1. Status of SiD Silicon Tracker and Tracking Performance
Richard Partridge
SLAC
U. Oregon SiD Workshop

2. Tracking Performance Overview
LOI shows excellent tracking performance for baseline SiD tracker
>99% track finding efficiency over most of the solid angle
~98% in core of 500 GeV light quark jets
Momentum resolution typically ~0.2% for |cos(q)| < 0.65
s(pT) / pT < 0.5% over most of solid angle for 1 GeV < pT < 100 GeV
DCA resolution typically ~15mm for pT = 1 GeV, |cos(q)| < 0.65
Most tracks multiple scattering limited – resolution approaches ~4mm at high pT
>99% of tracks have ≤1 mis-assigned hits
Fake track rate is 0.07% for tt events
New since the LOI:
Planar sensor geometry
Realistic charge deposition and digitization/clustering
Improvements to tracking performance for high occupancy
Richard Partridge

3. Switch to Planar Geometry
LOI geometry consisted of cylinders and disks with virtual segmentation
New geometry models each silicon sensor – rectangular detectors in barrel, trapezoidal detectors in endcaps
Richard Partridge

4. Realistic Detector Geometry
Blow-up of vertex detector showing hits on planar sensors
Richard Partridge

5. SiD LOI Geometry – CAD Drawing
Richard Partridge

6. SiD LOI Geometry – Event Display
Richard Partridge

7. Realistic Hit Digitization
Charge deposition for strip detectors based on CDF Si sensor simulation algorithm implemented by Tim Nelson
TKN + RP extended strip charge deposition model to pixel detectors, while Nick Sinev has developed a detailed modeling using electric field maps
Strip/pixel charges clustered by a nearest neighbor algorithm
Hash maps used to achieve approximately linear scaling of clustering time
Settable parameters for noise, readout and clustering thresholds
Form tracker hits from clusters with expected hit errors
Richard Partridge

8. Track Finding Improvements
No significant changes to track finding algorithm
See LOI and RP talk at 2008 LCWS for details of tracking algorithm
Some structural/implementation changes to improve tracking performance, especially with large numbers of hits
Layout of strips / pixels on a sensor was separated from the geometry code allowing for more user control
Developed a set of tracking strategies for the planar geometry
Strategies used to guide track reconstruction
Created a “standard” driver that to run hit digitization / tracking for the sidloi geometry:
org.lcsim.recon.tracking.seedtracker.trackingdrivers.sidloi3.MainTrackingDriver
Richard Partridge

9. A Partial List of Outstanding Issues
Forward tracking efficiency has gotten worse since the LOI
Small charge asymmetry in helix parameters
Pixel hit resolution is larger than expected
Occasionally produce duplicate tracks
Track reconstruction is very slow for some events
CLIC people report hits sometimes located incorrectly
More on these issues in the following slides
Richard Partridge

10. Forward Tracking Efficiency
Look at 100 GeV single muons in sidloi3
Efficiency for reconstructing “findable” tracks drops precipitously in the forward region
Angle
Momentum
Generated
Findable
Found 90
100
5000
5001
5026
110
5002
120
130
5003
140
150
160
4906
3844
170
4519 1
Richard Partridge

11. Source of the Problem
To speed up the track finding performance, the FastCheck class is used to see if a given pair of hits is consistent with the pT and DCA requirements for the strategy
This algorithm was “improved” when I finally figured out how to solve for the circle(s) passing through 2 hits that is tangent to a circle whose radius is the maximum DCA
Can use this to determine the allowed pT range for these two hits
Reject hit pairs inconsistent with the minimum pT cut
New algorithm also gave accurate determination of range in arc lengths s1, s2
Richard Partridge

12. Consistency in the s-z View
Hits passing the checks on pT and DCA in the x-y plane were then checked for consistency with the s-z impact parameter z0
With the improved x-y fits, the range in arc length s became quite narrow (high pT tracks from the origin have s ~ r)
The z coordinate for endcap disks is fixed, so the calculated range of z0 for a hit pair can be quite small
The geometry of the forward region magnifies multiple scattering errors such that the calculated z0 range is frequently not consistent with the maximum s-z impact parameter
To address this problem, the range in s was increased by the maximum allowed x-y impact parameter (1 mm)
Kludged solution for now – ultimately should properly account for MS errors
Richard Partridge

13. Updated Tracking Efficiency
Repeat the tracking efficiency measurements for single muons
100 GeV muons
Richard Partridge

14. Check Low-pT Tracking Efficiency
“Kludge” to FastCheck class appears to be working
1 GeV muons
Richard Partridge

15. Charge Asymmetry
First noticed >2 years ago by summer student at SLAC
Observed in all “circle fit” parameters (curvature, f0, DCA)
Likely an artifact of fitting algorithm
Karimake algorithm has no MS correlations, is a non-iterative approximation
Proposed solution is development of a Kalman fitter to perform a true helix fit
1 GeV muons
100 GeV muons
pT Pull Distributions
Richard Partridge

16. Kalman Fitter Status
Stanford summer student worked on adapting the trf code to lcsim track fitting during summer 2010
Developed by Dave Adams and Norman Graf for the DØ experiment
Kalman filter portion of trf ported to Java / lcsim by Norman
Project turned out to be more difficult than expected
Large and complex code base
Some difficulty in understanding how to properly use trf toolkit
Some success…
Ran forward (inside out) track fits using “XY Plane” hits (e.g., barrel) from SeedTracker tracks for the proposed Heavy Photon Search experiment
…but work is not complete
Need to translate endcap hits into trf “Z-Plane” hits
Existing track fitter (FullKalmanFit) works on a list of hits – needs to be modified / extended to include dead material surfaces
Further work needed to track down dead material along track path
Richard Partridge

17. Pixel Hit Resolution
The charge deposition model adapted from CDF drifts segments of charge to the sensor surface
Width of charge distribution on suface depends on track angle, Lorentz drift, and diffusion parameterization
The Gaussian charge density from diffusion is integrated over the pixel array
If the charge spreads to more than one pixel, a center-of-gravity algorithm is used to determine the hit position
In this case, the electronic noise assigned to the pixel readout can potentially have a big effect on hit position resolution
If the pixel is very thin, there is little drift/diffusion and this algorithm may tend towards single-hit clusters
A more sophisticated pixel charge deposition model has been developed by Nick – someone would need to interface this into the planar geometry/digitization/hit making code
Richard Partridge

18. Duplicate Tracks
Occasionally find two tracks with nearly identical track parameters
This was not a problem for the LOI where we used virtual segmentation of cylinders / disks
SeedTracker only allows 1 hit to be shared between two tracks and has strong bias towards maximizing the number of hits on a track
Planar geometry can produce multiple hits per layer for tracks passing through regions where there is sensor overlap
Tracking code currently only tries to add 1 hit per layer
If the track produces multiple hits in enough layers, can form two separate tracks that do not share hits
Solution requires some thought – want to add overlap hits to tracks – but not clear how to do this while performing exhaustive search of all possible track candidates
Richard Partridge

19. Slow Track Reconstruction
Does not appear to be an infinite loop
Evidence to date points towards it being a “feature”, not a bug
May just be a combinatoric increase in the number of possible tracks to check in the core of high energy jets
Additional hits from overlaps have a multiplicative effect
Tracker design may play a role
3D measurements are less prone to combinatorics
Track finding strategy may also have an influence
Outside-in strategies vs inside-out
Hard problem to address / debug
Especially if the code is working “as designed”
Richard Partridge

20. Mis-Placed Hits
Past experience suggests the following possible causes:
Not enough bits assigned to the strip/pixel readouts to uniquely identify each strip/pixel
First thing to check when you see bizarre hit behavior
More than once have endured long and tedious tracking of problem through complex digitization code to find the problem was in the compact.xml file
Jeremy may be able to add some checking in his identifier code
Position of stereo hits depends strongly on track angle due to gap between stereo planes
Can give bizarre hit positions, especially for small angle stereo
Not a problem during tracking since the hit position and covariance matrix account for the track angle
Reports seemed to be correlating this with poor efficiency in the forward direction, but that problem appears to have been due to fast hit check algorithm
Richard Partridge

21. Summary
LOI demonstrated that an all-silicon tracker with ~10 hit measurements would give excellent performance at the ILC
Substantial effort in developing a more realistic detector simulation is bearing fruit
Forward tracking efficiency problem appears to have a “kludge” fix that is working
Other issues:
Charge asymmetry probably requires the Kalman filter fitter to be completed
Large hit error in pixels may be due to noise setting being too large
Duplicate tracks appear on occasion – no easy solution
Some events take a long time to complete – no easy solution
Misplaced hits – too few strip / pixel readout bits??
Some of these issues require focused / concentrated effort – difficult to achieve with present manpower situation
Richard Partridge