Intelligent experiments through real-time AI: Fast Data Processing and Autonomous Detector Control for sPHENIX and future EIC detectors

A multi-institutional collaboration involving LANL, MIT, FNAL, NJIT, ORNL, and GIT, supported by the DOE Office of Science Nuclear Physics AI/ML initiative, is advancing the use of artificial intelligence to address data processing challenges at RHIC and the future Electron-Ion Collider (EIC). The primary objective is to develop a real-time processing demonstrator for high-rate data streams from the sPHENIX tracking detectors, aimed at identifying rare heavy-flavor events in proton-proton (p+p) collisions. Our innovative approach combines a streaming readout architecture of key detectors with an intelligent control system, leveraging FPGA hardware to accelerate AI inference. This integration enhances the efficiency of collecting rare heavy-flavor events in high-rate (~ 1MHz) p+p collisions while optimizing the use of limited data acquisition bandwidth (~15 kHz). To achieve this, we employ Graph Neural Network (GNN)–based trigger algorithms trained on sPHENIX p+p simulation data, utilizing the hls4ml and FlowGNN toolkits to translate AI models into firmware. These models are deployed on FELIX-712 boards equipped with Xilinx Kintex Ultrascale FPGAs.

Beyond sPHENIX, our methodology is extensible to other applications requiring high-throughput data handling and real-time detector control, including future experiments at the EIC. This talk will highlight the AI-enabled heavy-flavor triggering system developed for sPHENIX and introduce DIS electron tagger algorithms for the EIC, demonstrating the transformative impact of AI and FPGA technologies in real-time data processing for high-energy nuclear and particle physics.

(LANL LA-UR-25-26719)