Pre-Production Validation of the ATLAS Level-1 Calorimeter Trigger System
description
Transcript of Pre-Production Validation of the ATLAS Level-1 Calorimeter Trigger System
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Pre-Production Validation of the ATLAS Level-1 Calorimeter Trigger System
ATLAS Level-1 Calorimeter Trigger Collaboration:
R. Achenbach1, C. Ay2, B.M. Barnett3, B. Bauss2, A. Belkin2, C. Bohm4,I.P. Brawn3, A.O. Davis3, J. Edwards3, E. Eisenhandler5, F.
Föhlisch1,C.N.P. Gee3, C. Geweniger1, A.R. Gillman3, P. Hanke1, S.
Hellman4,A. Hidvégi4, S.J. Hillier6, E.-E. Kluge1, M. Landon5, K. Mahboubi1,
G. Mahout6, K. Meier1, A. Mirea3, T.H. Moye6, V.J.O. Perera3,W. Qian3, S. Rieke2, F. Rühr1, D.P.C. Sankey3, U. Schäfer2,
K. Schmitt1, H.-C. Schultz-Coulon1, S. Silverstein4, R.J. Staley6,S. Tapprogge2, J.P. Thomas6, T. Trefzger2, D.Typaldos6,
P.M. Watkins6, A. Watson6, G.A. Weber2, P. Weber1 1Kirchhoff-Institut für Physik, University of Heidelberg, Heidelberg, Germany2Institut für Physik, University of Mainz, Mainz, Germany
3CCLRC Rutherford Appleton Laboratory, Oxon, UK4Fysikum, University of Stockholm, Stockholm, Sweden
5Physics Department, Queen Mary, University of London, London, UK6School of Physics and Astronomy, University of Birmingham,
Birmingham, UK
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Pre-Production Validation of the ATLAS Level-1 Calorimeter Trigger System
o Trigger Architecture
o Module Design and Challenges
o Testing Methodologies:o Software data verificationo Real-time link stability
measurementso Test-beam performance
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Level-1 triggering in ATLAS
o All data buffered at bunch-crossing rate of 40 MHz for 2.5 s
o Three-stage triggering systemo Level-1: custom built hardware,
fixed latency – target rate 75 kHzo Level-2: software, RoI based
selection – target rate 1000 Hzo Event Filter: software, full
detector – target rate 200 Hz
o Level-1 has three sub-systems:o Calorimeter Triggero Muon Triggero Central Trigger (CTP)
CalorimeterTrigger
e/γtau
jetET
ΣET
MuonTrigger
μ
Level-1Trigger
CentralTrigger
Processor
Trigger toFront-end
Buffers
Calorimeters Muon Detectors
Regions ofInterest (RoI)
To Level-2
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
CalorimeterTrigger
Calorimeter Trigger Requirements
CentralTrigger
~100 bits(0.5 Gbyte/s)
Intermediatedata
ReadoutSystem
Region ofInterest data
Level-2Trigger
< 1μs
Real time(40 MHz)
Triggeredread out(75 kHz)
~2000 Tile Calorimeter
Trigger Towers
300 Gbyte/s
~5000 Liquid Argon Calorimeter
Trigger Towers
50 ns
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Features:
Realtime Path: Fixed Latency Pipelined
Many stage processing
Massive parallelism
Dual purpose modules
Heavily FPGA based
Calorimeter Trigger Architecture
Preprocessor124 modules
Cluster Processor
56 modules
Jet/Energy Processor
32 modules
Merging8 modules
Merging4 modules
Digitized
Energies
Calorimeter
Signals
Merged
Results
To CTP
Real-time
Data Path
ReadoutDriver (ROD)14 modules
Readout
Data
Region of Interest ROD
6 modules
Region of
Interest Data
PPM CPM JEM CMM RODFive Types of
Custom 9U Modules
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Generic Module Design
Readout
InputStage
1-20 FPGA
ProcessingStage
1-8 FPGA
MergingStage
1-2 FPGA
High DensityBack-planeConnector
(~1150 pins)
Signal Speeds up to:400 Mbit/s differential
160 Mbit/s single ended
o Challenges:o Many FPGAso High connectivity
o Internalo External (via
backplane)o High speed
signals
Readout outputup to 800 Mbit/s
PlaybackMemory
SpyMemory
Pro
ce
ss
ing
…
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
A real example: theCluster Processing Module
o Backplane Connector i/o
o Input: ~58 Gbit/so Output: ~28 Gbit/s
o Input Stage:o 20 FPGAs
o Processing Stage:o 8 FPGAs
o Merging Stage:o 2 FPGAs
o Modules Needed: 56
InputProcessingMerging
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Module Testing Goals and Tools
o High speed link stability and performanceo Use data integrity to establish good timing
windowso High statistics mode measurements in real-time
o Parity error countingo Dedicated firmware loads
o Algorithm correctnesso Use specially designed test-vectors, eg:
o Physics-likeo Boundary conditions
o Data formattingo Read-out must conform to external expectations
o System Integrationo Does the complete chain work as a trigger?
Detailed Module SimulationRuns in parallel with hardwarePredicts output data at any stageRequires generation of
synchronised trigger patterns
Test-beam operationAnalogue inputs vs Calorimeter
dataDigital processing integrity
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Simulation architecture and usage
Required Output
o Generic C++ frameworko VHDL inspired
o Processes, Ports, Linkso Hierarchical, Scalableo Output ‘Driven’
Online Database:
Modules present
Cabling
Configuration
Test Vectors
Trigger Pattern
Configure hardware
(data in playback memories)
Run hardware
Run simulation
Read hardware
(spy memories, readout)
Simulation Results
Compare
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Artificial trigger generation
o To predict readout data need to control trigger generation
o Also need to test hardware with high rate and ‘difficult’ trigger patterns
o Done via a custom module plus trigger pattern generation (under user control)
o Triggers synchronised to playback memories
o Simulation also knows about trigger pattern
o Many types of pattern possible
Continuous:eg 50 kHz
‘Bursty’:eg Bursts of 5
triggers
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
CPM input LVDS timing windows
o Input FPGAs receive 40 Mbit/s signalso Input strobing requires timingo Timing window: 20ns over whole moduleo Once optimised, check with firmware
o No error in 10 minutes in whole module = bit error rate < 1 in 1013
Strobe Phase (ns) 0 5 10 15 20 25
Err
or
Co
un
t
400
0
800
1200Error Free
Window ~20 ns
Strobe Phase (ns) 0 5 10 15 20 25
0
4
8
12
16
20
Inp
ut
FP
GA
Id
en
tifi
er
Timing structure
derived from
PCB layout
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Processor FPGA timing windows
o Jet/Energy Processor FPGAs:385 inputs at 80 MHz210 on-board, 165 from
backplane
1600
800
400
1200
0
2000
Err
or
Co
un
t
Strobe Phase (ns) Strobe Phase (ns) 0 5 10 15 20 25 0 5 10 15 20 25
800
400
200
600
0
1000
Err
or
Co
un
t
0
o Cluster Processor FPGAs:108 inputs at 160 MHz60 on-board, 48 from
backplane
Error Free
Window ~8 ns
Error Free
Window ~2.5 ns
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Pre-processor
Calorimeter Signals:32 in 8x2x2 array
12 Cables:8 CPM4 JEM
5 ReadoutLinks
ATLAS Combined Test-beam 2004
o Level-1 Calorimeter Trigger present September to mid-October
o Triggered successfully for part of the time by L1Calo
o A full slice through the trigger systemo ~1% of final capacity
Cluster P
rocessor
Jet Processor
Ene
rgy Merge
r
Hit M
erger
Readout
RoI R
eadout
Readout
Readout
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
The RealityPre-processor
Receivers
Cluster/Jet/Energy
Processors
Readout Drivers
Interface to Test-beam
Trigger, Timing ....
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Checks on Test Beam data
o Internal Consistency checkso Assume full granularity input data is correcto Are all other data (energies, hits etc) consistent?o Performed in ~500,000 events
o Only minor problems, identified as firmware featureso No evidence of data integrity problems
o Comparisons with Calorimeterso Good correlation was seen
o With some problems – overlapping pulses?o Triggers generated on genuine physics events
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Cluster processor hit results
o Four thresholdso 20, 50, 100, 200
GeV
o Hit results are as expected
o RoIs also checked
o Positions, hits all formed correctly
Cluster Energy (GeV)
10
100
1000
Ev
en
t C
ou
nt
events with
no cluster hits
just one
threshold passed
just two
thresholds
passed
three or more
thresholds
passed
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Jet/energy algorithm results
o Only one threshold used o 170 GeV
o No errors seen in hits
o No errors seen in energy sums
Jet Energy (GeV)
Ev
en
t C
ou
nt
Ev
en
t C
ou
nt
Jet Energy (GeV)
102
104
103events with
no jet hits
events with
jet hit
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
Correlation with calorimeters
o Tile comparison:
o Very encouragingo Saturation at ~225
GeV understood
o Liquid Argon comparison:
o ~factor 2 scaling (ET)o Lose energy in some
events: overlapping events? bunch-crossing identification problem?
100 50 0 150 200 250300 350Trigger Energy (GeV)
Ele
ctr
om
ag
ne
tic
En
erg
y (
Ge
V)
Ha
dro
nic
En
erg
y (
Ge
V)
Validation of ATLAS Level-1 Calorimeter Trigger, Stephen Hillier
o Run with Cluster Threshold of 20 GeV as CTP triggero Clear cut-off in Electromagnetic Energy as Measured by
Liquid Argon Detector
E.M. trigger threshold
Did the Trigger Work?
Calorimeter Energy (GeV)
Tri
gg
er E
ner
gy
(GeV
) Jet and Energy Triggersalso used in other runs