Speaker
Description
Summary
The ATLAS Level-1 Calorimeter Trigger system reduces the LHC bunch-crossing rate of
40 MHz down to a rate of 75 kHz. About 300 Gbytes/s of input data are processed and
events are selected within a fixed time of 2s. The algorithms of selection are
FPGA-based to add flexibility to the system. Data are digitised and pipelined by
using three subsystems, namely the Preprocessor, electron/photon and tau/hadron
Cluster Processor (CP), and Jet/Energy-sum Processor (JEP). The CP and JEP receive
their digitised calorimeter trigger-tower data from the PreProcessor, and provide
trigger multiplicity information to the Central Trigger Processor via Common Merger
Modules (CMM). Using Readout Driver (ROD) modules, the CP and JEP will also provide
Region-of-Interest (RoI) information for the Level-2 Trigger, and intermediate
results to the data acquisition (DAQ) system for monitoring and diagnostic purposes.
The CP and JEP system receive their data through backplane connections. A custom
backplane has been designed in order to cope with the high connectivity demands of
the system and to route data between boards in order to accommodate the trigger
decision algorithms. The backplane also needs an adaptor for a VME CPU board and an
additional module to broadcast the 40 MHz TTC encoded information to the modules in
the crate.
The CPM and JEM have common design features. Both received digitised data in LVDS
format at a rate of 480 MBaud. Single-ended serialised data between boards across the
backplane is processed up to a speed of 160 MHz in case of the CPM. The CMM gathers
information through the backplane, but also between crates, and transfers the trigger
information to the overall Level-1 Central Trigger Processor.
All Calorimeter Trigger modules transmit their readout data using G-link
transceivers. These are connected to RODs, capable of processing up to 18 modules
individually.
All trigger modules, with the exception of the VME CPU, are custom designs. Although
every module had been developed and tested and seemed ready for production, it was
important to validate the final versions of the different modules in harsh
conditions. The production was therefore divided in two phases: a first batch was
manufactured with an adequate number of pre-production boards to perform a full-crate
test, and only after seeing the results would the remainder of the modules be
produced. Up to 14 CPMs and 16 JEMs have been built, representing nearly one third of
the final digital part of the trigger system. The cables delivering the digitised
signals were used, together with the final custom backplane. The source of data was
provided by specially designed boards that emulate the total input of one CPM or JEM,
corresponding to around 70 Gbytes/s of transferred data when the crate was fully
configured. All readout information was fed to a ROD.
Previous tests had been run using the ATLAS online software. It was therefore easily
expandable to a system of such a scale by extending the database. With a full crate
in operation, bit-error rate measurements or parity checks were performed over
long-term runs. A simulation framework was already available, and individual readout
data were compared against the simulation to verify the correct behaviour of the
different algorithms. Domains of validity of the data were measured by shifting the
40 MHz clock over its period. To detect any timing misalignment, Level-1 pre-flagged
events were sent to a ROD. Trigger rates up to 100 kHz were investigated. Stressful
data patterns were created in order to test the reliability of the system with an
ATLAS occupancy rate greater than 10 %. The system was also monitored using the PVSS
tools to check for current consumption and temperature hot spots inside the crate.
The results of the tests were successful and the ATLAS level-1 calorimeter system is
now proceeding to full production.
