Edge Machine Learning for Cluster Counting in Next-Generation Drift Chambers

TL;DR

The paper tackles the challenge of achieving effective particle identification via cluster counting in future high-granularity drift chambers, where data-rate constraints motivate edge-front-end processing. It compares traditional derivative-based cluster counting with a compact dense neural network that directly regresses the number of primary ionization clusters per track, trained on Garfield++-based simulations of CEPC-inspired drift cells. Results show that ML approaches outperform $dN/dx$-based traditional methods in pion-kaon separation across momenta, with FPGA-synthesized implementations achieving latency in the tens of nanoseconds range, especially after pruning and 10-bit quantization ($O(10)$ ns). The work demonstrates a viable path toward edge deployment of ML-based cluster counting for real-time PID and data compression, while noting limitations related to toy detector design and the need for power and full-system throughput assessments in future studies.

Abstract

Drift chambers have long been central to collider tracking, but future machines like a Higgs factory motivate higher granularity and cluster counting for particle ID, posing new data processing challenges. Machine learning (ML) at the "edge", or in cell-level readout, can dramatically reduce the off-detector data rate for high-granularity drift chambers by performing cluster counting at-source. We present machine learning algorithms for cluster counting in real-time readout of future drift chambers. These algorithms outperform traditional derivative-based techniques based on achievable pion-kaon separation. When synthesized to FPGA resources, they can achieve latencies consistent with real-time operation in a future Higgs factory scenario, thus advancing both R&D for future collider detectors as well as hardware-based ML for edge applications in high energy physics.

Figures (3)

• Figure 1: Example waveform from Ref. Tian_2025, including primary and secondary electron truth labels, truncated to 500 samples.
• Figure 2: Performance of the baseline ML model for the task of predicting the truth number of clusters, comparing the predicted vs. truth cluster number distribution across the test set (left) and the correlation between the predicted and truth cluster number per test set event (right).
• Figure 3: Performance of cluster counting methods for the task of separating pions and kaons. The ML-based methods outperform both traditional derivative-based methods across the full track momentum range, and the application of compression techniques can achieve a model that has $\mathcal{O}$(10) ns latency.