Return to search

Energy reconstruction on the LHC ATLAS TileCal upgraded front end: feasibility study for a sROD co-processing unit

Dissertation presented in ful lment of the requirements for the degree of:
Master of Science in Physics
2016 / The Phase-II upgrade of the Large Hadron Collider at CERN in the early 2020s
will enable an order of magnitude increase in the data produced, unlocking the
potential for new physics discoveries. In the ATLAS detector, the upgraded Hadronic
Tile Calorimeter (TileCal) Phase-II front end read out system is currently being
prototyped to handle a total data throughput of 5.1 TB/s, from the current 20.4 GB/s.
The FPGA based Super Read Out Driver (sROD) prototype must perform an energy
reconstruction algorithm on 2.88 GB/s raw data, or 275 million events per second.
Due to the very high level of pro ciency required and time consuming nature of
FPGA rmware development, it may be more e ective to implement certain complex
energy reconstruction and monitoring algorithms on a general purpose, CPU based
sROD co-processor. Hence, the feasibility of a general purpose ARM System on Chip
based co-processing unit (PU) for the sROD is determined in this work.
A PCI-Express test platform was designed and constructed to link two ARM
Cortex-A9 SoCs via their PCI-Express Gen-2 x1 interfaces. Test results indicate that
the latency of the PCI-Express interface is su ciently low and the data throughput is
superior to that of alternative interfaces such as Ethernet, for use as an interconnect
for the SoCs to the sROD. CPU performance benchmarks were performed on ve ARM
development platforms to determine the CPU integer,
oating point and memory
system performance as well as energy e ciency. To complement the benchmarks,
Fast Fourier Transform and Optimal Filtering (OF) applications were also tested.
Based on the test results, in order for the PU to process 275 million events per
second with OF, within the 6 s timing budget of the ATLAS triggering system, a
cluster of three Tegra-K1, Cortex-A15 SoCs connected to the sROD via a Gen-2 x8
PCI-Express interface would be suitable. A high level design for the PU is proposed
which surpasses the requirements for the sROD co-processor and can also be used
in a general purpose, high data throughput system, with 80 Gb/s Ethernet and
15 GB/s PCI-Express throughput, using four X-Gene SoCs.

Identiferoai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:wits/oai:wiredspace.wits.ac.za:10539/20358
Date10 May 2016
CreatorsCox, Mitchell Arij
Source SetsSouth African National ETD Portal
LanguageEnglish
Detected LanguageEnglish
TypeThesis
Formatapplication/pdf

Page generated in 0.0019 seconds