cms tracking trigger studies for the slhc

Marcello MannelliApril 2010 CMS sLHC Tracking Trigger

CMS Tracking Trigger studies for the SLHC


CMS from LHC to SLHCMotivation for L1 Tracking Trigger

10331033

10351035

1032 cm-2 s-1 1032 cm-2 s-1

10341034

I. OsborneI. OsborneAt SLHC CMS faces new challenges, in

particular for both Tracking and Triggering At SLHC CMS faces new challenges, in

particular for both Tracking and Triggering

Marcello MannelliFebruary 2010 Intelligent Trackers as L1 Trigger Providers for the SLHC

Scope of this Discussion:Outer Tracker

• The region of the inner-most Pixel Layers is fundamentally challenging at the SLHC, especially for the Sensor Technology

– One may speculate as to the most promising way forward– B-tagging, e/discrimination remain Very Important

• Assume 4 or 5 Layers of Fine-Pitch Pixels– To be better defined– Possible L1 Trigger primitives formation under study, but no viable

schemes identified yet

• Here focus on Outer Tracker as providing also L1 Trigger Primitives– Assume boundary between inner-most Pixel Layers and Outer Tracker is

somewhere between 20 ~ 40cm– In any future baseline layout, Outer Tracker and inner-most Pixel Layers

will have to make a coherent Tracking System

Marcello Mannelli

• Electrons/Photons

– Rate dominated by high Pt 0 from jets

– Is there a high Pt charged track consistent with L1 ECAL Cluster?• Is it Isolated?

• Muons

– Rate dominated by miss measured muons from jets, below the Pt threshold

– Is there a matching high Pt charged track above the Pt threshold?• Is it Isolated?

• Taus

– Rate dominated by ~ narrow high Pt jets

– Is there a narow low multiplicity jet associated to the L1 Calo Cluster?– Is it Isolated?

SLHC L1 Tracking Trigger:Reminder of Required Functionality

April 2010 CMS sLHC Tracking Trigger

Marcello Mannelli

Required FunctionalityL1 Trigger: Taus

• Confirmation of High Pt Narrow Jet Candidates– Tracks with Pt above ~ 10 GeV– Fast, Efficient high Pt Tracking– Good Pt resolution

• Isolation– Tracks with Pt above ~ 2 GeV– Fast, Efficient and Clean Tracking

down to Pt ~ 2GeV

• Longitudinal Primary Vertex association– Required to maintain efficiency at 1035

– Tracks with Pt above ~ 2 GeV– Good Z Vertex resolution


Marcello Mannelli

An additional Desirable Feature:rejection of Uncorrelated Combinations

• Rejection of Uncorrelated Combinations, from different primary vertices?

– At 1035 expect ~ 200 pile-up interactions for each 25ns bunch crossing

– Uncorrelated Trigger Primitives from different primary vertices may force higher thresholds Combined Triggers

– Longitudinal of Trigger Primitives with Tracks at Vertex would avoid this

– May be possible if track Longitudinal Vertex Association, required for Isolation cuts, is available


Marcello MannelliOctober 2009 Marcello Mannelli & Anders Ryd SLHC Tracking Trigger

~ 60kHz at 34GeV(vs ~ 4kHz at 1034)

Isolated Electron-Photon L1 Trigger with 200 event Pile Up

Optimized for 1034

Retuned for ~ 90% Efficiency at 1035


Marcello MannelliOctober 2009 Marcello Mannelli & Anders Ryd SLHC Tracking Trigger

~ 40KHz at 100GeV(vs ~ 5kHz at 1034)

Isolated TauL1 Trigger with 200 event Pile Up

Optimized for 1034

Retuned for ~ 90% Efficiency at 1035


Marcello Mannelli

Local Occupancy Reduction

• Cannot possibly transfer all Tracker data at 40MHz !

• Crossing Frequency / Event Read-Out ~ 40MHz / 100kHz ~ 1 / 400– L1 Data reduction by a factor of 10 ~ 100 is required

• For L1 Trigger propose to transfer only hits from tracks with

Pt > ~ 2 GeV– The aim is to provide useful Isolation information– Tracks with Pt > ~ 2 Gev are less than 1% of the Tracks inside acceptance– This corresponds to the maximum plausibly manageable L1 data rate

• In addition, must provide means of rapidly & reliably identifying high Pt (isolated) tracks ( Pt > 10 ~ 25 GeV)



Local Occupancy Reduction

Tracks with Pt > 1 GeV< 10% of Tracks in acceptance

Tracks with Pt > 2.5 GeV< 10% of the remaining Tracks


α

Local Occupancy Reductionwith Local Track Vectors

• Pairs of Sensor Planes, for local Pt measurement

• High Pt tracks point towards the origin, low Pt tracks point away from the origin

• Use a Pair of Sensor Planes, at ~ mm distance

– Pairs of Hits provide Vector, that measure angle of track with respect to the origin

– Note: angle proportional to hit pair radius

• Keep only Vectors corresponding to high Pt Tracks

J. Jones (~2005)CMS Tracker SLHC Upgrade Workshops

Marcello Mannelli

Possible Tracker LayoutsIncorporating Stacked Module Layers

• Several potential layouts for an SLHC Outer Tracker are under study– Want to provide both Tracking Trigger & Optimal Tracking in future Tracker– Options under study range from minimal Tracking Trigger with

conventional (much improved) Tracker, to fully integrating Tracking Trigger and Tracker functionality

• No final decision on layout of tracker until final requirements determined, and feasibility and performance potential of different options understood

February 2010 Intelligent Trackers as L1 Trigger Providers for the SLHC

Minimal Hybrid Layout Long Barrel Straw-Man

Marcello Mannelli

Possible Tracker LayoutsIncorporating Stacked Module Layers

• Several potential layouts for an SLHC Outer Tracker are under study– Want to provide Tracking Trigger & optimal Tracking in future tracker– Options under study range from minimal Tracking Trigger with

conventional (much improved) Tracker, to fully integrating Tracking Trigger and Tracker functionality

• Here focus on the Long Barrel Straw Man as it allows evaluating the full range of possible Tracking Trigger architectures which build on the Stacked Module approach

February 2010 Intelligent Trackers as L1 Trigger Providers for the SLHC

Minimal Hybrid Layout Long Barrel Straw-Man

Marcello Mannelli

A Hierarchical Scheme withDouble Stack Layers

• Stubs from the Stacked Modules– Combine clusters across Stacked Module to form Track Vectors, or Stubs– Reduce rate by rejecting stubs with Pt below ~ 2GeV

• Low threshold driven by Isolation cut requirement

• Tracklets from the Double Stack Layers– Combine Stubs across Double Stack Layer Stubs to form Tracklets, in a

“local” and computationally efficient way– Tracklets are 4 Vectors with useful Pt, and resolution

• Use Tracklets as seeds for L1 Tracks which combine stubs from other Double Stack Layers

– Allow for hit loss, to provide robust system

• A Long Barrel Straw Man Layout– Which uses Double Stack Layers for both L1 Tracking Trigger and Full

Track Reconstruction



Long Barrel Straw Man Layout:3 Double Stack Layers + 2 Short Fwd Barrels

Marcello Mannelli

Tracking Trigger Simulations:Goals and Program of Work

• Consider three levels of L1 Tracking Trigger Primitives

– Hits above threshold -> Valid L1 Clusters– Valid L1 Clusters -> Accepted L1 Stubs– Accepted L1 Stubs -> Accepted L1 Tracklets

• Build L1 Tracking Trigger Objects for High Pt Leptons & Isolation– Using combinations of Tracklets and/or L1 Tracks

• Combine with Calo and/or Muon to build Combined L1 Trigger Objects– Using first High Pt Lepton Tracking Trigger Primitives– Then seeing effect of adding Tracking Trigger Isolation criteria

• Validate and study Efficiency vs Rate at each step– Standard Plots


Marcello Mannelli

• Simulated threshold curves for Stubs, with 2GeV pt acceptance cut, for various sensor separations within a Stack, in different Barrel Layers

– The Pt thresholds are sharper at larger radii, as expected


Track Trigger Primitive Simulations

Marcello Mannelli

• Simulated threshold curves for Stubs, with a 1mm Sensor separation within a Stack, for various Pt acceptance cuts, in different Barrel Layers

– The Pt thresholds are sharper at larger radii, as expected


Track Trigger Primitive Simulations

Marcello Mannelli

Agenda Tracking Trigger Working GroupCMS Upgrade Week, April 2010

• Benchmark code for Clusters & Stubs– Rates and FAST versus FULL Sim Comparisons

• L1 Tracking Triggers, and performance of Benchmark code– Muon Triggers– Electron Triggers– Tau Triggers

• L1 Pixel Jet Triggers

• Proposed Architecture for Tracklet and L1 Track formation

• Discussion– Implementation of Tracklet and L1 Track Algorithm



Track Trigger Primitive Simulations:Benchmark Code for Clusters & Stubs

Marcello Mannelli

• Comparison of– Full Simulation– Full Sim no ou-of-time PU– Fast Simulation

• Detailed Simulations in place

• There still remain– Substantial Discrepancies– Peculiar Features

• => Uncertainties in Rates…– Work with Fast Simulation– Present thinking based on

conservative assumptions


Hits

Clusters

Stubs

Track Trigger Primitive Simulations:Benchmark Code for Clusters & Stubs

Marcello Mannelli

Muon BarrelTracking Trigger Studies


Marcello Mannelli

Muon End-CapTracking Trigger Studies


Marcello Mannelli

ElectronTrack Trigger Studies


Marcello Mannelli

Why Tracklets?

• Consider 3 layers and no tracklets (track hits but no Pt)

• To find tracks need to test all combinations of tracks from all layers (minimum Pt covers entire sector)

– Must be completed in 25 ns.

• Tracklets allow projection of the track hit from one layer to another

– reduces the range to check in the destination layer

• All hits can be compared simultaneously in an FPGA

– Practical limit is ~10 hits/layer

– Subdivide layer to stay within this limit

3 track 3 layer example

Tracklets allow reduced scan range


Marcello Mannelli

• For example, project Tracklets from inner 2 layers to outer layer segment (independently)

– Projection from an Inner Layer requires complete Tracklet• 2/2 Stubs, 4/4 Clusters

– Outer layer uses stubs– Reduces effect of hardware failures

• Segments must overlap due to imprecise projection


M. Johnson, U. Heintz et alSee presentation by U. Heintz

Why Tracklets?

Marcello Mannelli

• Repeat for the other Layers (independently)

– With this scheme Tracklets in a given layer can be put into coincidence with Tracklets or Stubs in the other Layers

– Recover possible missing stubs in up to two Layers– Minimize sensitivity to hardware failures and inefficiency

• Must remove duplicates


M. Johnson, U. Heintz et alSee presentation by U. Heintz

Why Tracklets?

Marcello Mannelli

• Benchmark code for Clusters and Stubs exists– Studies ongoing to understand rates etc

• Combinations of Stubs from different layers demonstrate encouraging performance potential across a wide range of studies

• Now moving to implement benchmark code for Tracklets and L1 Tracks

– Inspired by proposed back-end architecture proposed by M. Johnson et al

• This will make available the full suite of possible Tracking Trigger primitives across the e, and channels

– Other studies may follow

• Work is also continuing on the feasibility of inner pixel based jet triggers


Conclusions and Prospects

Marcello Mannelli

Back-Up slides


cms tracking trigger studies for the slhc

Documents