Speaker
Description
Future high-energy physics detectors must process increasingly complex signals under stringent constraints on bandwidth, latency, power, and hardware resources. We present a hardware-aware machine-learning approach for general real-time waveform processing in detector readout systems, with applications to drift-chamber cluster counting and scintillation/Cherenkov waveform decomposition in homogeneous dual-readout calorimetry.
For drift-chamber readout, an ultra-compact streaming neural network is co-designed with FPGA constraints for real-time primary-cluster counting, improving particle-identification performance over conventional peak finding while using very limited hardware resources for on-detector application. For dual-readout calorimetry, the same hardware-aware model architecture is applied to implement the ML method for real-time scintillation/Cherenkov (S/C) waveform decomposition, achieving competitive performance with conventional template fitting while remaining effective at reduced sampling rates and compatible with low-latency on-detector FPGA deployment. Together, these applications demonstrate the broader applicability of the approach to real-time analysis of one-dimensional detector waveforms.
We further discuss preliminary studies using knowledge distillation to transfer capabilities from large pre-trained time-series models into compact edge models for real-time waveform analysis, together with open-source workflows that accelerate the path from model design to RTL implementation. These developments provide a practical and generalizable route toward intelligent real-time waveform processing for future detector readout systems.
This presentation includes results from the following papers and related recent studies:
L. Wu, Q. Liu, M. T. Lucchini, J. Gonski, M. Campajola, and S. Moneta, “Machine Learning Enables Real-Time Waveform Decomposition for Dual-Readout Calorimetry,” arXiv:2604.26090 (2026).
G. Ancone, Q. Liu, L. Wu, and J. Gonski, “Leveraging Industrial Foundation Models at the Edge of Particle Physics Detectors via Distillation Learning and Hardware Co-design,” arXiv:2609.23385 (2026).