1. Comparing geometries
Kenneth Wraight on behalf of UK ATLAS ITk Pixel Upgrade
Cluster based comparison of alternative ITk geometries
Alternative geometries
Mapping pixels to readout with strong interactions
Comparison of geometries
Clusters
Charge ‘bleeding’
Residuals
AC/DC comparison 1

2. Alternative geometries
Cern Pixel 5 (CP5) production involved alternative geometry pixel sensors
Possible alternative for improved vertex resolution  b-tagging
Charge sharing, capacitance, biasing
Pitch (µm2)
Ncols:Nrows
Channels
250x50
80:336
26880
125x100
160:168
167x125
120:134
16080 (3/5)
500x25
40:672
Data taken over several test beams
DESY 2013/14/15
SLAC 2014
CERN 2015
Devices come from Micron, bump-bonded at Advacam
FEI4 readout chip: 80x336 channels: 20x16.8mm2
Pixel array may not match readout channel array
Problematic for clustering: pixel-readout mapping required
AC/DC readout possible (only DC readout devices studied here)
300 μm, 10 kOhms/cm, DC, ~80V depletion voltage
Pixel efficiency comparison (Milovanovic et al, JINST):

3. Alternative geometries
500x25µm2
125x100µm2
Each readout pattern specific to sensor
Square pixels have long readout strips to accommodate geometry
167x125µm2

4. Strong scattering at CERN
SPS pion beam
Pions interact strongly with nucleus (moreso with >Z)
Secondary interactions seen
in detectors
Isotropic decays crossing
many pixels
See increased pixel
occupancy w.r.t. DESY
 Larger cluster sizes
rate >100 (250x50µm2) pixel clusters: 5 in 10,000
Not enough to upset reconstruction
Reference sensors
Pixels hit/event
250x50µm2
500x25µm2
167x125µm2

5. Visual Mapping: 500x25µm2
Check mapping in events with >100 pixels firing
unmapt
mapt v2
mapt v2
zoom
mapt v1

6. Mapping by Cluster Parameters 500x25µm2
Visual analysis matches mapped clustering results
Mapping version1 has correct ordering of cluster sizes and looks reasonable w.r.t. reference  good
Version2 has almost equal cluster sizes and doesn’t compare well to reference  bad
So can make comparison of AC & DC (THL=3000, -100V)
DUT
Sig
cluX
cluY
mult
250x50
9.07
1.06
1.41
1.27
500x25 v1
8.75
1.02
1.81
1.25
500x25 v2
6.93
1.38
1.44
1.58
DUT
Sig
cluX
cluY
mult
500x25 DC
8.75
1.02
1.81
1.25
500x25 AC
7.93
1.03
1.69
1.16

7. Mapping by Cluster Parameters 167x125µm2
TB Itinerary
Mapping by Cluster Parameters 167x125µm2
From comparing mapped cluster parameters:
Versions 2,3&5 have wrong order of cluster size  bad
Versions 2,3&4 have parameters which are too low bad
Versions 1,6 seem to have correct cluster parameters…
DUT
Sig
cluX
cluY
mult
250x50
9.07
1.06
1.41
1.27
167x125 v1
9.93
1.31
1.35
1.08
167x125 v2
7.99
1.11
0.888
167x125 v3
7.90
0.915
167x125 v4
6.76
1.03
1.04
0.539
167x125 v5
9.66
1.26
1.14
167x125 v6
1.25
1.33

8. Visual Mapping: 167x125µm2 Check any events with >100 pixels firing
unmapt
zoom
unmapt
mapt v1
zoom
mapt v6
zoom

9. The Imitation Game “Imitate” pixel sizes to get expected response
Use smallest pitch available: 25µm
Group pixels to similar sizes as other sensors: n*25µm
DUT
Sig
cluX
cluY
mult
250x50
9.07
1.06
1.41
1.27
500x25
8.75
1.02
1.81
1.25
500x“50”
8.83
1.03
1.33
1.24
500x“100”
8.9
1.17
1.23
500x“125”
8.92
1.18
500x“150”
8.93
1.12
1.22
500x“175”
8.94
1.13
See agreement between simulated and normal pitch parameters
500x“125”, 500x“150”, 500x“175” cases give approximate values expected for 167x125 clusters

10. 167x125µm2 Mapping
With expected values one mapping looks most successful
From imitation: 167x125 cluster size should be ~1.13 and ~1.18 for X & Y respectively
unmapt
zoom
mapt v5
zoom
DUT
Sig
cluX
cluY
mult
167x125 v1
9.93
1.31
1.35
1.08
167x125 v5
9.66
1.26
1.14
1.11
167x125 v6
1.25
1.33

11. Evidence of charge leaking
Vertical track suggests reason for X cluster problem
mapt v5
167x125µm2
167x125µm2
Unlikely charge made zig-zag track
Looks like straight vertical track with
alternate adjacent pixels firing due to
charge leakage
 More trouble for 125x100µm2 device
2/4 pixels have long readout connections
125x100µm2

12. Cluster Size Map 167x125µm2 (CoG) map of per pixel cluster ave. size
Global ave. cluster size X,Y = 1.18, 1.15
See pattern in X absent from Y
1/3 pixels cooler than rest
Correspond to pixels without long readout connection and not adjacent to long RO
Structure found in profile
Average X cluster size per pixel
Average X cluster size per pixel X Profile
Average Y cluster size per pixel

13. Cluster Size Map 125x100µm2 (CoG) map of per pixel cluster ave. size
Global ave. cluster size X,Y = 1.31, 1.18
No pattern seen in X map as all pixels either have long readout connections or are adjacent to long RO
Structure found in profile
Average X cluster size per pixel
Average Y cluster size per pixel
Average X cluster size per pixel X Profile

14. Cluster Size Map 125x100µm2 no CoG
Using cluster midpoint to populate map, see column structure: bias introduced in calculation of cluster coordinate
Also seen in using seed pixel coordinate
Profile plots show structure from increased entries per pixel
In seed position case this matches long RO pixels
Average X cluster size per pixel
Cluster midpoint
Seed postion

15. Cluster size comparison
Comparison based on DESY Nov’13 & CERN V=-100V
Mapping specific to each device
Expected mapping “imitated” using multiples of 25µm pitch
Based on 500x25µm2 sensor at DESY
Two sets of points for 125µm pitch
X & Y dimensions
See effect of charge “bleeding” in X dimension square pixel sensors at 125µm & 167µm
See CERN cluster size consistently smaller than DESY
less scattering?
charge deposition?
Cluster size Vs pixel pitch
Cluster size follows trend at DESY & CERN
Cluster size most sensitive at smaller pitch sizes

16. See Alex’s talk GBL Form triplets on both upstream and downstream arms
Triplet: form doublet from first and third planes (with cut), add middle plane (with cut)
Triplets meet in middle of telescope with cut on coincidence position and slope
Each track is parameterised as straight line. The track is extrapolated or interpolated to form a trajectory
Plots available to tune cuts

17. Initial Residual comparison
Residuals calculated from mapped clusters
Residuals for all cluster sizes shown
Ideal residual pitch/√12
CERN residual consistently smaller than DESY
scattering?
charge deposition?
See effect of charge “bleeding” in X dimension square pixel sensors at 125µm & 167µm
Sins in sensors masked by sins in set-up
Residual Vs pixel pitch
Residuals follows ideal trend at DESY & CERN
Some deviation (separation) of trends at (lower)extremes

18. DESY 250x50µm2 AC/DC comparison
250x50µm2 AC/DC comparison Relative signal drop for AC across biases from testbeam
Similar magnitude to lab tests: 10-20%
Pixel efficiency comparison shows both perform well
Response comparison with SPS desirable
250x50µm2 DC, AC
ToT
THL=3000
Bias
250x50 AC
250x50DC (Ref)
20V / /
50V / /
80V / /
100V / / AC DC
Efficiency Analysis from H. Hayward

19. July SPS AC/DC signal comparison
ToT
500x25µm2
167x125µm2
THL=3000
Comparison with Liverpool lab tests
Same: 500x25µm2 has lower response than 167x125µm2
Diff: No change in AC/DC order with bias present in lab test
Diff: smaller discrepancy seen (same tuning)
Need to look into details of ToT calculations:
cluster signal vs. pixel source scan

20. Summary & Next steps Reconstruction comparison of DC devices
Residuals vs pitch
2 separate testbeam environments
Next complete analysis: single/double pixel cluster comparison
Move to Tbmon-II analysis
Initial look into charge sharing
Possible metric
Initial comparison of AC & DC devices suggest differences in response and charge evolution
Next testbeams will focus of AC comparison of pitch sizes

21. Beyond the Standard Slides
Back-up
Beyond the Standard Slides

22. EUTelescope mapping recipe
Here, used EUTelescope trunk version ~May 2015
In order to read in alternative geometry
Define DUTs with correct pixel array
edit $EUTELESCOPE/src/FEI4Single.cc as template
edit $EUTELESCOPE/src/EUTelescopeGenericPixGeoMgr.cc
Add new geometries to gear file
Define as plane (appropriate Nrows & Ncols etc.)
Use new libraries e.g. FEI4Single_250x50.so
Additional reconstruction step to map converter pixels before clustering
No changes made to clustering
Clustering by $EUTelescope/src/EUTelProcessorGeometricClustering.cc
General Broken Lines used
patternRecognition and GBLAlign

23. GBL Form triplets on both upstream and downstream arms
Triplet: form doublet from first and third planes (with cut), add middle plane (with cut)
Triplets meet in middle of telescope with cut on coincidence position and slope
Each track is parameterised as straight line. The track is extrapolated or interpolated to form a trajectory
Plots available to tune cuts

24. GBL steps Usual converter, clustering & hitmaker steps
Alignment takes place in a couple of steps between hitmaker & track fitting
DAFfitter combines steps in single processor
Hitlocal step produces augmented gear file with basic plane orientation and XY offset calculated based on correlations
… input to …
Pattern recognition creates track candidates and filters based on supplied cuts (viewable in histo file)
GBLAlign makes final alignment of telescope. New gear file
GBLTrackFit produces residual plots and can produce TBmon readable root file for further APIX analysis

25. TB Itinerary Seed Fraction
Maps of cluster_seed_charge/cluster_charge: only 2 hit clusters considered
167x125 X
167x125 Y
Derive global seed fraction maps: fraction of cluster charge found in hottest pixel in cluster
For non-leaky devices (250x50, 500x25) see larger pitch leads to smaller seed fraction
Leaky devices have reverse trend, i.e. smaller pitch has smaller seed charge
DUT
Ave. seedFrac X
Ave. seedFrac Y
250x50
0.675, 0.676
0.686, 0.690
500x25
0.677, 0.643
0.707, 0.705
167x125
0.788
0.691
125x100
0.823
0.670
"500x125"
0.687 , 0.645
0.712, 0.705

26. Leak rate Suppose charge falls homogenously across devices
Neglect cluSize>2 (~1% or less from measurements)
125x100 example
Imitation of 125 pitch estimates cluster size = 1.16
Suppose (Ax+By)=C, where
A,B are cluster sizes from different processes
C is total cluster size (for one dimension)
x,y are fraction of time clusters produced by process
 x+y=1  y=(A-C)/(A-B)
Solving how often normal clustering occurs
(2-1.31)/(2-1.16)=0.8214
So charge leaking occurs ~18% of all hits (far below 50%)
2/4 are do not have long connection to readout
So, other 50% is liable for the leakage
leakage twice as likely  frequency of pixels leaking: ( )*(2/1)=0.3572
i.e. about a third of the time pixels with long connection leak
125x100
500x“125”
THL=3ke, -100V
125x100
167x125
estimate
1.16
1.13
measure
1.31
1.18
Leak rate
18% 5%
Next pixel leak rate
36%
17%