm) as the AGE. The conventional PL spectra of both samples were measured at a temperature of 12 K. The output power of the AGE laser was kept constant at 16 mW and the neutral density filters were used for changing the excitation density. A temporally switched BGE light by an optical shutter with the internal 500 seconds ON/OFF was superposed on the AGE light to excite the sample at the same point. The BGE light of energy 1.27 eV (980 nm), 1.17 eV (1064 nm), and 0.93 eV (1340 nm) were used for the TWEPL measurement. The modulated PL signal from the sample was fed to a monochromator through a set of objective lenses, converted to photocurrent by a photomultiplier and recorded by a computer after the lock-in amplification. Here, an optical chopper was used with the digital lock-in-amplifier for improving the S/N ratio of the PL signal. By measuring PL intensity with and without the BGE, IAGE+BGE and IAGE, respectively, the normalized PL intensity IN = IAGE+BGE/IAGE was determined.
3. Results and Discussion
3.1. PL Intensity Comparison
The PL spectra of n-GaN samples measured at 12 K under the irradiation with the AGE light only are shown in Figure 3. Both spectra consist of near band-edge (NBE) luminescence (3.47 eV), shallow donor (oxygen) to valence band transition (Iox at 3.41 eV), donor acceptor pair (DAP) transition (3.27 eV), and yellow luminescence (YL) (2.3 eV) peaks. Same type of emission spectra has
Figure 3. PL spectra of n-GaN samples grown on LT-GaN buffer layer (Sample A) and AlN buffer layer (Sample B).
been also reported for GaN in earlier studies  . The NBE luminescence peak intensity of sample B is 8 times higher than that of sample A while the YL intensity of sample B is lower. The PL intensity is used to measure the quality of samples, and here it elucidates that quality of sample B is better than that of the sample A.
3.2. TWEPL Measurement
In this study, we focus on the NBE luminescence as the principal component for comparative analysis of two samples. The normalized PL intensity (IN) of the NBE peak has been measured at a fixed temperature of 12 K and the AGE density of 1.10 mW/mm2 is shown in Figure 4 as a function of the BGE power density for both samples.
With the addition of the BGE light, the value of IN quenches from unity for all BGE energies of 0.93, 1.17 and 1.27 eV. The quenching of the IN can be explained by two levels model schematically shown in Figure 5 indicating the presence of a pair of NRR centers in the sample whose energy difference correspond to that of the BGE energies     . Determination of the NRR parameters becomes possible quantitatively by fitting experimental results based on Shockley-Read-Hall (SRH) statistics   .
From Figure 4, it is observed that the amount of PL quenching becomes pronounced with increasing the BGE energy and density. It also exhibits that the value of IN for sample A is always lower than that of sample B throughout the
Figure 4. The normalized PL intensity (IN) of the NBE emission as a function of the BGE power density.
Figure 5. Two levels model of NRR process which explains the PL intensity quenching after irradiation of the BGE.
experimental BGE density range. The minimum values of IN, 0.60 and 0.74 are obtained for samples A and B, respectively at highest BGE density (1.37 W/mm2) of 1.17 eV BGE. The degree of change of the IN values from unity represents the density of NRR centers in the samples. Thus, this result implies that the density of NRR centers is higher in sample A compared to that in sample B. It is consistent with the relative intensity of conventional PL for these samples.
When the BGE energy matches the energy difference between two coexisting below-gap NRR levels, electrons in NRR level 1 are excited to NRR level 2 from which they recombine nonradiatively with holes in the valence band of GaN. Hence, the hole density in the valence band decreases. Similarly, the electron vacancies in the NRR level 1 allow an increase of NRR process from conduction band. Thus, the electron density in the conduction band decreases. The combination of both effects reduces the number of electron-hole pairs available for radiative recombination and resulting in the PL intensity quenching. In the region of low BGE densities, the electron occupation function of NRR level 2 remains much lower than 1 and the PL quenching proceeds with the increase in the BGE density. In the region of higher BGE densities, on the other hand, the electron occupation function of NRR level 2 approaches unity and the PL quenching shows saturation tendency with further increase in the BGE density.
The AGE density dependence of IN has been measured at a fixed BGE density and temperature by utilizing BGE energies of 0.93 and 1.17 eV shown in Figure 6. With increasing the AGE density from 1.10 mW/mm2 to 4.60 mW/mm2, the value of IN approaches to unity for both samples. At lower AGE density, the excitation of electrons via below gap states relative to band-to-band excitation
Figure 6. AGE power density dependence of the normalized PL intensity (IN) for samples A and B.
increases which results in higher BGE effect due to the enhancement of the non-radiative recombination. Similar AGE density dependence of the IN was observed in our previous studies of TWEPL     .
The temperature dependence of IN for samples A and B has been also examined at a fixed AGE (1.10 mW/mm2) and BGE (1.37 W/mm2 and 0.95 W/mm2) densities shown in Figure 7. It has been observed that the IN value increases for both 1.17 and 0.93 eV BGE, with increasing temperature from 12 K to 70 K. The IN value of sample A enhances from 0.60 to 0.90, and that of sample B from 0.74 to 0.91, for 1.17 eV BGE. Further increase in temperature brings little change in the IN values, showing a saturating tendency up to 130 K. This type of temperature dependency was observed in previous studies and attributed to the thermal emission of electrons en from NRR level 2 to the conduction band in Figure 5 in the two levels model   . This type of thermal emission reduces the electronic population in the NRR level 2 even under the below-gap excitation and decreases the BGE effect.
3.3. Rate Equation Analysis
In order to corroborate our qualitative interpretation by the two levels model, a semi-quantitative simulation for the TWEPL results of 1.17 eV BGE energy has been carried out. The rate equations of the two levels model as shown in Figure 5 can be written below with charge neutrality condition (CNC)    .
Figure 7. The Normalized PL intensity (IN) as a function of temperature observed for samples A and B.
where G1 [cm−3・s−1] and G2 [cm3・s−1] are generation rate for the AGE and the BGE, respectively, B [cm3・s−1] is the radiative recombination coefficient, Nt is the density of NRR levels, n0 is the density of free electrons, ft1 andft1 are the electron occupation function of NRR level 1 and NRR level 2, respectively.
For simplicity, we assumed that the electron capture coefficient Cn1 is equal to radiative recombination coefficient B, as 1.2 × 10−11 cm3・s−1 for GaN  . Such consideration has been taken by other researchers   . Reshchikov et al.   have reported that the hole capture coefficient of Cp1 is in the order of 10−6 cm3・s−1 for GaN. The hole capture coefficient of Cp2 has been also reported in the order of 10−9 cm3・s−1  . The procedure of estimating generation rate of the AGE (G1) and BGE (G2) have been explained in our earlier study  .
The density of free electrons is assumed as n0 ≈ 1 × 1016 cm−3 for both samples considering the Si doping concentration. The system of rate equations can be solved numerically and the dependencies of n, p, ft1, and ft2 on G2 was found for the constant parameters of G1, B, n0, Nt, Cn, and Cp. By systematically solving and fitting the simulated results with experimental data, the defect parameters have been chosen as G1= 4.0 × 1020 cm−3・s−1, Cp1 = 1 × 10−6 cm3・s−1, Cn1 = 8.5 × 10−11 cm3・s−1 and Cp2 = 6.5 × 10−9 cm3・s−1 for both samples A and B, respectively. The densities of two NRR levels are obtained as Nt1 = 8.0 × 1015 cm−3, Nt2 = 3.0 × 1017 cm−3 for the sample A, and Nt1' = 6.1 × 1015 cm−3, Nt2' = 6.0 × 1016 cm−3 for the sample B. The value of IN is calculated as a function of generation rate of BGE (G2) under fixed AGE generation rate of 4.0 × 1020 cm−3・s−1 and shown in Figure 8. The broken and solid lines represent the simulated result together with experimental points for both samples. The simulated IN value shows a reasonable agreement with the measured points.
The dependence of the IN as a function of the electron-hole generation rate of the AGE (G1) at 12 K has been calculated by setting G2 = 1.0 × 10−14 cm3・s−1 and keeping all the other parameters as constant as previous. Figure 9 shows the IN value of the NBE peak for both samples as a function of the electron-hole generation rate of the AGE (G1).
Here, a set of parameters give the insight of below-gap states acting as NRR centers in samples A and B, and a reasonable fitting with experimental data. The estimated result shows that the densities of NRR centers are lower in sample B than that in sample A. From both fitting results, it is concluded that the interpretation based on the two-levels model is valid and the use of the AlN buffer layer is more effective for reducing the density of NRR centers in n-GaN layer than the LT-GaN buffer layer. The TWEPL study of NRR centers guides us to optimize growth conditions further.
Figure 8. Variation of the normalized PL intensity as a function of BGE density (G2). The broken and solid lines represent the simulated results for sample A and B, respectively.
Figure 9. AGE density (G1) dependence of the normalized PL intensity in samples A and B. The broken and solid lines represent the simulated results.
Defect States acting as NRR centers in n-type GaN layers grown on a LT-GaN buffer layer and aAlN buffer layer have been studied by TWEPL method. The near band-edge PL peak intensity quenches after the irradiation of BGE energies of 0.93, 1.17 and 1.27 eV. The quenching of the PL intensity has been interpreted by the two levels model and indicates the presence of a pair of NRR centers in the samples which are activated by the BGE. The dominant quenching of the PL intensity for the sample A (with LT-GaN buffer layer) indicates a direct evidence for the higher density of NRR centers compared to the sample B (with AlN buffer layer). A simulation of rate equations agreed well with our experimental data with a set of NRR parameters. The use of AlN buffer layer is more effective for reducing the NRR density in n-GaN layers than the LT-GaN buffer layer.