IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 55, NO. 1, FEBRUARY 2008 145
Performance of the Proposed Fast Track Processor
for Rare Decays at the ATLAS Experiment
Erik Brubaker, Catalin Ciobanu, Francesco Crescioli, Monica Dunford, Paola Giannetti, Young-Kee Kim,
Tony Liss, Mauro Dell’Orso, Giovanni Punzi, Mel Shochet, Giulio Usai, Iacopo Vivarelli, Guido Volpi, and
Kohei Yorita
Abstract—The Fast Track processor (FTK) has been proposed
for high–quality track finding at very high rates (Level–1 output
rates) for the LHC experiments. Fast, efficient and precise pattern
recognition has been studied using a silicon 7-layer sub-detector,
including a subset of the pixel and SCT layers. We tested the FTK
algorithms using the ATLAS full simulation. We compare the FTK
reconstruction quality with the tracking capability of the offline
iPatRec algorithm. We show that similar resolutions and efficien-
cies are reached by FTK at a speed higher than iPatRec by or-
ders of magnitude. With FTK full events are reconstructed at the
Level–1 output rate. events are fully simulated to-
gether with background samples. We show that a low Level–2 rate
is allowed by FTK, even using a single 6 GeV Level-1 muon selec-
tion trigger. FTK provides the full-resolution track list ready for
the Level–2 identification. All selection cuts performed by the
Event Filter can be easily anticipated at Level–2. We present the
efficiency gain and related Level–2 rates.
Index Terms—ATLAS, fast track, FTK, trigger.
I. INTRODUCTION
T
HE Fast Track processor (FTK) [1] is a dedicated hard-
ware processor for on-line pattern recognition of tracker
detector data. FTK is an evolution of SVT [2], the Level–2
tracker currently running in the CDF experiment. FTK is a pow-
erful processor that, in combination with the Level 2 Farm is
capable of reconstructing offline–quality tracks for all particles
of transverse momentum above 1 GeV or even less. This work
can be performed at the very high event rates accepted by the
Level–1 trigger, i.e., up to 50–100 kHz.
Fig. 1 shows how FTK can be integrated in the data acquisi-
tion system of a LHC experiment. Tracking data are collected at
the Level–1 trigger rate in the front end, then stored into large
memory buffers. These buffers are interfaced to a large CPU
Manuscript received: 10 May, 2007; revised 15 November, 2007.
E. Brubaker, M. Dunford, Y. K. Kim, M. Shochet, G. Usai, and K. Yorita
are with Chicago University, Chicago, IL 60637 USA (e-mail: brubakee@fnal.
gov; mdunford@hep.uchicago.edu; ykkim@fnal.gov; shochet@hep.uchicago.
edu; giulio.usai@pi.infn.it; kohei@fnal.gov).
C. Ciobanu and T. Liss are with the University of Illinois, Urbana, IL 61801
USA (e-mail: catutza@lx1.hep.uiuc.edu; tml@uiuc.edu).
F. Crescioli, M. Dell’Orso, and G. Punzi are with the University of Pisa,
INFN Pisa, 43-56126 Pisa, Italy (e-mail: francesco.crescioli@pi.infn.it; gio-
vanni.punzi@pi.infn.it; mauro.dellorso@pi.infn.it).
P.Giannetti and I. Vivarelli are with the INFN Pisa, 43-56126 Pisa, Italy
(e-mail: paola.giannetti@pi.infn.it).
G. Volpi is with the University of Siena, 53100 Siena, Italy, and also with the
INFN Pisa, 43-56126 Pisa, Italy (e-mail: guido.volpi@pi.infn.it).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TNS.2007.913476
Fig. 1. The FTK processor has access to the tracker data of level-1 selected
events flowing through the DAQ system. The FTK output is put into standard
DAQ buffer memories. In the DAQ buffer FTK data are merged with the Trigger
Read-out (RO) data and then sent to the read-out interface (RoI).
farm performing higher level trigger selections. The FTK looks
at the tracker data flowing to the memory buffers without in-
terfering with the operation of the DAQ system, reconstructs
high–quality tracks, and stores its compact output into an ad-
ditional memory buffer that can be easily accessed at high rate
by the Event Filter CPUs. This implementation scheme allows
a high input bandwidth for FTK with minimal interference with
the rest of the DAQ. It can even be added after the baseline has
been built, as an upgrade, provided the possibility of adding a
splitter at the output of the pipeline memory was already evicted.
A detailed technical description of the FTK can be found else-
where [1]. In this document we describe the results of a detailed
simulation of operation of a FTK within the ATLAS detector
and first results on its performance in tracking precision and
speed. This tracking performance can be exploited to great ad-
vantage by many high–level trigger algorithms; in this document
we discuss in detail the application to a specific example: trig-
gering of rare decays in the mode . This is a
challenging process to trigger on, requiring extensive, high-pre-
cision tracking at low-Pt.
0018-9499/$25.00 © 2008 IEEE