Edge Machine Learning for Smart Detectors in Future Colliders
Poster, Bridging the Farm: AI for Science at SLAC and Stanford, Stanford, CA
In next-generation high-energy physics experiments, detectors face several daunting requirements: high data rates, radiation exposure, and strict constraints on latency and power. Machine learning (ML) has been explored for use in readout application specific integrated circuits (ASICs) to perform intelligent inference and data reduction at the source. We demonstrate low-latency, resource-efficient ML models, such as boosted decision trees and autoencoders, for front-end processing that reduce off-detector data rates and utilize latent space information for real-time sensor defect monitoring. We also present embedded field programmable gate arrays (eFPGAs) as a potential hardware technology that can provide reconfigurable digital logic to deploy ML at the edge in high energy collider experiments.