Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 41: Trigger+DAQ 1

T 41.5: Vortrag

Dienstag, 5. März 2024, 17:00–17:15, Geb. 30.23: 3/1

Convolutional Neural Networks on FPGAs for Processing of ATLAS Liquid Argon Calorimeter Signals — Anne-Sophie Berthold, •Anna Franke, Nick Fritzsche, Markus Helbig, Rainer Hentges, Arno Straessner, and Johann Christoph Voigt — IKTP, TU Dresden

During the Phase-II upgrade of the ATLAS Liquid Argon Calorimeter, over 500 high-performance FPGAs will be installed to allow for the energy reconstruction of all 182468 detector cells at the LHC bunch crossing frequency of 40 MHz.

We trained 1-dimensional convolutional neural networks (CNNs) to improve the energy reconstruction under high-luminosity conditions with respect to the currently used Optimal Filter. In particular, the performance for overlapping pulses is demonstrated for 6 representative detector cells. The network architecture has been optimized with a hyperparameter search, where the network size is constrained to 100 parameters to be able to fit onto the FPGA.

The inference code of these networks has been implemented in VHDL targeting an Intel Agilex FPGA. This firmware can run at 480 MHz and applies 12-fold time-division multiplexing to reduce the resource requirements. This allows the design to process the readout of up to 384 detector cells per FPGA, while meeting the latency constraints of the ATLAS trigger. Quantization aware training using QKeras is used to adapt the CNNs to 18 bit fixed point numbers. To better evaluate the physics performance, the networks are being integrated into the ATLAS ATHENA detector simulation.

Keywords: LAr; CNN; detector; calorimeter; readout