Tracker in the Trigger: from CDF experience to S-CMS Fabrizio Palla INFN – Pisa Vertex 2007 Lake...

Tracker in the Trigger: from CDF experience to S-CMS

Fabrizio Palla INFN – Pisa

Vertex 2007Lake Placid, NY, USA

Vertex 2007, 23-28 September 2007

Tracker in the Trigger: from CDF to S-CMSF. Palla INFN Pisa

The Tracker and the Trigger Trigger rates control is extremely challenging in high luminosity

hadron collider experiments As the luminosity increases, physics goals change in response to new

discoveries and the detector ages.

It is thus essential that the trigger system be flexible and robust, and have redundancy and significant operating margin

Providing high quality track reconstruction over the full detector can be an important element in achieving these goals.

This has certainly been the case in the CDF experiment where the Silicon Vertex Trigger (SVT) has significantly extended the experiment’s physics capability

B-physics and Bs oscillations Search for the Higgs boson

Even more challenging will be the trigger for the foreseen upgrade of LHC, the so-called SuperLHC (SLHC)

Luminosity increase of a factor 10 wrt “standard” LHC



3cm15cm150cm

Outerdrift

chamber

Silicon stripdetector

Siliconclose-up

Impact parameter

Beam spot

1mm

Zoom-inInput (every Level 1 accept):XFT trajectories silicon pulse height for each channel

Output (about 20 s later): trajectories that use silicon pointsr- tracksimpact parameter: (d)=35 m

AMalgo

SVT at CDF Level-2 Trigger



The Event

...The Pattern

Bank

The pattern bank is flexibleset of pre-calculated patterns:

can account for misalignmentchanging detector conditionsbeam movement …

Pattern matching in CDF (M. Dell’Orso, L. Ristori – 1985)



The Displaced Track Trigger @ CDF Striking difference wrt D0 thanks to SVT But also vs CDF:

Run II ~2000 BsDs vs 1 in Run I Compare with only 10 times increased

integrated luminosity The trigger had a big impact

Online mass

Phys. Rev. Lett. 97, 242003 (2006).



On the opposite side: FPGA for the same AMchip

P. Giannetti et al. “A Programmable Associative Memory for Track Finding”, Nucl. Intsr. and Meth., vol. A413/2-3, pp.367-373, (1998).

AM chips from 1992 to 2005

• (90’s) Full custom VLSI chip - 0.7m (INFN-Pisa)

• 128 patterns, 6x12bit words each

• 32k roads / wedgeF. Morsani et al., “The AMchip: a Full-custom MOS VLSI Associative memory for Pattern Recognition”, IEEE Trans. on Nucl. Sci., vol. 39, pp. 795-797, (1992).

In the middle: Standard Cell 0.18 m(INFN-Pisa) 5000 pattern/chip AMchip

L.Sartori, A. Annovi et al., “A VLSI Processor for Fast Track Finding Based on Content Addressable Memories”, IEEE Transactions on Nuclear Science, Volume 53, Issue 4, Part 2, Aug. 2006 Page(s):2428 - 2433NEXT:

NEW VERSIONFor both L1 & L2



SLHC Level-1 Trigger issues @ 1035

~400 Minimum Bias events/bx (50 ns) [16xLHC]

Occupancy Degraded performance of algorithms

Electrons: reduced rejection at fixed efficiency from isolation Muons: increased background rates from accidental coincidences

Larger event size to be read out New Tracker: higher channel count & occupancy large factor Reduces the max level-1 rate for fixed bandwidth readout

Trigger Rates Try to hold max L1 rate at 100 kHz by increasing readout bandwidth Implies raising ET thresholds on electrons, photons, muons, jets and use

of less inclusive triggers Need to compensate for larger interaction rate & degradation in

algorithm performance due to occupancy



Needs for a Tracker Trigger at SLHC

Note limited rejection power (slope) without tracker information

Tracker needs to be rebuild due to radiation aging

Move some HLT algorithms into Level-1 or design new algorithms reflecting tracking trigger capabilities

See previous talk by Marcin Konecki

Proposed boundary conditions Rebuild L1 processors Maintain 100kHz limit Increase latency to 6.4µs

ECAL digital pipeline holds 256 samples @ 40MHz

From CMS-DAQ TDR

L =1034 cm-2 s-1



Main issues for Tracker Trigger Data Rate in Tracker Volume at SLHC

About 12,000 tracks per bunch crossing (50 ns) in the Tracker volume ||<2.5 and ~400 primary vertices/bx

Needs high granularity information

Innermost radii (~<10 cm) needs new technology to cope high radiation doses (>3x1015 neq cm-2)

At 10 cm radius the rate is ~108 cm-2s-1 giving ~6000 links @ 2.5 Gbps link speed– simply impossible to cope with

Push towards larger radii Benefits for momentum measurement

based trigger Ideal to match with muon and

calorimeter triggers ~1 mm pointing resolution for z- larger

than average 2 pileup interactions (~0.4 mm)

Still fairly nice impact parameter resolution in transverse plane (~100 m for 10 GeV muons) to get rid of muons from and decays

Without pixel

With pixel

(z0)

Without pixel

With pixel

(d0)



Tracking Trigger driving ideas Design considerations:

Main usage for pT reconstruction Need low occupancy and large lever-

arm, rather than brilliant space-point resolution

Need to decrease Tracker material budget by a sizeable amount

multiple scattering will drive the momentum resolution below ~10 GeV

Reduce to a minimum (possibly null) Tracker trigger-only layers

Impact on the Tracker design Two possible approaches

Tracker primitives Full readout for trigger Limited information

Detector level data reduction

Selective readout using Mu or Calo information

Like CDF XFT



Tracker primitives from CDF SVX approach

1 AM for each enough-small Patterns

Hits: position+time stampAll patterns inside a single chip

N chips for N overlapping events(identified by the time stamp)

Event1AMchip1

Event2AMchip2

Event3AMchip3

EventNAMchipN

Main problem: input Bandwidthdivide the detector in thin sectors. Each AM searches in a small

OFF DETECTOROFF DETECTOR

Data links



Full readout option Exercise using CMS layout but modified strip pitch:

From 40 to 90 cm radius Conservative approach Barrel coverage up to ||<1.5

R=40 and 50 cm macro-pixel like sensors of 0.2x5 mm2 cell size R>50 cm strip like sensors 50 m x 10 cm cell size

Subdivide the detector in many sectors Keep data volume limited in each sector

Combine information from at least 3 layers out of 4 in each sector

Momentum resolution of ~ few (<10)% at 10 GeV/c Granularity driven by the minimum measurable pT for triggering

purposes, without loosing efficiency 70 sectors at the innermost radius

Well covering the bending of a track of 10 GeV pT and above



Sensors and signal routing

hybrid

5 cm

6 cm

200x5000 m2

Readout chips

Fiber(s)

Smaller radius sensors

Larger radius sensors

Fiber(s)

32 bits sufficient to locate hit positionin the sector and bx time stamp

hybrid

6 cm

10 cm

50 m x 10 cm

Readout chips

fibers



Occupancy GEANT4 simulation of tracking layers

Includes material budget and loopersRadius (cm) Hit/module/bxa Rate*/module (Gbps) Rate*/sector (Gbps) No. data links†/layer

40 43.6 28 170 3200

50 33.5 21 130 3100

60 25.5 16 175 5300

70 20.2 13 110 4600

80 16.6 11 80 5300

90 13.9 9 60 3900

ª average number on minimum bias events, 50 ns bx*32 bits/hit † for a data link speed of 5 Gbps Pros: same fiber links for data and trigger

Simple and elegantCons : large number of links needed

Need >=5Gbps laser drivers: 90 nm technology required Allow for fluctuations: increase by ~1.5 ? Laser driver power: commercial range from 330mW/fiber (4Gbps) to 700 mW/fiber (10Gbps)

Current links in CMS Silicon Strip: 1300 @ 60 cm and 2600 @34 cm



Conceptual design

AM EV0

AM EV1

AM EV50

...

Layer 0: ~35 fibers bringing ~20 Hits/50 ns

1 Hit/5 ns

1 Hit/5 ns

1 Hit/5 ns

From otherlayers

From otherlayers

From otherlayers

Distribute hits into different sets of storage units

depending on EVent #

Parallel INSerial OUT

...Parallel INSerial OUT

Parallel INSerial OUT

1 FPGA

From Detector



Reducing the data rate Keeping all information

Send only cluster position Need simple ASIC on chip to perform clustering Reduction of a factor ~2

Reducing the number of bits per hit Possible if time stamp not needed

Only strips detectors (larger radii) Reduction of a factor ~1.8 Needs fast readout electronics

Overall a factor ~3.5 reduction feasible Current GBT chip (Marchioro group) 2.56 Gbps for data

Local data reduction for trigger purposes only Coarser granularity joining readout channels for trigger

purposes Exploit high vs low pT tracks patterns



Board dimensions Current board dimensions

About 30,000 patterns per AM chip required Needed ~ 80 boards with ~40 AM chips each

3200 AM chips The current AM for CDF holds ~5000 patterns/6 planes in 0.18m technology If developed in the 90 nm technology one could accommodate ~4 times more

patterns/AM chip hence 30,000 for 4 planes

In order to evaluate the board dimensions need to define the granularity and the lowest pT threshold

For instance, a 50 m pitch on 4 detector planes and a pT threshold of 10 GeV needs ~90,000 patterns (needs 3 AM chips)



MIPMIP

Cluster width discrimination

90 cm70 cm50 cm30 cm

Discrimination of low pT tracks made directly on the strip detector by choosing suitable pitch values in the usual range for strip sensors.



Cluster width results In this region, using 50µm pitch,

about 5% of the total particles leave cluster sizes with ≤2 strips

Huge reduction in trigger links Once reduced to ~100 KHz, it

would only need few fast readout links to readout the entire Tracker

R(cm) Pitch

(m)

Z coverage

(cm)

Data Rate /Layer (Gbps)

40 40 170 350

60 60 200 160

70 90 200 160

90 90 200 95

No. of links (5Gbps) ~ 150 for whole tracker

Data rate per layer could be reduced by more than 1 order of magnitude wrt full readout case



Coarser granularity Calorimeters and muon system is

=0.087x0.087 Compromise with momentum

resolution Sufficient to use O(500m) at

innermost radius (R~50 cm) for a precision of ~15% at 10 GeV

OR-ed readout of 10 strips – simple logic

Reduction of a factor 10

ButpT/pT pT solve reducing promotion of

low pT tracks and kill photons from , but not brilliant momentum resolution

Loss in precision recovered by adding in AND a electron or a muon directly adding other information in the AM chip “layer”

Natural with AM chips

Silicon sensorOR

OR

OR

OR

AM chip

Muon pT,



A first idea on selective readout Provide a Muon Track fast Tag

(MTT) to be associated with hits in outermost Tracker layers

Need a new detector for MTT yet to be defined (2 layers of RPC?)

Still very preliminary need more thoughts

Electrons and Jets?



Stacked readout for Trigger Angle determines pT of track, assuming tracks

coming from origin Smaller a = greater pT

Pair of sensor planes at ~ mm distance for local pT estimate

Needs a correlator ASIC Fast data links

If located at ~20 cm needs 3 Gbps Stacks of 2 sensor planes at ~cm distance to be

correlated (off detector) for pT measurement Tight construction tolerances required for both

sensors and their alignment See M. Mannelli talk



Conclusions Tracker information helps reducing drastically the rate of

uninteresting events CDF SVT experience has become cornerstone for Tracker Trigger

B physics reach has been boosted after the usage of displaced vertex triggers and now Higgs search uses it

S-CMS will make use of tracking information at SLHC and has started to discuss several options

Triggering at SLHC is challenging due to fantastic data rate Concentrate on momentum resolution, muon and electron matching SVT approach seems feasible, especially if complemented by a data

reduction at the module level The trigger design and the detector layout are completely interleaved

and are influencing each other Next steps will involve simulating trigger with different layouts and are

going to define which strategy will be used, stay tuned!



BACKUP



L at end of year

time to halve error

integrated L

radiationdamage limit~700 fb-1

(1) LHC IR quads life expectancy estimated <10 years from radiation dose(2) the statistical error halving time will exceed 5 years by 2011-2012(3) therefore, it is reasonable to plan a machine luminosity upgrade based on new low-

IR magnets by ~2016

design luminosity

ultimate luminosity

courtesy J. StraitModified (start date) by F. Palla

ultimatevs. design

Time Scale of LHC Upgrade

2010 2012 2014 2016 2018 2020



switch board numbersAll info here TBC with simulation and R&D!

•80 switch boards• 1 / -sector

• 80 fibers / board• assume 5Gbps each

• 40 AMchip / board• now we can fit 32 AMchips in one 4th of a 9U VME board

• 4 FPGA switches (1/layer)•Each receiving ~20 fibers, i.e. ~100Gbps•40 outputs: one per Amchip•Possible with today’s FPGAs

32 AMchips



•Dedicated device: maximum parallelism•Each pattern with private comparator•Track search during detector readout

• If you can read it out you can track it!

AM: Associative Memory

Bingo scorecard

AM = BINGO PLAYERS

HIT # 1447

PATTERN NPATTERN 1PATTERN 2

PATTERN 3

PATTERN 5

PATTERN 4



Associative Memory (AM) for pattern matching ~ Bingo game

BINGO PLAYERS

Dedicated device:

Maximum parallelism !

Each pattern with

private comparator

Tracks found during

detector readout !

M. Dell'Orso and L. Ristori, “VLSI structures for track finding”,

Nucl. Instr. and Meth., vol. A278,

pp. 436-440, (1989).



Level 1 drift chamber trigger (XFT)

1 1.5 2 2.5 3 3.5 4

offline transverse momentum (GeV)

XFT

effi

cien

cy

Finds pT>1.5 GeVtracks in 1.9 s

For every bunchcrossing (132 ns)!

(1/pT) = 1.7%/GeV

(0) = 5 mrad

96% efficiency



PDPD

TIBTIB

TOBTOB

TOBTOB

TIDTIDTIBTIB

TECTEC

PDPD

• 210m2 micro-strip silicon detectors 15.232 modules• 6136 320m thick and 18.192 500m thick sensors (all from 6” wafers).• 7.136 APV chips• 9.648.128 analog strip channels.• About 25M wire bonding.

The CMS Silicon Tracker at LHC 22

0 cm

270 cm



SLHC Trigger Requirements

High-pT discovery physics Need high thresholds

Completion of LHC physics program Example: precise measurements of Higgs sector Require low thresholds on leptons/photons/jets

Use more exclusive triggers since final states will be known Still an issue for L1

Control & Calibration triggers W, Z, Top events Low threshold but prescaled



Too large AM? 2 step approach

Roads1. Find low resolution track candidates called “roads”. Solve most of the pattern recognition

2. Then fit tracks inside roads.Thanks to 1st step it is much easier

Super Bin (SB)

OTHER functions are needed inside SVT: Hit Buffer + Track fitter + Hit Finder



Latency ~100 m fiber 300 ns [6 bx] Switch + AM ~1s [20 bx] Sensor read-out latency budget should be less than ~20 bx TOTAL ~ <50 bx [2.5 s]



Muon L1 Trigger Rate at L=1035 cm-2 s-1

1.E+02

1.E+03

1.E+04

1.E+05

1.E+06

1.E+07

0 5 10 15 20 25

pT (GeV/c)

Ra

te (

Hz)

Muons+Tracker

Muons



Stacked trigger

Tracker in the Trigger: from CDF experience to S-CMS Fabrizio Palla INFN – Pisa Vertex 2007 Lake...

Documents

Transcript of Tracker in the Trigger: from CDF experience to S-CMS Fabrizio Palla INFN – Pisa Vertex 2007 Lake...