The Phase-2 Upgrade of the Hardware Trigger of CMS at the LHC

The CMS experiment uses a two-level triggering system consisting of the Level-1, instrumented by custom-design hardware boards, and the High Level Trigger, a streamlined version of the offline reconstruction software running on a computer farm. The upgrade of the collider to the ?High-Luminosity LHC? that will deliver a luminosity of $5-7 \cdot 10^{34} cm^{-2}s^{-1}$, corresponding to 140--200 pile-up events, also necessitates a substantial upgrade of the trigger system to make optimal use of the data. An important difference from the present system will be that after the upgrade, information from the silicon strip tracker will be available already for the Level-1 Trigger. This will allow CMS to use so-called ?particle flow? objects, i.e. signals seen not only in one subdetector but put together from all available subdetectors, resulting in much sharper cuts on trigger objects. Also, trigger rates will rise both at Level-1 (from 100 kHz to 750 kHz) and at the High-Level Trigger. At the same time, more sophisticated algorithms will be available at Level-1. To make this possible, the latency - the processing time available for arriving at the Level-1 trigger decision - will increase significantly. Machine-learning techniques such as Boosted Decision Trees have already started to be implemented in the trigger electronics and will occupy a more important place in the future. The use of High-Level Synthesis (HLS) tools will allow physicists to formulate trigger requirements in a language closer to that of data analysis. To avoid missing unexpected signatures from New Physics, studies are underway to employ anomaly detection using autoencoders.