Integration of a novel FPGA-based system for real-time pattern recognition in the LHCb experiment

The LHCb experiment at CERN pursues precision studies of heavy-flavour processes to investigate potential manifestations of physics beyond the StandardModel. The increased luminosities envisaged for future running necessitate novel real-time data-processing strategies capable of operating reliably under high detector occupancies and elevated data rates. This thesis reports on my contributions to an FPGA-based system designed to alleviate the computational load of the LHCb trigger-level reconstruction by providing low-level partial reconstruction directly at the detector readout stage. It first details the design, implementation, and operation of the Retina demonstrator, an instance of the “Artificial Retina” architecture for track reconstruction realised on a network of interconnected FPGAs. Optimisation of the associated hit-distribution network enabled a throughput of 19.6 MHz on simulated data samples, while a dedicated software interface allowed parasitic operation on real detector data, thereby validating the robustness and reliability of the system under realistic experimental conditions. Second, this work addresses the prototype interface of the future DoWnstream Tracker (DWT), which will be installed in Run 4, with the LHCb online DAQ system, demonstrating the feasibility of integrating an FPGA-based track-reconstruction architecture within the existing data-taking infrastructure. The DWT is specialised in enhancing the tracking performance for particles originating downstream of the vertex subdetector acceptance. Finally, thanks to an exploratory deployment of a GPU-based downstream tracking algorithm in Run 3, the physics impact of downstream reconstruction is assessed by deploying a dedicated selection line in the first-level trigger (HLT1) for D0→K0sK0s decays with K0s mesons reconstructed from downstream pions and by studying the corresponding data sample collected in 2025. The results indicate that those selections are indeed viable in the LHCb HLT1 trigger, and have the potential to significantly enhance the physics output of the experiment. To quantify this effect, the CP asymmetry ACP(K0sK0s) is adopted as a benchmark observable and the corresponding improvement in expected precision is estimated. The gain is particularly significant with the DWT being installed in Run 4, allowing the computational cost of trigger-level reconstruction to be reduced and enabling increased selection efficiency. Under these assumptions, the achievable statistical precision is expected to improve by a factor of two relative to that achievable with a selection strategy that does not exploit downstream tracks at the HLT1 trigger level. This would place the measurement in a regime of genuine discovery potential, allowing possible deviations from the Standard Model expectation to be probed with unprecedented sensitivity.