Performances of the upgraded SVT

1
The Silicon Vertex Trigger upgrade at
CDF J.Adelman1, A.Annovi2, M.Aoki3, A.Bardi4,
F.Bedeschi4, S.Belforte5, J.Bellinger6,
E.Berry1,M.Bitossi2, M.Bogdan1, M.Carlsmith6,
R.Carosi4, P.Catastini9, A.Cerri8, S.Chappa7,
W.Chung6,M.A.Ciocci9, F.Crescioli2, M.Dell
Orso2, B.Di Ruzza11, S.Donati2, I.Furic1,
S.Galeotti4, P.Giannetti4, C.M.Ginsburg6,
P.Giovacchini4, R.Handler6, Y.K.Kim1, J.D.Lewis7,
T.Liu7, R.Mahlum7, T.Maruyama3, F.Morsani4,
G.Ott6, I.Pedron10, M.Piendibene4, M.Pitkanen7,
L.G.Pondrom6, G.Punzi2, B.Reisert7, M.Rescigno11,
L.Ristori4, H.Sanders1, L.Sartori10,
F.Schifano10, F.Sforza9, M.Shochet1, B.Simoni2,
F.Spinella4 , P.Squillacioti9, F.Tang1, S.Torre9,
R.Tripiccione10, G.Volpi9, U.K.Yang1,
L.Zanello11, A.M.Zanetti5 1University of
Chicago,Illinois,USA, 2University of Pisa, Italy,
3University of Tsukuba,Japan, 4INFN Sezione di
Pisa, 5INFN Sezione di Trieste, 6University of
Wisconsin,USA, 7Fermilab,Batavia,Illinois,USA,
8LBL,California,USA, 9University of Siena,Italy,
10University of Ferrara and INFN,Italy,
11University of Rome and INFN,Italy
7.6 MHz Crossing rate
CDF DAQ Trigger

• The Silicon Vertex Trigger reconstructs in real
  time tracks precise enough to measure b quark
  decay secondary vertices.
• The tracks reconstructed by SVT are used for the
  selection of events at the Collider Detector at
  Fermilab (CDFII)
• The CDF DAQ and Trigger system is organized in
  three levels. The Level 2 uses the SVT tracks for
  the event selection
• The Level-2 Trigger processing time at present
  limits the Level-1 bandwidth depending on
  instantaneous luminosity. The SVT takes a
  significant fraction of the total Level-2
  processing time whose fluctuations cause deadtime
  and limit the Level-2 processing rate

Finding tracks in the silicon
Detector Raw Data
Design goals

• Level 1
• 7.6 MHz Synchromous Pipeline
• 5544 ns Latency
• 50 KHz accept rate

20 kHz actual

• 40 kHz accept rate

SVT here
Level 1 pipeline 42 clock cycles
Level 1 Trigger

• The task of the Silicon Vertex Trigger (SVT) is
  very complex
• Links hits from five layers of the Silicon Vertex
  Detector (SVX) to segments observed in the
  Central Outer Chamber (COT)
• The task proceeds through steps of increasing
  resolution.
• Associate hits to tracks at low resolution
  (roads) strongly reducing the combinatorics
• Fit tracks and precisely determine their
  parameters solving the residual combinatorics
• Thanks to the use of Associative Memories the
  first step is performed in parallel during the
  detector readout

• Level 2
• Asynchronous 3 Stage Pipeline
• 20 ?s Latency
• 300 Hz accept rate

• 20 ?s average Latency

35 ?s actual
COT tracks
SVX hits
L1 Accept

• 2 steps
• Find tracks _at_ low resolution not time consuming
• Fit hits at full res. time consuming depending
  on the number of fits

Level 2 Trigger
Low res track
Level 2 buffer 4 events
300?m
L2 Accept
Tails are important
DAQ buffers
L3 Farm
To Mass Storage (50100 Hz)
1st Pulsar Sequencer Road warrior (AMS/RW)
Associative Memory 512 kpattern (AM)

• Why was the SVT upgrade necessary?
• SVT processing time is well described by this
  model c1(35ns)N(Hit) (300ns)N(Comb.). Left
  plot shows comparison between this
  parameterization (blue line) and data (red
  histogram) taken at 5x1031cm-2s-1. The two
  histograms agree.
• The peak luminosity the Tevatron is expected to
  provide is 30x1031cm-2s-1 (6 times the luminosity
  used to train the model). Middle plot shows the
  expected performances of the SVT at the maximum
  luminosity. 56 of events would take longer than
  50?s to be processed a time long enough for all
  Level 2 buffers to be filled. Impossible to run
  SVT at that luminosity.
• To reduce the processing time
• Thinner patterns ? less fits but bigger AM
  (AM)
• Bigger AM ? Need faster HIT-PATTERN
  association ? new Hit Buffer (HB)
• Faster Fits ? new Track Fitter (TF)
• Rightmost plot shows how the 512k pattern AM bank
  and TF reduce the tails
• The two steps upgrade
• First install AM, AMS/RW, TF allow for 128k
  pattern bank. AM inherited from FTK. TF and
  AMS/RW implemented in Pulsar
• Second step faster HB in another pulsar to
  support the final 512k pattern bank.

3rd Pulsar Track Fitter (TF)
2nd Pulsar Hit Buffer (HB)

• Phased installation of the SVT upgrade
• The commissioning of the SVT upgrade occurred
  during data taking
• Need to reduce the impact on the data
  acquisition proceed in three phases
• Install AMS/RW and AM with 128k patterns
  enabling only 32k patterns major changes to the
  SVT crate layout
• Install TF and after few days of data taking
  without problem enable the whole 128k patterns
  bank (July 2005)
• Install HB and after few days of data taking
  without problem enable the whole 512k patterns
  bank (February 2006)
• System fully tested before installation of any
  board
• Standalone test of each board check firmaware
  functionality and develop the software for
  monitoring and debugging
• Vertical slice tests create a whole SVT crate
  with new boards and feed it with data coming from
  one SVT wedge to compare the output of old and
  new system
• Take data with one upgraded wedge before
  proceeding to the full installation we install
  the new boards in one wedge and take data for at
  least 100 hours
• Most of the data taken during the commissioning
  were good

Performances of the upgraded SVT
Effect of the upgrade on the DAQ The deadtime as
a function of the rate of events accepted by the
Level 1 (L1A) shows the upgrade impact on the
performance of the DAQ. The upgrade reduces the
deadtime allowing for higher output rates at
Level 1

• Mean processing time
• The average processing time of old SVT used to
  have a large growth at high luminosity.
• The faster hardware
• allows for smaller mean processing time
• reduces the dependence on the instantaneous
  luminosity
• The new system allows for a smaller latency at
  Level 2

• Fluctuations of the processing time
• Large processing times measured by the
  distribution RMS are due to complex events
• They are reduced by improving the fitting stage
• The TF fits each hit combination in less time,
  reducing the dependence on the number of
  combination (175 ns instead of 300 ns)
• The larger pattern bank allow for thinner road
  and consequently a smaller number of combination
  to be fitted per road
• The upgrade reduce the dependence of the
  fluctuations on the luminosity providing a larger
  Level 1 bandwidth over a wide luminosity range

B triggers

• Fraction of long processing time events
• Events with processing time higher than 50?s can
  cause all the Level 2 buffers to be filled and
  therefore deadtime
• The percentage of this kind of events used to be
  strongly dependent on the instantaneous
  luminosity
• The upgrade reduces the fraction of long
  processing time events and its dependence on the
  instantaneous luminosity
• With 512k patterns at luminosity of
  1.5x1031cm-2s-1 less than 2 of events require
  more than 50?s to be processed

• At low luminosity the bandwidth is mostly filled
  by B physics triggers
• The 128 kpattern bank already allowed to
  increase the minimum Level 1 Accept rate at low
  luminosity from ?20 kHz (blue) to ?25 kHz
  (violet).
• With the 128 kpattern bank we can already
  collect 20 more of B decays than in the past
  with negligible deadtime
• Because of the shutdown no significant
  comparison with fully upgraded SVT is possible
  yet at high luminosities, but the power of the
  system has been strongly improved (see plots on
  the left) to be ready for the highest
  luminosities .
• Thanks to the upgrade, CDF will be able to fully
  exploit the increase of the Tevatron luminosity
  and efficiently select events containing
  displaced vertexes