alberto annoviftk meeting - september 30, 2004 ideas for a fast-track trigger processor - ftk... an...
DESCRIPTION
Alberto AnnoviFTK meeting - September 30, 2004 30 minimum bias events + H->ZZ->4 Tracks with P t >2 GeV Where is the Higgs? FTK 30 minimum bias events + H->ZZ->4 Tracks with P t >2 GeV Where is the Higgs? Help! Online tracking: a tough problemTRANSCRIPT
Alberto Annovi
FTK meeting - September 30, 2004
Ideas for a Fast-Track trigger processor - FTK... an evolution of the CDF Silicon Vertex Trigger
(SVT)
A. Annovi for the Fast-Track group
Work during 1998-2003: INFN, University of Pisa, SNS - PisaUniversity of ChicagoUniversity of Geneva
Co-operating on standard cell AMChip:INFN, University of Ferrara
FAST-TRACK COLLABORATION
Offline-quality tracks @LHC Level 1 output rate
Alberto Annovi
FTK meeting - September 30, 2004
Fast-Track working principles
FTK performances overview• speed & size• track quality
FTK can grow with the experiment• Proposed plan
Possible applications and physics reach:• b-tagging • e/ selection
More details on Trans. on Nucl. Sci. papers:http://www.pi.infn.it/~orso/ftk
Outline
Alberto Annovi
FTK meeting - September 30, 2004
30 minimum bias events + H->ZZ->4
Tracks with Pt>2 GeV
Where is the Higgs?
FTK
30 minimum bias events + H->ZZ->4
Tracks with Pt>2 GeV
Where is the Higgs?Help!
Online tracking: a tough problem
Alberto Annovi
FTK meeting - September 30, 2004
Where could we insert FTK?
Fast Track + few(Road Finder) CPUs
Track dataROB
high-qualitytracks:Pt>1 GeV
Ev/sec = 50~100 kHz
Very low impact on DAQ
PIPELINE
LVL1
Fast network connectionCPU FARM (LVL2 Algorithms)
CALO MUON TRACKER
BufferMemory
ROD
BufferMemory
FEFE
Raw dataROBs
2nd output 1st output
No changeto LVL2
Alberto Annovi
FTK meeting - September 30, 2004
Tracking in 2 steps
Roads1. Find low
resolution track candidates called “roads”. Solve most of the combinatorial problem.
2. Then track fitting inside roads.Thanks to 1st step it is much easier.
Alberto Annovi
FTK meeting - September 30, 2004
...
The Pattern Bank
1st step: pattern recognition with the Associative Memory (AM)
• Dedicated device with maximum parallelism
• Store all patterns corresponding to interesting tracks
• Road search happens during detector readout
• How to send all hits to the AM?
SVT’s AMChip
Alberto Annovi
FTK meeting - September 30, 2004
Pixels barrel SCT barrel Pixels disks
1/2
AM
1/2
AM
Divide into sectors
6 buses 40MHz/bus
ATLAS Pixels + SCT
Feeding FTK @ 50KHz event rate
6 Logical Layers: full coverage
Allow a small overlapfor full efficiency
Simple configuration for the beginning
Alberto Annovi
FTK meeting - September 30, 2004
AM input bandwidth = 40 MHz cluster/busAM input buses = 6
Logical layer <cluster/event> cluster rate
Pix 0 1300 64 MHzPix 2 + extra 1200 61 MHzSC0 + extra 1000 50 MHzSC1 + extra 1300 65 MHzSC2 + extra 1200 61 MHzSC3 + extra 1300 64 MHz
Ev/sec 50kHz
2 FTK processorsworking in parallel for the whole Pix+Si tracker
More processors as a backup option
ATLA
S-TD
R-11
Alberto Annovi
FTK meeting - September 30, 2004
Track dataROB
Raw dataROBs
~Offline quality Track parameters
~75 9U VME boards – 4 types
SUPER BINSDATA ORGANIZER ROADS
ROADS + HITS
EVENT # NPIPELINED AM
HITSDO-board
EVENT # 1AM-board
2nd step: track fitting
Inside Fast-TrackPixels & SCT
DataFormatter
(DF)
50~100 KHzevent rate
RODs
cluster findingsplit by logical layer
overlap regions
GB
Few CPUs
S-links
CORE
Alberto Annovi
FTK meeting - September 30, 2004
proposed R&D program
2nd output 1st output
• Soon: in order to have the FTK in the future the only short term issue is the availabilityof the dual output HOLA. (also usefulfor diagnostic and commissioning)Alternative: use optical splitters
• 2008: @ very low luminosity minimal R&D FTK system
very cheap using low density CDF AMChip
(barrel only: ~40 boards)• 2009 ?: increase the R&D system to include disks
new AMChip for 2*1033 lumi (barrel+disks: ~75 boards)• 2011 ?: upgrade for high lum.
Alberto Annovi
FTK meeting - September 30, 2004
How FTK core will look like?
AM-B
7AM
-B8
AM-B
1AM
-B0
DO5
DO4
DO3
DO2
DO1
DO0
CUSTOM BACKPLANE
Ghos
t Bu
ster
FTK INPUT
CPU
0 C
PU1
O(50 106) patterns
AM-B
2AM
-B3
CPU
2 C
PU3 AM
-B4
AM-B
5AM
-B6 • ~offline
quality tracking 50 kHz event (2*1033 lumi)
• 2 core crates
• + 3DF crates
128 AMChips/board
Alberto Annovi
FTK meeting - September 30, 2004
ATLAS Barrel (~CERN/LHCC97-16) 7 layers: 3 Pixel + 4 strip (no stereo)Cylindrical Luminosity Region: R=1mm, z=±15cmGenerate tracks (Pt>1 GeV) & store NEW patterns
1/4BARREL15M
patterns Thin Road Width (r z): pixel 1mm6.5cm Si 3mm12.5cmMedium Road Width: pixel 2mm6.5cm Si 5mm12.5cmLarge Road Width: pixel 5mm6.4cm Si 10mm12.5cm
The Associative Memory can store any kind of tracks: Conversions, delta-rays, ks decays …Including them just requires a lager Associative Memory
These kind of tracks have not been studied.BUT we can do the exercise again.
Alberto Annovi
FTK meeting - September 30, 2004
Fit/trk <Nfit/Road>x<Nroads/track>
13 comb x 34 roads= 440 comb/track QCD Pt>401.4 fit x 4 roads = 6 comb/track QCD Pt102.3 fit x 6 roads = 14 comb/track QCD Pt407.8 fit x 9.5 roads = 74 comb/track QCD Pt10027 fit x 25 roads = 658 comb/track QCD Pt200
thinthin
large
large
Track fitting workload<N
fit/ro
ad>
Low luminosity: 2*1033
Alberto Annovi
FTK meeting - September 30, 2004
Step 2: Software Linear FitNfit/trk
65874146
Ntrk/ev1716108
L1 Trigjetjet
soft jetsoft
L1 Rate<100Hz<3KHz~5KHz
~40KHz
Pt 200Pt 100Pt 40Pt 10
Fits/sec<1.1MHz<3MHz750KHz3MHz8MHz
Pulsar TFfit/s 10 MHz
PIII 800MHzfit/s 1.1 MHz
Htt 130 comb/trk 34 trk/ev <latency> = 1ms max latency = 100ms
only 8 CPUs (barrel)
Latency Test
Pulsar TF + new mez.fit/s >30 MHz
Alberto Annovi
FTK meeting - September 30, 2004
Is 2nd step as good as offline?
/N
ATLAS Genova: M. Cervetto, P. Morettini, F. Parodi, C. Schiavi, presented on 20-Nov-2002 at PESA
• Track finding within a road is fast
• Fitting in linear approximation
• Testing the linear fit with a fast simulation of ATLAS Silicon TrackerTrack parameter residuals:(d0) = 17 m
Alberto Annovi
FTK meeting - September 30, 2004
FTK R&D status
3 DF crates:cluster findingsplit by layer
2 “core” crates:road findingtrack fitting
S-linksRaw data
ROBsTrack data
ROB
AMChipAMBoard
Data OrganizerGhost BusterTrack Fitter
Data Formatter boardPixel cluster finder
TODO:
Pixels& SCTRODs
Alberto Annovi
FTK meeting - September 30, 2004
TODO list• DF board have some ideas
• Pixel cluster finder need R&D work
• AMChip new design for 2*1033 lumi
• AMBoard modify prototype
• Data Organizer modify prototype / new R&D
• Ghost Buster Pulsar ??
• Track Fitter CPU or FPGA ???
• FTK simulation needed for design studies
Alberto Annovi
FTK meeting - September 30, 2004
FTK R&D status FTK AMBoardModifing it for CDF SVT upgradeWill learn from CDF experiencethen modify it for ATLAS
FTK Data Organizer1st prototype never fully testedNeed a lot of RAM on boardBuffers up to 16 events more complex than SVT HB
Alberto Annovi
FTK meeting - September 30, 2004
How to use Fast-Track to capture as much PHYSICS as
possible b
e
b
b
FT
K
e hb
b
Hard life for all LVL2 objects!
Alberto Annovi
FTK meeting - September 30, 2004
ATL-DAQ-2000-033
Offline-quality b-tagging for
events rich in b-quarks
with Fast-Track offline b-tag performances early in LVL2
ATLA
S TD
R-01
6
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 bbbbH+- tb qqbb
Z0 bb
Alberto Annovi
FTK meeting - September 30, 2004
ATLA
S +
FTK
4SE200 + J70 + J50 + J15 (||<2.5)
““
““
2.6MU6 + J25 + J10 (||<2.5)
50mini ev.
2 b-jets +Mbb > 50
160mini ev.
2 b-tags +Mbb > 50
13b leading
43 b-tags
ATL-
COM
-DAQ
-200
2-02
2 F.
Gia
notti
, LH C
C , 0
1/0 7
/ 20 0
2 &
CM
S T D
R 6
Triggers w/o and with FTKScenario: L= 2 x 1033 deferral
ATLA
S
CMS 5b-jet 237Inclusive b-jet
0.20.20.2
J2003J904J65
4020210
0.80.2
MU202MU6
HLT rate (Hz)
HLT selection
LVL1 rate (kHz)
LVL1selection
25
j4003j1654j110
Alberto Annovi
FTK meeting - September 30, 2004
bbH/A bbbbAT
LAS-
TDR-
15 (1
999)
MA (GeV)
tan
200
Analysis:4 b-jets |j|<2.5 PT
j > 70, 50, 30, 30 GeV efficiency 10%Effect of trigger thresholds(before deferrals)
ATLAS + FTK triggers
13%3b leading3J + SE2008%3 b-tagsMU6+ 2J
Effic.LVL2LVL1 As efficient as offline selection:full Higgs sensitivity
ATL-
COM
-DA Q
-200
2-02
2
Alberto Annovi
FTK meeting - September 30, 2004
Electron Identification
Swapping trigger algorithms can reduce trigger rate while increasing efficiency!
CERN
/LHC
C/20
00-1
7
L2 tracking
EF tracking
ATLAS
With FTK tracks are ready on the shelf: using tracks could be even faster than using calorimeter raw data!
Efficiency & jet rejection could be enhanced by using tracks before
calorimeters.
Alberto Annovi
FTK meeting - September 30, 2004
L=2x1033 cm-2 sec-1
HLT selection @ CMS H(200,500 GeV) 1,3h± + X
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
0,50
0 Ge
V)
1,3
h+X)
mH=500mH=200
TRK tau on first calo jetsPix tau on first calo jet
Staged-Pix tau on first calo jet
TRK tau on both calo jetsCalo tau on first jet
0.0070.004
Efficiency & jet rejection could be enhanced by using tracks before
calorimeters.
Alberto Annovi
FTK meeting - September 30, 2004
FTK can find offline quality tracks @LVL1 output rate!
FTK is very compact: 2 “core” crates + 3 DF’s crates (for a first barrel only R&D system)
More efficient LVL2 triggers: Lower LVL1 & LVL2 thresholds and save CPU power!
b-jet, -jet tagging at rates 10-20 KHz: more Higgs physics !
Conclusion
Alberto Annovi
FTK meeting - September 30, 2004
Mbb(GeV)
Even
ts
Z0 b-bbarImportant b-jet calibration tool
CDF RunIIpseudo exp. (with SVT)
Cdf/anal/top/cdfr/4158
ATL-
COM
-DAQ
-200
2-02
2
ATLAS + FTK 20fb-1
20Mbb > 503J + SE200
60Mbb > 50MU6+ 2JS/BLVL2LVL1
(S/B = 35)
2fb-1
Mbb(GeV)
Even
ts
Alberto Annovi
FTK meeting - September 30, 2004
Standalone program to produce hits from tracks; it includes:• multiple scattering• ionization energy losses• detector inefficiencies• resolution smearing• primary vertex smearing: xy=1mm z=6cm
Detector hits generated from: (Pythia) • QCD10 sample: QCD Pt>10 GeV L1 • QCD40 sample: QCD Pt>40 GeV L1 soft jet• QCD100 sample: QCD Pt>100 GeV L1 jet • QCD200 sample: QCD Pt>200 GeV L1 jet
all samples + noise + <5 MB>. Road finding 6 layers/7 (FTK simulation)
Alberto Annovi
FTK meeting - September 30, 2004
Data Organizer
Hits
Tracks parameters(d, pT, , z)
Roads
AssociativeMemory
Hits
Pattern recognition with Associative Memory (AM) using up to 12 layers no need for initial seed
highly parallel algorithm using coarser resolution to reduce memory size
Roads + hitsTrack Fitter
Track fittingusing full resolution of the detector
Use CPUs for maximum flexibility
FTK Basic Architecture
Alberto Annovi
FTK meeting - September 30, 2004
Step 1: Pattern Recognition
Hardware + CPU:•4 AM (40M patterns)•8 CPUs•Ev/sec 50KHz
AM Simulation:•107 CPUs•Ev/sec 50 KHz
Software future:better algorithms(region of interest)
Barrel