Speaker
Description
Multiplexed readout of large cryogenic detector arrays is essential for next-generation particle-physics experiments such as direct neutrino-mass measurements and dark-matter searches. In these systems, online triggering is a key component of the readout chain, enabling real-time pulse detection and selective event storage under low latency and resource utilization constraints. We present a modular FPGA-based event-detection framework for RFSoC-based cryogenic detector readout. The framework integrates interchangeable trigger filters, including classical online filters, optimum-filter-based triggering, and a quantized machine-learning-based filter. Using low-amplitude pulse streams, we validate the implemented filters in terms of trigger efficiency, false-trigger rate, latency, throughput, and FPGA resource usage.
Summary (500 words)
Cryogenic detectors provide the energy resolution required for rare-event searches and precision neutrino-mass measurements. Scaling modern frequency-multiplexed Kinetic Inductance Detectors (KIDs), Transition Edge Sensors (TESs), or Metallic
Magnetic Calorimeters (MMCs) based systems to thousands of channels requires online data reduction directly in the room-temperature electronics. At very low signal amplitudes, the trigger stage becomes a limiting factor: thresholds must be low enough to preserve weak physical events, but high enough to suppress noise and background induced triggers.
Traditional online FPGA-based trigger systems usually rely on simple filtering approaches, such as differentiation or moving-average filtering, due to their low resource cost they can be easily implemented within the programmable logic. However, they provide limited discrimination against non-ideal pulse shapes and events induced by secondary particles. More selective filters, such as optimum-filter or machine-learning-based triggers, can improve event selection, but must comply with the latency, throughput, and resource constraints of the FPGA readout system.
In this work, we present the development and extension of a modular FPGA-based event detection module for RFSoC-based cryogenic detector readout. This module is designed around a common streaming interface, allowing different trigger filters to be exchanged and evaluated under the same firmware and data conditions. The implemented filtering options include digital differentiation, moving-average filtering, low-order IIR filtering, an optimum-filter-based trigger, and a quantized machine-learning-based filter. The optimum-filter implementation provides a classical reference for low-threshold triggering and selective pulse discrimination, while the machine-learning approach investigates whether quantized data-driven filters can improve the performance of the triggering system while still being a viable option for deploying in resource constrained scenarios.
The filters are validated using low-amplitude pulse streams with configurable pulse parameters. The evaluation focuses on metrics relevant for deployment in real-time readout electronics: trigger efficiency, false-trigger rate, latency, and FPGA resource usage. This comparison provides a hardware-level benchmark of classical, optimum-filter, and machine-learning trigger strategies within a common event-detection framework. The resulting system establishes a flexible basis for future online trigger development in multiplexed cryogenic detector experiments.