Acceleration of Physics Algorithms on FPGA Platforms to perform Online Analysis within the CMS Phase-2 Level-1 Data Scouting System

The upcoming High-Luminosity Large Hadron Collider (HL-LHC) era will present unprecedented computational challenges for particle physics research, with instantaneous luminosities reaching up to 7.5 × 1034 cm–2 s–1 and pile-up of up to 200 simultaneous collisions per bunch crossing. The CMS Phase-2 upgrade introduces the Level-1 Data Scouting (L1DS) system, which proposes to analyze all reconstructed events at the full bunch crossing rate of 40 MHz without preselection, requiring revolutionary approaches to real-time data processing that go beyond traditional CPU-only server architectures.

This thesis investigates the feasibility of Field-Programmable Gate Array (FPGA) acceleration for par- ticle physics online analysis algorithms within the L1DS framework. Specifically, this work presents the design, implementation, and evaluation of FPGA-based realizations of the W → 3π and W → πγ decay analyses on the AMD Alveo U55C platform. The research investigates whether FPGAs can meaningfully complement and extend CPU-based systems to accommodate additional compu- tational functions and enhance processing capacity for real-time particle physics analysis while maintaining necessary computational precision and algorithmic flexibility.

The implementation shows the development of modular, scalable hardware functions for funda- mental computational primitives, including angular distance calculations, isolation computation, and combinatorial operations. It is demonstrated that through the implementation of iterative optimization cycles, the final FPGA implementation achieves competitive performance metrics. The results show that the FPGA implementation delivers the best total per-orbit runtime for bulk processing and nearly matches GPU performance levels for analysis kernel execution time. Perfor- mance evaluation reveals that the FPGA implementation consumes significantly less power than the GPU version while utilizing approximately 30% of available computational resources, highlighting substantial energy efficiency benefits. Additionally, a small library with ROOT-like user interface was developed to generate code for both software and hardware-accelerated analysis execution, enabling seamless comparison and validation between computational approaches.

This work provides practical insights into the viability of FPGA acceleration for next-generation trigger and particle physics data acquisition systems, demonstrating that specialized hardware can successfully enhance computational performance for high-energy physics applications while offering energy-efficient alternatives to traditional processing paradigms.