Status of
L1 Pixel Trigger implications on ROC data
rates
Fabrizio Palla (INFN Pisa and CERN)
On behalf of the TTI group
F. Palla INFN Pisa
Pixel Trigger framework
• Two possible schemes depending on rates and latencies (two
options under considerations: 10 and 20 µs)
• “Push” architecture: pixel data readout at 40 MHz and provided to the L1 trigger
• “Pull” architecture: Calorimeter/Muon L0-triggers (at a latency of <6 µs?) and/or L1Tracks from outer tracker (at a latency of <10 µs?).
See R. Horisberger talk at DESY CMS Upgrade week
https://indico.cern.ch/getFile.py/access?contribId=5&resId=0&
materialId=slides&confId=253700
F. Palla INFN Pisa
2
Introduction
• Track Trigger Integration group is currently investigating the
usage (and usefulness) of a L1-Trigger based on pixels
• At the moment three use cases studied
• Primary vertex reconstruction for jets
• https://indico.cern.ch/getFile.py/access?contribId=11&resId=0&materialId=slides&confId=277737
• Tau → 3 prongs reconstruction
• https://indico.cern.ch/getFile.py/access?contribId=7&resId=0&materialId=slides&confId=288250
• Electrons
• https://indico.cern.ch/getFile.py/access?contribId=1&sessionId=3&resId=0&materialId=slides&confId=290480
• No data rate estimates yet
• None of the above have (yet) demonstrated a compelling evidence about the
necessity of a L1-pixel trigger
• L1-Tracks with outer tracker have same rejections powers for similar (or better
efficiencies).
• However, have smaller precision on impact parameter (secondary vertices) or
cannot reach pT<2 GeV (might result needed for isolation)
• Other use cases being investigated (b-tagging, secondary vertex tagging)
F. Palla INFN Pisa
3
Jets primary vertex reconstruction
F. Palla INFN Pisa
4
Jets primary vertex reconstruction
• Algorithm uses pixel cluster information only from chips in the
line-of-sight of jets
• Two possible use cases
1. Make clusters on chip@40 MHz and read out reduced information to keep the
•
data rate low
2. Read full pixel information for a subset of the detector (reduction x 5-10) only
for Calo-jets with some threshold as L0-trigger (and if latency allows)
Performance depends on Jet ET. Typical reduction factors ~5 with ~90%
efficiency.
• CAVEAT: does not do better than L1 Tracks
~4 mm resolution (RMS)
Similar resolution with L1-Tracks
F. Palla INFN Pisa
5
Data rate estimate
Numbers will scale down depending on the L0 rate. E.g. for 500 kHz L0 rate
they will be reduced by 80, but the complication will be to add extra latency.
Data rate
(Gbps/cm2)
1 (r-phi) x1 (r-z) cm2 segmentation
2 (r-phi) x1 (r-z) cm2 segmentation
All clusters
Only clusters with
size-X=2 *
All clusters
Only clusters with
size-X=2 *
3
1.8
3
1.8
Read clusters grouped per
cluster size
1.3
0.95
0.88
0.7
As above but only ≥ 2
clusters of the same width
within a 1x1 cm2 area
0.68
0.38
0.57
0.39
LAYER 1
Read all clusters with coarse
position
* At all effects a pT cut on tracks
Data rates too large at 40 MHz: from 2.5 to 20 Gbps per chip (2.56x2.56 cm2).
Rates could become affordable (a few hundred Mbps) at the expenses of a longer latency.
F. Palla INFN Pisa
6
Tau to 3 prongs
• One “solution” using pixel-stand alone “tracking” at 40 MHz. (results presented here)
• Another one is being investigated to use L1-track seeded regions (no results yet)
Feasibility of a ! trigger at Level 1 using Pixel Detector
!
Average decay angle of ! with transverse momentum (PT) greater than 40 GeV/c is ~2°
!
It means that all three prongs can be found in a single chip, in fact, to keep most of ! is enough to have a signal
region in the 1st Layer of diameter of 4 mm
!
Expected similar cluster* size along the beam line for the prongs originated from the same vertex
!
Expected 3 isolated clusters (triplet) above a given PT threshold. Eventually merging for higher PT.
Example picture of a ! ! 3 prongs +
PU events on a single chip
Global Coordinate System
Global z²
Chip
Layer 1
&R
background
Global r%²
Layer 2
y
Background hits around
signal and isolation region
&z
&z = zcluster_barycenter z! ProductionVertex
6
*
Isolated charge deposit made by one or more tracks
x
Androsov, Grippo - L1 Pixel Tau Trigger - Worshop INFIERI
Signal
Region
Isolation
Region
20/01/14
F. Palla INFN Pisa
7
Tau to 3 prongs
1. Identify triplets in a pixel chip of compatible cluster size in first two layers (using beam
spot as a third point is in effect a pT cut)
2. Look for compatible triplets in the third layer and form all track combinations,
requiring ≥2 GeV track pT.
3. Estimate tau decay vertex position in both R-Z and R-Phi planes.
Trigger Rate vs. Tau Detection Efficiency
! Reasonable
trigger
rate of 40 KHz
corresponds for
each curve to a !
detection efficiency
!
Feasible L1
trigger rate
14 Pisa
F. Palla INFN
For ! with
observable PT [40,50]
GeV, ! detection
efficiency= 11.2% ,
chosen as Reference
Point (see next slide)
L1 rate 300 kHz and 75% efficiency with L1-tracks
Or factor
reduction of L1 Calo Trigger at 40 GeV
Androsov, Grippo - L1 Pixel Tau Trigger - Worshop
INFIERI 4
20/01/14
8
Data rate
Data rate estimation with acceptable max rate ~O(1Gb/s)
Layer
Average Rate
per chip
[Gigabit/s]
Max Rate
per chip
[Gigabit/s]
4
1 (separate clusters)
6.06
15.12
6
5
1 (regional packing)
5.60
13.72
5
6
! (separate clusters)
5.36
12.60
! (regional packing)
5.28
11.80
" (separate clusters)
4.40
9.24
" (regional packing)
4.72
9.24
Layer
Average
number of
clusters per
chip
Average
number of
triplets per chip
1
7
2
3
!
Per triplet packing is not efficient,
because amount of triplets is the
same order of # clusters
Separate cluster representation requires 21 bits per cluster
(16 bits per position + 5 bits per size)
! Regional packing representation requires: 11 bits per region
header and 16 bits per cluster
! Data Rate produced by a Single Layer Algorithm is too high
to be implemented at hardware level
!
Sending separate clusters
or clusters packed per
region
27
F. Palla INFN Pisa
Androsov, Grippo - L1 Pixel Tau Trigger - Worshop INFIERI 20/01/14
9
Electrons
• Match a E/gamma cluster with pixel stand alone tracks
To be compared with L1 track E/gamma
rejection at 20 GeV ~6 with an
efficiency of ~90%
F. Palla INFN Pisa
10
Conclusions
• At the moment no clear “smoking gun” for the L1 pixel trigger
• L1-Tracks Algorithms have similar (or better) performances than those
tried with the pixels
• Still a large phase space of investigations with pixels
• “Pull” architecture seems more favorable in terms of data rates
• Need however firm statements of what is needed from the Physics
• Imply data to be available at ~few µs latency.
• Isolation with low pT tracks (especially in taus)
• Secondary vertex reconstruction/impact parameter
• B-tagging (Higgs?)
• B-hadron reconstruction (low pT for B-physics?)
F. Palla INFN Pisa
11