Dr Jean-Baptiste Sauvan (Laboratoire Leprince-Ringuet (LLR), Ecole Polytechnique, France)

The high-luminosity phase of the LHC (HL-LHC) is planned to start in 2029, and should accumulate an integrated luminosity of 3000fb-1. Due to the high level of radiation and pile-up coming from this increased luminosity, many sub-systems of the CMS experiment will need to be upgraded or changed. Among others, the endcap calorimeters will be replaced by a highly-granular and radiation-hard detector, the HGCAL. This calorimeter will have around 6 million readout channels, which combined with the high level of pile-up expected at the HL-LHC, poses enormous challenges on its readout and on the reconstruction of the collision events.

This is especially true for the real-time reconstruction performed within the Level 1 (L1) trigger of CMS. This L1 trigger will be based on custom electronics boards containing FPGAs interconnected with fast optical links. It will process collision events at a rate of 40MHz and with a latency of at most 12.5 microseconds. The very first step is a strong reduction of the data in the frontend electronics followed by the reconstruction of local clusters of energy in three dimensions in the off-detector electronics. Higher-level physics objects are then built from these local clusters and used to make the trigger decision. In addition to standard compression and reconstruction algorithms, machine learning algorithms could be used in several places along this processing chain, from the compression of the data in the HGCAL frontend to the identification of the reconstructed objects.

In this seminar I will present the HGCAL detector and the algorithmic chain developed to reconstruct the HGCAL data at the L1 trigger. I will also show how machine learning algorithms could be integrated within this chain.