invited talk for the Samba2 conference

1
X-ray ImagingUsingSingle Photon
ProcessingwithSemiconductor Pixel Detectors
2
The Origins...

• High energy physics
• unambiguous reconstruction of particle patterns
  with micrometer precision
• low input noise due to tiny pixel capacitance

WA 97, RD 19 (CERN) 208Pb ions on Pb target 7
planes of silicon pixel ladders 1.1 M pixels
3
Hybrid Pixel Detectors

• Electronics
• CMOS technology advances steadily Moores law

• Sensors
• new materials to increase stopping power and CCE
  main problem inhomogeneities!

4
Single Photon Processing

• Quantum imaging
• Example photon counting

? Q has to correspond to a single particle!
5
Quantum Imaging - Advantages

• Noise suppression
• high signal-to-noise ratio dose reduction
• low rate imaging applications
• Linear and theoretically unlimited dynamic range
• Potential for discrimination of strongly Compton
  scattered photons (for mono-energetic sources) or
  e.g. fluorescence X-rays
• Energy weighting of photons with spectral sources
  possible
• higher dose efficiency dose reduction

6
Medical Imaging

• Detector requirements (sensor and electronics)
  depend on diagnostic X-ray imaging application.
• Example mammography
• spatial resolution 5-20 lp/mm
• high contrast resolution (lt3)
• uniform response
• patient dose lt3 mGy
• imaging area 18 x 24 (24 x 30) cm2
• compact and easy to handle
• stable operation
• no cooling
• digital
• cheap

Moore and direct detection quantum
processing sensors to be improved high DQE
(sensor q.p.) to be solved ???
7
Medipix1 / Medipix2

• Medipix1
• square pixel size of 170 µm
• 64 x 64 pixels
• sensitive to positive input charge
• detector leakage current compensation columnwise
• one discriminator
• 15-bit counter per pixel
• count rate 1 MHz/pixel (35 MHz/mm2)
• parallel I/O
• 1 ?m SACMOS technology (1.6M transistors/chip)

• Medipix2
• square pixel size of 55 µm
• 256 x 256 pixels
• sensitive to positive or negative input charge
  (free choice of different detector materials)
• pixel-by-pixel detector leakage current
  compensation
• window in energy
• discriminators designed to be linear over a large
  range
• 13-bit counter per pixel
• count rate 1 MHz/pixel (0.33 GHz/mm2)
• 3-side buttable
• serial or parallel I/O
• 0.25 ?m technology (33M transistors/chip)

8
Medipix1 / Medipix2
the prototype
the new generation!
9
Medipix1 Applications

• Examples
• Dental radiography
• Mammography
• Angiography
• Dynamic autoradiography
• Tomosynthesis
• Synchrotron applications
• Electron-microscopy
• Gamma camera
• X-ray diffraction
• Neutron detection
• Dynamic defectoscopy
• General research on photon counting!

10
Applications
Dynamic Autoradiography (INFN Napoli)
Mammography (INFN Pisa, IFAE Barcelona)
Mo tube 30 kV Medipix1 part of a
mammographic accreditation phantom
Medipix1 14C L-Leucine uptake from the solution
into Octopus vulgaris eggs (last slice in time
80 min)
11
Applications
Dental Radiography (Univ. Glasgow, Univ.
Freiburg, Mid-Sweden Univ.)
Sens-A-Ray commercial dental CCD system (Regam
Medical)
Medipix1
160 ?Gy
80 ?Gy
40 ?Gy
12
Medipix1 - SNR

• Pixel-to-pixel non-uniformities
• optimum for counting systems Poisson limit ? N
• optimum SNR N / ? N
• determined SNR for
• Medipix1 taking flood fields
• (Mo tube) covering the entire
• dynamic range of the chip
• ?
• SNRuncorr(max.) 30
• using a flatfield correction
• ?
• Medipix1 follows perfectly
• the Poisson limit!

Red curve Poisson limit
SNRuncorr
13
Medipix1 - SNR
SNRuncorr
8.5 keV 11.7 keV 12.4 keV
with adj. (35V det. bias) 29.8 18.8
without adj. (35V det. bias) 7
with adj. (17V det. bias) 19.2
with adj. (80V det. bias) 30.7

• differences in the raw SNR, but with flat field
  correction the Poisson limit is ALWAYS reached
• BUT flat field correction dependent on energy
  spectrum!
• working in over-depletion reduces charge sharing
  effects

?? flat field corrects mainly sensor
non-uniformities!
14
Medipix1 Flat Field Studies
2 kinds of non-uniformities waves and fixed
pattern noise
17 V detector bias (under-depleted)
35 V detector bias (fully depleted)
waves due to bulk doping non-uniformities
raw image
wrong flat field inverse waves, BUT single
pixel inhomogeneities smeared out ? fixed
pattern noise!
flat field corrected
15
Si Wave Patterns

• vary detector bias voltage from under- to
  over-depletion
• divide flat field map _at_Vbias with map _at_100 V

16
Si Wave Patterns

• Section of the correction map for different
  detector bias
• waves move in under-depletion stable in
  over-depletion
• amplitude decreases with bias, but waves dont
  disappear completely
• Remark images can be corrected for these
  non-uniformities

17
Dose Optimization

• Dose optimization for specific imaging tasks
• example accumulation of single X-ray signals
  during X-ray of an anchovy

18
Summary Medipix1

• The Medipix1 prototype chip allows to study the
  photon counting approach
• Comparison to charge integrating systems turned
  out to be sometimes difficult due to the larger
  pixel size of Medipix1
• Most of the problems encountered were due to
  sensor non-uniformities (e.g. locally varying
  leakage currents) and bump-bonding quality
• Medipix1 turned out to be a tool to study the
  attached sensor even silicon sensors show
  non-uniformities
• The flat field correction was intensively studied
  and allows to minimize the pixel-to-pixel
  variations down to the Poisson limit over the
  full dynamic range of the chip. The energy
  dependence of the flat field correction has to be
  further investigated.
• The experience with Medipix1 lead to many
  improvements implemented in the Medipix2 ASIC.

19
Medipix2 Characterization

• all the reported measurements were done using the
  electronic calibration (injection capacitor
  external voltage pulse).
• The 8 fF injection capacitor nominal value has a
  tolerance of 10.
• The dedicated Muros2 readout system had been used

20
Medipix2 Characterization
adjusted thresholds 110 e- rms
unadjusted thresholds 500 e- rms
21
Medipix2 Characterization

• Threshold linearity in the low threshold range

22
Medipix2 Characterization

• threshold at 2 ke-
• injection of 1000 pulses of 3 ke-
• matrix unmasked

23
Summary of the Electrical Measurements
Electron/Hole Collection
Gain 12 mV/ke-
Non-linearity lt3 to 80 ke-
Peaking time lt200 ns
Return to baseline lt1?s for Qin lt50 ke-
Electronic noise ?nTHL 100 e- ?nTHH 100 e-
Threshold dispersion ?THL 500 e- ?THH 500 e-
Adjusted threshold dispersion ?THL 110 e- ?THH 110 e-
Minimum threshold 1000 e-
Analog power dissipation 8 ?W/channel at 2.2 V supply
24
Conclusions

• Miniaturization of CMOS technology allows for
  small pixel sizes and increased functionality.
• A new single photon processing chip Medipix2
  consisting of a 256 x 256 matrix of 55 ?m square
  pixels has been produced and successfully
  characterized.
• The potential of quantum imaging for various
  applications is still far from being fully
  explored.
• Quantum imaging in the medical domain
• rather complete systems are required to convince
  end users
• MTF and DQE curves as well as comparative phantom
  images are necessary for approval (see e.g. FDA)
• A lot of progress has been made to achieve large
  areas as yet no satisfactory solution for most
  medical applications
• There is a trend in some applications towards
  object characterization in addition to simple
  transmission images ?need energy information
• ? colour X-ray imaging

25
Wishlist

• sensors high absorption efficiency and improved
  homogeneity
• reliable ASIC-to-sensor connections
• tiling large areas without dead space
• ASIC
• small pixel size with charge sharing solutions
  (modern CMOS technologies!)
• low-noise front-end with appropriate sensor
  leakage current compensation sensitive to
  electron and hole signals
• very fast front-end for time-resolved studies
• a precise threshold above noise
• a multi-bit ADC/pixel for energy information
  (optimum weighting!)
• large dynamic range
• ???
• cost!

26
Medipix1 Flat Field Studies
a phantastic tool to study sensor inhomogeneities

• vary detector bias voltage from under- to
  over-depletion
• calculate corresponding flat field from flood
  images (1st row)
• divide with correction map from 100 V detector
  bias data (2nd row)

27
Medipix1 Flat Field Studies
28
Medipix2 Characterization
adjusted thresholds 110 e- rms
mean 1100 e- spread 160 e- rms
unadjusted thresholds 400 e- rms