Göttingen 2025 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 77: Data, AI, Computing, Electronics VIII (Fast ML, Triggers)
T 77.8: Talk
Thursday, April 3, 2025, 18:00–18:15, VG 2.102
Performance of a Quantized Neural Network on an FPGA for Next Generation Radio Array DAQ Systems — •Adam Rifaie for the IceCube-Gen2 collaboration — Bergische Universität Wuppertal, Wuppertal, Deutschland
The IceCube neutrino observatory is a cubic kilometer neutrino detector built into the Antarctic ice at the geographical Southpole. As a result of its success, the next-generation detector for IceCube, IceCube-Gen2, is currently being planned. This will extend the optical array to approximately 10 cubic kilometers and will include a ∼ 500 km2 radio array, sensitive to Ultra High Energy neutrinos. The state-of-the-art, with respect to phased radio arrays, is the Radio Neutrino Observatory Greenland (RNO-G), where currently 7 of the planned 35 stations have been deployed. These stations enable hardware testing and optimization for the DAQ system of RNO-G. A novel idea for a DAQ system would consist of an FPGA with a trained and Quantized Neural Network implemented. The neural network will read the datastream and discriminate between background and signal in real-time. This will improve the effective volume of the detector by a factor of 3 compared to a standard threshold trigger at certain energies. The performance and comparison of a quantized neural network with a regular neural network will be discussed, followed by the next steps to an all-digital DAQ system for Radio arrays.
Keywords: IceCube-Gen2; RNO-G; Ultra High-Energy Neutrinos; Quantized Neural Networks; Deep Learning