Hadronic Tau Reconstruction in the CMS Phase-2 Level-1 Trigger using NNs with Calorimetric Information

The CMS Collaboration is planning to entirely replace its trigger and data acquisition systems to match the ambitious physics program of the High-Luminosity LHC. In the harsh environment of 200 pileup collisions per bunch crossing, efficiently collecting datasets in Phase-2 will be a challenging task. The already difficult implementation of an efficient trigger targeting hadronically decaying $\tau$ leptons ($\tau_h$) will become, in such conditions, an even more crucial and demanding task. The enhanced capabilities of the Phase-2 Level-1 (L1) triggering system can be exploited to design new complex machine learning trigger algorithms that are not yet implementable in the current Phase-1 L1 system. Moreover, the foreseen high-granularity endcap calorimeter (HGCAL), and the amount of information it will provide, play a key role in the design of novel $\tau_h$ triggering methods. This DPS Note presents the design, first firmware implementation, and performance evaluated by means of a firmware-based emulator for a newly designed algorithm for the reconstruction, identification, and calibration of $\tau_h$ candidates in the Phase-2 L1 trigger: the TauMinator. The algorithm is designed to exploit only calorimeteric information and to ensure a unified treatment of all trigger primitives from the barrel and endcap of the CMS detector. At its core, the TauMinator is composed of two neural networks, that can be implemented in the state-of-the-art FPGAs foreseen for the Phase-2 L1 trigger. This algorithm is a new development that was not presented in the Level-1 Phase-2 TDR. Reaching the typical physics thresholds targeted by the L1 Menu at HL-LHC ( $\approx50$ GeV) requires using the tracker information. Nevertheless, algorithms exploiting only calorimetric information have a major role in increasing the trigger efficiency at higher thresholds ( $\approx100$ GeV). This is the context in which the TauMinator has been developed.