Anne-Sophie Berthold (IKTP) & Nick Fritzsche (IKTP): Convolutional Neural Networks for processing of ATLAS Liquid Argon Calorimeter signals with FPGAs

Links:

• with ZIH login
• without ZIH login

Abstract:

Starting in 2027, the enhanced performance of the High-Luminosity LHC
will increase the number of particle collisions in the ATLAS detector
significantly. Since up to 200 pile-up events will emerge within one
bunch crossing, one important part of the so-called Phase-II upgrade
will be the processing of the Liquid-Argon Calorimeter signals. It has
been shown that the conventional signal processing, which applies an
optimal filtering algorithm, will loose its performance due to the
increase of overlapping signals and a trigger scheme with trigger accept
signals in each LHC bunch crossing. That is why more sophisticated
algorithms such as neural networks come into focus. This talk will deal
with the development and performance of convolutional neural networks
and their resource-optimized implementation on FPGAs.