Electron paramagnetic resonance (EPR) spectroscopy was used to study the point defects in 2 × 1017–1019 cm−3 C-doped GaN substrates grown by hydride vapor phase epitaxy. The intensity of an isotropic signal with g = 1.987 ± 0.001 increased monotonically with the carbon concentration, indicating that the EPR signal represents a carbon-related defect. In each sample, the signal intensity increased under illumination with photon energy greater than 2.75 eV, and the photo-induced signal decreased with subsequent illumination at 0.95 eV. A second signal, well-documented to be a shallow donor, appeared along with the g = 1.987 signal in the most lightly doped samples. The appearance of the donor confirms that the photo-induced increase is caused by excitation of an electron to the conduction band and implies that a defect level for the carbon-related center is about 1 eV above the valence band edge, consistent with temperature-dependent Hall measurements.

Although GaN and related alloys are widely used in successful commercial devices, such as LEDs, realization of high-power and high-frequency applications has been limited.1–3 The slow development of nitride-based power electronics is due, in part, to the limited availability of semi-insulating (SI) substrates. As-grown GaN is n-type due to unintentionally added donor impurities, such as O and Si, and highly resistive material is achieved by doping with acceptors that can electrically compensate the unintentionally added donors.4,5 Fe and C are proven effective in creating highly resistive GaN, but the dopant with the deeper defect level, C, will be necessary for extreme power applications.5–11 GaN doped with Be yields a material with a larger activation energy than either Fe or C, but the toxicity associated with a Be source will likely prevent the widespread use of the dopant.12 

Many groups have investigated the theoretical defect level and the formation energy of carbon and carbon complexes in GaN.6,13,14 One of the most recent prediction was when carbon substitutes for nitrogen (CN), the impurity acts as a deep acceptor with a negative-to-neutral charge state (−/0) transition 0.9 eV above the valence band edge.6,15 Consistently, Iwinska et al. reported that 1018–1019 cm−3 carbon-doped GaN grown by hydride vapor phase epitaxy (HVPE) exhibits hole conductivity with an activation energy of 1 eV.9 To date, however, there is little other experimental support for the deep acceptor nature of carbon in GaN. Only recently, our electron paramagnetic resonance (EPR) studies showed that a defect in heavily C-doped GaN exhibits an optical response consistent with the theoretical predictions of the deep carbon acceptor.6,16 Although the relationship between the defect detected and the semi-insulating property of the material is clear, the identity of the defect could not be discerned from the limited sample set. Here, we show that the defect is directly related to the presence of carbon and, through observation of an optically induced shallow donor, strengthen the deep acceptor model developed from the photo-EPR data.

C-doped GaN grown by hydride vapor phase epitaxy (HVPE) onto an ammonothermal GaN seed were used for our investigation. These samples are free-standing, removed from the substrate by mechanical polishing. The samples were doped with carbon to concentrations ranging from 2 × 1017 to 1019 cm−3 by adjusting CH4 flow rate in the growth zone. The concentration of the impurities measured by secondary ion mass spectrometry (SIMS) is shown in Table I. The values recorded are those measured about 0.5 μm below the surface to avoid ambient contaminants, and the value remained constant throughout the depth measured as 6 μm. SIMS was performed on two pieces each of samples D-G to estimate nonuniformity. For all except D, the variation within a sample was less than the absolute uncertainty of the SIMS measurement of 20%. For sample D, the value varied between 11 × 1017 and 40 × 1017 cm−3, and the average of the two measurements is reported in Table I. The value for the more heavily doped samples was that measured in a single piece grown at the same time as the piece used for this study.

TABLE I.

Impurity concentration (1017 cm−3) measured by SIMS in HVPE GaN samples.

SampleCHOSiFe
100 0.003 
60 0.004 
60 0.005 
25 0.02 
10 0.02 
0.01 
0.01 
SampleCHOSiFe
100 0.003 
60 0.004 
60 0.005 
25 0.02 
10 0.02 
0.01 
0.01 

The table indicates that the total of all the unintentional impurities in the samples is lower in the first three, compared to the remaining four. The variation is likely due to the use of different reactors for each set of samples. As will be seen below, although the Fe has a minimal effect on our results, the variation in donor impurities, such as O and Si, plays an important role.

The 0.2 mm thick samples were cut into 5 × 2 mm2 pieces suitable for EPR, and the measurements were performed at 3.5 K with the magnetic field parallel to the c-axis. The g-value, the EPR parameter used to distinguish the spectra, was obtained using the resonance condition, hf = gμBBo, where h is Plank's constant, f is the microwave frequency, μB is the Bohr magneton, and Bo is the magnetic field corresponding to the point where the EPR intensity passes through zero. To determine the relationship between spin density and carbon concentration, each sample was illuminated with a 4.1 mW LED with a peak energy at 3.1 eV to excite as many defects as possible. The number of EPR active centers was determined in the most heavily doped sample (A) by comparison of the double integral of the sample spectrum with a spectrum obtained from a calibrated Si:P powder.17 The number of centers in the remaining samples was estimated by comparison of their EPR amplitudes with that of sample A. The average defect concentration was determined by dividing by the sample mass and converting to volume concentration using a GaN density of 6.15 g/cm−3.

The photo-EPR measurements consisted of photo-excitation and photo-quenching, where the former refers to the increase and the latter refers to the decrease in the EPR signal intensity. For photo-excitation, the samples were illuminated with light of different photon energies from 1.5 to 3.3 eV. Above 2.7 eV, the EPR signal gradually increased with time and saturated after about 20 min. A saturated EPR spectrum was recorded before proceeding to the next wavelength. After the 3.3 eV illumination, samples were exposed to the UV LED until the EPR signal intensity maximized, then the LED was removed and a spectrum was recorded after the signal stabilized. The signal was gradually quenched in a manner similar to that described for the excitation process using light of photon energy from 0.5 to 1.7 eV. A 250 W quartz-tungsten-halogen (QTH) lamp along with monochromator and appropriate filters was used to illuminate the sample. The lamp power for each wavelength was on the order of 100 μW, and the photon flux was kept constant within ± 15% using neutral density filters.

An EPR spectrum representing 2 × 1017 cm−3 centers is present in the most heavily doped crystal (sample A). A similar signal, with intensity that decreases with decreasing carbon concentration, is observed in the remaining samples. Exposure to light of photon energy greater than 2.7 eV increased the signal in all GaN:C crystals. For example, Fig. 1 shows the effect of illumination on the EPR spectra of 1019 cm−3 (A, black solid line), 2.5 × 1018 cm−3 (D, blue dashed line), and 2 × 1017 cm−3 (G, red dotted line) GaN:C samples. The g-value, common to all spectra, is 1.987 ± 0.001. The slightly distorted line shape near B = 3350 G in sample D is caused by the presence of a second impurity. The measurements taken over a wide magnetic field range indicate that the feature is due to Fe3+.18 The analysis shows that about 2 × 1015 cm−3 Fe3+ is present in samples D through G, consistent with the SIMS results. Although the Fe signal complicated the analysis, the impurity did not affect overall interpretation of the data. The feature seen in the spectrum of the most lightly doped sample (G) at B = 3445 G is the well-documented shallow donor, as verified by g values measured with the magnetic field parallel and perpendicular to the c-axis: gpar = 1.951 ± 0.001 and gperp = 1.950 ± 0.001.19 Furthermore, spectra obtained at temperatures above 45 K exhibit the Dysonian line shape expected for a shallow donor, where the paramagnetic electron becomes increasingly delocalized with increasing temperature.20 The observation of the donor in the EPR-active neutral charge state in only the most lightly doped samples, F and G, is significant to interpretation of the photo-EPR data, as will be discussed later.

FIG. 1.

EPR spectra of GaN:C samples measured at 3.5 K after illumination with UV LED: 1019 cm−3 doped sample A (black, solid line), 2.5 × 1018 cm−3 doped sample D (blue, dashed line), and 2 × 1017 cm−3 doped sample G (red, dotted line).

FIG. 1.

EPR spectra of GaN:C samples measured at 3.5 K after illumination with UV LED: 1019 cm−3 doped sample A (black, solid line), 2.5 × 1018 cm−3 doped sample D (blue, dashed line), and 2 × 1017 cm−3 doped sample G (red, dotted line).

Close modal

Figure 2 shows the defect density calculated from the UV excited EPR signal compared to the carbon concentration measured by SIMS. The error bar noted on the most heavily doped sample represents the uncertainty in the double integration required to determine the number of defects, the uncertainty in the remaining data arises from the intensity comparison. The monotonic relationship of defect density with carbon doping indicates the defect is carbon-related. We point out that there is no correlation between the EPR defect density and other impurity concentration variations, thus eliminating the unintentional impurities as a source of the EPR signal. Although carbon is a common impurity in UID GaN, the EPR signal seen in Fig. 1 was not observed, either before or after illumination, in an unintentionally doped sample grown in the same furnace as the most heavily C-doped sample. The only EPR spectrum detected in the UID sample was the neutral shallow donor, as expected, since the typical unintentional carbon concentration in HVPE GaN is below that in any of the samples studied here.21 

FIG. 2.

Comparison of the average defect density calculated from EPR spectrum and carbon concentration obtained from SIMS measurement.

FIG. 2.

Comparison of the average defect density calculated from EPR spectrum and carbon concentration obtained from SIMS measurement.

Close modal

Having established that the defect is related to carbon incorporation, we further study the center by investigating the defect level. In Fig. 3, we compare the photo-EPR results reported earlier for the 1019 cm−3 doped sample (A, closed squares) with the results obtained on the 2.5 × 1018 cm−3 (D, open circles) and 2 × 1017 cm−3 (G, stars) GaN:C samples. Figure 3(a) shows the photo-excitation data and Fig. 3(b) shows the photo-quenching data. Each point represents the number of defects observed after illumination with a particular wavelength relative to that observed in the most heavily doped UV illuminated sample. All samples share the same threshold, confirming that the photo-excitation process is independent of carbon concentration. The working model based on the data of Fig. 3 is that the EPR signal represents an acceptor that is ionized in the as-grown material by unintentionally added donors. As shown in the inset of the figure, optical excitation releases the electron from the acceptor, making it EPR active. Then, during the quenching process, an electron is excited from the valence band to the defect and the EPR intensity is reduced. The presence, initially, of a negatively charged acceptor is supported by the relatively small intensity of the EPR signal before exposure to light. Furthermore, the pre-illumination signal is strongest in the most heavily doped samples, where the carbon concentration is greater than the total amount of donors such as O and Si. The signal intensity then decreases as the carbon concentration approaches the total donor concentration and is below the detection limit in the most lightly doped samples where the number of donors outnumbers the carbon impurities.

FIG. 3.

Steady state photo-EPR data for 1019 cm−3 (A, square), 2.5 × 1018 cm−3 (D, circle), and 2 × 1017 cm−3 (G, star) C-doped samples for excitation (a) and quenching (b). Each point represents the relative number of defects observed after illumination with a particular wavelength. The dashed lines denote excitation and quenching threshold. Insets: Simple band model for excitation (a) and quenching (b).

FIG. 3.

Steady state photo-EPR data for 1019 cm−3 (A, square), 2.5 × 1018 cm−3 (D, circle), and 2 × 1017 cm−3 (G, star) C-doped samples for excitation (a) and quenching (b). Each point represents the relative number of defects observed after illumination with a particular wavelength. The dashed lines denote excitation and quenching threshold. Insets: Simple band model for excitation (a) and quenching (b).

Close modal

Interpretation of photo-EPR data is often limited by the lack of knowledge regarding the fate of the excited carriers. In general, the excitation process could represent electron capture onto the defect from the valence band rather than excitation to the conduction band. However, in the present situation, observation of the shallow neutral donor, as indicated by the extra feature near B = 3445 G in Fig. 1, greatly strengthens the proposed scenario. Detection of the EPR signal indicates that the donor captured an electron from the conduction band. Since there would be no conduction band electrons in this heavily compensated material without illumination and only sub-bandgap light was used, the electron captured by the donor must have originated from the acceptor. Thus, the photo-excitation occurs from the defect to the conduction band, not valence band. Several features of the data indicate that the electron originated from the carbon-related acceptor. First, we note that the threshold for both excitation and quenching for the donor are the same as those found for the C-related center. Figures 4(a) and 4(b), which show the photo-excitation and the quenching data, illustrate the point. The data were obtained from sample G, where the closed triangles represent the C-related signal and the open stars represent the neutral donor. The overlap of the data is clear, indicating that the same process drives donor and acceptor excitation/quenching. The time dependence of the EPR signal intensity during illumination adds additional support to the model. Although not shown here, the time scale for the excitation and quenching processes at each wavelength is similar for the donor and carbon-related center. Finally, we note that the detection of the donor exclusively in the more lightly doped samples (F and G) also supports the proposed model if one assumes donor/acceptor (DA) pair recombination between the neutral donor and C-related acceptor. According to the model of Thomas, DA pair recombination time is reduced at least an order of magnitude when the concentration of one species becomes significantly larger than the other due to a reduction in the distance between donor and acceptor.22 Thus, in the heavily doped samples, all the donors and an equivalent amount of C-related defects would be ionized (not EPR active) within the EPR detection time of milliseconds. The observed C-related centers are those with concentration in excess of the donor species.

FIG. 4.

Steady state photo-EPR data obtained for the carbon-related center (closed triangles) and neutral donor (open stars) in the 2 × 1017 cm−3 C-doped sample: (a) excitation and (b) quenching. The dashed lines denote excitation and quenching thresholds.

FIG. 4.

Steady state photo-EPR data obtained for the carbon-related center (closed triangles) and neutral donor (open stars) in the 2 × 1017 cm−3 C-doped sample: (a) excitation and (b) quenching. The dashed lines denote excitation and quenching thresholds.

Close modal

Release of an electron by the acceptor and subsequent capture by the donor, as discussed above, suggests that the intensity of the photo-induced signal for the carbon-related center and the donor to be the same during illumination. However, as indicated by the axes in Fig. 4, the maximum donor concentration is 2.4 × 1015 cm−3, while the concentration of the C-related centers is 1 × 1016 cm−3. The apparent discrepancy could be accounted for by capture of an electron onto the Fe3+ as well as intrinsic, non-EPR active donors known to plague GaN. Indeed, 1015 cm−3 Fe3+ are removed during the photo excitation, suggesting that some of the electrons are captured by Fe.

To summarize, the dependence of the EPR signal intensity on carbon concentration, measured both before and after illumination, implies that the resonance represents carbon-related defects, some of which are compensated by donors prior to illumination. The photo-EPR excitation threshold, coupled with the observation of the shallow donor, indicate that photon energy greater than 2.75 ± 0.05 eV excites an electron from the negative acceptor to the conduction band. In the reverse process, the excitation of an electron from the valence band to defect (quenching) is supported by several experimental and theoretical studies. The most pertinent are the temperature-dependent Hall measurements performed on the samples studied here. The Hall results indicated hole conductivity at temperatures above 350 °C, and the data analysis yields an activation energy of 1 eV.9 Thus, as noted in the previous work, the 0.95 ± 0.05 eV quenching threshold is consistent with the Hall data and indicates that the EPR detected carbon-related center is responsible for the high resistivity measured in C-doped GaN. However, as has been pointed out by others, optical thresholds are an exact reflection of a defect level if and only if lattice interactions are neglected.23 Assuming that such interactions are minimized for the quenching process, 1 eV may be considered a first approximation for the defect level. The exact value, however, awaits analysis of the time dependent photo-EPR data and associated optical absorption spectrum with a model which incorporates lattice interactions.

While the EPR analysis limits identification of the spectrum to a carbon-related defect, comparison with theoretical predications for carbon centers in GaN hints at a more precise description. Of the types of carbon-related centers investigated in GaN, the one predicted to have a defect level closest to the 1 eV quenching threshold is the neutral-to-negative charge transition of CN.6,13,14 Thus, assuming 1 eV is a reasonable approximation of the defect level of the EPR detected center, we suggest that the carbon-related defect observed here represents CN.

In summary, on studying a set of seven GaN substrates with carbon concentrations varying from 2 × 1017 to 1019 cm−3, we conclude that the EPR spectrum represents a carbon-related acceptor with optical thresholds at 2.75 and 0.95 eV. Detection of the shallow donor simultaneously with the carbon-related center confirms that the former relates to excitation of an electron from the defect to the conduction band. The temperature-dependent Hall data support the assignment of the latter to electron capture from the valence band. Together, the thresholds provide an initial estimate for the defect level of the carbon-related center as approximately 1 eV above the valence band edge.

The work at UAB was supported by the National Science Foundation, NSF/DMF 1606765. This research in Poland was supported by the National Center for Research and Development PBS3/B5/32/2015 and the Department of the Navy, Office of Naval Research (ONRG-NICOP-N62909-17-1-2004).

1.
J. B.
Webb
,
H.
Tang
,
S.
Rolfe
, and
J. A.
Bardwell
, “
Semi-insulating C-doped GaN and high-mobility AlGaN/GaN heterostructures grown by ammonia molecular beam epitaxy
,”
Appl. Phys. Lett.
75
,
953
(
1999
).
2.
S.
Tomiya
,
T.
Hino
,
S.
Goto
,
M.
Takeya
, and
M.
Ikeda
, “
Dislocation related issues in the degradation of GaN-based laser diodes
,”
IEEE J. Sel. Top. Quantum Electron.
10
,
1277
(
2004
).
3.
I. C.
Kizilyalli
,
A. P.
Edwards
,
O.
Aktas
,
T.
Prunty
, and
D.
Bour
, “
Vertical power P-n diodes based on bulk GaN
,”
IEEE Trans. Electron Devices
62
,
414
(
2015
).
4.
M.
Bockowski
, “
High nitrogen pressure solution growth of GaN
,”
Jpn. J. Appl. Phys., Part 1
53
,
100203
(
2014
).
5.
J. L.
Lyons
and
C. G.
Van de Walle
, “
Computationally predicted energies and properties of defects in GaN
,”
Npj Comput. Mater.
3
,
12
(
2017
).
6.
J. L.
Lyons
,
A.
Janotti
, and
C. G.
Van de Walle
, “
Effects of carbon on the electrical and optical properties of InN, GaN, and AlN
,”
Phys. Rev. B
89
,
035204
(
2014
).
7.
S.
Heikman
,
S.
Keller
,
S. P.
Denbaars
, and
U. K.
Mishra
, “
Growth of Fe doped semi-insulating GaN by metalorganic chemical vapor deposition
,”
Appl. Phys. Lett.
81
,
439
(
2002
).
8.
A. Y.
Polyakov
,
N. B.
Smirnov
,
A. V.
Govorkov
,
V. I.
Vdovin
,
A. V.
Markov
,
A. A.
Shlensky
,
E.
Prebble
,
D.
Hanser
,
J. M.
Zavada
, and
S. J.
Pearton
, “
Properties of Fe-doped, thick, freestanding GaN crystals grown by hydride vapor phase epitaxy
,”
J. Vac. Sci. Technol. B
25
,
686
(
2007
).
9.
M.
Iwinska
,
R.
Piotrzkowski
,
E.
Litwin-Staszewska
,
T.
Sochacki
,
M.
Amilusik
,
M.
Fijalkowski
,
B.
Lucznik
, and
M.
Bockowski
, “
Highly resistive C-doped hydride vapor phase epitaxy-GaN grown on ammonothermally crystallized GaN seeds
,”
Appl. Phys. Express
10
,
011003
(
2017
).
10.
H.
Teisseyre
,
M.
Bockowski
,
I.
Grzegory
,
A.
Kozanecki
,
B.
Damilano
,
Y.
Zhydachevskii
,
M.
Kunzer
,
K.
Holc
, and
U. T.
Schwarz
, “
GaN doped with beryllium—An effective light converter for white light emitting diodes
,”
Appl. Phys. Lett.
103
,
011107
(
2013
).
11.
M. J.
Uren
,
M.
Caesar
,
S.
Karboyan
,
P.
Moens
,
P.
Vanmeerbeek
, and
M.
Kuball
, “
Electric field reduction in C-doped AlGaN/GaN on Si high electron mobility transistors
,”
IEEE Electron Device Lett.
36
,
826
(
2015
).
12.
T.
Suski
,
E.
Litwin-Staszewska
,
P.
Perlin
,
P.
Wisniewski
,
H.
Teisseyre
,
I.
Grzegory
,
M.
Bockowski
,
S.
Porowski
,
K.
Saarinen
, and
J.
Nissilä
, “
Optical and electrical properties of Be doped GaN bulk crystals
,”
J. Cryst. Growth
230
,
368
(
2001
).
13.
D. O.
Demchenko
,
I. C.
Diallo
, and
M. A.
Reshchikov
, “
Yellow luminescence of gallium nitride generated by carbon defect complexes
,”
Phys. Rev. Lett.
110
,
087404
(
2013
).
14.
A. F.
Wright
, “
Substitutional and interstitial carbon in wurtzite GaN
,”
J. Appl. Phys.
92
,
2575
(
2002
).
15.
J. L.
Lyons
,
A.
Alkauskas
,
A.
Janotti
, and
C. G.
Van de Walle
, “
First-principles theory of acceptors in nitride semiconductors
,”
Phys. Status Solidi Basic Res.
252
,
900
(
2015
).
16.
W. R.
Willoughby
,
M. E.
Zvanut
,
S.
Paudel
,
M.
Iwinska
,
T.
Sochacki
, and
M.
Bockowski
, “
A compensating point defect in carbon-doped GaN substrates studied with electron paramagnetic resonance spectroscopy
,”
J. Appl. Phys.
123
,
161547
(
2018
).
17.
J. A.
Weil
and
J. R.
Bolton
,
Electron Paramagnetic Resonance: Elementary Theory and Practical Applications
, 2nd ed. (
John Wiley & Sons, Inc
.,
Hoboken
,
2010
).
18.
E.
Malguth
,
A.
Hoffmann
,
W.
Gehlhoff
,
O.
Gelhausen
,
M. R.
Phillips
, and
X.
Xu
, “
Structural and electronic properties of Fe3+ and Fe2+ centers in GaN from optical and EPR experiments
,”
Phys. Rev. B
74
,
165202
(
2006
).
19.
W. E.
Carlos
,
J. A.
Freitas
,
M. A.
Khan
,
D. T.
Olson
, and
J. N.
Kuznia
, “
Electron-spin-resonance studies of donors in wurtzite GaN
,”
Phys. Rev. B
48
,
17878
(
1993
).
20.
A.
Wolos
,
Z.
Wilamowski
,
M.
Piersa
,
W.
Strupinski
,
B.
Lucznik
,
I.
Grzegory
, and
S.
Porowski
, “
Properties of metal-insulator transition and electron spin relaxation in GaN:Si
,”
Phys. Rev. B
83
,
165206
(
2011
).
21.
J. A.
Freitas
,
J. C.
Culbertson
,
N. A.
Mahadik
,
T.
Sochacki
,
M.
Iwinska
, and
M. S.
Bockowski
, “
HVPE GaN wafers with improved crystalline and electrical properties
,”
J. Cryst. Growth
456
,
113
(
2016
).
22.
D. G.
Thomas
,
J. J.
Hopfield
, and
W. M.
Augustyniak
, “
Kinetics of radiative recombination at randomly distributed donors and acceptors
,”
Phys. Rev.
140
,
A202
(
1965
).
23.
A.
Alkauskas
,
M. D.
McCluskey
, and
C. G.
Van De Walle
, “
Tutorial: Defects in semiconductors—Combining experiment and theory
,”
J. Appl. Phys.
119
,
181101
(
2016
).