1 FastTrack: Real Time Silicon Tracking for LHC Alessandro Cerri (borrowing from several talks…)

1

FastTrack: Real Time Silicon Tracking for LHC

Alessandro Cerri(borrowing from several

talks…)

2

Outline

• What are we talking about?– ATLAS trigger (quick!) overview– What’s missing?

• Does it work? How?– CDFII experience– Evolving towards LHC

• Why would one want to use it?– Selected physics cases

• Think outside the box!

3

The ATLAS Trigger

• High rate pp collisions force us to throw away events: 40MHz ~100Hz

• You want to throw away uninteresting* stuff

• How?• Combine trigger primitives:

“crude” approximations of analysis objects, like:– Jets– e/– Tracks

– Et (and lack thereof)

– EM

• Where is the 3rd generation???

4

FastTrack• L2 is designed to be basically a

commercial CPU farm• …not enough time to reconstruct tracks

at full resolution• Why would I want to do that?

– b tagging – … but keep your mind open: you can do a

lot more with a little fantasy!• Is there money (physics reach) to gain?

3rd generation is the closest to new physics!

5

30 minimum bias events + H->ZZ->4

Tracks with Pt>2 GeV

Where is the Higgs?

FTK

FastTrack to the rescue!

Where is the Higgs?

Help!

6

ATL-DAQ-2000-033

with Fast-Track offline b-tag performances early in LVL2. You can do things 1 order of magnitude better

ATLA

S T

DR

-016

0.6

100

10

1000

b

Ru Calibration sample

bbH/A bbbb

tt qqqq-bb

ttH qqqq-bbbb

H/A tt qqqq-bb

H hh bbbb

H+-

tb qqbb

Z0 bb

The case for offline-like b-tagging

7

FastTrack/LHC: access to the 3rd generationA

TLA

S +

FTK

4ET200 +

j70 + j50 + j15 (||<2.5)

““

““

2.66 + j25 + j10 (||<2.5)

ATL-

CO

M-D

AQ

-20

02-

02

2

F. G

ianott

i, L

HC

C, 0

1/0

7/2

002

CM

S T

DR

6 &

Scenario: L= 2 x 1033 deferralA

TLA

S

CMS 5b-jet 237Inclusive b-jet

50mini ev.

2 b-jets +Mbb > 50

160mini ev.

2 b-tags +Mbb > 50

0.20.20.2

j2003j904j65

4020210

0.80.2

2026

HLT rate (Hz)

HLT selection

LVL1 rate (KHz)

LVL1selection

25

j4003j1654j110

13b leading

43 b-tags

Even better strategies: see ‘physics cases’

8

Is it feasible?• We are talking about something

capable of digesting 100000 evts/second and identifying tracks in the silicon

• What on earth would be able to do that?

~ 48 m

Single Hit

Superstrip

Road

Dete

ctor

Laye

rs

•… it turns out CDFII has been doing something similar since day 0•The recipe uses specialized hardware:

1)Clustering Find clusters (hits) from detector ‘strips’ at full detector resolution

2)Template matching Identify roads: pre-defined track templates with coarser detector bins (superstrips)

3)Linearized track fitting Fit tracks, with combinatorial limited to clusters within roads

9

Is it effective?

2 b-jets (Zbb)MET + disp. tracks (ZH)lepton + disp. track (SUSY)gamma + disp. track (SUSY)

Many high-pt triggers based on SVT are taking

data.

SVT

SVT rejection:3 orders of magnitude

10

Can we scale to the ATLAS complexity?

• Not easy:– 500K channels O(100M)– 20s2s

• But feasible:– SVT has been designed in

~1990 with (at the time) state of the art technology

– We have been thinking a lot on how to improve the technology

– The SVT ‘upgrade’ (2005) is in fact partly done with hardware capable of LHC-class performance!

1998: Full custom VLSI “Associative Memory” chip:

128

patterns

2004: Standard Cell “Associative Memory” chip:

~5000

patterns

11

1/2

A

M1

/2

AM

Divide into sectors

6 buses 40MHz/bus

ATLAS Pixels + SCT

Feeding FTK @ 50KHz event rate

Pixel barrel SCT barrel Pixel disks

6 Logical Layers: full coverage

2 sectors

~70MHz cluster/layer(Low Luminosity, 50KHz ev.)

ATLA

S-T

DR

-11

Allow a small overlapfor full efficiency

12

How to pick the ATLAS data?

Fast Track + few(Road Finder) CPUs Fast Track + few(Road Finder) CPUs

ROBROB

offlinequalitytracks:Pt >1 GeV

Ev/sec = 50~100 kHz

~NO impact on DAQ

PIPELINE

LVL1LVL1

Fast network connectionFast network connection

CPU FARM (L2 Algorithms)CPU FARM (L2 Algorithms)

CALO MUON TRACKERCALO MUON TRACKER

BufferMemory

ROD

BufferMemory ROB

FEFE

S-link

Two outputs!

~40 9U VME boards

13

Selected Physics Cases

14

Lots of ideas, limited energy:

Zbb Better acceptance (calibration samples)

bbH/A Hbb,

Low Pt b-jets

Hhh bb bb

l Lower thresholds (calibration sample)W

Multi-prong triggers Improved acceptance

B Lower thresholds

15

Example 1: Zbb•Important calibration tool to measure jet response/resolution (-jet and z-jet balance have theo/exp issues)

•Standard trigger: Large L1 rate higher Et threshold high Mjj turn-on

•With FastTrack: qgZqbbq (3jet + btag) advantages:

•Better Mjj acceptance, improved rejection

•Highest Et jet needs not be tagged!LVL1 LVL2 S/B

MU6+ 2J 2.6 KHz Mbb > 50 160 Hz 60 (@20 fb-1)

3J + SE200 4 KHz Mbb > 50 50 Hz 20 (@20 fb-1)

J190 5 KHz 1 non-b, 2b

10 Hz 21 (@30 fb-1)

16

Example 2: bbH/A bbbbA

TLA

S-T

DR

-15

(1

99

9)

MA (Gev)

tan

200

Optimized Analysis (not very recent though):4 b-jets |j|<2.5 PT

j > 70, 50, 30, 30 GeV efficiency 10%Effect of trigger thresholds (70,50,30,30)->4x110 !!!

ATLAS + FTK triggers

13%3b leading3j + ET200

8%3 b-tagSoft6 + 2j

Effic.LVL2LVL1 As efficient as offline selection:full Higgs sensitivity

ATL-

CO

M-D

AQ

-20

02

-0

22

Standard trigger limits tan reach at low MA !!!

17

Example 3: @ CMS

L=2x1033 cm-2 sec-1

0.4

0.5

0.6

0.7

0.8

0.9

1.

0 0.02 0.06 0.1 0.14

(QCD 50-170 GeV)

(H

(20

0,5

00

GeV

)

1

,3h+

X)

mH=500

mH=200

TRK tau on first calo jets

Pix tau on first calo jet

Staged-Pix tau on first calo jet

TRK tau on both calo jets

Calo tau on first jet

0.0070.004

Efficiency & jet rejection could be enhanced by using tracks before

calorimeters.

L2/L1

Default algorithm: calorimetric search first, then tracking

Isol: R~0.2-0.45

Tag tracks: R~0.07

Lead. Track: R~0.1

18

Q: Which of these represents an actual trigger rate vs luminosity?

19

Be careful!

•CDF misunderestimated( GWB) the background rates by large (~2x) factors. Not for ingenuity but for lack of better ways of extrapolating to the High Energy Frontier! Expect something similar!

•Rates and rejections must be understood at our best NOW:

•Anything too loose will be cut out/removed

•Trigger rates are *not* dominated by physics:

20

Where would I put effort•Simulating background requires HUGE resources: billions of MC events @ 5 minutes/event ?!??

•Revert to fast simulation

•Calibrate (e.g. jet response and trigger efficiencies) from full simulation

•Parameterize in AtlFast!

•Need to strengthen the physics case:

•Ideas

•Other physics cases

•Applications

•Tools

•Fast simulation is basically there (but still not 100%)

•There is a substantial setup time: the sooner the better

•Brainstorming!

21

Beyond b tagging?• FastTrack is extremely modular• With little interfacing, any detector can in principle

be used as seed for FastTrack objects:– Muons– Calorimetry– TRT

• What would you be able to do with those at trigger level?

• Any other wild dream of yours?• Mine: FastTrack can do more complicated pattern

recognition than just tracks– Vertices?– Topological triggers?

22

perform b tagging. ol that allows good

trigger

Some References:

http://www.pi.infn.it/~orso/ftk/

http://www.pi.infn.it/~annovi/

http://hep.uchicago.edu/cdf/shochet/ (under ftkxxx)

http://www-cdfonline.fnal.gov/svt/

physics

23

IEEE Trans. Nucl. Sci. 51, 391 (2004)