IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 58, NO. 1, FEBRUARY 2011 323
Characterization of Medipix3
With Synchrotron Radiation
Eva N. Gimenez, Rafael Ballabriga, Member, IEEE, Michael Campbell, Ian Horswell, Xavier Llopart,
Julien Marchal, Kawal J. S. Sawhney, Nicola Tartoni, Member, IEEE, and Daniel Turecek
Abstract—Medipix3 is the latest generation of photon counting
readout chips of the Medipix family. With the same dimensions
as Medipix2 (256
256 pixels of
55 m 55 m
pitch each),
Medipix3 is however implemented in an 8-layer metallization 0.13
m
CMOS technology which leads to an increase in the function-
ality associated with each pixel over Medipix2. One of the new
operational modes implemented in the front-end architecture is
the Charge Summing Mode (CSM). This mode consists of a charge
reconstruction and hit allocation algorithm which eliminates
event-by-event the low energy counts produced by charge-shared
events between adjacent pixels. The present work focuses on the
study of the CSM mode and compares it to the Single Pixel Mode
(SPM) which is the conventional readout method for these kind
of detectors and it is also implemented in Medipix3. Tests of a
Medipix3 chip bump-bonded to a 300
m
thick silicon photodiode
sensor were performed at the Diamond Light Source synchrotron
to evaluate the performance of the new Medipix chip. Studies
showed that when Medipix3 is operated in CSM mode, it generates
a single count per detected event and consequently the charge
sharing effect between adjacent pixels is eliminated. However in
CSM mode, it was also observed that an incorrect allocation of
X-rays counts in the pixels occurred due to an unexpectedly high
pixel-to-pixel threshold variation. The present experiment helped
to better understand the CSM operating mode and to redesign the
Medipix3 to overcome this pixel-to-pixel mismatch.
Index Terms—Medipix3, synchrotron radiation, X-ray detec-
tors, X-ray imaging.
I. INTRODUCTION
T
HE brilliance now being achieved at the 3rd generation of
synchrotrons has increased the potential for high quality
X-ray diffraction experiments. Hence, new high-performance
detectors with a high spatial resolution, very high dynamic
range and high detection efficiency are required. Hybrid silicon
photon-counting detectors can meet these requirements [1].
Manuscript received June 30, 2010; revised October 07, 2010; accepted Oc-
tober 12, 2010. Date of publication December 03, 2010; date of current version
February 09, 2011. This work was supported by Diamond Light Source.
E. N. Gimenez, I. Horswell, J. Marchal, K. J. S. Sawhney, and N. Tartoni are
with Diamond Light Source, Harwell Science and Innovation Campus, Didcot,
Oxfordshire OX11 0DE, U.K. (e-mail: Eva.Gimenez@diamond.ac.uk; Ian.Hor-
swell@diamond.ac.uk; Julien.Marchal@diamond.ac.uk; Kawal.Sawhney@di-
amond.ac.uk; Nicola.Tartoni@diamond.ac.uk).
R. Ballabriga, M. Campbell, and X. Llopart are with CERN, European Organ-
ization for Nuclear Research, CH-1211 Genève 23, Switzerland (e-mail: Rafael.
Ballabriga@cern.ch; Michael.Campbell@cern.ch; Xavier.Llopart@cern.ch).
D. Turecek is with the Institute of Experimental and Applied Physics, Tech-
nical University in Prague, 128 00 Prague 2, Czech Republic (e-mail: Daniel.
Turecek@cern.ch).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TNS.2010.2089062
These detectors normally consist of monolithic arrays of silicon
photodiodes bump-bonded to CMOS electronic readout chips.
Currently both fields, silicon photodiodes and CMOS tech-
nology, are experiencing considerable improvements which are
leading to an increase in detector performance. New generations
have also emerged in other families of hybrid pixel detectors,
such as Pilatus II [2] and XPAD3 [3], [4]. The advantages of
these new photon-counting detectors were quickly appreciated
in scattering experiments. However, their pixel pitch of 172
and 130 respectively, limited their satisfactory use in other
experiments as for example with coherent X-ray diffraction.
When a survey of the requirements for future detectors was car-
ried out in conjunction with beamline scientists at the Diamond
Light Source (DLS) synchrotron [5], the need for detectors
with a pixel pitch smaller than 100
was highlighted with
the additional property of being radiation hard [6].
The present work was focused on the new Medipix3 readout
chip [7], [8] which takes advantage of the advances in CMOS
technology to allow a high level of functionality to be imple-
mented in each pixel of size
. The Medipix3 chip
enables the development of hybrid pixel detectors which ac-
complish the aforementioned requirements and can improve the
quality of data acquired in synchrotron experiments. The eval-
uation of the new Medipix3 readout chip was done at beamline
B16 [9] of the DLS synchrotron. Experiments were performed
in order to study the detector response and its image quality.
II. M
EDIPIX3DETECTOR
The detector consisted of a 300
thick pixel array silicon
sensor of resistivity 13
bump-bonded to the Medipix3
readout chip. The system was operated with a bias voltage of
90 V resulting in full depletion of the sensor and was readout
using Medipix3 USB interfaces and Pixelman software, pro-
duced by the IEAP [10], [11].
Medipix3 readout chip, the latest in the Medipix family,
has the same dimensions as Medipix2 (256
256 pixels of
size each) but it is manufactured in an 8-metal
0.13
CMOS technology which enables further functions to
be implemented for each pixel. Amongst others, Medipix3 has
a new front-end architecture aimed at eliminating the spectral
distortion produced by the charge sharing seen in highly seg-
mented semiconductor detectors, as observed in the Medipix2
chip [12]–[14]. In the new architecture, the charge generated
by a photon is reconstructed by grouping pixels in clusters of
four and summing the charge collected in each cluster. The
event is then assigned as a single hit to the summing circuit
with the largest charge deposit. This operating mode, referred
to as Charge Summing Mode (CSM), is complementary to the
0018-9499/$26.00 © 2010 British Crown Copyright
324 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 58, NO. 1, FEBRUARY 2011
operating mode in conventional detector systems, referred to as
Single Pixel Mode (SPM) [15]. The Medipix3 chip may still be
programmed for SPM by disabling the charge reconstruction
and the communication between neighbouring pixels. Another
of the new operational modes enables the size of the pixel
readout channel to be configured. It is also possible to group
the readout pixels in clusters of four and the cluster to become
a single detection unit allowing up to 8 thresholds for energy
binning, this is referred to as Spectroscopic Mode (SM). In this
mode, the chip should be connected to a sensor with a pixel
pitch of 110
.
Each pixel cell of the readout chip occupies an area of
and contains an analog front-end and dig-
ital processing circuitry. The analog front-end consists of a
preamplifier, a shaper and two threshold discriminators. The
first discriminator is used to define the global lower threshold
(THL) and as an input to the arbitration logic when charge sum-
ming is enabled. The second discriminator can be used to define
a global high threshold (THH). The digital circuitry contains
control logic, arbitration modules for hit allocation, circuitry
for storage of the pixel configuration data and two registers
that can be configured as two 1-bit, 4-bit or 12 bit counters or
as a single 24 bit counter. In 24-bit counting mode only one
discrimination level is used [16]. Medipix3 also exhibits two
operational gain modes: low gain mode and high gain mode. In
the tests performed at the synchrotron, only the high gain mode
was used, since the energies of the experiment went from 6 keV
to 20 keV. Furthermore, techniques were used in the design of
the chip in order to minimize the radiation damage.
III. B
EAMLINE SET-UP
The DLS synchrotron is made up of three stages: a linac, a
synchrotron booster and a storage ring. For the tests, the energy
of the electrons circulating in the storage ring was 3 GeV and
were grouped in bunches 10 ps long with a minimum spacing
between contiguous bunches of 2 ns. Up to 936 contiguous
bunches can travel in the storage ring of DLS; however gen-
erally DLS is operated in gapped mode where not all 936
bunches are present. During our experiment 685 contiguous
electron bunches were present in the storage ring and 251
missing. Therefore, the generated X-ray beam had an overall
burst duration of 1370 ns followed by 502 ns period without
photons. The cycle was repeated every 1.872
, which is the
revolution period of the electrons in the storage ring of DLS.
The storage ring was operated at a current of 150 mA in top-up
mode, i.e. a small amount of electrons were injected every ten
minutes to compensate for losses and keep the current constant
to within 1%.
Beamline B16 aims at developing novel techniques and
characterizing optics, detectors and other instrumentation. It
provides both white and monochromatic X-ray beams in the
4–25 keV photon energy range. The Medipix3 detector test con-
sisted of several experiments. In all of them a monochromatic
beam achieved by a Si(111) double crystal monochromator
was used for the characterisation of the detector. Depending on
the experiment, the dimension and characteristics of the X-ray
beam varied as described in the next subsections.
Fig. 1. Experimental set-up at B16 test beamline for monochromatic micro-
focused beam.
A. Unfocused Monochromatic Beam
An unfocused monochromatic beam (set at several energies
between 6 keV and 20 keV) was used to study the threshold
equalization procedures, the linearity and the homogeneity of
the detector’s response. Appropriate beam sizes were generated
using the beamline slits.
B. Micro-Focused Monochromatic Beam
A special feature of this beamline is the ability to deliver a
micro-focused beam a few micrometers in size. The charge col-
lection and readout process of the CSM and SPM modes of
the Medipix3 detector were studied by scanning the pixels in
the detector with a micro-focused X-ray beam. The micro-fo-
cused monochromatic beam was generated using a compound
refractive lens (CRL) and delivered photons at 15 keV. The ex-
perimental set-up is shown in Fig. 1. The CRL comprises 98
lenses stacked in line in a He filled chamber. All the lenses
were identical: made of beryllium, of parabolic shape, with a
radius of 0.2 mm at the apex and a geometrical aperture of 1
mm. The parabolic shape provides 2 dimensional focusing. The
Medipix3 detector was installed approximately 0.67 m down-
stream of the CRL lenses, on a versatile optic table where the
detector and CRL lenses were aligned to the synchrotron beam.
The size of the micro-focused beam was measured by taking
transmission scans of 200
diameter Au cross-wires. The
derivatives of the wire scans gave the beam size. The raw data
and the derivative, for vertical and horizontal scans, are shown
in Fig. 2. The measured FWHM size of the micro-focused beam
was
in and , respectively.
IV. T
HRESHOLD EQUALIZATION
The Medipix3 chips contain two 9-bit Digital to Analog
Converters (DACs) for setting the global low and high thresh-
olds (THL and THH respectively) used to determine whether or
not a hit is registered. However CMOS transistors experience
mismatch, which is the process that causes random variations
in the physical quantities of identically designed devices. This
mismatch causes pixel-to-pixel gain and offset variations in
the analog front-end circuits. To correct for this mismatch,
threshold equalization needs to be performed before using
the detector for measurements. A simplified schematic of the
threshold equalization circuitry for SPM mode is shown in
Fig. 3. Transistor mismatch in the front-end causes a DC level
GIMENEZ et al.: CHARACTERIZATION OF MEDIPIX3 WITH SYNCHROTRON RADIATION 325
Fig. 2. Transmission signal from a Au wire scan (dashed line) in the focal plane
of the compound refractive micro-focusing lens measured at 15 keV photon en-
ergy. The derivative of the raw data (dots) and a Gaussian fit (solid line) are also
shown. Plots (a) and (b) are for vertical and horizontal beam profiles, respec-
tively.
shift of the shaper output current whose amplitude contains
the information of the induced charge in the pixel input pad.
Each pixel of the Medipix3 has DACs to adjust this DC level
which connects to the crossing discriminator. The DAC uses
5 equalization bits: one bit controls the
current source,
which is activated when the DC offset is negative; and the other
4 bits are used for threshold trimming by means of the current
. The global threshold is set by the current source
. Two 8-bit DACs in the periphery are used to generate
the bias for the
and current sources (THN and
DAC_Pixel respectively) [15], [16].
Two threshold equalizations of the pixel matrix for the
THL (THH was not used during these tests) were performed:
one using the noise edge of the THL threshold scan (see
Section IV-A); the other with X-rays of a given energy (see
Section IV-B) using the inflexion point of the integral spectrum
obtained by plotting the total number of counts when scanning
along the THL DAC (referred to as S-curves).
A. Threshold Equalization on Noise
The THL threshold equalization on noise starts by defining a
THL DAC target value where the noise edge of all pixels will
be set after threshold tuning. The edge was defined by the first
Fig. 3. Schematic of the front-end circuitry in Single Pixel Mode including the
threshold adjustment mechanism and the zero-crossing discriminator.
I
sets
the global threshold.
I
and
I
adjust the threshold in the matrix.
point in the S-curve where each pixel detects three events in the
sample time. Subsequent THL threshold scans using the equal-
ization algorithm are used to optimize the settings of two of the
periphery DACs (THN DAC and DAC_pixel). After the optimal
value of these DACs is found, the on-pixel 5-bit DAC equaliza-
tion coefficients are tuned to locate the pixel noise edge as close
as possible to the THL DAC target value.
B. Threshold Equalization on 8 keV X-rays
In order to flood the detector with X-rays, a Cu sheet was
placed at 45 degrees to the unfocussed monochromatic beam to
produce fluorescence X-rays of energy 8 keV. The beam energy
was tuned to 9.3 keV (above the K-edge of Cu). The detector
was placed at a distance of 20 cm from the Cu target and per-
pendicular to the direction of the 8 keV X-rays. The threshold
equalization followed the same process as the one described is
Section IV-A. for the threshold equalization on noise keeping
the previously found settings of THN and DAC_pixel periphery
DACs. The definition of the initial THL DAC target value was
set according to the energy of the incoming radiation, in this
case 35 DAC units higher than the noise edge for this energy.
C. Threshold Equalization Precision in Single Pixel Mode
The precision of the threshold equalization procedures was
investigated by flood illuminating the detectors with 8 keV
X-rays from the Cu sheet and scanning the THL threshold
across the X-ray signal down to the noise for the two types of
threshold equalization masks loaded in the detector. Fig. 4(a).
and Fig. 5(a). show the individual pixel S-curves for the 8 keV
X-ray illumination obtained with the equalization on noise and
on 8 keV X-rays, respectively. The mean of the selected pixels
response is plotted as a black wide line.
In these figures, S-curves of individual pixels are less widely
spread around the mean in the region of the inflection point
when the detector THL was equalized with X-rays instead of
with the noise edge. This is due to the fact that the equalization
done with X-rays compensates for gain and offset variations be-
tween pixels whereas equalization done on noise compensates
only for offset variations.
An estimation of the threshold dispersion over the detector
can be obtained by plotting the threshold value corresponding
to the edge (see Section VI-A) of the X-ray signal S-curve for
each pixel. A histogram of these values is shown in Fig. 4(b). and
326 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 58, NO. 1, FEBRUARY 2011
Fig. 4. THL equalized on the noise edge for a detector operated in Single Pixel
Mode. (a) A ROI of 10
2
10 individual pixel S-curves are displayed around the
mean (black line) corresponding to X-ray signal. (b) Distribution of pixel THL
values after equalization on noise.
Fig. 5. THL equalized on 8 keV X-rays edge for a detector operated in Single
Pixel Mode. (a) A ROI of 10
2
10 individual pixel S-curves are displayed around
the mean S-curve (black line) corresponding to X-ray signal. (b) Distribution of
pixel THL values after equalization on 8 keV X-ray edge.
TABLE I
S
TANDARD DEVIATION VALUES
(
)
OF THE
THRESHOLD DISPERSION
HISTOGRAM FOR THRESHOLD EQUALIZATION ON NOISE AND ON 8-keV
E
NERGY PHOTONS IN SPM MODE
Fig. 5(b). A Gaussian distribution was fitted to the threshold dis-
persion histogram for the threshold equalization on noise and on
8 keV energy photons in SPM. Table I summarizes the standard
deviation values of the Gaussian fit. The measured threshold
variation was 3.5 times larger than that expected from simu-
lations, this discrepancy is being investigated. The equivalent
values in keV and e-rms were calculated using the conversion
factor of 0.36 keV/DAC step obtained from the energy calibra-
tion of the detector which will be discussed in Section V and the
energy to electrons conversion factor of 3.6 eV/e-hole pair for
silicon.
This study was repeated by detecting 16 keV X-rays from a
Zr sheet with each of the equalization masks, on noise and on
8 keV X-rays. The set-up and procedure was the same as the
previously described but with a Zr sheet instead of a Cu one.
Results show a 30% increase in the dispersion of the pixel THL
values for 16 keV X-rays compared to the dispersion measured
for 8 keV X-rays.
Fig. 6. A ROI of 10
2
10 individual pixel S-curves are displayed around the
mean (black line) for THL equalized on the noise edge (a) and 8 keV X-rays
edge (b) for a detector operated in Charge Summing Mode.
D. Threshold Equalization in Charge Summing Mode
Threshold adjustment was also performed on the noise edge
and Cu X-ray fluorescence edge for the detector operated in
CSM mode. Fig. 6. shows the individual pixel S-curves for
the Cu flood illumination obtained after equalization on noise
and on 8 keV X-rays, respectively. The S-curves for individual
pixels show a larger dispersion in the pixel response compared
to SPM mode. This is explained by the observation that some
pixels are never collecting any X-rays because their hits are
wrongly being allocated to their neighbours (see Section VIII).
This effect is present independently of the threshold equal-
ization method used (noise or X-rays). The strong distortion
observed in CSM mode further confirms the unexpectedly high
component mismatch which was already observed in SPM
mode (see Section IV-C).
V. L
INEARITY
Unfocused monochromatic X-ray beams at 6 keV, 10 keV, 15
keV and 20 keV photon energies were used to study the linearity
of the detector. Additionally, the monochromator was adjusted
in order to provide the 3rd harmonic of these energies. Using
these 3rd harmonics, energies at 18 keV, 30 keV, 45 keV and 60
keV were also available. Before acquiring data, the threshold
equalization procedure on noise was used to trim the pixel ma-
trix. The same threshold equalization file was used for all en-
ergies except for 6 keV. For this energy, a new THL threshold
equalization on noise was performed, since this energy was very
close to the noise edge (see Fig. 7.).
For each one of the aforementioned X-ray energies, a scan of
the THL threshold was performed. The total number of counts
was calculated for each image (excluding noisy pixels), pro-
ducing an integral spectrum of counts vs low threshold (THL)
setting. Fig. 7. shows the S-curves for 6 keV, 15 keV and 20
keV photon energy for SPM and CSM modes. In all energies,
there is a slope related to charge-shared events for SPM mode. In
CSM mode instead of this slope there is a flat region which im-
plies that the charge sharing effect is being corrected when op-
erating the chip in this mode. Fig. 8. shows the differential spec-
trum obtained from the S-curves for SPM and CSM modes at 15
keV energy photons. For SPM mode, additional counts at low
energy are observed, which are due to the charge cloud being
shared between pixels. Whereas in CSM mode, a clear photo
peak spectrum is visible. In this mode, the charge reconstruc-
tion and hit allocation algorithm implemented in the front-end
GIMENEZ et al.: CHARACTERIZATION OF MEDIPIX3 WITH SYNCHROTRON RADIATION 327
Fig. 7. Global S-curves for 6 keV, 15 keV and 20 keV monochromatic X-ray
beams for a Medipix3 operated in Single Pixel Mode (a) and in Charge Sum-
ming Mode (b). Charge sharing events contribute to the slope in the S-curves
for Single Pixel Mode. In Charge Summing Mode, the flat region implies that
charge-shared events are eliminated.
Fig. 8. Differential spectrum for Single Pixel Mode and Charge Summing
Mode at 15 keV X-rays. Single Pixel Mode spectrum shows collection of
counts at low energies due to the charge shared effect between pixels. For
Charge Summing Mode this effect is corrected.
design of the Medipix3 chip eliminates event by event these low
energy counts produced by charge-shared events.
For each differential spectrum, a Gaussian was fitted to the
peak centroid, obtaining a correspondence between the THL
value and the beam energy for each of the aforementioned eight
energies. The linearity response of the detector for each mode
under study is shown in Fig. 9. The detector is linear up to 30
keV for both SPM and CSM modes. A linear least-squares fit
was applied to these points to find the calibration THL-energy
for each mode. Results show that a step of 1 in THL corresponds
to a step in energy of 0.360 keV in SPM mode and 0.350 keV
in CSM mode, showing very good agreement between both op-
erational modes. These conversion factors were then applied to
the spectra. Table II shows the standard deviation value of the
Fig. 9. Linearity response of the detector at different energies for Single Pixel
Mode and Charge Summing Mode. Solid lines show the fits.
TABLE II
S
TANDARD DEVIATION
VALUES
(
)
OF THE
DIFFERENTIAL SPECTRUM
FIT AT
DIFFERENT
ENERGIES IN
SPM
AND CSM M
ODES
Gaussian fit performed on the differential spectrum for each of
the energies.
VI. D
EAD
TIME
The dead time of the electronic chain of the individual pixels
(preamplifier, shaper, discriminator, counter) is an important pa-
rameter of the detector to evaluate. The measured counting rates
have to be corrected in order to get the actual photon flux im-
pinging on the pixel when the intensity is very high. An estima-
tion of the dead time of the detector was obtained by exposing
the detector to the unfocused direct beam. The counting rate of
selected pixels was measured as aluminium attenuation foils,
each of the same thickness, were placed in the beam. The re-
sulting counting rates, referred to as output counting rate (ocr),
were fitted with the well known
paralyzable model [17], [18].
The free parameters were: the input count rate (icr) with no at-
tenuation, the transmission of a single foil
and the dead time
of the detector
. The formula fitted was (1):
(1)
where
is the number of foils.
Dead time was evaluated for the SPM operational mode. In
order to have the best estimate of the dead time, the model
was fitted to seven pixels whose counting rate was higher than
200,000 counts per second when the beam was not attenuated.
The transmission of a single aluminium foil was a parameter
common to the seven sets of data and gave a value of 0.958; this
is consistent with the thickness of the foils being 25
. The
results for the dead times of the seven pixels were between 1.1
and 1.4 . Fig. 10. shows the counting rates for the seven
