The T-line luminescence system is created in Si by annealing at
400-600 
C. Shifts and splitting of the spectral features with
C and D isotope substitution identify the presence of two C
atoms and one H atom in the center. Uniaxial stress and magnetic field
measurements show that the T-center has monoclinic-I symmetry and
possesses an acceptor (-/0) level at 0.2 eV below the conduction band.
Ab-initio cluster calculations lead to a structure in which an
interstitial C-H defect binds with a substitutional C atom. The
calculated vibrational modes are in good agreement with those
observed.
It is well established now that hydrogen is a common
contaminant in silicon
[1, 2, 3, 4]. It can interact
with both shallow and deep level centers and eliminate their
electrical [1] and optical [1, 4]
activity. However, hydrogen can have a quite different effect. In
this letter we report the structure and properties of an optical
center created in silicon which is stabilized by the interaction
between hydrogen and interstitial carbon. This center gives rise to
the T-line (0.9351 eV) luminescence system which is created in either
float-zone (FZ) or Czochralski (CZ) grown silicon by radiation damage
and subsequent thermal treatment in the temperature range
400-600 C, or by thermal treatment alone in carbon-rich CZ Si
[2, 3, 4, 5, 6]. It can also
be created in some [5, 6], but not all [7],
FZ Si by thermal treatment without radiation damage. This center has
been long known to be carbon related, a conclusion originally based on
sample statistics [5], but subsequently confirmed by a shift
in the zero-phonon line with C isotope substitution
[6]. However, recent investigations of silicon deliberately
doped with hydrogen-deuterium mixtures have shown that hydrogen is
also involved in this center
[2, 3, 4]. We present the effects of
isotope substitution on the local vibrational mode features observed
in the luminescence spectra, the effects of uniaxial stress and
magnetic field perturbations on the zero phonon line and temperature
dependence measurements. These show that the center contains two
inequivalent carbon atoms and one hydrogen atom, has monoclinic-I
symmetry and creates an electron trap at 0.2 eV below the conduction
band. Within these constraints, we use ab initio local density
functional cluster theory to show that the structure consists of a
100 
oriented pair of carbon atoms which share a
lattice site, together with one hydrogen atom attached to one of the
carbon atoms. The calculation also allows us to assign the four
observed local vibrational modes and account for their isotope shifts.
The material employed in the work reported here was mainly FZ silicon
with [C] 
2 
10 
cm 
; one sample had a
C/ 
C ratio of 1.4. All the material was
saturated with 1.5 10 
cm H, D, or a H+D
mixture by heating in flowing gas for 30 minutes at 1300 C,
and rapidly cooling in silicone oil [8]. The material was
then irradiated with a flux of either 10 cm 
neutrons,
with a Cd ratio of 25, or 2 10 cm 2 MeV
electrons, followed by annealing for 30 minutes at 450 C.
The photoluminescence (PL) was excited by a 514 nm Ar 
laser and
detected with a Fourier transform spectrometer fitted with a North
Coast cooled Ge diode detector. Uniaxial stress measurements were
made with stresses up to 160 MPa. Magnetic fields up to 5T were employed.
Fig. 1 shows a PL spectrum of the T-line system obtained at 35 K from
a sample saturated with deuterium. The line at 1.75 meV higher energy
than the T-line is associated with a zero-phonon transition from a
higher excited state. The phonon-assisted region at lower energy
contains a broad sideband of perturbed lattice modes, typical of
carbon-containing defects [9], together with several local
mode satellites L 
-L 
and resonance modes L 
and L 
.
Above 10 K, all the L mode features are doublets mirroring the two
zero-phonon transitions, showing that these features are all
associated with the same defect.
We first discuss the isotopic shifts of local mode features which
reveal the chemical composition and atomic structure of the center.
In C+ C+H and C+H+D mixed isotope
materials, the T-line splits into two components establishing that the
center contains carbon and hydrogen (Fig. 1). Similar isotopic
splitting can be observed for the zero-phonon transition associated
with the higher energy excited state. However, unlike the zero-phonon
line, the L local mode splits into four components in the mixed
carbon isotope material (Fig. 2) which unambiguously shows that the
defect contains two inequivalent carbon atoms, contrary to a previous
conclusion [6]. The splitting of a zero-phonon line is
dependent on the nature of the electron-phonon coupling [10]
and, although this effect is extensively used for chemical
characterization, it does not unambiguously reflect the atomic
structure of the center. The modes and their isotope shifts are
listed in Table 1. They give no indication that more than one
hydrogen atom is involved in this center.
Next we discuss the results of external field perturbation measurements,
which reveal the symmetry of the center and the origin of the luminescence
transitions. Both the zero-phonon lines split into 2, 3, and 4 components
under uniaxial stresses along 100 , 111 ,
and 110 directions, respectively [7].
The relative amplitudes of the peaks do not depend on temperature,
which strongly suggests that the splitting of the line is due to the
lifting of orientation degeneracy. The number of components and the
absence of thermalization suggest that the symmetry of the center is
monoclinic-I [11]. This is not in agreement with the results of
earlier study [6], where a smaller number of split components
was observed, probably due to the lower spectral resolution employed.
Temperature-controlled measurements have shown that with increasing
temperature in the range 20-60 K, the photoluminescence (PL) intensity
decays with an activation energy of 32 meV, which is much smaller than
the total binding energy of the electron-hole pair,
235 meV. This is consistent with the T-line luminescence
being related to the recombination of an exciton bound to a neutral
center, where one particle is bound into a deep level state and other
is weakly bound in the Coulomb potential produced by the first. The
thermal dissociation of the bound exciton and PL decay, in this case,
is associated with an excitation of the weakly bound particle to the
nearest band. A detailed analysis of the uniaxial stress and magnetic
field measurements has shown that the shallow effective-mass-like
particle in the bound exciton is a hole, contrary to a previous
suggestion [6], and consequently the defect possesses a deep
acceptor level (-/0) at 0.2 eV below the conduction band
[7]. It has been found that the behavior of the hole
under external field perturbations, as well as the strong anisotropy
of the g-tensor, can be successfully described by the
characteristics of the valence band maxima, in agreement with the hole
being bound by the Coulomb potential [7]. In addition, the
analysis of the magnetic data has shown that the defect in its neutral
ground state has an uncoupled electron with spin 
and a
g-factor close to 2. Upon photo-excitation, when an exciton is
bound to the defect and an additional electron occupies the acceptor
level, the two electron are coupled in a singlet state with S =0, so
that the magnetic splitting of the excited state is determined by the
weakly bound hole with spin and strong anisotropic
g-factors while the splitting of the final state is determined by
the electron. This fully accounts for the observed structure of the
Zeeman splitting, which corresponds to transitions between states with
spin and explains the absence of thermalization between
the two sub-levels of the final ground state and the absence of
exchange coupling between the bound exciton particles [6].
Note that the paramagnetic nature of the center in its neutral state
is consistent with the presence of one hydrogen atom in the defect.
Having established the chemistry and symmetry of the center, we now
describe the results of ab initio cluster calculations which
lead us to a unique model of the defect. Details of the method and
applications to substitutional carbon (C 
), interstitial carbon
(C 
), and the di-carbon defect (C -C ), as well as defects
containing hydrogen have been given previously
[12, 13, 14] and will not be repeated here.
A trigonal 88 atom cluster C HSi 
H 
was used in all
calculations presented here, all atoms were relaxed until the
equilibrium structure was determined.
The second derivatives of the energy with respect to atomic positions
were calculated for the C and H atoms, as well as their nearest Si
neighbors. Energy second derivatives for the remaining atoms were
taken from the Musgrave-Pople potential found by a previous ab
initio calculation [15], and the dynamical matrix of the
cluster, and hence its vibrational modes, were calculated.
We first note that the C-related L mode at 1056 cm 
lies
within 130 cm of modes due to C and more than 400 cm
above those of C [16]. This suggests that the T-center
contains C . We now argue that the observed C-isotope shifts of
the L mode suggest the presence of a C-C bond. This follows as
the reduced mass of an isolated C-C unit implies that its stretch
frequency would shift downwards by 20, 20 and 41 cm for
C- C, C- C, and C- C
respectively. These shifts roughly agree with the observed values of
18, 27 and 45 cm for the H case, and 16.5, 20.5 and
38 cm for D. The different shifts in the two mixed C-isotope
cases imply that the C atoms are inequivalent, and the distinct shifts
in the H and D cases suggest that H is bonded with one of the C atoms.
Thus we look for models with 
symmetry containing C and
C-C-H units.
Several models were examined [17]
and the ground state structure was found to be one shown in Fig. 3 where
a 100 oriented C-C pair share a substitutional site.
The H atom is attached to one of the carbon atoms, denoted C 
, and
the second undercoordinated carbon atom is labeled C .
The calculations show that the C -C , C -H, C -Si and
C -Si bond lengths are 1.46, 1.11, 1.91 (2) and 2.05 (2) Å , respectively.
The calculated vibrational modes and their isotope shifts
are given in Table 1. Now, PL only detects
modes of A-symmetry and we concentrate on those at 2913, 1180, 1098,
744, 558 and 542 cm . The first four modes correspond
predominantly to C -H stretch, C -H wag, C -C stretch and
C -Si stretch respectively. The last two modes at 558 and
542 cm are close to the Raman frequency, and involve predominantly
the motion of C and C , respectively, as well as their Si
neighbors. There is close agreement between the modes L to L
observed by PL and the calculated vibrational modes which enables them
to be identified (Table 1).
The two highest energy modes at 2913 and 1180 cm are undetected
by PL. However, the intensities of the phonon-assisted transitions are
critically dependent on the exciton-phonon coupling, and not
necessarily all local modes can be detected by PL. In addition, there
is an experimental problem in observing the highest energy local mode
since the Ge detector is insensitive in the region where the position
of the corresponding luminescence line is expected.
The C -C stretch mode lies close to
L and has carbon isotope shifts 19, 21, 41 cm in the H case
and 17, 24, 42 cm in the D case. These shifts are in good
agreement with the experimental values 18, 27, 45 cm and 16.5,
20.5, 38 cm , respectively, observed for L . We therefore
identify the L mode with C -C stretch. The C -Si
stretch mode at 744 cm involves the movement of H as well, but
has little amplitude on C . As a consequence, this mode displays
large isotope shifts with both C and D -- 23 cm and
30 cm respectively -- but essentially no additional modes appear
in the mixed C-isotopic cases. These results allow us to identify the
mode with L at 796 cm , which decreases by 25 and
37 cm for C and D, respectively, and involves the motion of
only one of the C atoms. Finally, the two lower A modes at 558 and
542 cm (H) are in good agreement with L and L , at 567.5
and 531.5 cm , respectively. These modes are more delocalized
and have only small shifts with D and the various combinations of the
C-isotopes.
The neutral defect has a deep mid-gap donor level occupied by one electron,
although
the theory is unable to locate it exactly. This donor level has not
been reported so far.
The acceptor level at 
eV would then be explained by a
large Hubbard-U term of at least 0.4 eV.
The structure and electronic properties of the center are similar to
the stable form of the P-C defect which possesses acceptor and donor levels at
eV and 
eV respectively [18].
The wavefunction for the singly occupied level is
localized on a non-bonding p-orbital on the C
atom and has spin in agreement with the
Zeeman measurements described above. Only small isotropic hyperfine
couplings with C and H are, however, to be expected. The
saturation of the dangling bond related to C by a second H atom
would eliminate the optical activity of the center, which might account
for the
observed loss of the PL intensity when the H concentration is large
[3, 4].
The calculations presented here favor the 100
oriented C -C -H split-interstitial as a candidate for the
T-center, although additional support would arise from observations of
the C-H stretch and wag modes at 2914 cm and 1180 cm , as
well as the mode of B-symmetry. The presence of a C-C bond is
surprising as the C -C defect, which is stable to about 300
C, only contains C-Si bonds [19]. We suggest that
the T-center is formed when C traps H in the BC sited defect.
This in turn diffuses to C where conversion to the T-center takes
place. A calculation shows the trapping of C -H by C leads to
an energy reduction of 1.4 eV [17]. This formation mechanism
is more likely than the trapping of H by the di-carbon defect as the
latter is not stable at 400-600 C.
The identification of this defect is important for several reasons.
It is commonly produced in FZ and CZ silicon samples by irradiation
and/or thermal treatment, showing that H is present as an unintentional
impurity in silicon [2, 3]. The defect
structure and the mechanisms of formation provide further evidence
that thermal treatment of CZ Si leads to a generation of Si , which
can be subsequently trapped by C to produce mobile C .
The stability of the T-center over an important temperature range
(400-600 C) where thermal donors are generated may enable
it to be used to monitor Si production. The results reported in this
paper show that the presence of hydrogen can lead to the formation of
unusually stable electrically and optically active defects.
In conclusion, a combination of high resolution photoluminescence experiments, Zeeman and uniaxial stress studies, and ab initio theory have successfully elucidated the structure and properties of an interstitial carbon-hydrogen defect in silicon.
This work was carried out with funding from the Engineering and Physical Sciences Research Council. R.J and P. L. thank the HPCI for computer time on the T3D at Edinburgh. S. Öberg thanks the Swedish National Scientific Research Council for financial support as well as the PDC for computer time on the SP2 at Stockholm.