- The paper demonstrates a 5,376-channel VMM3a/SRS readout for the HYDRA time projection chamber, adapting front-end electronics to fit inside the GLAD magnet while supporting high-rate pion tracking.
- Custom thinned Hybrids, adapter boards, external power distribution, and heimtime timestamp injection preserve approximately 1,000-electron baseline noise and enable synchronization with the R³B timing system.
- Validation shows simulated crosstalk below 0.1% of the driving signal and clean TPC–scintillator timing correlations, while in-magnet cooling and full in-beam performance remain to be demonstrated.
Overview
This paper describes the implementation of a VMM3a-based Scalable Readout System (SRS) for the HYDRA (HYpernuclei Decay at R3B Apparatus) time projection chamber, a dedicated pion tracker for hypernuclear studies at the R³B experiment of GSI/FAIR. The central engineering problem is twofold: reading out more than 5,000 pad channels with high-rate capability and precise timing, while fitting the front-end electronics inside the constrained bore of the GLAD superconducting dipole magnet. The authors address this through physical modification of the standard VMM3a Hybrids, custom adapter and power-distribution PCBs, and a timestamp-injection scheme (the heimtime protocol) that circumvents the VMM3a/SRS lack of native White Rabbit timestamp receiver support. Validation consists of SPICE-based crosstalk analysis, S-curve noise measurements, and cosmic-ray tests demonstrating time correlation with the scintillator wall trigger.
Physics context and requirements
HYDRA is a small-size TPC with a 300 mm drift length, a GEM + Micromegas two-stage amplification stack, and a pad plane of 44 × 128 square pads of 1.9 × 1.9 mm², yielding 5,376 signal pads after grounding the two outer rows. It operates inside the GLAD magnet (±80 mrad acceptance, field integral up to 4.8 T·m, ~2.2 T central field) and is paired with a 16-bar plastic scintillator wall providing timing and trigger.
The physics requirement drives the readout specification: hypernuclei are produced with cross sections of order 1 μb and reconstructed through their weak two-body decay into π⁻ and a residual fragment. A pion momentum resolution of 1% (σ) is required to reach an invariant-mass resolution of ~1.5 MeV/c², which in turn demands a 300 μm spatial resolution. Beam intensities of O(10⁶) particles/s imply roughly 200 kHz of charged particles crossing the drift volume, so the front-end must sustain continuous operation with minimal dead time. The VMM3a ASIC, developed for the ATLAS muon upgrade, provides O(1 ns) time resolution and per-channel rates up to 4 MHz in continuous readout mode, making it well matched to these requirements.
Front-end adaptation
The baseline readout unit is the VMM Hybrid, a 5 × 8 cm² PCB hosting two 64-channel VMM3a ASICs with a Spartan-7 FPGA, connected via HDMI to the DVMM adapter and then to the FEC card (Virtex-6 FPGA), which performs zero suppression and Gigabit Ethernet transmission. One FEC handles up to eight Hybrids (1024 channels), with a Clock and Trigger Fanout module distributing a common 44 MHz base clock.
Three custom developments adapt this system to HYDRA:
- Hybrid modification: the standard Hybrid (~15 mm thick) exceeded the space budget. Removal of heatsinks, their connectors, and the power connector, replaced by 2.54 mm pin headers, reduced the thickness to 8.5 mm — nearly a 50% reduction. This necessitates an alternative cooling solution; fans were used for the measurements presented, and a dedicated cooling system for in-magnet operation is still under development.
- Pad-plane adapter module: an eight-layer HDI PCB with 125 μm controlled-impedance traces, one per TPC section, routing 672 pads to six Hybrids (112 signal channels per Hybrid, 16 spares). Because HDMI power delivery is infeasible over the several-meter cables required (the SRS crate sits outside the magnet), the boards are powered externally via Molex Micro-Fit 3.0 connectors with local 100 μF tantalum polymer decoupling on the 1.8 V and 3.3 V rails. Assembly of the 2-mm-pitch THT pin sockets required a non-standard reflow procedure using low-melting-point bismuth–tin solder with mechanical weighting, reflecting the absence of commercial BGA-style connectors at this pitch.
- Card adapter and flex add-on: an eight-layer FR4 PCB with via-stitched coplanar ground planes for crosstalk suppression, a 1-Wire temperature sensor (TMP104) per ASIC, over-current fuses, and a two-layer flex PCB that actively buffers the two synchronization clocks into eight outputs, AC-coupled into spare ASIC channels via 500 fF capacitors.
Synchronization with the R³B DAQ
The R³B timing architecture distributes timestamps via the White Rabbit protocol from a master timekeeper, but the VMM3a/SRS has no native WR timestamp receiver. The authors adopt the heimtime protocol: a periodic timing signal generated by the VULOM module is recorded locally in dedicated VMM3a channels, and the global timescale is decoded during offline processing. Trigger signals are distributed alongside, and by recording both leading and trailing edges of the trigger, cross-detector synchronization can be verified. The signal distribution tree fans out from 16 signal pairs at the distribution board, through 1:3 clock buffers on the pad-plane adapters, to 1:4 buffers on each card adapter feeding the two ASICs per Hybrid, with one redundant output per signal. The scheme is scalable and fully integrated into the adapter architecture.
Validation measurements
Crosstalk: LTspice transient simulations of the card-adapter traces (0.1 mm width, ~60 mm length, 350 μm spacing, via-stitched ground pour), using lumped-element Pi-network models with extracted per-unit-length parasitics, show a NEXT peak of 1.3 mV and FEXT below 0.2 mV under a 3.3 V, 3 ns edge-rate aggressor — induced noise below 0.1% of the driving signal. This confirms that the 3W design rule plus ground shielding adequately suppresses both capacitive and inductive coupling.
Electronics noise: using the S-curve method with the VMM internal pulser, the ENC was measured at ~1000 electrons per channel without the detector connected — consistent with prior measurements on unmodified, adapter-free Hybrids, indicating the modifications and adapter boards do not degrade baseline noise. With TPC high voltage applied, the ENC rises to ~5000 electrons, a factor-of-five increase attributable to the detector itself rather than the readout chain. The implication is that the modified architecture preserves the intrinsic VMM3a noise floor, and further noise reduction must come from the detector side.
Cosmic-ray test: with 50% of the readout chain installed (24 Hybrids, 4 pad-plane adapters), gain 3 mV/fC, peaking time 200 ns, and an Ar–CO₂ (90–10) mixture at O(10⁴) gain, muon tracks were reconstructed in 3D using pad-plane positions and drift times referenced to the scintillator wall start time. The correlation plot of trigger signals recorded independently by the TPC and the scintillator wall shows a clean diagonal with no off-diagonal events, validating the heimtime-based synchronization.
Limitations and open questions
Several caveats are stated or evident. The cooling solution for operation inside the GLAD magnet is not yet finalized; the reported results rely on fans, and thermal management of 48 Hybrids (~4 W each, ~190 W total) in the magnet bore remains an open engineering task. The cosmic-ray validation used only half of the readout chain, so full-system performance with all 48 Hybrids is not yet demonstrated. The ENC of ~5000 electrons with high voltage applied, while attributed to the detector, has not been decomposed further, and its impact on the target 300 μm spatial resolution is not quantified in this work. The paper also notes that in-beam validation of rate capability and position resolution is the subject of separate, forthcoming work. Finally, the crosstalk assessment combines SPICE simulation with "direct measurements" but the paper does not present independent measured crosstalk data beyond this statement.
Conclusion
The paper demonstrates a complete, validated VMM3a/SRS readout solution for the 5,376-channel HYDRA TPC, achieved through Hybrid thinning, custom adapter and power-distribution hardware, and a scalable heimtime timestamp-injection scheme that integrates the CERN-developed SRS into the GSI/FAIR White Rabbit timing infrastructure. The key quantitative results — baseline ENC of ~1000 electrons unchanged by the modified electronics, sub-millivolt simulated crosstalk, and clean TPC–scintillator time correlation in cosmic-ray data — support the conclusion that the system meets the requirements for hypernuclear physics measurements at R³B. Remaining work centers on in-magnet cooling and full-chain, in-beam commissioning.