Development of a real-time FPGA-based CNN trigger operating on full-granularity calorimeter images

Abstract

We present a real-time field-programmable gate array (FPGA) implementation of a convolutional neural network (CNN) for event classification on full calorimeter images that retain crystal-level shower information. To meet the resource and latency constraints of trigger electronics, the network parameters and hidden activations were quantized to 8 bits with quantization-aware training, and a lossless space-to-depth reorganization rearranged the input image into fewer spatial positions, reducing the estimated CNN inference latency by more than an order of magnitude. The approach was evaluated with the J-PARC KOTO detector, recorded data, and data-acquisition (DAQ) constraints as a benchmark, targeting the online separation of the rare K⁰_L→π⁰e⁺e⁻ decay from its dominant K⁰_L→π⁺π⁻π⁰ background. As a proof of principle, the firmware was implemented and validated on an AMD Versal Premium VPK120 evaluation board. On 100,000 test events, inference on the board rejected (99.34±0.04)% of the recorded background sample enriched in K⁰_L→π⁺π⁻π⁰ decays at a nominal K⁰_L→π⁰e⁺e⁻ efficiency of 95%, with a measured CNN inference latency of 17.9 μs. Combining the measured image-assembly and CNN inference latencies with the estimated link-transfer and score-return times gives a projected end-to-end latency of 25.9 μs, within the adopted 32 μs DAQ latency budget. These results demonstrate the feasibility of using full-granularity calorimeter images for CNN-based event selection that achieves high background rejection while keeping the trigger latency within tens of microseconds.

Real-Time SiPM Pulse Deconvolution for a High-granularity Dual-readout Calorimeter with Neutral Networks
Spencer Allen, Matteo Cremonesi, Jordan Damgov +6
arXiv9 Sep 2026PreprintarXiv licenceFull textApplied physics & instrumentation Computational physicsarXiv
Development of a high-granularity, high-precision timing readout electronics system for large-area MRPC detectors
J. N. Tang, W. H. Wu, Y. Q. Tan +4
arXiv28 Aug 2026PreprintarXiv licenceFull textApplied physics & instrumentationarXivHAL
Using Graph Neural Networks for the segmentation of overlapping objects in high granularity calorimeters
Matthieu Melennec, Frédéric Magniette
HAL29 Sep 2026PreprintCC BY 4.0Full textHigh-energy & particle physicsHALarXiv
End-to-end Particle Reconstruction with Transformers in High-Granularity Calorimeters
Qiming Bao, Haotian Bi, Benedikt Maier +2
arXiv27 Oct 2026PreprintCC BY 4.0Full textApplied physics & instrumentation High-energy & particle physicsarXiv
Design and Simulation Study of the Hadronic Calorimeter for the EicC Zero Degree Calorimeter
Maiyu Wang, Yuan Li, Hao Wang +6
arXiv15 Sep 2026PreprintarXiv licenceFull textApplied physics & instrumentationarXiv

Cite

BibTeX RIS

BibTeX

@misc{wu2026development,
  title = {{Development of a real-time FPGA-based CNN trigger operating on full-granularity calorimeter images}},
  author = {Jia-Jen Wu and Yun-Tsung Lai and Yu-Chen Tung and Sin-Ting Chen and Yu-Sheng Liu},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.36720},
  eprint = {2610.36720},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.36720},
}

RIS

TY  - UNPB
TI  - Development of a real-time FPGA-based CNN trigger operating on full-granularity calorimeter images
AU  - Jia-Jen Wu
AU  - Yun-Tsung Lai
AU  - Yu-Chen Tung
AU  - Sin-Ting Chen
AU  - Yu-Sheng Liu
PY  - 2026
DA  - 2026/10/27
PB  - arXiv
DO  - 10.48550/arxiv.2610.36720
UR  - https://arxiv.org/abs/2610.36720
AB  - We present a real-time field-programmable gate array (FPGA) implementation of a convolutional neural network (CNN) for event classification on full calorimeter images that retain crystal-level shower information. To meet the resource and latency constraints of trigger electronics, the network parameters and hidden activations were quantized to 8 bits with quantization-aware training, and a lossless space-to-depth reorganization rearranged the input image into fewer spatial positions, reducing the estimated CNN inference latency by more than an order of magnitude. The approach was evaluated with the J-PARC KOTO detector, recorded data, and data-acquisition (DAQ) constraints as a benchmark, targeting the online separation of the rare K^0_L→π^0e^+e^− decay from its dominant K^0_L→π^+π^−π^0 background. As a proof of principle, the firmware was implemented and validated on an AMD Versal Premium VPK120 evaluation board. On 100,000 test events, inference on the board rejected (99.34±0.04)% of the recorded background sample enriched in K^0_L→π^+π^−π^0 decays at a nominal K^0_L→π^0e^+e^− efficiency of 95%, with a measured CNN inference latency of 17.9 μs. Combining the measured image-assembly and CNN inference latencies with the estimated link-transfer and score-return times gives a projected end-to-end latency of 25.9 μs, within the adopted 32 μs DAQ latency budget. These results demonstrate the feasibility of using full-granularity calorimeter images for CNN-based event selection that achieves high background rejection while keeping the trigger latency within tens of microseconds.
ER  -

Abstract

Related

Cite