ParticleNet for Jet Tagging in Particle Physics on FPGA

Jet tagging is a crucial classification task in particle physics experiments. In recent years, the introduction of deep learning methods has significantly improved the accuracy of jet tagging classification tasks, with graph neural networks like ParticleNet demonstrating outstanding performance in this domain. Regarding model deployment, common hardware options include CPUs, GPUs, FPGAs, and ASICs. Presently, due to FPGA’s advantages such as low power consumption, low latency, and hardware programmability, it has become a cutting-edge research focus for accelerating AI deployment. Compared to CPUs, FPGAs can achieve better parallel operations, while compared to GPU platforms, using FPGAs can enhance computation efficiency and reduce power consumption. Hence, porting and optimizing ParticleNet on FPGAs can enable rapid and low-power execution of classification tasks in particle physics, thereby reducing economic costs and expediting data processing in particle physics research. This study will adopt the CPU+FPGA heterogeneous computing model, offloading compute-intensive tasks to the FPGA for faster execution and reduced algorithm latency.