Speaker
Description
SLAC’s LCLS-II delivers attosecond X-ray pulses at high repetition rates, targeting 1 MHz. Meeting this challenge requires hardware-optimized, low-latency pipelines for real-time, single-shot diagnostics. We present a heterogeneous data processing approach for the Multi-Resolution Cookiebox (MRCO) detector—an array of 16 electron time-of-flight spectrometers with tunable flight lenses and dedicated amplifiers for time, angle- and energy-resolved spectroscopy.* MRCO full data pipeline integrates analog-digital co-design with FPGA-accelerated algorithms** and edge-deployed machine learning to perform denoising, spectral feature extraction, and temporal reconstruction of the X-ray pulses from LCLS-II.*** FPGA-optimized peak finding algorithms** enable online feature extraction while component neural networks—including long short-term memory models and ResNets—are trained on synthetic data to recover attosecond temporal substructure of the X-ray pulses.*** By co-optimizing algorithm design with traditional and emerging hardware (e.g., Groq inference cards, FPGAs), we achieve high-throughput, low-latency inference suitable for shot tagging and feedback. These efforts represent a state-of-the-art step toward closing the gap to MHz operation and highlight the growing need for deeper algorithm–hardware co-design in attosecond XFEL science.
Footnotes
Walter, et al. J. Synchrotron Rad. 28.5, DOI 10.1107/S1600577521007700 (2021).
Hirschman, et al. SMC Conf., DOI 10.1007/978-3-031-23606-8_7 (2022).
**Hirschman, et al. arXiv preprint arXiv:2502.16141 (2025).
Funding Agency
Supported by US DOE Contract No. DE-AC02-76SF00515, DE-SC0022559 and DE-SC0022464; NSF under Contract No. 2231334; and US DOD under AFOSR FA9550-23-1-0409.
| I have read and accept the Privacy Policy Statement | Yes |
|---|---|
| Please consider my poster for contributed oral presentation | No |
| Would you like to submit this poster in student poster session on Sunday (August 10th) | No |
