Performance of an HRPPD in Tesla-scale magnetic fields

This paper tests the performance of a representative EIC-HRPPD (High-Rate Picosecond Photodetector) tile, a microchannel-plate (MCP) based photosensor, in Tesla-scale magnetic fields up to 1.8 T, with the goal of quantifying its performance for use in the ePIC experiment at the future Electron Ion Collider (EIC) at Brookhaven National Laboratory. The EIC-HRPPD is designed to operate in a 1.5 T magnetic field with inclination angles up to ±15° relative to the normal of the MCP surface. The study explores optimal operating parameters, mainly voltages applied across the gaps and MCPs, for single photon detection at various inclination angles in a uniform magnetic field.

The paper finds that magnetic field-induced distortions of collected charge can be compensated by tuning operating parameters. Specifically, increasing the MCP bias voltage by only a few tens of volts is sufficient to recover the nominal operating gain (>10 6) in magnetic fields up to 1.8 T for most angles. At the nominal HV setting of 200-675-200-675-200, the HRPPD achieves a gain exceeding 10 6 in magnetic fields up to 1.8 T over most inclination angles, including those envisioned for the pfRICH at ePIC.

Key results include:

• Gain spectra were measured under six magnetic field configurations (0 T, 0.37 T, 0.75 T, 1.14 T, 1.51 T, 1.81 T), with the gain spectrum in the absence of a magnetic field horizontally scaled by a factor of 0.3 for visual comparison. In the presence of a magnetic field, the gain spectra exhibit a non-Gaussian shape.

• A pronounced gain minimum is observed near 13° and a minor minimum near −13° when scanning inclination angle about the X-axis, corresponding to alignment of the magnetic field with the capillaries of the first and second MCPs, respectively. The gain dip associated with alignment with the first MCP capillaries is significantly more pronounced than that with the second MCP. No significant gain dip is observed when scanning about the Y-axis, which is perpendicular to the MCP capillary direction.

• The afterpulsing rate at the nominal operating voltage of 675 V was measured at (1.34 ± 0.15)% in a 1.3 T magnetic field. The afterpulsing rate exhibits markedly higher sensitivity to the bias voltage of the first/top MCP compared to the second/bottom, indicating the first MCP is the dominant contributor to afterpulsing. A distinct peak in the time separation between primary signal and afterpulse occurs at approximately 6.4 ns when the first MCP is operated at a higher bias voltage, consistent with the flight time of H+ ions originating from the bottom of the first MCP.

• The dark-count rate decreases monotonically with increasing magnetic-field strength and increases monotonically with increasing HV bias. A cross-check using the CAEN V1742 digitizer found a dark-count rate of 28 ± 9 Hz/cm2 in a 1.3 T magnetic field with 675 V applied to each MCP, consistent with the oscilloscope result of 24 Hz/cm2.

• The timing resolution was measured using both digitized waveforms (CAEN V1742) and a high-precision oscilloscope (Tektronix MSO66B). The timing resolution includes non-negligible timing uncertainty from the trigger signal (12 ps) and the PiLas laser pulse width (35 ps). The intrinsic timing resolution of the HRPPD is expected to be better than 40 ps under the experimental conditions of ePIC, excluding broadening effects from laser and trigger timing jitter. The timing resolution evolves with magnetic field strength, showing a degradation from 0 to 0.37 T, improvement above 0.37 T, and deterioration beyond approximately 0.7 T, attributed to competing effects of signal amplitude reduction and helical motion suppressing transit-time fluctuations.

• A signal rate scan showed a reduction in gain at higher signal rates, indicating a saturation effect on the order of several percent due to microchannel dead time.

• The right-hand tail of gain distributions shrinks in the presence of a magnetic field, indicating the onset of signal saturation, attributed to the influence of the magnetic field on photoelectron trajectories and the reduction in the number of pores participating in amplification in the second MCP, funneling excessive electrons into the same pores.

The paper concludes that the EIC-HRPPD demonstrates reliable performance in Tesla-scale magnetic fields, achieving a gain exceeding 10 6 in magnetic fields up to 1.8 T over most inclination angles, with a low afterpulsing rate of 1.34% and a dark-count rate of 24 Hz/cm2. The results also suggest that applying asymmetric bias voltages across the two MCPs will effectively mitigate the observed saturation effects, at the cost of an increase in afterpulsing.

Improvements to AI Systems:

1. Physics-Aware Sensor Calibration Models

• Train AI models to predict optimal MCP bias voltage adjustments (e.g., +tens of volts) as a function of magnetic field strength and inclination angle, using the paper’s gain vs. field/angle data.

• The improved AI can auto-compensate for magnetic-field-induced gain distortions in real time, maintaining gain >106 across 0–1.8 T without manual tuning.

2. Afterpulsing Source Identification and Mitigation

• Use the measured afterpulsing rate sensitivity to first vs. second MCP bias to build a predictive model that separates afterpulse contributions (e.g., H+ ion flight time 6.4 ns) from primary signals.

• The improved AI can dynamically adjust asymmetric MCP voltages to minimize afterpulsing while avoiding saturation, based on real-time pulse-shape analysis.

3. Magnetic-Field-Dependent Timing Resolution Correction

• Leverage the observed non-monotonic timing resolution (degradation at 0.37 T, improvement to 0.7 T, then deterioration) to train a regression model that predicts intrinsic timing jitter from field strength and signal amplitude.

• The improved AI can subtract magnetic-field-induced timing broadening from detector outputs, achieving <40 ps intrinsic resolution under ePIC conditions.

4. Saturation and Dead-Time Compensation

• Use the gain reduction vs. signal rate data (several percent at high rates) to train a neural network that estimates microchannel dead-time occupancy and corrects for gain droop.

• The improved AI can linearize detector response at high rates, enabling accurate single-photon counting in high-luminosity environments.

5. Gain Spectrum Shape Analysis

• Model the non-Gaussian gain spectra (right-hand tail shrinkage) in magnetic fields to infer the number of active MCP pores and electron-funneling effects.

• The improved AI can detect early saturation onset and flag when asymmetric bias voltages are needed, based on real-time spectral shape deviations.

6. Automated Angle-Scan Optimization

• Use the pronounced gain minima near ±13° (MCP capillary alignment) to train a reinforcement-learning agent that actively steers the detector orientation or adjusts voltages to avoid these dips.

• The improved AI can autonomously maintain stable gain during detector rotation in magnetic fields, critical for pfRICH operation.

7. Dark-Count Rate Prediction and Filtering

• Correlate dark-count rate with magnetic field strength and HV bias (monotonic decrease with field, increase with bias) to build a predictive model for background noise.

• The improved AI can subtract expected dark counts from signal data in real time, improving signal-to-noise ratio in low-light measurements.

8. Multi-Parameter Co-Optimization

• Integrate all measured trade-offs (gain, afterpulsing, dark count, timing, saturation) into a multi-objective optimization framework.

• The improved AI can recommend the best operating point (voltages, angles, field) for any given experiment, balancing performance metrics automatically—e.g., minimizing afterpulsing while keeping gain >106 and timing <40 ps.