Characterization of Epitaxial GaN Thin Films by Slow [612586]

Characterization of Epitaxial GaN Thin Films by Slow
Positrons
Doru Dinescu1,2,a)and Nikolay Djourelov1,b)
1Extreme Light Infrastructure – Nuclear Physics, Horia Hulubei National Institute for Physics and Nuclear
Engineering, 30 Reactorului Street, 077125 Magurele, Ilfov county, Romania
2University Politehnica of Bucharest, Faculty of Applied Sciences, 313 Splaiul Independentei, 060042, Bucharest,
Romania
a)Corresponding author: [anonimizat]
b)[anonimizat]
Abstract. GaN presents all the characteristics to be considered a suitable material for field assisted e+moderation. Even though
typical e+diusion lengths, L+, in semiconductors are known to be between 200 and 300 nm, the values reported in the literature
up to now for GaN are below 60 nm. Due to the interest in high quality GaN wafers for various applications, there have been
improvements in reducing the defects in GaN. Positron annihilation experiments by a slow positron beam are used to qualify four
epitaxial GaN thin films, mass produced for device manufacturers, by determination the L+. The longest L+=967 nm was
obtained for a 700 nm film grown on Si.
INTRODUCTION
Since the 1990’s GaN has been one of the most promising materials in the development of semiconductor based
devices [1]. The wide direct band gap, high displacement energy and high thermal stability made GaN a key material
in the development of optoelectronics, high temperature and high power devices [2]. Perfect lattice semiconductors
such as Si or Ga are known to have di usion lengths L+=200300 nm [3]. The wide band gap of GaN (3.4 eV),
due to the epithermal positron di usion, should translate to a longer L+than the ones in typical semiconductors . The
expected long e+diusion length, the fact that GaN presents a negative e+workfunction and the high breakdown
voltage (1 kV) means that GaN might be a suitable material for field assisted e+moderation [4]. On the other hand,
L+reported in the literature for GaN are below 60 nm [5]. It is well known that during the di usion process e+
get trapped in defects [6]. This a ects the L+and implicitly the moderation e ciency, so it is very important for
the moderation material to have a low defect concentration. Positron annihilation spectroscopy is a powerful non-
destructive method of studying the defects in a material. By using the slow positron beam (SPB) technique depth
profiling of the studied sample can be achieved. GaN epitaxial wafers are produced for the manufacturing of high-
electron-mobility transistors (HEMT). Due to the constant development in HEMT research, over the years, there have
been improvements in reducing the defects in the crystal structure of the GaN thin films. The aim of the present study
is to investigate by the Doppler broadening spectroscopy (DBS) and coincidence DBS (CDBS) using a slow positron
beam (SPB) the progresses made in the fabrication of GaN thin films which might be useful for positron moderation.
SAMPLES AND EXPERIMENT
The GaN thin films studied in the present work are produced by NTT Advanced Technology Corporation which claims
to provide high quality GaN wafers for device manufacturers, with high uniformity and high breakdown voltage. The
epitaxial GaN thin films were grown on three kinds of substrates. The studied samples will further be referred to
as it follows: GaN300 /Si, GaN700 /Si, GaN500 /SiC, GaN300 /Sp, where the number in the name represents the GaN
thickness in nm, as given by the producer, and Sp stands for sapphire.

The DBS and the CDBS experiments were performed at the slow positron beam line at the Institute of High
Energy Physics in Beijing, China. For the DBS measurements, the gamma energy spectra were recorded by a HPGe
detector, with a resolution of FWHM =0.97 keV estimated for 511 keV . The detector was placed perpendicularly in
respect to the positron beam axis at a distance of 20 cm from the sample. The incident positron energy was controlled
from E+=0.5 to 25 keV . Each of the 35 experimental spectra was collected over a period of 8 minutes for a fixed
E+, resulting in statistics of 5105counts in the 511 keV region. The shape of the 511 keV peak was analyzed by
the sharpness parameter, S, and by the wings parameter, W. Due to the correlation between the mean e+implantation
depth, zm, and E+(zm=(36=)E1:62
+nm, whereis the sample density in g cm3), the experimental data represents
the depth profile of S. The VEPFIT software was used to fit the S(E+) and W(E+) data [7]. We applied the model 5,
available in VEPFIT, with 2 or 3 layers in addition to the surface e+states.
The CDBS experiments were performed with two HPGe detectors facing each other in order to detect the energies
E
1andE
2of the annihilation
-rays in coincidence. The detectors were set up at distance of 20 cm from the sample
perpendicular to the e+beam direction. The resolution of the second detector was FWHM =1.02 keV . The CDBS
spectra were acquired for only two fixed E+per sample over a period of 14 hours for each measurement. The total
statistics in each spectrum is 1107counts. The distribution, M(pL), of the longitudinal momentum, pL=2
E=c,
of the annihilation pair was extracted from the 3D CDBS data by histograming the di erence
E=E
1E
2on the
conditionj2m0c2(E
1+E
2)j<2:4 keV . In order to magnify the di erences in the high momentum region, the
momentum distributions are often presented as a ratio to a reference sample by R(pL)=Msample(pL)=Mreference(pL).
RESULTS AND DISCUSSION
The depth profiles and best fits, represented in Fig. 1a, are obtained with two layer model, i:e:a single GaN layer
and the Si substrate. The fit results are summurized in Table 1. The e ective L+in Si(100) was fixed to 245 nm
in order to reduce the uncertainties [8]. It can be seen that the film thickness, d, for the films on Si as obtained by
the fit overestimates the one given by the producer of the thin films. One of the reason for the overestimation is the
existence of an interfacing layer between the film and the substrate. Introducing additional bu er layer in the VEPFIT
analysis leads to higher uncertainties so we preferred to stick to the simpler model. In the case of GaN film on SiC
the scatter of the data of the measured depth profiles is large (in particular for S(E+)) (see Fig. 1a), which is a result
of the relatively close values of the film and substrate characteristic SandW(see Table 1). This caused problems
in the VEPFIT analysis and we were forced not only to fix the L+in SiC to 150 nm and also to introduce a highly
defective layer, i:e:to apply a fit by a 3-layer model. It should be mentioned that according to our experience, the
introduction in the VEFIT analysis of an interface layer with small thickness and short di usion length both of few
nm (similarly to [9]) leads to unreliable best fit results since the last are strongly dependent on the initial parameters.
Whether the data for GaN500 /SiC, presented in Table 1, obtained with the mentioned constraints is trustworthy or not
will be discussed below. A good fit of the experimental data of GaN film on sapphire was also obtained by a 3-layer
model. The di usion length in sapphire was fixed to 80 nm [10].
TABLE 1. Best fit parameters obtained by VEPFIT fitting of S(E+) and W(E+) depth profiles.
Sample GaN Substrate
L+(nm) S W d (nm) L+(nm) S W
GaN300 /Si 66(10) 0.4550(3) 0.0931(2) 347(7) 245 0.5257(3) 0.0447(1)
GaN700 /Si 96(7) 0.4540(3) 0.0944(1) 754(6) 245 0.5273(6) 0.0440(3)
GaN500 /SiC 60(6) 0.4562(3) 0.0929(2) 500(6) 150 0.4658(7) 0.0709(4)
1 0.4775(25) 0.0783(14) 1
GaN300 /Sp 37(3) 0.4504(20) 0.0950(25) 109(20)80 0.4232(3) 0.1077(2)
26(19) 0.4656(25) 0.0857(14) 173(16)y
near-surface.ynear-substrate.
In Fig. 1b are plotted the characteristic SWpoints. The points for GaN are closely grouped except for the
point of the near-substrate layer in GaN300 /Sp indicating a higher defect concentration. Although the di erences in
the characteristic S(and W) of the GaN films are rather small, the parameters show an ordinary correlation with L+,
i:e:the higher the Sthe lower the WandL+(see Table 1). The only exception is the relatively short L+=37 nm in the

FIGURE 1. a) The depth profiles S(E+) and W(E+) for the GaN thin films on all four substrates. The lines represent the best fit
obtained by VEPFIT; b) SWpoints of the characteristic best fit parameters for the GaN layer and the substrates.
near-surface layer for GaN300 /Sp which does not correspond to the high S(low W) parameter. Lack of correlation
between SandL+was reported also in [5]. They suggested that the observed low L+values are due to e+interaction
in dislocations. Even though such an explanation sounds reasonable, we cannot draw firm conclusion if it is relevant
for our case or the short L+in the near-surface layer for GaN300 /Sp is an artificial e ect due to the complexity of the
applied VEPFIT model. CDBS spectra were measured at selected E+=7.5 keV (and 12.5 keV for GaN700 /Si) with
a goal to get information for the GaN films avoiding at the same time the surface influence as much as possible, and
atE+=25 keV (the maximum for the beam) in order to get information for the substrates. Due to the fact that the e+
implantation profile width increases significantly with E+, due to the long L+, the information for the film is mixed
with the one of the substrate, i:e:
S(E+)=fepi(E+)Sepi
GaN+fsurf(E+)Ssurf+fGaNSGaN+fsub(E+)Ssub: (1)
FIGURE 2. Fractions of epithermal and thermal e+annihilating on the GaN surface, in the GaN film and in the Si substrate as
determined by VEPFIT with a 2 layer model.
The same equation is valid for the Wparameters and for the momentum distributions, as well. If we consider that
the momentum distributions for GaN in GaN300 /Si and GaN700 /Si are undistinguishable, which is supported by the
very close SWpoints for GaN (Fig. 1b), we easily find the characteristic momentum distributions MGaN(pL) and
MSi(pL) by solving the following system of linear equations:
MGaN300=Si(pL;)=fGaN300=Si
GaN()MGaN(pL)+fGaN300=Si
Si()MSi(pL); (2)
MGaN700=Si(pL;)=fGaN700=Si
GaN()MGaN(pL)+fGaN700=Si
Si()MSi(pL); (3)
wherestands for E+=25 keV and the coe cients fare extracted from Fig. 2.
Further, by using the CDBS data for E+=7.5 (and 12.5) keV in an analogical way, the momentum distributions
which comes from thermal e+annihilation on the GaN surface is extracted. The extracted characteristic momentum

distributions are shown in Fig. 3a. The momentum distributions shown in Fig. 3a are in good agreement with the data
available in the literature for perfect GaN and it definitely di ers from the momentum distribution in case of saturation
trapping of e+in Ga vacancies [11]. Consequently, we may consider that the studied GaN films on Si present a low
defect concentration. The momentum distributions for E+=7.5 keV in GaN300 /Si, GaN300 /Sp and GaN500 /SiC
as a ratio to the data taken at E+=12.5 keV in GaN700 /Si are shown in Fig. 3b. The fraction from the thermal e+
annihilating at the GaN surface ranges from 0.07 to 0.17. The lines represent the experimental points corrected for the
influence from the substrates. Di erences in the quality of the GaN films on Si and SiC is not observed. However, the
shape of the ratio for GaN on sapphire resembles the ratio shape for e+trapped in GaN defects like V Ga, VGa-VNand
VGa-ON[11].
0 5 10 15 20 25 3010−510−410−310−210−1
pL (10−3 m0c)Probability

Si perfect
GaN perfect
VGa in GaN
Si extracted
GaN extracteda)
0 5 10 15 20 25 300.70.80.911.11.2
pL (10−3 m0c)Ratio to corrected GaN700/Si ( E+= 12.5 keV)

GaN700/Si ( E+= 12.5 keV)
GaN300/Si ( E+= 7.5 keV)
GaN300/Sp ( E+= 7.5 keV)
GaN500/SiC ( E+= 7.5 keV)b)
FIGURE 3. a) Momentum distributions for GaN and Si extracted from the experimental CDBS data together the data reproduced
from literature represented by the lines [11]; b) Momentum distributions obtained for samples GaN300 /Si, GaN500 /SiC, and
GaN300 /Sp measured at E+=7.5 keV and for GaN700 /Si at E+=12.5 keV as a ratio to corrected GaN700 /Si data. The lines
represent the pure GaN data, i:e:after the correction of the measured momentum distribution for the small contributions from the
substrates.
CONCLUSIONS
DBS and CDBS on SPB was used to characterize four epitaxial GaN wafers. The short L+of GaN thin films compared
to the typical L+in semiconductors was confirmed. However, improvements in the crystal structure of GaN thin films
on Si were found ( L+(GaN700 /Si)=967 nm). The defect concentration in GaN300 /Sp is higher than in the films
on Si and SiC.
ACKNOWLEDGMENTS
The authors wish to acknowledge the support from the ELI-NP – Phase II a project financed through the European Re-
gional Development Fund – the Competitiveness Operational Programme (1 /07.07.2016, COP, ID 1334), the contract
27-ELI /2016 and the help of the Positron group at IHEP, Beijing, especially Prof. B. Wang, Prof. X. Cao.
REFERENCES
[1] S. Nakamura, T. Mukai, and M. Senoh, Japanese Journal of Applied Physics 30, L1998–L2001 (1991).
[2] J. Wang, P. Mulligan, L. Brillson, and L. R. Cao, Applied Physics Reviews 2, p. 031102 (2015).
[3] P. J. Schultz and K. G. Lynn, Rev. Mod. Phys. 60, 701–779 (1988).
[4] L. Jørgensen and H. Schut, Applied Surface Science 255, 231 – 233 (2008).
[5] S. Saleh and A. M. Elhasi, American Journal of Modern Physics 3, 24–28 (2014).
[6] P. Coleman, Positron Beams and their Applications (World Scientific Publ., Singapore, 2000), pp. 14–21.
[7] A. van Veen, H. Schut, M. Clement, J. M. M. de Nijs, A. Kruseman, and M. R. Ijpma, Appl. Surf. Sci. 85,
216–224 (1995).
[8] P. J. Schultz, E. Tandberg, K. G. Lynn, B. Nielsen, T. E. Jackman, M. W. Denho , and G. C. Aers, Phys.
Rev. Lett. 61, 187–190 (1988).
[9] R. Ferragut, A. Calloni, A. Dupasquier, and G. Isella, Nanoscale Research Letters 5, p. 1942 (2010).
[10] A. Zubiaga, J. A. Garc ´ıa, F. Plazaola, F. Tuomisto, J. Z ´u˜niga P ´erez, and V . Mu ˜noz Sanjos ´e, Phys. Rev. B 75,
p. 205305 (2007).
[11] F. Tuomisto and I. Makkonen, Rev. Mod. Phys. 85, 1583–1631 (2013).