Workshops Archive

This Data Will Self-Destruct in One Microsecond: Live Data Reduction and Steering at LCLS-II


Workshop: ISAV26: In Situ AI, Analysis and Visualization

Authors: Jana Thayer (SLAC National Accelerator Laboratory)

Abstract: The LCLS-II X-ray free electron laser at SLAC fires at repetition rates up to 1 MHz, generating raw data rates that start at 280 GB/s and will approach 1,120 GB/s by the end of the decade. At this scale, the traditional “write everything, analyze later” approach to science breaks down. Most raw data must be reduced, characterized, or discarded within moments of acquisition. At LCLS-II in situ processing is built into the architecture. This talk presents the overall design of the LCLS-II Data System, from detector readout through a distributed data reduction pipeline to fast feedback describing how LCLS measures, analyzes, and decides to steer experiments on the timescales of seconds, minutes, or hours. It also covers how the ILLUMINE project extends that pipeline with ML models that analyze high-rate data streams and steer experiments in near real time. I will discuss where AI/ML delivers today including event vetoing, rapid feature extraction and digital-twin aided reconstruction, and where it’s a work in progress. There are engineering realities that need to be reckoned with: latency budgets measured in microseconds, model deployment on heterogeneous edge hardware, coupling to remote HPC facilities, and earning scientists’ trust in decisions that are made before a human ever sees the data.


Back to ISAV26: In Situ AI, Analysis and Visualization Archive Listing

Back to Full Workshop Archive Listing