arXiv:astro-ph/0509080v2 5 Dec 2005
First limit on WIMP cross section with low
background CsI(Tℓ) crystal detector
H.S. Leea,∗, H. Bhanga, J.H. Choia, I.S. Hahnd, D. Hef,
M.J. Hwangc, H.J. Kime, S.C. Kima, S.K. Kima,∗∗,
S.Y. Kima, T.Y. Kima, Y.D. Kimb, J.W. Kwaka, Y.J. Kwonc,
J. Leea, J.H. Leea, J.I. Leeb, M.J. Leea, J. Lif,gS.S. Myunga,
H. Parka,1, H.Y. Yanga, J.J. Zhuf,
aDMRC and School of Physics, Seoul National University, Seoul 151-742, Korea
bDepartment of Physics, Sejong University, Seoul 143-747, Korea
cPhysics Department, Yonsei University, Seoul 120-749, Korea
dDepartment of Science Education, Ewha Womans University, Seoul 120-750,
ePhysics Department, Kyungpook National University, Daegu 702-701, Korea
fDepartment of Engineering Physics, Tsinghua University, Beijing 100084, China
gInstitute of High Energy Physics, Chinese Academy of Sciences, Beijing 100039,
The Korea Invisible Mass Search (KIMS) collaboration has been carrying out WIMP
search experiment with CsI(Tℓ) crystal detectors at the YangYang Underground
Laboratory. A successful reduction of the internal background of the crystal was
done and a good pulse shape discrimination was achieved. We report the first result
on WIMP search obtained with 237 kg·days data using one full-size CsI(Tℓ) crystal
of 6.6 kg mass.
Key words: WIMP, dark matter, CsI(Tℓ) Crystal, pulse shape discrimination,
PACS: 29.40.Mc, 14.80.Ly, 95.35+d
1Present address : Division of Chemical Metrology and Materials Evaluation, Ko-
Preprint submitted to Elsevier Science5 February 2008
Although the existence of dark matter as a major portion of the matter in
the universe has been well supported by various astronomical observations,
its identity is unknown yet. Weakly Interacting Massive Particle (WIMP) is
regarded as one of the strongest candidates for cold dark matter particle  and
many experimental searches for WIMPs have been performed. An indication of
an annual modulation of the signal reported by DAMA using NaI(Tℓ) crystal
detectors may be a possible evidence of WIMP signal . However, stringent
limits set by cryogenic detectors, CDMS  and EDELWEISS  seem to rule
out the DAMA signal region. Still, ways of interpreting both results without
conflict are not completely excluded because of the difference in experimental
techniques and target nuclei [5,6,7].
The Korea Invisible Mass Search (KIMS) collaboration has been carrying out
the WIMP search with CsI(Tℓ) crystals. Low threshold suitability for WIMP
search and pulse shape discrimination (PSD) superiority to NaI(Tℓ) has been
demonstrated [8,9,10,11]. Even if the CsI(Tℓ) crystal has the advantage of a
good PSD and the ease of getting a large detector mass, the internal back-
ground including137Cs ,134Cs and,87Rb has been a major hurdle to apply the
crystal to WIMP search [11,12]. We have successfully purified CsI(Tℓ) crystals
after extensive studies on the contamination mechanism. A pilot experiment
with a low background CsI(Tℓ) crystal of 6.6 kg mass has been carried out at
the YangYang Underground Laboratory (Y2L) in Korea. We report the first re-
sult obtained with 237 kg·days of data taken with one crystal of 8 x 8 x 23 cm3
2YangYang Underground Laboratory
We have established an underground laboratory at YangYang utilizing the
space provided by the YangYang Pumped Storage Power Plant currently un-
der construction by Korea Midland Power Co. The underground laboratory
is located in a tunnel where the vertical earth overburden is approximately
700 m. The muon flux measured with the muon detector is 2.7×10−7/cm2/s,
which is consistent with the water equivalent depth of 2000 m . The lab-
oratory is equipped with a clean room with an air conditioning system for
a constant temperature and low humidity. An environment monitoring sys-
tem is installed for continuous monitoring of temperature and humidity. The
temperature inside the CsI(Tℓ) detector container is stable within ± 0.2oC.
rea Research Institute of Standards and Science, Yuseong, Daejeon, 305-600, Re-
public of Korea
The rock composition surrounding the laboratory was analyzed with the ICP-
MASS method and the contamination of238U and232Th is reported to be
at a level of < 0.5 ppm and 5.6 ± 2.6 ppm respectively. The relatively low
contamination of238U in the rocks of the tunnel results in a low level of radon
contamination in the air of the tunnel. A radon detector  was constructed
to monitor the level of radon in the experimental hall. The contamination
level of222Rn in the tunnel air was 1-2 pCi/ℓ which is slightly lower than
other underground laboratories such as Gran Sasso  and Kamiokande .
The neutron flux in the experimental hall is continuously measured with two
one-liter BC501A liquid scintillation detectors inside and outside of the main
shield. The estimated neutron flux in the experimental hall is 8×10−7/cm2/s
for 1.5 MeV < Eneutron < 6.0 MeV which is much lower than that in the
Cheongpyung Underground Laboratory .
3Reduction of the internal background in CsI(Tℓ) crystal
The major radioisotopes in the CsI(Tℓ) crystal contributing to the internal
background are137Cs,134Cs and87Rb [12,18].134Cs has a half-life of 2 years
and has a cosmogenic origin - neutron capture by
is unavoidable unless the crystals are stored underground for many years.
However the signal from its decay can be easily removed by the γ-ray that
follows immediately after the β decay. The signal is large and beyond the
energy range of our interest for WIMP search. Additional reduction can be
made by using coincidence signals from neighboring crystals. The average
contamination level of134Cs is measured to be 20 mBq/kg, while 1 mBq/kg can
contribute only less than 0.07 counts/(keV·kg·day) (CPD) at 10keV. Therefore
the134Cs contamination is not a major problem.
137Cs, the half-life of which is 30 years, comes from a man-made origin, mainly
due to nuclear bombs and nuclear reactors. It decays to an excited state of
137Ba by emitting an electron with a Q value of 514 keV, followed by γ-ray
emission to the ground state137Ba with a half-life of 2.55 minutes. Therefore
the Compton scattering of this γ-ray as well as the β-ray can cause background
in the low energy range where WIMP signals are expected. Simulation stud-
ies show that 1 mBq/kg can contribute 0.35 CPD background in the 10 keV
region. We investigated the contamination process and the reduction method
of137Cs. Most of the suspected intermediate products from pollucite, an ore
of Cs, to CsI powder has been measured with a low background HPGe de-
tector installed at the Y2L. We also investigated the processing water and
conclude that the main source of137Cs is the processing water used for the
powder production, in which the contamination level of137Cs was found to be
0.1 mBq/ℓ. By using purified water, we succeeded in reducing the137Cs in the
final powder and the crystal .
Rb, which includes 27.8% of87Rb, exists in the pollucite at a 0.7% level and
can contaminate easily because it is chemically similar to Cs. The87Rb under-
goes β decay to the ground state of its daughter nucleus,87Sr, with emission
of an electron whose end point energy is 282 keV. This can be a very serious
background, and 1 ppb contamination can contribute 1.07 CPD at 10keV.
Contamination of many CsI powders in the market has been measured with
the ICP-MASS method and we found that it varies from 1 ppb to 1000 ppb.
The reduction technique of Rb, a repeated recrystallization process, has been
widely known. We performed the recrystallization to the powder that we ob-
tained with pure water and reduced the Rb contamination below 1 ppb. The
reduction of the internal background is discussed in detail elsewhere .
We have installed a shielding structure in the experimental hall to stop the
external background originating mainly from the surrounding rocks. The shield
consists of 10 cm thick Oxygen Free High Conductivity (OFHC) copper, 5 cm
thick polyethylene (PE), 15 cm thick Boliden lead and 30 cm mineral oil (liquid
parafin) from inside out. The mineral oil is mixed with 5% of a pseudocumene-
based liquid scintillator and mounted with PMTs so that it can perform as a
muon detector . Inside the copper chamber, N2gas is flown at a rate of
4 ℓ/min to reduce the radon contamination as well as to keep the humidity
The CsI(Tℓ) crystal (full-size crystal) used for the experiment has a dimen-
sion of 8 x 8 x 23 cm3and a mass of 6.6 kg. The crystal is attached with two
low background quartz window PMTs with RbCs photocathode. RbCs pho-
tocathode enhances quantum efficiency in the green wavelength region and
gives 50% more photoelectron yield for CsI(Tℓ) crystal than a normal bialkali
PMT does. As a result, the number of photoelectrons is about 5.5/keV for the
The signal from the PMT is amplified with a preamplifier mounted outside the
main shield and brought to the FADC module through a 20 m-long coaxial
cable. The homemade FADC module is designed to sample the pulse every
2 ns for a duration up to 32 µs so that one can fully reconstruct each photo-
electron pulse as shown in Fig. 1. The trigger is formed in the FPGA chip on
the FADC board. For low energy events, it is required to have more than five
photoelectrons in two µs for the event trigger. An additional trigger is gen-
erated if the width of the pulse is longer than 200 ns for high energy events
where many single photon signals are merged into a big pulse. The FADC
located in a VME crate is read out by a Linux-operating PC through the
VME-USB2 interface with a maximum data transfer rate of 10 Mbytes/s. The
0 1020 30
5.56 6.57 7.5
Fig. 1. (a) shows typical low energy γ signal from CsI(Tℓ) crystal for one PMT
obtained by Compton events from137Cs source. Zoomed pulse shape of this event
from 5.5 µs to 7.5 µs is shown at (b). The same pulse spectrum with clustering is
shown in (c). The neighboring cluster is separated by a different style (solid line
and dashed line).
DAQ system is based on the ROOT  package.
During the two-month period starting from July 2004, we have taken data
for WIMP search using one crystal with a background level of approximately
7 CPD at 10keV. The amount of data was 237 kg·days.
5 Calibration data
The different timing characteristics between nuclear recoil and electron recoil
in CsI(Tℓ) crystal make it possible to statistically separate the nuclear recoil
events from the γ background using the mean time distribution [9,10,11]. In
order to have a good reference distribution of the mean time for them, we took
calibration data of electron recoil from the γ source and nuclear recoil from
the neutron source.
The γ calibration data was obtained with a full-size crystal by a137Cs source in
the copper chamber of Y2L. Identical setup and conditions as for the WIMP
search data were used. For one week irradiation, we took low energy γ-ray
events equivalent to approximately 3000 kg·days WIMP search data.
Neutron calibration data were obtained by exposing a small-size test crystal
(3 x 3 x 3 cm3) to neutrons from 300 mCi Am-Be source, prepared at Seoul
National University . In order to identify neutrons scattered from CsI, we
used neutron detectors, made of BC501A contained in a cylindrical stainless
steel vessel. Each neutron detector is shielded by 5 cm lead and 10 cm paraffin,
and set up at various angles with respect to the incident neutron direction.
The Am-Be source is surrounded by liquid scintillator (LSC) composed of 95%
mineral oil and 5% of pseudocumene with a collimation hole to the direction
of the CsI(Tℓ) crystal. The LSC acts as a tagging detector of 4.4 MeV γ’s
which are simultaneously generated with neutrons from the Am-Be source as
well as a neutron shield for the low surface neutron flux outside of the source.
In order to identify neutron-induced events, we required a coincidence be-
tween any one of the neutron detectors and the CsI(Tℓ) crystal. With a good
neutron separation capability, we took neutron data, whose amount depended
on energy, equivalent to approximately 1200 kg·days WIMP search data at
We also took electron recoil data using a137Cs source for the test crystal
used for the neutron calibration. This data is compared with the electron
recoil calibration data obtained for the full-size crystal to confirm that neutron
calibration data can be used for the full-size crystal.
6 Data analysis
Single photo-electrons (SPEs) in an event is identified by applying a clustering
algorithm to the FADC data. The energy deposition is evaluated from the sum
of charges of all SPEs in the event. Also, using the time information of SPEs
we calculate Mean Time (MT). The MT distribution of events above 3 keV
and up to 11 keV is used to extract the fraction of nuclear recoil events for
As one can see in Fig. 1 (b), the single clusters of low energy events are well
reconstructed in our DAQ system. A clustering algorithm to identify each SPE
signal is applied for the data analysis. The clustering algorithm includes the
identification of local maximum to form isolated cluster using the FADC bins
above the pedestal and the separation of neighboring cluster in the case two
local maximum is found in a cluster. A threshold is applied to the pulse height
to select SPE candidate. Additionally the SPE candidate with an unusually
narrow pulse width out of 3σ is rejected. Fig. 1 (c) shows result of clustering
of (b). The sum of the single cluster charges for whole window (32 µs) are
used to calculate the deposited energy. Low energy calibration is done using
a 59.5 keV γ peak from241Am.
Fig. 2 shows the charge distribution of single cluster after clustering of 5.9 keV
183.4 / 82
Number of clusters
183.4 / 82
Fig. 2. Single cluster charge spectrum. The distribution is fitted with two Poisson
peak from a55Fe source without height threshold. The distribution is fitted
by two superimposed Poisson functions (one for the SPE and the other for the
SPE-overlapped signal) with exponential noise component:
f = A
Γ(r + 1)+ Bµ′re−µ′
Γ(r + 1)+ Ce−x/λ
r = xg, µ = mg, µ′= m′g
where m(m′) is the mean of the Poisson distribution, and g is the gain factor
of the PMT. The fitting function is overlaid in the figure as a solid line. The
ratio of the contribution of the two Poisson distributions is 9.3%. Therefore,
we conclude that ∼90% of SPEs make single clusters and ∼10% makes over-
lapped clusters. The ratio of the mean values of the two Poisson distributions,
m′/m, is 2.11 ± 0.03 which is consisted with a expectation considering over-
lap of up to 3-SPEs. In this fit, the most probable value (MPV) of the SPE is
obtained as 0.86 pC. From the total charge of 59.5 keV from a241Am source,
the photoelectron yield of this crystal is obtained as 5.5/keV.
The photoelectron yield is calibrated at the beginning and at the end of a run.
The results show stability of the light output within 1% for the whole period.
The time-dependent gain variation is corrected by the MPV which is obtained
from the SPE charge spectrum of low energy (4-8 keV) WIMP search data
and its Poisson fit. Every one week’s WIMP search data are accumulated to
the fit. The result shows that the gain is stable within 5% for each PMT in
the whole period.
02468 1012 14
The biggest cluster charge(PE)
Mean Charge of clusters(PE)
02468 10 1214
The biggest cluster charge(PE)
Mean Charge of clusters(PE)
Fig. 3. (a) shows the charge of the biggest cluster normalized by the MPV of the
SPE charge obtained from Fig. 2 fit, versus the measured energy for the calibration
data with a137Cs γ source. Two solid lines indicate -1.65σ(lower solid line) and
1.28σ band (upper solid line). The vertical line is the 3keV analysis threshold. (b) is
a similar spectrum for the mean charge of clusters for the events within the signal
band of the biggest cluster cut. (c) and (d) are the corresponding plots for PMT
The PMT noise, which was also detected without the crystal, is seen by both
PMTs which have very fast timing characteristics. A similar noise was re-
ported by another group . It seems to be originated from a spark in the
dynode structure . The PMT noise usually induces an abnormally big
cluster. Therefore, the charge of the biggest cluster and the mean charge of
clusters for each event can be used to reject these events. We construct a
good event-band using the Compton scattering events from137Cs, and they
are compared with PMT noise events in Fig. 3. PMT noise events were taken
from the same system without the CsI(Tℓ) crystal in the copper chamber. The
distance between the two PMTs is maintained equal to that for the crystal-
attached set-up. In the 25.4 days data, equivalent to 167 kg·days WIMP search
data, only two events passed all the cuts with the 3keV energy threshold. We
conclude that the PMT background after the cuts is negligible. The same cut
is applied to the Compton scattering data and the efficiency was found to be
approximately 60% independent of energy as shown in Fig. 4.
We applied the same cuts to the calibration data taken with the test crystal.
The efficiency of137Cs and the neutron calibration data for the test crystal
Fig. 4. Efficiency calculated with γ calibration data for full-size crystal (filled circles),
test crystal (open circles), and neutron calibration data for test crystal (open square)
where only statistical errors are included.
are compared with137Cs data for the full-size crystal in the Fig. 4. About
a 10% difference between the full-size and test crystals is observed. This is
mainly due to the different PMTs used. We assign a systematic uncertainty
in efficiency to account for this difference. With a similarity of efficiency for
neutron and the γ calibration data in test crystal, we can conclude that we
can use γ calibration data for efficiency calculations of WIMP search data. A
slight difference in efficiency between γ and neutron data is also included as
a systematic uncertainty. The systematic error for the efficiency calculation is
where σcrystal diff is efficiency difference between the full-size crystal and the
test crystal, σrecoil diff is the efficiency difference between the nuclear recoil
and the γ recoil events. Efficiency corrected energy spectrum of events before
and after the cuts are shown in Fig. 5. Events below 11 keV are used for the
crystal diff+ σ2
To estimate the WIMP signal fraction in the WIMP search data, we introduce
a mean time (MT) value which is defined as
< t >=
where Aiand tiare the charge and the time of the ith cluster respectively,
and t0is the time of the first cluster (assumed as time zero).
Fig. 5. Energy spectrum in WIMP signal region before applying cuts (filled circles),
the big cluster events rejection with efficiency correction (open squares), and fitted
nuclear recoil rate (open circles) where the errors include systematic uncertainty of
efficiency for the latter two cases. A 90% upper limit on the nuclear recoil rate is
shown with a solid line.
Because we use different crystals for the neutron calibration, we need to con-
firm whether the two different crystals show the same MT characteristics. The
137Cs calibration data of the test crystal is compared with that of the full-size
crystal. As one can see in Fig. 6, the MT distribution of Compton electrons
in the test crystal is well matched with that of the full-size crystal. The mean
value of the log(MT) distribution as a function of energy is shown in Fig 7.
An excellent agreement between the test crystal and the full-size crystal for
the Compton electron allows us to use the neutron signal from the test crystal
as a reference for nuclear recoil signal for the full-size crystal. A slight MT
difference is adjusted by the assumption of a constant Rτ= τn/τe. Where
τnis the MT of the nuclear recoil and τeis the MT of the electron recoil.
Since the MT distribution depends significantly on the measured energy in
the low energy region, the log(MT) distribution in each keV energy bin is
fitted to the reference distribution for the same energy bin. The fitted nuclear
recoil event rate after the efficiency correction is given in Fig. 5. The fitted
nuclear recoil event rates are consistent with zero within one standard devi-
ation error for all energy bins. A 90% confidence level (CL) upper limit on
nuclear recoil event rates are shown with a solid line. Since below 3 keV the
PMT background contributes significantly and the pulse shape discrimination
power is less effective, we do not use events below 3 keV. In order to evaluate
Number of Events
Fig. 6. Mean time distributions of Compton electrons for the test crystal (filled trian-
gles) and full-size crystal (open circles) in the 4-5 keV energy range are compared.
Also, we include nuclear recoil(open squares), and the WIMP search data(filled
squares) for comparison.
Mean of Log(Mean Time)
0123456789 10 1112
RMS of Log(Mean Time)
Fig. 7. (a) The mean value of log(MT) as a function of measured energy for Compton
electrons with the test crystal (open circles), with the full-size crystal (filled circles),
and for the nuclear recoil with the test crystal (filled squares). (b) Root Mean
Square (RMS) of log(MT) as a function of measured energy for Compton electrons
with the full-size crystal.
nuclear recoil energy one needs to know the quenching factor (QF) defined
by the γ equivalent measured energy divided by the nuclear recoil energy. We
used the QF measured in our previous beam test . Our threshold of 3 keV
corresponds to 20 keV nuclear recoil energy.
7Result and discussion
Assuming a Maxwellian dark matter velocity distribution with a spherical halo
model discussed in Ref. , the total WIMP rate is obtained as
where R0 is the event rate per kg·day for vE = 0 and vesc = ∞, vesc =
650 km/sec is the local Galactic escape velocity of WIMP, mtis the mass of a
target nucleus, ρχ= 0.3GeV/cm3is local dark matter density, v0= 220 km/sec
is a Maxwell velocity parameter, and c1, c2 are constants, as discussed in
In order to estimate the expected event rates for each energy bin, we use MC
simulation. The MC simulation based on GEANT4  takes into account the
recoil energy spectrum, the QF, and the light transportation to the PMTs.
Then the simulated events are analyzed in the same way as the data except
for the applying any analysis cuts. The energy is tuned to provide good agree-
ment with the calibration data using 59.5 keV γ-rays from a241Am source.
Fig. 8 (a) shows good agreement between Monte Carlo and calibration data
for the energy distribution. The Monte Carlo generated electron equivalent
energy distributions (Eee) for several WIMP masses are shown in Fig. 8 (b).
Fig. 8. (a) Distribution of Eeefor the MC simulation (solid line) and calibration
data (filled circles) for a241Am source are compared. (b) Simulated Eeespectra for
several WIMP masses (20 MeV - dotted line, 50 MeV - dashed line, 100 MeV - solid
line, 1000 MeV - dotted dashed line) are shown.
From the rate of nuclear recoil in each energy bin, we can estimate the total
WIMP rate in comparison with the simulated Eeedistribution for each WIMP
mass by the following relation.
R(E0,E∞) = REkNtotal/NEk
where REkand NEkare the measured nuclear recoil rate and the simulated
WIMP events for each energy bin Ekrespectively and, Ntotalis the total num-
ber of WIMP events generated by simulation. With Eq. (1) and Eq. (2), we
can convert the rate of nuclear recoil in each energy bin to the WIMP-nucleus
cross section for each WIMP mass.
The limits on the cross-section for various energy bins and targets (Cs and I)
have been combined following the procedure described in Ref.  assuming
the measurements for different energy bins are statistically independent. The
combined result from energy bins for a WIMP-nucleus cross section is obtained
from this expression
where σW−Ais a combined WIMP-nucleus cross section, σW−A(Ek) is a WIMP-
nucleus cross section calculated in an energy bin Ek. As one can see in Fig. 5,
the rate of nuclear recoil events is consistent with zero. Therefore, we can set
the 90% CL upper limit on the WIMP-nucleus cross section with Eq. (3). In
this process, we assign zero as the mean value for the event rate for the bins
with negative means. The WIMP-nucleon cross section can be obtained from
WIMP-nucleus cross section by following equation
where µn,Aare the reduced masses of WIMP-nucleon and WIMP-target nu-
cleus of mass number A and CA/Cn= A2for spin independent interaction.
The limit on the WIMP-nucleon cross section for each nucleus can be com-
bined by this expression
A 90% CL upper limit on the WIMP-nucleon cross section from CsI for spin
independent interaction is shown in Fig. 9, together with the limits obtained
from two NaI(Tℓ) crystal based WIMP search experiments with similar pulse
shape analyses, NAIAD(UKDMC)  and DAMA . Although the amount
of data used to get our limit is 10 times less than that of NAIAD, we achieved
a more stringent limit than that of NAIAD due to the better pulse shape
discrimination and lower recoil energy threshold.
WIMP Mass (GeV)
Fig. 9. The KIMS limit on a WIMP-nucleon cross section for a spin independent
interaction with 237 kg·days exposure (solid line), DAMA positive  annual mod-
ulation signal (closed curve), NAIAD limit  with 3879 kg·days exposure (dashed
line) and, DAMA limit  with 4123 kg·days (dashed-dotted line) are presented.
Also the KIMS projected limit with 250 kg·year exposure of 2 CPD background
level(dashed-double-dotted line) is presented. Dotted line shows current best limit
by CDMS group .
The KIMS collaboration has developed a low background CsI(Tℓ) crystal for
the WIMP search. We set the first limit on the WIMP cross section using the
237 kg·days data taken with a 6.6 kg crystal. Our limit already partially ex-
cludes the DAMA 3σ signal region. The current experimental setup is designed
to accommodate about 250 kg of CsI(Tℓ) crystals without any modification. Download full-text
We expect that the background rate will be reduced to the level of approx-
imately 2 CPD or less for the new powder produced with purer water. The
projected limit for 1 year of data taken with 250 kg crystals of 2 CPD back-
ground is shown in Fig. 9 in comparison with the current best limit set by
CDMS . With the 250 kg setup, KIMS can explore the annual modulation
This work is supported by the Creative Research Initiative program of the
Korea Science and Engineering Foundation. We are grateful to the Korea
Middleland Power Co. and the staff members of the YangYang Pumped Power
Plant for their providing us the underground laboratory space.
 G. Jungman, M. Kamionkowski, K. Griest, Phys. Rep. 267 (1996) 195.
 R. Bernabei et al., Phys. Lett. B 480 (2000) 23.
 D. S. Akerib et al., Phys. Rev. Lett. 93 (2004) 211301.
 S. Marnieros et al., Nucl. Instrum. Methods A 520 (2004) 101.
 R. Bernabei et al., Int. J. Mod. Phys. D 13 (2004) 2127.
 R. Foot , Mod. Phys. Lett. A 19 (2004) 1841.
 F. Giuliani, Phys. Rev. Lett. 93 (2004) 161301
C. Savage et al., Phys. Rev. D 70 (2004) 123513.
 H. J. Kim et al., Nucl. Instrum. Methods A 457 (2001) 471.
 H. Park et al., Nucl. Instrum. Methods A 491 (2002) 460.
 S. Pecourt et al., Astropart. Phys. 11 (1999) 457.
 V. A. Kudryavtsev et al., Nucl. Instrum. Methods A 456 (2001) 272.
 T. Y. Kim et al., Nucl. Instrum. Methods A 500 (2003) 337.
 J. J. Zhu et al., High Energy Physics and Nuclear Physics 29 (2005) 721.
 M. J. Lee et al., paper in preparation.
 A. Bassignani et al., Radiation Measurements 25 (1995) 557