Low Noise Photo-Detectors for Application in Nuclear Imaging P. Weilhammer / CERN representing CERN* Institute of Nuclear Physics, Cracow - University of Ljubljana University of Michigan* University of Valencia, IFIC Collaboration *I would like to thank in particular: Andre Braem, Christian Joram and Neal Clinthorne from whom I borrowed a number of transparencies used in this presentation.
The Collaboration has recently been joined by other institutes INFN, Bari, University Hospital, Geneva, HUG (1), University of Geneva, Dept. of Physics, INT Lisbon, Ohio State University, INFN and University Perugia, University of Oslo, Dept. of Physics, ISS, Rome (INFN), Ioffe Institute, St. Petersburg Funding request to Brussels in FP6
1. INTRODUCTION: Motivation and Purpose 2. TWO PROJECTS USING DIFFERENT HIGH SENSITIVITY PHOTO-DETECTIORS THE PROSTATE PROBE PROJECT THE CONCEPT OF A NEW PET SCANNER 3. REQUIREMENTS ON Silicon Sensors Front-end Chip Connectivity 4. SOME RESULTS FROM DETECTOR COMPONENTS
MOTIVATION AND PURPOSE Imaging in Nuclear Medicine has had an important place in Health Care. Single Photon Emission Computed Tomography ( SPECT ) is a standard method in cancer diagnostics. Many attempts have been made to introduce new techniques which would have better performance than the classical Anger Camera which has been the mainstream technology over the last 35 years. In the pursuit of R&D to introduce new concepts one has to be aware that both medical doctors and industry are very conservative: e.g. most effort these days goes into replacing the detection method in Anger cameras but not the principle.
Instead of scintillator arrays read out with photo tubes big efforts go into readout of scintillator by Si photo diodes or replacing scintillator by heavy semiconductors like CdZnTe. But still keep the collimator! Another important factor is that any new technology in a given application must not be more costly than what is used at present in hospitals.
ANGER CAMERA: The standard method in most hospitals in Nuclear Medical Imaging to produce Single Photon Emission Computed Tomography (SPECT) images is the Anger camera based on the principle of mechanical collimators. Principle: Cell distributions of interest ( e.g. tumor) in the body are labeled with specifically engineered radio-tracers doped with radioactive isotopes, like 111 In, 99m Tc or 131 I. Photons emitted from a source have to pass a mechanical collimator, usually a few cm thick lead plate perforated with small diameter holes, in front of a suitable photo detector.
Photons which pass the holes are fully absorbed in e.g. NaI or BGO scintillation detectors and the signals are readout by photo multiplier tubes. Collimators impose hard physical constraints: detection sensitivity and spatial resolution are coupled in an inverse relationship. Limit of about 6 mm spatial resolution with a sensitivity of 10-4 to 10-5!
I will discuss in this presentation two specific R&D projects with the goal to achieve performance improvements in medical nuclear imaging applications. This R&D is based on experience gained in the past years in developments for HEP. The main ingredients are: Highly segmented silicon sensors Low noise front-end electronics Concept of a hybrid photon detector Two applications will be discussed: A COMPTON Camera for SPECT applications A fully 3D parallax free PET detector In both ( all) cases high quality photon-detection, both for γ rays and for scintillation photons, is of primordial importance.
1. Compton Camera Principle of the COMPTON Camera: The idea is to use Compton scattering of the primary γ ray in a first detector and the detection of the scattered γ ray in a second detector to measure the direction of the primary γ ray..
Spatial Resolution as a Function of Scattering Angle θ and Energy Resolution T (Compton Kinematics) 140 kev 511 kev The most important quantity is the energy resolution of the recoil electron! Higher g energy gives better resolution.
First results have been obtained by the CERN- Lubljana- Michigan-Valencia Collaboration about 3 years ago, using a first series of proto-type Si pad sensors and available front end electronics ( VA chips with backplane trigger). Results are encouraging but: Improvement both in pad sensors and selftriggering, sparse readout front-end electronics are required. 10 cm x 10 cm Multiple 131 I Point Sources Two 131 I point sources were placed at a distance of approximately 3cm. Resolution is about 7mm FWHM
A POSSIBLE FIRST APPLICATION: A Prostate Imaging Probe Illustration of possible prostate Imaging probe concept. Endo-rectal probe can view prostate internally. Can also be used externally. Si detector array of endorectal probe and possible shielding and septa against direct radiation from patient Second detector could be an Anger camera head without the collimator. ~ 4 cm Second Detector Lead Shielding First Detector Array Trans-Rectal Probe Imaging Table External Probe Second Detector ~ 12 mm high Lead Septa The proposed Prostate Probe
Some figures of merit expected from simulation for the Prostate Probe
Efficiency is very high ( in comparison with conventional Anger SPECT)
Comparison of Compton Prostate Probe with conventional SPECT ( from simulation!)
2. PET Detector with Parallax-free Compton Enhanced 3D Gamma Reconstruction PCT Patent Filed Existing PET scanners are limited in resolution, sensitivity, rate. Limitations are partly due to - parallax error, no DOI information - modest energy resolution - coarse segmentation, low readout speed (no data spying during acquisition) The main ingredient for this R&D is the Hybrid Photon Detector (HPD) which has originally been developed at CERN as the photo-detector for the LHCb RICH
Hybrid Photon Detector photocathode light quantum focusing electrodes e - V principle Developed and built @ silicon sensor + FE electronics real device segmented silicon sensor Readout logic Single photon imaging with 2048 channels Pad HPD 127mm Ø
HPD performance HPD combines single photon sensitivity of PMT with spatial and energy resolution of silicon sensor. Q.E. (%) 32 28 24 20 16 12 8 Sensitivity like classical PMT HPD PC87 (produced Easter Sunday 2001) Electronics noise well separated from signal Signal definition and energy resolution 4 0 200 300 400 500 600 lambda (nm) counts 1 p.e. 2 p.e. x x silicon silicon (mm) 30 20 10-0 -10 Imaging properties: 1 to 1 or linear demagnification m = 2.7 3 p.e. signal amplitude (a.u.) -20 114 mm -30 0.00 31.75 63.50 95.25 127.00 x cathode (mm) x cathode (mm)
HPD fabrication Facilities and infrastructure for the fabrication of large HPDs (up to 10 Ø) have been developed at CERN. Turbo Pump All ingredients for photodetector production are available: Design/simulation Photocathode processing (bialkali, Rb 2 Te, CsI) Glass / ceramic tube manufacturing Indium sealing technique
PET concepts Conventional PET geometry The proposed new PET geometry Every PMT reads 4 crystals. Slotted blocks of scintillator crystals are read out by 4 PMT. Hit cell identified by charge ratio of PMTs. No DOI. Axial arrangement of individual long scint. crystals Readout by HPDs on both sides. 1 crystal = 1 HPD channel 3 rd co-ordinate from from difference of signal strength on both sides
Features and main advantages of the concept Full 3D reconstruction of γ quanta without parallax error x,y from silicon pixel address, z from amplitude ratio of the 2 HPD s Precise Depth of Interaction DOI measurement Measurement of light yield on both sides of crystals Negligible statistical fluctuations in HPD Good γ energy resolution Reduced random coincidence rate due to fine granularity 3D reconstruction provides possibility of recuperating part of γ s which underwent Compton scattering in the detectors Compton enhanced sensitivity
Critical issues of the concept How to obtain optimum z resolution? How to arrange modules for very long (full body) scanners? The PET-HPD device may also be interesting for a conventional SPECT/PET geometries and other imaging applications.
The photodetector basic considerations HPD needs about 200 channels. Electronics encapsulated in vacuum envelope. focusing electrodes photocathode silicon sensor + FE electronics light quantum e - V HPD window must be flat and as thin as possible proximity focused electrostatics Segmentation of Si sensor matches crystal matrix Electronics must be autotriggering. segmented silicon sensor Minimize dead space Rectangular HPDs with rectangular Si sensor would give the best filling factor
Energy resolution R = FWHM E E ( ) = R Sci R stat R noise 2.5 % negligible (discussed later) R stat (6.5 7.1)% E γ 511 (kev) R 7 7.5% (FWHM) at E g = 511 kev 16% at 100 kev
Photoelectron distribution on Si plane ( for the usual case of a γ conversion with total deposited in 1 crystal) 75% of p.e. hit the central pad 95% are concentrated on 5 pads 75 Typical pattern on Si pads 1 5 1 5 5 1...
Reconstruction of the interaction point x-y: dimension of crystal determine precision L/2 σ x = σ y = 1 12 s 2.4 mm (FWHM) With light absorption length in scintillator λ α one can deternine z: ratio of light detected by the 2 HPDs z = 1 L + k 2 g λ a Q log Q R L k g is a geometrical factor eff k = 0. 8 Good linearity g z z
z - resolution σ z = k g λ a 2Q [ e ] z / λ + e ( L z )/ λ 1/ 2 a a (E γ = 511 kev, L = 100 mm) λ a = 75 mm λ a = 100 mm λ a = 125 mm λ a = 150 mm σ z (mm) σ z ~2.5 mm.
The HPD anode (round prototype) Base plate 5 (existing) 52 mm Si sensor (2 independent halves) 8 x 13 pads each (208 total) pad size 4 x 4 mm 2 Ceramic PCB 32 mm 2 VaTagp3 chips underneath Chips encapsulated in vacuum envelope A 24 wafer production run with these sensors is under production at SINTEF
The PET HPD Round prototype PCR5 127 mm Ø overall Proximity focused Sapphire window (d=1.8 mm) Ceramic body Nb skirt Nb electrodes Bialkali photocathode QE(370 nm) 25% U C 12 kv Gain 3000 Body construction by ceramic / metal brazing technique (under vacuum). Technology available at CERN.
PET camera prototype module 2 HPD PCR5 Scintillator array (208 crystals) Axial FOV = 10 cm
Proposed geometry for proof of principle 2 or 4 camera modules R=230 mm R=160 mm
Towards an optimized PET HPD design rectangular prototype (very preliminary) d = 1.3 mm Use of same Si sensor as PCR5 Ceramic PCB needs to be re-designed. 2 VATAGP underneath. Vacuum feedthroughs hidden under PCB Tricky but feasible connectivity (wire bonding) ~80 mm ~100 mm
Requirements on photo detection The most severe requirements on the sensor and the front-end electronics performance for single γ detection are those for the Compton Camera application. In particular for the lowest energy ( and most used isotope) 99m Tc (140 kev). To obtain good image resolution: σ(e e- ) < 1 1.2 kev FWHM 120-140 e - ENC Series Noise: Capacitance of sensor element (assume routing line resistance small( ~O(10 Ω)) Pad size: 1.4mm x 1.4mm needed to get sufficient σ x and σ y Back plane cap. Inter-pad cap. Routing line cap. 0.2 pf.3 -.5 pf ~ 2 pf
C total ~ 4pF For VATAGP chip at τ P = 1 µs ENC = 70e - + 12e - /pf ENC = 120 e - expected. Only possibility ( when using these simple sensors) to gain is to go to longer shaping times. Parallel noise: a. The sensors are dc coupled preamp feed-back resistor has to be high ( O(GΩ)) b. Leakage current of sensor: For 1 na at τ s = 1 µs ENC = 100 e- To have negligible contribution from leakage current: require I l < ~60 pa/pad! For 1 mm thick sensors and overdepletion
Other important requirements are Good timing resolution. Fast charge collection in sensor: over-depletion, electron collection On-chip time walk compensation All detector components have to be cheap and available in industry
Can one hope to get this performance with the simple devices proposed?: A test was done with an array of 25 diodes of 1 mm 2 DC-coupled Si diodes connected to VA3 chip. Determine electronic DE from 14 kev line produced by a 57 Co source. The 2 COMPTON edge shoulders from 122 kev and 136 kev can be well distinguished in the spectrum. DE = 500 ev FWHM obtained for 1 channel 1Mcts Run2 showing approx. theoretical values for the lines, continua, and backscatter peaks 100kcts Cobalt calc. Cobalt data Backscatter peaks kcts 300 200 Cobalt peak at 14.4keV FWHM=500eV 10kcts 100 0 Counts 1kcts 13.0 14.0 kev 15.0 16.0 100 cts 10 cts 1 cts 0 20 40 60 kev 80 100 120 140
Tests with Silicon Pad Sensors
Photograph of a 512 pad sensor with double metal routing to the periphery :
Microscope photograph of pad area of sensor Vias connecting metal 1 to metal 2 Gap between p + implants: 20 µm p + n pad 1.4 x 1.4 mm 2
Routing of pads to the peripheral bond pads for connection to readout chip. Alu lines are narrow (10µm) and thick (~3 µm) Bond pads meed to be on top of field oxide
RESULTS FROM FIRST MEASUREMENTS
I-V Characteristics of 1 mm pad sensors Produced by SINTEF, Oslo dc-coupled sensors! Measure a corner pad and ground surrounding pads and guard ring Leakage current of all sensors between 20pA and 45 pa at V= 500V Leakage Current [A] 50x10-12 40 30 20 10 waf35 waf45 I-V PAD in CORNER waf 38 waf39 waf41 waf42 waf 46 0 0 100 200 300 Bias Voltage 400 500
I-V Characteristics of 0.5 mm pad sensors Produced by SINTEF, Oslo 6 out of 7 sensors have I l between 10 and 20 pa @ 300 V; 7 th sensor has 140 pa 50x10-12 40 I-V CURVES EC PAD SENSOR CORNER PAD 2 ADJACENT PADS GROUNDED Leakage Current 30 20 10 0 0 50 100 150 Depletion Voltage 200 250 300
1/C 2 versus V depl for test diode on wafer 38 1mm thick 1/C 2 600 500 x10 21 400 300 Full Depletion Voltage ~ 135 Volts 200 0 100 200 300 400 500 Bias Voltage [Volts]
The Self-triggering Front-end Chip VATAGP3 Description of Architecture and Functionality Results from First Tests
t p = 170 nsec Gain-stage: 1 to 8 times t P = 4 ms 4 bit DAC for individual threshold correction
In this architecture 3 read-out modes are forseen: Serial readout of all 128 channels Sparse readout of all channels hit ( address and pulse-height) Sparse readout of channels hit plus n neighbors left and right
A Micro Photograph of parts of the actual chip
First Results on Performanc of one Chip assembled with a Si Pad Sensor on a G10 Hybrid
The Test Module
NOISE IN SLOW AND FAST SHAPER Measurements were done on a single channel with oscilloscope Analysis of the RMS of the trace without the Am signal results in ENC = 140 e - For single channel
Scope trace of fast shaper output Rise time ~ 160 ns Shaping time ~ 200 ns Noise of fast shaper ENC ~ 620 e -
Sparse readout works Ag Luminescence X-rays recorded in sparse readout mode. No pedestal subtraction Gaussian fit σ corresponds to about 220 e - ENC Still a factor 2 to go Cannels are not yet very uniform
Multiplicity in sparse readout with an Am source: nearly 100% of events have only one channel hit Same with a 90 Sr source: the betas scatter often over several pads as expected. Simultaneous signals in neighboring pads
OUTLOOK In 2 years: Full ring scanner avaiable Possible configuration for a Brain PET R = 170 mm 34 cm inner diameter 10 cm axial length 2496 crystals 24 HPDs total detection volume 2556 cm 3 F coverage 66% W coverage 18%