Annealing improvement on the localized states of plasma grown boron nitride film assessed through admittance measurements

Article abstract of DOI:10.1016/j.jallcom.2008.08.014

Boron nitride (BN) thin film was grown by plasma enhanced chemical vapor deposition (PECVD) technique and was investigated by UV-Visible transmission, Fourier transform infrared (FTIR) and ac conductance spectroscopies. Mainly the density of electronic lo

Full text of DOI:10.1016/j.jallcom.2008.08.014

Journal of Alloys and Compounds 475 (2009) 794–803
Contents lists available at ScienceDirect
Journal of Alloys and Compounds
journal homepage: www.elsevier.com/locate/jallcom
Annealing improvement on the localized states of plasma grown boron
nitride film assessed through admittance measurements
Orhan Özdemira,∗, Mustafa Anutganb, Tamila Aliyeva-Anutganb,
b
b
˙
a
Ismail Atilgan , Bayram Katirciog˘lu
˙
Yildiz Technical University, Physics Department, Davutpas¸a, Istanbul, Turkey
b Middle East Technical University, Physics Department, Balgat, Ankara, Turkey
a r t i c l e i n f o
a b s t r a c t
Article history:
Received 8 April 2008
Received in revised form 1 August 2008
Accepted 7 August 2008
Available online 23 September 2008
Boron nitride (BN) thin film was grown by plasma enhanced chemical vapor deposition (PECVD) technique
and was investigated by UV–Visible transmission, Fourier transform infrared (FTIR) and ac conductance
spectroscopies. Mainly the density of electronic localized states (Dit) at BN/Si interface was obtained
by continuum and statistical models of ac conductance through an MIS structure (Al/BN film/Si). The
origins of the electronic defects have been outlined and discussed within the frame of a nitrogen deficient
turbostratic structure where more or less parallel hexagonal crystallites of distributed size would be
embedded in a disordered phase. The nitrogen deficiency of the film was tried to be restored by annealing
treatment under nitrogen atmosphere at two temperatures.
Keywords:
Turbostratic boron nitride
Admittance
© 2008 Elsevier B.V. All rights reserved.
DOS
Annealing
Dangling bond
Improvement
1. Introduction
investigation [11,12]. For that purpose, metal–semiconductor (MS)
and metal–insulator–semiconductor or metal (MIS or MIM) devices
The boron nitride compound (BN) is a polymorphous mate-
rial. Its most stable and natural phase being hexagonal form
(h-BN) is relatively easier to deposit in thin film form by plasma
enhanced chemical vapor deposition (PECVD) system which is
widely used for production of large area low cost opto-electronic
materials [1]. In addition, h-BN film is a potential material for
thin film opto-electronic applications [2]. Its possible negative
electron affinity (NEA) allows some applications like displays,
detectors, electron microscopes, cold cathode emitters etc. [3,4].
Moreover, its chemical inertness and dielectric behavior render it
a candidate for passivation or insulator layer in both VLSI circuits
assigned to reactive media and high-speed integrated circuits of
the metal–insulator–semiconductor structure [5–9].
h-BN films grown by PECVD upto nowadays are very defective
[1,10]. Huge density of localized electronic states (DOS), some of
which are related to the stoichiometric problems, constitutes a seri-
ous threat to the possible applications of h-BN films despite their
low cost and easy production. These localized states are studied
by admittance measurements on a device involving the film under
must be available for simplicity of both production and interpreta-
tion of the measured data.
The charges of the localized states within h-BN film may be
changed by storing free charge carriers on the electrode by apply-
ing externally dc and ac bias voltages on MIM and MIS structures.
These charges are followed by admittance measurements (both
capacitance and conductance components). MIM structure has the
advantage over MIS one in its ability to measure exclusively the
geometric capacitance of the dielectric film, avoiding the con-
tribution of the semiconductor; this geometric capacitance may
be used to obtain the film thickness if its dielectric constant is
known or vice versa. On the other hand, MIS structure is prefer-
able over MIM one since the second electrode (semiconductor) may
constitute a reference for determining the charge state of the sys-
tem at flat band voltage (VFB). The capacitance CFB corresponding
to this reference voltage VFB may be determined experimentally
and the amount of charges in the film (bulk and surface) is then
evaluated from VFB. The Fermi level of the semiconductor may
scan an energy interval corresponding approximately to its half
gap (here for p-type silicon) from midgap to about the valence
band edge by changing the gate electrode voltage, VG. Conse-
quently, the distribution of DOS may be obtained along this energy
interval.
∗
Corresponding author. Tel.: +90 212 383 4279.
E-mail address: ytu.orhan@gmail.com (O. Özdemir).
0925-8388/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jallcom.2008.08.014

O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
795
In this work, mainly an MIS structure was used since BN film is
relatively resistive and then may be taken as a dielectric. The admit-
tance measurements are utilized in the analysis of DOS in BN films
deposited on p-type silicon crystal. In this frame, the continuum
model [12] was first used for retrieving the DOS at the BN film/Si
interface. The statistical model was parallelly used for determining
whether or not the interface local charges are inhomogeneously
distributed [12].
2. Preparation and measuring tools
Two (1 0 0) oriented p-type silicon wafers (50 and 2 ꢀ cm) for
FTIR and the admittance analyses, respectively, were cleaned with
standard RCA (Radio Corporation of America) cleaning procedure.
Quartz plates, glass and aluminum (Al) coated glass microscope
slides (cleaned by boiling in degreasing agent and dipping in diluted
hydrogen peroxide (H2O2) and trichloroethylene (TCE) solution)
were used as substrates for UV–Visible transmission, mechani-
cal thickness and admittance measurements, respectively. Prior to
placing the substrates on the grounded lower electrode of PECVD
system (␮P 80, Plasmalab), backside of the substrate for MIS struc-
ture was coated by Al and the sample was annealed in nitrogen
ambient at 530 ◦C to form ohmic contact, followed by front side
cleaning via floating on diluted hydrofluoric acid (HF) solution.
Then, BN film was deposited by 13.56 MHz PECVD technique, using
0.55 W/cm2 RF power, flow rates of 20 cm3/min ammonia (NH3)
and 4.5 cm3 diborane (B2H6, diluted 15% in Hydrogen, H) under
pressure of 0.5 Torr at 300 ◦C. After deposition, Al as gate electrode
was evaporated through a shadow mask of diameter 0.15 cm to
fabricate MIS (Al/BN/p-c-Si/Al) and MIM (Al/BN/Al coated glass)
structures, respectively. The thickness of BN film is determined as
375 nm by a profiler (Ambios XP-2), and hence deposition rate is
determined to be 4.2 nm/min. The film was annealed under nitro-
gen atmosphere for 90 min at two different temperatures (475 and
650 ◦C).
Fig. 1. Transmission spectrum of as-grown BN film in the UV–Visible region. Inset
(i) shows the corresponding (˛hꢁ)1/2 and ˛hꢁ vs. hꢁ plots for the estimations of band
gap Eg and Urbach energy E0, respectively. Weak absorption (a), medium absorption
(b) and high absorption (c) regions are also indicated. Inset (ii) gives out Eg, E0 and
refractive index for the as-grown and 475 and 650 ◦C annealed samples.
By means of optical transmission measurements obtained by a
UV–Visible spectrometer (␭2S, PerkinElmer), refractive index (n),
optical energy gap of the film (EG), and characteristic energy width
of the tail states (=Urbach energy, E0) were obtained. On the other
hand, measurement of IR spectrum (FTIR, Nicolet 520) gives infor-
mation on the film structure since each phase of BN gives distinct
vibrational absorption spectrum. Finally, distributions of localized
DOS, main contribution of this work, were investigated via admit-
tance spectroscopy performed by an LCR meter (HP 4192A) within
a frequency range from 3 kHz to 100 kHz through the MIS structure.
semiconductors. The energy band gap slightly increases from 5.20
to 5.30 eV with increasing annealing temperature (Fig. 1). These
values are close to the gap of h-BN [13]. Finally, the log(˛hꢁ) vs.
(hꢁ) graph was plotted to see the degree of tail state density from
the inverse slope of the linear fit (i.e., Urbach energy, E0). E0 is
found to increase gradually from 0.17 towards 0.20 eV as annealing
temperature increases (Fig. 1).
Infrared spectra of PECVD grown BN film is depicted in Fig. 2. In
Fig. 2(a) peaks around 795 cm−1 (B-N-B out-of-plane bending) and
broad one around 1388 cm−1 (B-N stretching) indicate that films at
hand contain poorly crystalline hexagonal phase, i.e., turbostratic
BN (t-BN) [14]. Broad peak at 1388 cm−1 has a shoulder which may
be assigned to B-O-N vibration mode [15]. Besides, with increasing
annealing temperature B-O stretching mode around 1250 cm−1 can
be detected. In Fig. 2(b) it is clearly seen that with annealing, hydro-
gen evolves from the film. Using calibration constants for B H and
3. Results
3.1. Optical characterization
Optical transmittance measurements were taken in the
UV–Visible region (200–1100 nm) as depicted in Fig. 1. The opti-
cal path, n·d of the light within the BN films was deduced from the
fringes of the transparent region of the transmittance spectrum.
Then, the refractive index, n was found by using the previously
estimated thickness, d in the optical path. The refractive index
N
H bonds [16,17], a decrease of hydrogen content is calculated
(inset of Fig. 2(b)).
was found about 1.75 and remaining almost the same after the
3.2. Admittance analysis
√
annealing treatments. Moreover, coinciding n (≈ ε where ε =
permittivity) was obtained as 1.78 when compared with the one
obtained through capacitance–voltage (C–V) measurement at accu-
mulation for the highest frequency, i.e., C = εA/d with A = electrode
area. The absorption coefficient as a function of incident photon
energy, ˛(hꢁ) was calculated in the absorption region of the spec-
trum. The optical gap was found from the energy axis intercept of
the linear fit in the (˛hꢁ)1/2 vs. hꢁ graph, adequate for disordered
Admittance spectroscopy consists of measuring capacitance, Cm
together with conductance, Gm as a function of various parameters
such as dc bias voltage VG, ac voltage modulation frequency, ω. It has
been extensively used for investigating the electronic properties of
MIS structure due to its charge sensitivity. The measurements of
Cm and Gm/ω of MIS (Al/BN film/Si) capacitor vs. VG at room tem-
perature under various ω are given in Fig. 3a. Before proceeding

796
O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
Fig. 2. (a) B-N-B out-of-plane (op) bending and B-N in-plane (in) stretching modes: characteristic peaks of sp2 phase. Possible oxygen containing modes are indicated. (b)
Hydrogen containing bonds: B H out-of-plane bending, B H and N H stretching. Inset shows the calculated hydrogen atom density as a function of annealing temperature.
further, let us resume the basic points of the admittance analy-
ses considering both capacitance and conductance components for
better seizing the proposed interpretations.
located within the film and/or at the film/substrate interface can
be estimated [11,12,18,19]
Qit + Qf ≈ VFBCBN
(1)
Three distinct regions are evident when Cm–VG measurement
is performed: at high negative bias, accumulation of majority car-
riers is formed at the interface (holes for p-type silicon) and this
dc bias independent capacitance value CBN is used to obtain the
insulator film geometric capacitance after correcting the eventual
effect of series resistance (Rs). When applied gate bias is reduced
(i.e., dc gate bias scanned from accumulation towards more positive
values), the bias dependent capacitance region defines the second
bias regime as depletion. Further change of dc gate bias towards
more positive values accumulates minorities at the interface and
forms the third bias independent capacitance region, namely inver-
sion. The inversion capacitance in this work, as given in Fig. 3a,
remains below the accumulation one since, despite the availability
of minorities at the interface for dc bias, their ac modulation does
not occur within the 3–100 kHz frequency interval, due to relatively
lowergeneration/recombinationrates oftheminorities; instead the
required ac charge modulation occurs at the depletion edge and the
overall capacitance is the series combination of the film and silicon
depletion capacitances.
Although Cm and Gm/ω branches of the measured admittance
Ym(=Gm + jωCm) may be equally used for extracting interface state
density, the conductance component is preferable since it is more
accurate and sensitive due to its peak behavior. However, inter-
face conductance effects are not detectable in the accumulation and
strong inversion regions due to the large amount of stored charge
carriers. Conductance peaks along the depletion region, due to the
exchange of majority carriers of low quantity, between the interface
localized states and the relevant band are preferable for easiness of
interpretation (here, the conductance is due to the time delay exist-
ing between the demand and supply of the majority carriers during
the exchange).
The interface trap admittance Yp(=Gp + jωCp) and specially Gp/ω
may be retrieved from the experimentally measured admittance
Ym(=Gm + jωCm), first, by removing the contribution of both the
dielectric film capacitance, CBN and the series resistance, Rs accord-
ing to the equivalent circuit given in Fig. 3b. For that purpose, the
expressions of CBN and Rs are deduced from the high frequency
measured accumulation admittance Yma(=Gma + jωCma) (which is
approximately the inverse of the series combination of these two
impedances (1/jωCBN) and Rs [12]):
From the accumulation to the depletion, the polarity of silicon
band bending q changes, being flat ( s = 0) at V (≡VFB) = 0 for
s
G
ideal structure. However, in practical conditions, VFB =/ 0, due to
the existence of localized charges both in the bulk of the dielectric
(Qf) and at the interface (Qit) apart from a negligibly small effect of
work function difference between gate metal and silicon materials.
In other words, these charges shift the Cm–VG curves along the volt-
age axis. Consequently, the determination of VFB is crucial for the
overall evaluation of local charges. From the magnitude of the flat-
band shift of Cm–VG curve, roughly the effective density of charges
ωCB2N(Gm − ω2Cm2 Rs − RsGm2
)
Gp
ω
=
(2)
2
(ω2CBNRsCm − Gm
)
2 + ω2(CBN − Cm − CBNRsGm
)
with Rs = Gma/(Gm2 a + ω2Cm2 a) and CBN = Cma(1+(Gma/ωCma)2).
The retrieved Gp/ω vs. dc gate voltage and frequency curves are
given in Fig. 3c and d respectively.

O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
797
Fig. 3. (a) Admittance (Cm and Gm/ω) as a function of gate bias (VG) at room temperature on both Al/BN/p-Si/Al MIS and Al/BN/Al MIM structures, respectively. (b) Equivalent
circuit used for retrieving the interface state density Dit. (c) Gp/ω as a function of VG for various time frequencies after insulator film capacitance (CBN) and series resistance
(Rs) compensation. (d) Gp/ω vs. time frequency f for several bias voltages (corresponding to the depletion and onset of weak inversion regimes) after insulator film capacitance
(CBN) and series resistance (Rs) compensation.
where ꢂp−1 = hole emission/capture time = ꢃpvthNA exp(q s/kT)
The localized electronic states around the BN/Si interface should
be distributed throughout the BN energy band gap as outline in
Section 4. When ac gate voltage modulation is applied for admit-
tance measurement, a modulation of the Fermi level position at the
interface will be induced between E and E + dE and then the occu-
pancy of a certain amount Dit(E) dE of interface traps is affected
with Dit(E) = number of interface traps/(area·energy). This leads
to the following expression of Gp/ω for the so-called continuum
model:
with ꢃp = trap capture cross section,
vth = thermal velocity,
NA = majority doping density, s = silicon interface band bending.
Dit(E) around the interface level E may be evaluated from Gp/ω peak
value:
(Gp/ω)max
0.402q2A
Dit
=
(4)
The energy distribution of Dit(E) may be obtained by changing
the gate voltage VG along the depletion region where the probe trap
energy ET, defined by the interface Fermi level EF and measured
from the valence band edge EV as ꢄE is scanned from the midgap
q2Dit
Gp
=
A ln(1 + ω2ꢂp2)
2ωꢂp
(3)
ω

798
O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
4. Discussion
The PECVD deposited BN film, at least a relatively thin interfa-
cial layer, seems extremely defective as probed by the admittance
spectroscopy. In this section, the structural origins of the electronic
defects have been outlined and discussed.
4.1. Atomic configuration and energy band structure
One of the approaches for studying bonds is to consider the
bonds as “molecular” orbitals which are “built” of overlapping
atomic orbitals. Both for boron and nitrogen, 3 equivalent (hybrid)
orbitals may be obtained by combining one 2s-orbital with two 2p-
orbitals (planar 2px and 2py) as three sp2 hybrid orbitals (Fig. 5i).
Three planar sp2-like ␴ bonds are formed around each of the com-
ponents. An unhybridized free pz atomic orbital remains on each
component, which are perpendicular to the plane of the three sp2
orbitals (Fig. 5iv). Due to the greater nuclear charge of nitrogen,
the nitrogen orbital (pz)N is lying lower in potential energy than
that of the boron (pz)B. This is schematically depicted in Fig. 6d.
Consequently, (pz)N is expected to be occupied by two electrons (of
opposite spin) and (pz)B remains empty (it is assumed here that one
of the 2s2 electrons of nitrogen atom participates in sp2 bonding,
the other is excited to (pz)N orbital). In other words, the B N bond
is slightly polar (B+ı←→ N−ı), leading to a finite dipole moment
Fig. 4. Distribution of effective density of states (Dit) along the half energy gap of Si
obtained by continuum and statistical ac conductance models: for as-grown (open
and filled circles) and annealed (open and filled rectangle) at 475 and 650 ◦C (open
and filled triangular) films, respectively.
d
ꢆ = dı.
The possibility of double bonding in BN might be speculated as
follows [20]. The two (pz)N electrons around the nitrogen as shown
in Fig. 5iv might be partially available for ␲-bonding, by the partial
transfer of (pz)N electrons to (pz)B at the boron site, without totally
compensating the transfer along the sp2 ␴ bond from boron to
nitrogen due to the differences in electronegativity between boron
and nitrogen. However, this double bond character could not be
comparable to that of a carbon–carbon double bond [21]. Therefore,
the sharing of nitrogen (pz)N electron pair in the bond building
remains limited and depicted in Fig. 5iv.
to the proximity of valence band edge:
ꢄE = EV − EF = q s + kT ln(NA/ni)
(5)
where n = silicon intrinsic carrier concentration. Here and NA
s
i
are evaluated from the measured high frequency Cm–VG curve with
Cm = CDCBN/(CD + CBN); CD = εsA/W; s = qNAW2/2εs; W = depletion
width. The experimentally determined distributions of Dit(E) are
reported in Fig. 4.
If the eventual traps are inhomogeneously distributed through-
out the interface plane, the above continuum model supplies wrong
Dit(E) values since an inhomogeneous distribution of interface traps
would lead to a distribution of interface silicon band bending s.
The statistical model was developed taking into account this phe-
The above described boron–nitrogen combination where each
component atom of one sort has three planar trigonal sp2 ␴ bonds
120◦ apart to three atoms of second sort, may easily lead to six sided
planar ring structure.
These six-fold B3N3 rings have the ability to be arranged into a
hexagonal 2-dimensional infinite layers which are held together by
weak Van der Waals or dipolar interaction. As depicted in Fig. 5vi,
these hexagons are packed directly on top of each other where a
boron atom in one layer remains closest to each corresponding
nitrogen atom in the neighboring layers with ABAB. . . sequence
of planes. However, omitting the diatomic structure, sequence of
hexagonal lattice is as AAAA. . ., exhibiting the second difference
apart from the polarity of bonds from the packing of graphite where
the hexagons are shifted by one bond length along the neighbor-
ing planes and leading to ABAB. . . sequence. The analysis so far
establishes the theoretical basis of hexagonal boron nitride (h-BN)
crystals. Experiments have carried out that h-BN is the only sta-
ble phase found in nature although BN may be forged in, more or
less stable, polymorphous microscopic arrangements (i.e., rhom-
bohedral, cubic, and wurtzite symmetries [22]) in the practical
laboratory conditions.
Instead of the ideal single crystalline forms, the BN compound
appears practically at various (or microcrystalline or nanocrys-
talline) structures. In this respect, the qualification turbostratic is
commonly used where hexagonal crystallites of distributed sizes
are embedded in a disordered phase. Each crystallite may be more
or less perfect; most of them may contain a varying number of point
defects such as vacancies or dangling bonds. When BN material
nomenon where a Gaussian distribution P( s) of around its
s
average ꢀ sꢁ was assumed with ꢃs being the variance of band bend-
ing in terms of kT/q [12] and Gp/ω turns into ꢀGpꢁ/ω:
ꢀGpꢁ
= q2DitfDA(ꢃs)
(6)
ω
with
ꢁ
ꢂ
ꢀ
+∞
(2 ꢃs2)−1/2
Á2
2ꢃs2
fD(ꢃs) =
exp
−
exp(−Á)
2ꢅ
−∞
× ln(1 + ꢅ2 exp(2Á)) dÁ
(7)
where Á = us − u¯ s (with us = eq /kT ), ꢅ = ωꢂp and ꢅp is the value of
ꢅ corresponding to the maximum of the Gp/ω vs. ln ω curve.
From the experimentally obtained ꢀGpꢁ/ω vs. ln(ω) curve
(Fig. 3d), the ratio of two ꢀGpꢁ/ω values at two discrete frequen-
cies (peak frequency ωp and 5 ωp or 1/5 ωp for example [12]) is
evaluated and this ratio is used to find ꢃs by solving numerically
the relation (7). Finally, Dit is determined from the relation (6) at
the ꢀGpꢁ/ω vs. ln(ω) peak for the gate voltage VG at hand. Dit(E)
vs. ET–Ev distribution is obtained (Fig. 4) by the same procedure,
outlined previously for continuum model.
s

O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
799
Fig. 5. Schematic illustration of BN atomic organization from atomic orbitals to a turbostratic structure.
is produced in thin film form at low temperature without special
precautions, a turbostratic structure is unavoidable [23]. Relatively
far from the substrate, these sp2 ␴ bonded hexagonal BN planar
crystallites, which are relatively defect free, can have random ori-
entations (i.e., rotated more or less around c-axis). These layers
of planar hexagonal grains may not be perfectly parallel to each
other and the ABAB. . . sequence of layers might be no longer main-
tained. However, the “grain” boundary regions might be extremely
defective (a typical diagram depicting the turbostratic structure is
given in Fig. 6vii). The degree of porosity and contamination by
impurities may, of course, depend on the production conditions
and surroundings and then the resulting physical properties are
affected. At the extreme limit, this turbostratic structure may be
qualified as “amorphous” where the coherence length is reduced
to around the interatomic distance. During the film growth, at the
initial stage, due to mainly the effects of substrate surface, the first
layer on the substrate of thickness 2–3 nm may be amorphous. This
layer might contain defects of all sorts leading to a huge density
of electronic states which are expected to be distributed in energy
from valence band to the conduction band.
The three dimensional tight-binding (TB) band structure for h-
BN has been calculated [24] by transferring the TB interactions
from the analogous and better understood graphite. The results
are partially reported in Fig. 6a and b from ref. [24]. These fig-
ures are in reasonable agreement with the experimental electronic
structure extracted from X-ray and photoemission measurements
[24]. Contrary to the graphite without forbidden gap, h-BN exhibits
a net direct gap at the Brillouin Zone (BZ) boundary along the

800
O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
Fig. 6. (a) Composition of density of states around the band edges for both valence and conduction bands [24], (b) energy vs. quasimomentum along the indicated direction
in the Brillouin Zone for both valence and conduction bands [24], (c) first BZ of h-BN structure (where energy gap occurs at the BZ boundary KPH [24]), and (d) schematic
representation of atomic orbital interaction for producing both valence and conduction bands.
quasimomentum line KPH (Fig. 6c). Moreover, the valence band
region around the maximum is eventually constituted by (pz)N
orbital dominant bonding ␲ states, while the region of the conduc-
tion band around minimum is rather constituted by (pz)B orbital
dominant anti-bonding ␲ states (Fig. 6a). This behavior might be
qualitatively depicted in Fig. 6d.
ber of states remaining conserved, the width of the density of
states (DOS) of a band may be broadened more or less on either
side depending on the strength of the local perturbations.
The overall amount of a trapping potential, reflecting the
strength of local distortion, is related to the defect formation
energy such that larger defect formation energy decreases the
thermodynamical existence probability of this defect.
*
Consequently, the density of localized states, N(E) vs. E should
decrease (probably exponentially) towards the energy gap as a
tail from the relevant band edge, N(E) = Nedge exp(−|E|/E0) with
E0 being the so-called Urbach energy, interpreted as disorder
parameter. This is depicted in Fig. 7.
4.2. Eventual distribution of localized states throughout the gap
In real cases (Fig. 5vii), there are unavoidable deviations from the
ideal configurations leading to localized states, existing within lim-
ited regions of the material. These localized states may be roughly
divided into two classes: (i) tail states and (ii) deep states. The
energy levels of these states are expected to be distributed in energy
due to the possible differences in the surrounding of each state.
(ii) The coordination defect states, mainly related to the broken
or dangling bonds (DB) constitutes a second family of deep
traps. As the surrounding of each atomic site might be different
in this randomly distributed hexagonal BN clusters (Fig. 5vii),
the density of energy levels associated with these DBs of each
component, exhibits Gauss-like distribution within the energy
gap. In other words, the boron and nitrogen related two dif-
(i) Bond angle and bond length distortions lead to a local trapping
potentials, each one in its turn, localizes one of the band energy
levels E, measured from the relevant band edge. The total num-

O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
801
Fig. 7. Expected distribution of localized states across the (pseudo) gap of tur-
bostratic or amorphous BN film.
ferent Gaussians are located near the conduction and valence
bands respectively following the atomic orbital picture, shown
in Fig. 6d. The eventual Fermi level should be pinned between
these two distributions, defining the unoccupancy and occu-
pancy of them respectively. Taking into account the amphoteric
nature of each sort of DBs, the doubly occupied sites lead to
a shifted Gauss-like DOS distribution due to the correlation
energy. Thus, these DB distributions together with the localized
DOS of weakened bonding (valence band tail) and anti-bonding
states (conduction band tails) should cover the whole forbidden
gap at least within defective “amorphous” layer neighboring the
substrate.
Fig. 8. Kinetics of the high frequency Cm vs. VG shift under the stress of 10 V gate
voltage. The whole Cm vs. VG shift in time was followed by the gate voltage VHH
corresponding to the half value of the capacitance step.
VHH symbolizes the shift of whole C–V curve along the voltage axis
and reflects the injected amount of charges into the BN film. For a
very short time (less than 0.2 s.), voltage stress sequence was inter-
rupted and the resulting C–V curve was fitted with cubic spline
regression approach and the corresponding new VHH(t) was deter-
mined and applied to verify CHH. The cycle was resumed till stopped
[25]. Fig. 8 carries out that the annealing treatment seems to reduce
substantially the localized state density.
Opposite to both strong accumulation and inversion voltage
regions, along the depletion region a large frequency dependence
has been observed for the measured admittance (Fig. 3a). In other
words, under gate voltage modulation of frequency ω, the local-
ized states within the BN film, more or less distant from the
interface can exchange holes with the silicon valence band by
Shockley–Read–Hall (SRH) statistics in two step mechanism [26]:
first a hole is exchanged between the silicon valence band and the
interface states near the silicon Fermi level and then in series, the
interface state may exchange carriers with traps located inside the
BN film at the distance x from the interface by tunnelling (tun-
neling probability of a carrier at the interface to reach the depth
x ∝ exp(−Kx) where K2 = 2 mU/ꢀ2 with U is the effective potential
depth of the trap to the relevant band edge [11]). For a given cou-
ple of frequency (ω) and temperature (T), there exits a position xt
such that for x < xt the charge of all states can follow the voltage
modulation, for x > xt the opposite occurs and for boundary posi-
tion x = xt the states can only partially answer the modulation with
the following condition:
4.3. Speculation about the experimental results
The gate voltage region where the strong frequency dependent
capacitance steps and the corresponding Gm/ω peaks occur, has
been admitted as the depletion region (Fig. 3a). As the strong accu-
mulation voltage region could not be reached, especially for high
frequencies, due to the voltage limitation of the admittance meter
at hand (−35 to +35 V), the amount of frequency dependence of
its capacitance has not been clearly represented in Fig. 3a. Apart
from the effect of the series resistance Rs, eventual charge injec-
tion from the semiconductor side into the depth of the dielectric
would lead to a strong frequency dependence of the accumula-
tion capacitance [11]. This possibility was tested by the capacitance
measurement on the MIM (Al/BN/Al) structure. The result is shown
in Fig. 3a. The slight frequency dependence, remaining after Rs cor-
rection, points out that the charge injection from the silicon into
the BN film remains within a shallow thin region compared with
the whole film thickness (less than 1% which corresponds to about
2–3 nm). This probable interlayer/facial layer, although relatively
thin, should contain a very large density of localized charge states,
extended throughout the energy gap (one possible origin of these
defects was presented previously in Section 4.2). The detailed study
of charge carrier injection in BN film will be the subject of the next
work; but here a fragment of injection kinetics is reported in Fig. 8.
In the MIS structure at hand, holes and electrons to be injected
are stored at the interface by strong accumulation and inversion
respectively. As predicted by the relation (1), the increasing amount
of injected charges will gradually shift the whole Cm–VG curve along
the voltage axis (negative side for hole injection, positive side for
electron injection). The evolution of specific voltage (denoted as
VHH) through C–VG curves was monitored as follows: initially C–V
curve was measured and fitted to cubic spline. A specific capaci-
tance (CHH), corresponding to VHH, was defined as the middle value
of accumulation and inversion capacitance due to practical reasons
instead of the conventional flat band voltage, VFB. The measured
ωꢂt = 1
(14)
where ꢂt = (ep + cpps)−1 = emission time determined by SRH statis-
tics with ep = emission rate, cp = capture coefficient = ꢃpxvth with
ꢃpx = ꢃp exp(−Kx) = reduced capture cross section of the trap
viewed by a carrier at the interface, ꢃp = capture cross section of
the trap located in the film at a distance x from the interface,
vth = thermal velocity of the free holes, ps = p0 exp(−q s/kT) = free
hole concentration at the interface, p0 = free hole concentration in
the bulk of Si.

802
O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
If Fermi-Dirac distribution holds, ep = cpps the relation (14) leads
in this strained structure, the dangling bond density is reasonably
to:
expected to be increased. Unfortunately, the reduction of the local-
ized density of deep states to an acceptable amount seems avoided
by an accompanying structural change.
ꢃ
ꢄ
q
kT
const.
ω
s
Kxt +
= ln
(15)
The difference between the Dit distributions obtained by both
continuum and statistical models remains within the experimental
error for ‘as-grown’ and 475 ◦C annealed samples which might be
attributed to a relatively homogeneously distributed traps through-
out the interface plane. This result seems in agreement with the
existence of an interfacial BN amorphous layer of 2–3 nm thick
containing the majority of traps measured by the ac conductance
methods. However, a slight difference existing between the Dit dis-
tributions obtained by continuum and statistical models for 650 ◦C
annealing carries out that a partial turbostratic crystallization of
the amorphous interfacial layer occurs and then an inhomogeneous
trap distribution in such turbostratic BN film is expected leading to
the measured statistical distribution of Dit.
On the other hand, the effective density of interface state Dit
may be expressed as:
Dit = Ntxt
(16)
where Nt = number of localized trap per unit volume per unit energy
interval in the BN film. The relation (15) explains the strong fre-
quency dependence of the capacitance along the depletion region;
for a given VG (so a constant s):
- at higher frequency, the charge modulation on the gate electrode
is counterbalanced by the charge modulation at the depletion
edge since xt ≈ 0 and then Dit ≈ 0, leading to:
C
BNCD
Cm
≈
with CD = depletion capacitance
(17)
CBN + CD
- at lower frequency, xt is larger, and hence Dit is dominant and
5. Conclusion
Optical testings of plasma deposited BN thin film pointed out
an oxygen contaminated turbostratic structure (more or less par-
allel and disoriented hexagonal crystallites of distributed sizes
were embedded in a disordered phase). A complete admittance
analysis of BN film through an MIS (Al/BN film/Si) structure was
attempted for the first time. A very large amount of Dit was mea-
sured. The effects of annealing treatments on the Dit were assessed
and interpreted by assuming an amorphous interfacial layer of
2–3 nm thickness. Finally, for the first time, a detailed speculation
about eventual origins of electronic defects was proposed.
Cm ≈ CBN
(18)
- at intermediate frequency, the weighted average of Dit and
depletion modulation is effective for defining the measured
capacitance:
CBNCD
< Cm < CBN
(19)
CBN + CD
The physical origin of the measured Dit at the interface around
the midgap of the BN film may be associated with the dangling
bonds (DB) of both boron and nitrogen atoms. Considering boron
rich nature of BN film at hand (determined by XPS analysis [22]),
boron dangling bonds, (BN)B should be dominant; but this seems in
contradiction with the measured optical gap (≈5.2 eV) [13] indicat-
ing a rather stoichiometric hexagonal BN film. This inconsistency
may be eliminated by taking into account the contamination of
oxygen detected by IR spectroscopy (Fig. 2); these oxygen atoms,
replacing rather nitrogen site would enlarge the optical gap since
it is more electronegative than the nitrogen atom. Consequently
boron abundance and oxygen contamination seem to have com-
peting effects leading apparently to a gap value of stoichiometric
film.
The increase of Urbach energy (E0) by annealing may be
explained by the increase of bond angle and bond length stresses
due to the dehydrogenation effect of annealing; these stresses
would increase the tail state density corresponding to a larger
Urbach energy (E0).
However, on the contrary of the tail state density, the deep
state density should decrease due to the improvement of nitrogen
deficiency during the annealing treatment under nitrogen atmo-
sphere. As a result, with respect to the localized density of states,
the annealing may have opposite effects in the sense that the tail
state density is enhanced while the deep state density is reduced
by partially restoring the nitrogen deficiency and saturating boron
dangling bonds.
Acknowledgements
This work was carried out with the financial support of both the
Turkish Scientific and Technical Research Council (TUBITAK-TBAG-
Project No. 104M195) and State Planning Organization (DPT Project
No. DPT2002K120540-15).
References
[1] J.J. Pouch, S.A. Alterovitz (Eds.), Synthesis and Properties of Boron Nitride, Trans
Tech Publications, 1990, and reference therein.
[2] Z. Remes, M. Nesladek, K. Haenen, K. Watanabe, T. Taniguchi, Phys. Status Solidi
(a) 202 (11) (2005) 2229–2233.
[3] M.J. Powers, M.C. Benjamin, L.M. Porter, R.J. Nemanich, R.F. Davis, J.J. Cuomo,
G.L. Doll, S.J. Haris, Appl. Phys. Lett. 67 (26.) (1995) 3912–3914.
[4] F.J. Himpsel, J.A. Knaap, J.A. VanVechten, D.E. Eastman, Phys. Rev. B 20 (2) (1979)
624–627.
[5] T. Sogino, K. Tanioka, S. Kawasaki, J. Shirafuji, Jpn. J. Appl. Phys. 36 (1997)
L463–L469.
[6] G.R. Gu, et al., Chin. Phys. Lett. 20 (2003) 947.
[7] J. Robertson, M.J. Rutter, Diamond Relat. Mater. 7 (1998) 620–625.
[8] L. Diederich, O.M. Küttel, P. Aebi, E. Maillard Schaller, R. Fasel, L. Schlapbach,
Diamond Relat. Mater. 7 (1998) 660–665.
[9] K. Nose, H. Oba, T. Yoshida, Appl. Phys. Lett. 89 (2006) 112–124.
[10] A. Ahaitouf, A. Bath, P. Thevenin, E. Abarkan, Mater. Sci. Eng. B 77 (2000) 67–72.
˙
[11] Ismail Atılgan, Serhat Özder, Orhan Özdemir, Bayram Katırcıog˘lu, J. Non-Cryst.
Solids 249 (1999) 131–144.
[12] E.H. Nicollian, J.R. Brews, MOS (Metal Oxide Semiconductor) Physics and Tech-
nology, John Wiley and Sons, 1982.
[13] G.A. Battiston, D. Berto, A. Convertino, D. Emiliani, A. Figueras, R. Gerbasi, S.
Viticoli, Electrochim. Acta 50 (2005) 4600–4604.
[14] P.B. Mirkarimi, K.F. McCarty, D.L. Medlin, Mater. Sci. Eng. R21 (1997) 47–100.
[15] S. Su, J. Mol. Struct.-Theochem. 430 (1998) 137.
Although Dit is drastically reduced by an annealing of 475 ◦C
under N2 flow (Fig. 4), a further reduction of Dit has not been
observed for an annealing of 650 ◦C N2 atmosphere; on the contrary,
a slight increase is discernible. This might be interpreted as follows:
the dehydrogenation, by reducing the density of saturated B H and
[16] C. Godet, L. Schmirgeld, L. Zuppiroli, G. Sardin, S. Gujrathi, K. Oxoron, J. Mater.
Sci. 26 (1991) 6408.
[17] Y.B. Park, S.W. Rhee, J. Mater. Sci.: Mater. Electron. 12 (2001) 515.
[18] S.M. Sze, Physics of Semiconductor Devices, John Wiley and Sons, 1969.
[19] Özdemir Orhan, PhD Thesis, Middle East Technical University, 2004.
[20] E.L. Muetterties, The Chemistry of Boron and its Compounds, John Wiley and
Sons Inc., London, 1967.
N
H bonds, may increase the density of deep states due to the dan-
gling bonds, depicted in Fig. 7; in addition, the relaxing hydrogen
in the BN film continues to be evolved when annealing tempera-
ture increases, leaving behind a more stressed structure, and hence

O. Özdemir et al. / Journal of Alloys and Compounds 475 (2009) 794–803
803
[21] I. Bello, Y.M. Chong, K.M. Leung, C.Y. Chan, K.L. Ma, W.J. Zhang, S.T. Lee, A.
[24] J. Robertson, Phys. Rev. B 29 (4) (1984) 2131.
Layyous, Diamond Relat. Mater. 14 (2005) 1784–1790.
[22] Anutgan M., M.Sc. Thesis, Middle East Technical University, 2007.
[23] C. Ronning, E. Dreher, H. Felderrnann, M. Gross, M. Sebastian, H. Hofsass, Dia-
mond Relat. Mater. 6 (1997) 1129–1134.
[25] W.B. Jackson, J.M. Marshall, D.M. Moyer, Phys. Rev. B 39 (1989) 1164.
[26] Ismail Atılgan, Orhan Özdemir, Barıs¸ Akaog˘lu, Kıvanc¸ Sel, Bayram Katırcıog˘lu,
˙
Philos. Mag. B 86 (19) (2006) 2771–2796.