You are on page 1of 4

Temperature-dependent

29
Si NMR resonance shifts in lightly- and
heavily-doped Si:P
Ernesta M. Meintjes
a,1
, William W. Warren, Jr.
a
, James P. Yesinowski
b,n
a
Department of Physics, Oregon State University, Corvallis, OR 97331-6501, USA
b
Chemistry Division, Naval Research Laboratory, Washington, DC 20375-5342, USA
a r t i c l e i n f o
Article history:
Received 3 May 2013
Received in revised form
1 October 2013
Available online 14 October 2013
Keywords:
29
Si MAS-NMR
Silicon
Semiconductor
Dopant
Metalnonmetal
Chemical shift
Knight shift
Temperature dependence
a b s t r a c t
Careful NMR measurements on a very lightly-doped reference silicon sample provide a convenient highly
precise and accurate secondary chemical shift reference standard for
29
Si MAS-NMR applicable over a
wide temperature range. The linear temperature-dependence of the
29
Si chemical shift measured in this
sample is used to rene an earlier presentation of the paramagnetic (high-frequency)
29
Si resonance
shifts in heavily-doped n-type silicon samples near the metalnonmetal transition. The data show
systematic decreases of the local magnetic elds with increasing temperature in the range 100470 K
for all samples in the carrier concentration range from 2 10
18
cm
3
to 810
19
cm
3
. This trend is
qualitatively similar to that previously observed for the two-orders of magnitude larger
31
P impurity
NMR resonance shifts in the same temperature and concentration ranges. The
29
Si and
31
P resonance
shifts are not related by a simple scaling factor, however, indicating that impurity and host nuclei are
affected by different subsets of partially-localized extrinsic electrons at all temperatures.
Published by Elsevier Inc.
1. Introduction
In a comprehensive variable-temperature study of heavily-doped
Si:P Meintjes et al. [1] reported NMR data for a series of samples
doped in a range around the metalnonmetal (MNM) transition
(ca. 410
18
cm
3
). In this study, nominal phosphorus concentrations
ranged from 210
18
cm
3
to 810
19
cm
3
and the temperature
range was 100470 K. While the emphasis of this work was exploita-
tion of
31
P NMR to study the electronic state associated with the
phosphorus impurity, NMR results for the
29
Si host nuclei were also
presented. These included measurements of the resonance shifts,
linewidths, and spin-lattice relaxation rates over the above concentra-
tion and temperature ranges. The NMR data reported in Reference [1]
were obtained by conventional static NMR techniques.
This article reports new measurements that allow a correction
and renement of the
29
Si resonance shifts presented in Reference
[1]. In that study, the resonance shift was dened as the fractional
shift of the NMR frequency of
29
Si in a doped sample relative to
the frequency in a reference sample of undoped silicon at room
temperature. This approach isolates the contribution to the local
magnetic elds at the
29
Si nuclei due the paramagnetism of
extrinsic electrons introduced by the impurity. For analysis of the
earlier variable-temperature experiments, it was assumed that the
resonance frequency of the undoped reference sample, determined
essentially by a chemical shift, is independent of temperature.
Recently, one of us (JPY) has carried out high-resolution
29
Si
measurements at 11.7 T of a series of lightly- and heavily-doped
Si samples using magic-angle spinning NMR (MAS-NMR) methods
[2]. Detailed comparison of these new results with those of
Reference [1] has revealed the following. First, the resonance
shifts in Fig. 5 of that work were intended to represent the
hyperne shift due to conduction electrons (called the Knight shift,
at least in the case of degenerate electrons). These should be
measured from the position of the peak maxima of the doped
samples with respect to that of the undoped silicon reference
sample, which lacking conduction electrons experiences only
the normal isotropic chemical shift interaction of tetrahedrally-
bonded silicon in a diamond lattice structure. However, an incorrect
value for the room-temperature chemical shift of the undoped
silicon reference sample was used in Reference [1], resulting in the
resonance shifts of the doped samples being offset from their true
values. We have conrmed this with new MAS-NMR measurements
of the original reference sample (6 N Si, i.e. 99.9999% pure) as well
as other very lightly-doped silicon samples that have the same
resonance frequency and hence exhibit no resonance shifts from
hyperne interactions with conduction electrons.
Second, the recent high-resolution measurements [2] have
revealed a signicant temperature-dependence of the chemical
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/ssnmr
Solid State Nuclear Magnetic Resonance
0926-2040/$ - see front matter Published by Elsevier Inc.
http://dx.doi.org/10.1016/j.ssnmr.2013.10.002
n
Corresponding author. Fax: 1 202 767 0594.
E-mail address: yesinowski@nrl.navy.mil (J.P. Yesinowski).
1
Present address: MRC/UCT Medical Imaging Research Unit, Department of
Human Biology, Faculty of Health Sciences, University of Cape Town, Observatory
7925, South Africa.
Solid State Nuclear Magnetic Resonance 55-56 (2013) 9194
shift of the reference sample that was not resolved in the earlier
static NMR measurements. This observation allows us to apply
a correction to the resonance shifts due to conduction electrons
mentioned above that were obtained as a function of both
temperature and dopant concentration. This correction removes
the effects due to the changes in chemical shifts of the samples
with temperature, and thus yields resonance shifts due only to the
effects of conduction electrons. The new results allowcomparisons
to be made with previous
31
P resonance shift data over the same
temperature and concentration range.
2. Experimental
The
29
Si NMR spectra were acquired on a Varian/Agilent DDRS
500 MHz (
1
H frequency) spectrometer with a wide-bore 11.7 T
magnet. A triple-resonance T3 HXY MAS-NMR probe with 4 mm O.
D. zirconia rotors was tuned in double-resonance mode to the
29
Si
Larmor frequency of 99.36 MHz, with a typical 901 pulse length of
4.3 s. Variable-temperature operation was achieved by control-
ling the temperature of a stream of dry nitrogen gas entering the
rotor chamber, separate from the drive and bearing gas lines.
The thermocouple used for measuring this temperature was in
the variable-temperature stack in the magnet just above the probe
itself, which leads to sample temperatures in heated samples
being somewhat lower than the set temperature, and sample
temperatures in cooled samples being somewhat higher than the
set temperature [3]. For temperatures below room temperature a
FTS chiller unit was employed, and a copper-coil heat exchanger
immersed in liquid nitrogen was used in addition for the lowest
temperatures.
Modest spinning rates of typically 5 kHz were used, since the
weak
29
Si
29
Si homonuclear dipolar couplings (o1 kHz) resulted
in sharp peaks at even lower spinning speeds. The full-width at
half-maximum (FWHM) linewidth of the undoped 6N reference
sample was 25 Hz (0.25 ppm), and that of the very lightly-doped
Si:P reference sample (carrier concentration1.310
16
cm
3
)
only 7 Hz (0.07) ppm, reecting the high degree of crystallinity
of this sample. The latter sample was prepared by crushing a
portion of a silicon wafer from Silicon Quest International after
making 4-point conductivity measurements to establish the
carrier concentration. These linewidths were obtained for 4 mm
rotors lled with the sample. An even narrower linewidth
(FWHM4.7 Hz), shown in Fig. 1, was obtained for the latter
sample in a MAS probe with a 3.2 mm O.D. silicon nitride rotor in
which 6 mg of Si:P was sandwiched between 9 mg (bottom) and
7 mg (top) of adamantane (as a ller) in order to restrict the
sample to the most homogeneous region of the magnetic eld.
It is important to note that both of these samples, as well as
other, lightly B-doped, p-type Si:B samples having carrier concen-
trations of 2 10
15
cm
3
and 1.510
17
cm
3
, had identical
chemical shifts. This indicates that only the
29
Si chemical shift is
inuencing the peak position in these samples, and that there is
no measurable contribution of resonance shifts arising from
carriers such as is seen in more heavily-doped n-type or p-type
silicon samples [4]. Therefore we will consider both the 6N Si
sample and the Si:P sample with a carrier concentration of
1.310
16
cm
3
as secondary chemical shift reference standards.
The primary chemical shift reference at 0.00 ppm is the
29
Si
peak in TMS. The
29
Si chemical shift of the Si:P secondary chemical
shift reference sample at 293 K was measured to be 81.24 ppm
relative to a 22 mm long 3 mm O.D. cylindrical tube of pure bulk
TMS (effects from the isotropic bulk magnetic susceptibility of a
liquid sample vanish for an innitely long tube oriented at the
magic-angle, even without spinning [5]). While a
29
Si NMR signal
from the capillary of dilute TMS in CDCl
3
mentioned in Section 3
was too weak to observe, we measured the
1
H NMR frequencies in
both the bulk and dilute TMS samples in order to employ the
IUPAC-recommended unied scale for referencing heteronuclear
chemical shifts as the ratio of the exact heteronuclear frequency to
the
1
H frequency of dilute TMS in CDCl
3
[5]. For the bulk TMS
sample the measured /% for TMS was 19.867181, which differs
slightly from the value /%19.867187 reported for dilute TMS in
CDCl
3
[5]. If we use the lowering of the
1
H frequency at 11.7 T by
441 Hz upon going from 80% TMS to 1% TMS in CDCl
3
as a means
of accounting for most of the concentration dependence of the
1
H
chemical shift, we obtain a second value for /% of 19.867198
(using the
29
Si frequency in bulk TMS). Since the 441 Hz value was
measured with the eld locked on CDCl
3
, the changing bulk
susceptibility should not have been a factor, although solvent
effects upon the
2
H chemical shift of CDCl
3
might well be. We also
obtained a third value for /% of 19.867173 by taking the ratio of
our measured 203 Hz higher
1
H frequency of the dilute TMS in
CDCl
3
capillary (compared to that in the bulk TMS sample) to the
29
Si frequency of bulk TMS. These three calculated values of /%
represent deviations from the reported value [5] of 0.3 ppm,
0.6 ppm and 0.7 ppm respectively, giving a feel for the
accuracy of the shift determinations using TMS directly as a
reference. The /%
neat TMS
value (notation recommended in
Reference 4) of the above Si:P secondary chemical shift reference
sample at 293 K is thus calculated to be 19.865567.
3. Results and discussion
A better than order-of-magnitude increase in the resolution of
29
Si NMR for the chemical shift reference sample can be obtained
by using magic-angle spinning (MAS) to remove the homonuclear
dipolar broadening compared to the earlier static measurements
[1], as can be seen from Fig. 1. The full-width at half-maximum
(FWHM) MAS-NMR linewidth of the very lightly-doped Si:P
reference sample is only 4.7 Hz, considerably sharper than the
31 Hz value reported from MAS-NMR of a lightly-doped Si:Sb
sample [6].
This high degree of resolution achieved by MAS permitted a very
precise determination of the temperature dependence of the
chemical shift in this sample, which was extremely linear over
the range 250475 K, with a slope of 0.01272 ppm/K (Fig. 2).
Since the actual sample temperature was not measured directly, its
assumed value is subject to some error associated with two effects.
The rst is frictional heating from the spinning rotor itself, which
for the modest spinning speed used (5 kHz) is only about 2 1C at
Fig. 1.
29
Si NMR spectra of Si:P chemical shift reference sample (carrier con-
centration1.3 10
16
cm
3
) static (broad peak) and with MAS (sharp peak) at 8.32
kHz spinning speed, as described in Section 2. Recycle delays were 2 h and 8 and 32
scans respectively were acquired, with respective line-broadenings of 10 Hz and
2 Hz. The static FWHM linewidth of 152 Hz (1.53 ppm) is due to weak homonuclear
dipolar couplings (
29
Si N.A. 4.7%) that are removed by MAS to yield a 4.7 Hz
(0.047 ppm) FWHM linewidth.
E.M. Meintjes et al. / Solid State Nuclear Magnetic Resonance 55-56 (2013) 9194 92
room temperature in this type of probe [3] but somewhat greater at
lower temperatures due to the increased viscosity of the nitrogen
gas. The second is due to the fact that the temperature is measured
in the gas ow upstream of the rotor, as discussed in Section 2. We
estimate from measurements made at a similar spinning speed of
6 kHz in the same type of probe using the
207
Pb NMR chemical shift
of Pb(NO
3
)
2
as an NMR thermometer [3] that the measured slope of
0.01272 ppm/K may underestimate the true temperature coef-
cient of the
29
Si chemical shift by about 10%, meaning that the true
slope could be 0.0140 ppm/K, but we use the measured value in
the calculations below since there is negligible difference in the
nal results. From 172 K to 213 K the experimental data points were
slightly above the extrapolated linear t. Verication of the con-
stancy of the (unlocked) main magnetic eld at the sample over this
temperature range was accomplished by monitoring the
1
H NMR of
a small capillary of dilute tetramethylsilane (TMS) reference in
CDCl
3
, which is known to have a nearly-negligible temperature-
dependence of its
1
H chemical shift [7].
Since the FWHM linewidth of the Si:P chemical shift reference
sample was only 0.07 ppm (or less), and its peak maximum can be
located with even greater accuracy, we believe that very lightly-
doped (or pure undoped) silicon samples provide a valuable
29
Si
secondary chemical shift reference at a variety of temperatures.
One can use the 81.24 ppm (w.r.t. TMS) value at 293 K, or
alternatively the /%
neat TMS
value of 19.865567, reported in
Section 2. A previous MAS-NMR measurement of the
29
Si chemical
shift in silicon gave a value of 80.6 ppm [8], in reasonable
agreement. The sharp linewidth will enable accurate measure-
ment of very small resonance shifts due to hyperne interactions
with carriers in silicon at higher doping levels. The samples are
commercially-available as wafers which can be gently ground to
avoid broadening effects that have been observed upon ball-
milling [2]. As discussed in Section 2, carrier concentrations
(p-type or n-type) below about 10
17
cm
3
(in samples having
resistivities 4ca. 0.2 cm) show no measurable resonance shift
effects fromcarriers (although T
1
relaxation may be affected). Despite
a very long T
1
of ca. 4 h, a reasonable
29
Si MAS-NMR spectrum for
referencing can be obtained from a single scan after waiting 10 min.
As mentioned in Section 1, an incorrect value for the room-
temperature chemical shift of the undoped silicon reference sample
was used to obtain resonance shifts of the heavily-doped samples in
Fig. 5 of Reference [1]. In order to obtain the correct resonance
shifts at room temperature, a new shift measurement was made of
sample 10 from that work (the sample number here and in Fig. 3
represents the carrier concentration in units of 10
18
cm
3
). The
29
Si
MAS-NMR spectra of sample 10 at a spinning speed of 5.0 kHz were
obtained at 295 K using a saturation-recovery sequence. The T
1
of
15.4 s measured at the position of the peak maximum in these
spectra is close to the T
1
value of 12.7 s reported at a lower ( 0.68)
magnetic eld strength in Reference [1]. For a fully-relaxed
spectrum (recovery-time60 s) the position of the maximum of
the asymmetric broad peak (FWHM14.0 ppm, asymmetric tail
extending on the high-frequency side, see below) was 69.5 ppm.
This corresponds to a resonance shift of 11.73 ppm from the
position of the Si:P secondary chemical shift reference sample at
the same temperature.
We have thus obtained the correct chemical shift reference at
room temperature for the data shown in Fig. 5 of Reference [1].
However, in addition, we need to account for the temperature-
dependence of the chemical shift of the reference sample that was
neglected there. Corrections for both items have been incorpo-
rated in Fig. 3, which is based upon data presented in Fig. 5 of
Reference [1]. The rst correction is a simple offset accounting for
the incorrect original referencing at room temperature. The second
correction is for the temperature-dependence of the chemical
shifts of all the doped samples, which is assumed to be identical
to that of the secondary chemical shift reference sample in Fig. 2,
as justied below.
The procedure used to adjust the measurements in order to
account for the temperature-dependence of the chemical shifts in
the heavily-doped samples should be briey described, since the
recent experimental values shown in Fig. 2 extend down to only
172 K, whereas temperatures extended down to 103 K in the
original data. We believe that the shift to higher frequency at
higher temperatures, observed in other inorganic compounds in
the solid-state [9,10], may result from a combination of two
effects. There is a lattice expansion, described by a linear thermal
expansion coefcient in silicon for a wide range above 300 K [11],
that alters the orbital overlap and changes the paramagnetic
shielding in the Ramsey formulation of the theory of chemical
shifts. There is also increasing orbital overlap due to increasing
vibrational motions, which is important in some cases [9].
Regardless of the source of the temperature dependence in the
present case, it seems very reasonable to assume that dop-
ing by itself does not change the temperature coefcient of
the chemical shift, even though resonance shifts due to hyper-
ne interactions with conduction electrons have their own tem-
perature dependence. In correcting the data from Fig. 5 in Refer-
ence [1], we have therefore simply assumed that the linear
temperature dependence of the chemical shift (0.01272 ppm/
Fig. 2.
29
Si chemical shift w.r.t. TMS vs. temperature of the secondary chemical
shift reference sample of Si:P (carrier concentration1.310
16
cm
3
) measured by
MAS-NMR. The straight line is a linear t (slope0.01272 ppm/K) excluding the
four points at the lowest temperatures.
Fig. 3.
29
Si NMR resonance shifts versus temperature in phosphorus-doped silicon
(Si:P). Resonance shifts are calculated with respect to the temperature-dependent
shift of the (secondary) chemical shift reference standard. Samples are identied by
the nominal carrier concentrations in units of 10
18
cm
3
.
E.M. Meintjes et al. / Solid State Nuclear Magnetic Resonance 55-56 (2013) 9194 93
K) can be extended over the entire range of 103466 K. Extrapola-
tion of the curvature of the temperature-dependence of the
reference between 172 K and 254 K from Fig. 2 suggests that even
at the lowest temperature of 103 K the resonance shifts would be
too large by only about 1 ppm, comparable to the error bars.
The
29
Si resonance shift data in Fig. 3 differ qualitatively from
the original plots in Reference [1]. As in the original presentation
[1], the data show a systematic increase of the paramagnetic
(high-frequency) shift of the
29
Si resonance with increasing
phosphorus concentration. Further, the shifts decrease signi-
cantly with increasing temperature in all samples. This behavior
contrasts with the earlier presentation [1] in which shifts in the
most lightly-doped samples showed weak positive temperature
dependences, or were apparently independent of temperature.
These effects were artifacts of the underlying temperature depen-
dence of the chemical shift reference frequency, which decreases
at lower temperatures.
The data in Fig. 3 can also be compared to the two-orders of
magnitude larger temperature-dependent shifts for
31
P reported
in Fig. 3 of Reference [1]. There is a rough similarity in the fact that
for both nuclei the resonance shifts increase at lower tempera-
tures, with the dependence for sample 80 being much weaker
than for the samples with lower dopant levels. However, within
these latter three samples (10, 7 and 4) the
29
Si curves
remain separated at lower temperatures, but the
31
P curves
intersect and cross over. Furthermore, the
31
P curves show a much
stronger temperature dependence than do the
29
Si curves, espe-
cially for the latter three samples at lower dopant concentrations.
For example, the
31
P shift in sample 10 (carrier con-
centration110
19
cm
3
) varies by nearly a factor of 5 between
100 K and 470 K whereas the change in the
29
Si shift over the same
range is less than a factor of 2.
We call attention, however, to the essential challenge in deter-
mining an appropriate reference for impurity nuclei as compared
with NMR of a host species. In the case of
29
Si in doped Si, the
obvious reference is a sample of undoped (or extremely lightly
doped) Si in which conduction electron effects are absent. In
contrast, a dilute donor species such as
31
P in Si will carry an
unpaired electron in a localized impurity state and experience a
local hyperne eld and resonance shift except under concentration
and temperature conditions in which the donor is fully ionized.
Impurity NMR measurements under these conditions may be
prohibitively difcult, depending on the system. An extrapolation
procedure has been applied to this problem in the case of
69
Ga in
ZnO:Ga [12,4]. A more feasible approach is to select a reference
compound in which the species of interest is in a diamagnetic
environment such that one does not expect local paramagnetism
from either localized or conduction electrons. In Reference [1], a
dilute H
3
PO
4
aqueous solution was used for this purpose. Never-
theless, it is recognized that the environment and hence the
chemical shift of
31
P in this reference is not the same as the ideal
of an isolated P

ion in an otherwise pure Si host. Thus, the above-


mentioned factor of 5 for
31
P cannot be regarded as exact. Never-
theless, a comparison of the
31
P and
29
Si curves reveals that the
much stronger temperature dependences of the 10, 7 and 4
samples for
31
P vis-a-vis
29
Si result in the
31
P curves intersecting
and crossing over the
31
P curve of the 80 sample, in marked
contrast to the behavior seen for the
29
Si curves in Fig. 3. This
contrasting behavior reinforces a conclusion drawn from several
NMR parameters in Reference [1], that even in nominally metallic
samples with dopant concentrations well above the MNM transi-
tion, impurity and host nuclei experience markedly different local
magnetic elds produced by different subsets of extrinsic electrons.
These extrinsic electrons are partially-localized to a degree that
depends upon temperature, resulting in a strong temperature
dependence of the resonance shifts [1,2].
Finally, we should point out in connection with the broad
asymmetric
29
Si MAS-NMR peak observed above for sample 10
that for moderately or heavily doped semiconductors such as Si or
h-GaN broad asymmetric peaks have previously been observed
in static and MAS NMR experiments [1,2,4,13]. The origin of
this phenomenon is an active area of investigation and has been
attributed to a distribution of resonance shifts (referred to as
Knight shifts) reecting a spatially inhomogeneous distribution of
dopants over some distance scale. However, further discussion of
this intriguing phenomenon is beyond the scope of this paper.
4. Conclusion
We have shown that magic-angle spinning at moderate
spinning-speeds on carefully-shimmed samples of lightly-doped
silicon can yield
29
Si MAS-NMR linewidths with FWHM as little as
4.7 Hz (0.05 ppm). Such samples can provide a convenient, precise
and accurate secondary chemical shift reference for
29
Si MAS-
NMR. The measured value of /%
neat TMS
for the secondary
chemical shift reference sample at 293 K is 19.865567, correspond-
ing to a chemical shift from bulk TMS of 81.24 ppm. There is a
temperature-dependence to the chemical shift of this reference
sample, which is extremely linear over the range 250475 K. The
measured slope is 0.01272 ppm/K, but an estimate of the actual
sample temperature taking into account various sources of error
[3] suggests that the true value is closer to 0.0140 ppm/K.
These observations have enabled a signicant re-evaluation
of prior static
29
Si NMR data on heavily doped Si:P samples in
the region of the MNM transition [1], and have allowed a more
meaningful comparison to be made with
31
P NMR data on the
same samples. The effects of varying degrees of localization of
electron density in doped Si samples near the MNM transition can
now be probed as a function of temperature using
29
Si MAS-NMR.
The very high resolution obtained from the chemical shift refer-
ence sample, and the carefully-measured temperature depen-
dence of its chemical shift, should enable observation of much
smaller effects from carriers, both electrons and holes, in n-type
and p-type silicon samples than has previously been possible.
Acknowledgments
JPY acknowledges nancial support for this work from the
Ofce of Naval Research through the NRL 6.1 core program and
the NRL Nanoscience Institute Li-battery project headed by
Dr. Michelle Johannes. JPY thanks Dr. Steve Erwin (NRL) for
graphical gures of the silicon lattice.
References
[1] E.M. Meintjes, J. Danielson, W.W. Warren, Jr., Phys. Rev. B 71 (2005) 035114.
[2] J.P. Yesinowski, A.P. Purdy, C.T. Love, E.S. Epstein, S.M. Prokes, Presented at the
EUROMAR, University College Dublin, Dublin, Ireland, 2012. (unpublished).
[3] V. Hronsky, Acta Electrotech. Inform. 13 (2013) 95.
[4] J.P. Yesinowski, Top. Curr. Chem. 306 (2012) 229-312.
[5] R.K. Harris, E.D. Becker, S.M.C. de Menezes, R. Goodfellow, P. Granger, Solid
State Nucl. Magn. Reson. 22 (2002) 458.
[6] D. Li, Y.Q. Dong, R.G. Ramos, J.D. Murray, K. MacLean, A.E. Dementyev,
S.E. Barrett, Phys. Rev. B 77 (2008) 214306.
[7] R.E. Hoffman, E.D. Becker, J. Magn. Reson. 176 (2005) 87.
[8] J.S. Hartman, M.F. Richardson, B.L. Sherriff, B.G. Winsborrow, J. Am. Chem. Soc.
109 (1987) 6059.
[9] L.H. Ngai, J. Phys. Chem. Solids 30 (1969) 571.
[10] H. Eckert, J.P. Yesinowski, J. Am. Chem. Soc. 108 (1986) 2140.
[11] Y. Okada, Y. Tokumaru, J. Appl. Phys. 56 (1984) 314.
[12] N. Roberts, R.P. Wang, A.W. Sleight, W.W. Warren, Jr., Phys. Rev. B 57 (1998)
5734.
[13] J.P. Yesinowski, A.P. Purdy, H.Q. Wu, M.G. Spencer, J. Hunting, F.J. DiSalvo, J. Am.
Chem. Soc. 128 (2006) 4952.
E.M. Meintjes et al. / Solid State Nuclear Magnetic Resonance 55-56 (2013) 9194 94