ALLEGRO Ecal monthly meeting

→ Europe/Zurich
42/R-001 (CERN)

42/R-001

CERN

10
Show room on map
Zoom Meeting ID
66211778054
Host
Nicolas Morange
Passcode
00982207
Useful links
Join via phone
Zoom URL
AI-generated minutes

 

1. Introduction and News

  • Organizational, Administrative & Outreach Updates:

    • DRDCalo & FCC PED: Renewal documents and the MoU for DRDCalo are ready for the Nov 12–13 DRDC meeting. A new, streamlined organizational structure for the FCC PED Detector Concepts group establishes dedicated contacts for ALLEGRO and Calorimetry.
    • ALLEGRO Paper, Workshops & Outreach: The ALLEGRO ECAL paper targets a first draft by the end of November 2026. The next workshop is set for March 15–17, 2027, at CERN, with a late-November meeting planned for mini-prototype cryogenics and 2027 DESY testbeams. Additionally, an official CERN press visit from Finnish journalists is coming up, where Juska, Mikhail, and Gerard will showcase an ECAL PCB in the cryobox and a live signal-injection demo alongside a new overview poster by Martina.
  • Hardware & Cryogenics:

    • Electronics: Section 5 board production is nearly complete, a revised summing board prototype has been ordered, and the CALOROC1c chip passed its first cold test at IJCLab on October 29 with Christoph and Emmy from OMEGA—successfully configuring, communicating, and taking initial data (full results to be presented in December).
    • Mechanics: Gerard Aliana Cervera has taken over mechanics coordination from Fernando. Mini-proto manufacturing is underway at CPPM (parts expected early December), custom press tooling for lead absorbers is designed, and discussions are underway with a supplier for pre-cut lead plates under an inert atmosphere. Cool-down tests on $60\times 60\text{ cm}$ sandwiches will compare $0.1\text{ mm}$ and $0.08\text{ mm}$ steel thicknesses. The upgraded cosmic ray cryostat (V0.2) held liquid nitrogen ($-196^\circ\text{C}$) for 9–10 hours (~8 hours for a submerged PCB); during Q&A, Juska and Gerard noted exterior ice buildup at a corner/lid joint that needs leak inspection, and Chris Rasmussen cautioned that switching to liquid argon will reduce hold time to roughly 3 hours due to density and latent heat differences.
  • Software & Simulation: Key4hep has transitioned to LCGCMAKE (which limits nightly builds to a 7-day rolling window) and Ubuntu 26.04. Surface manager initialization optimizations reduced DIGI-RECO initialization time from 330 s to 45 s. In Q&A, Brieuc clarified that Beam-Induced Background (BIB) samples are available at the GEN level (thousands of BX for IPC, ~30 BX for single-beam synchrotron radiation), and Giovanni noted that the newly merged conformal tracking still awaits a k4geo PR (pending a DD4hep tag) to extrapolate tracks to the calorimeter surface.

2. Update on the CHARMS250 ASIC

  • Current Prototype (CHARMS250V1):

    • Designed in 65 nm CMOS (using thick-oxide transistors to suppress gate leakage), this 16-channel cryogenic front-end ASIC operates from 77 K to 300 K with low power consumption (< 10 mW/channel).
    • Characterization at liquid nitrogen temperature (LNT) shows stable gain (~12.4 mV/fC at the 14 mV/fC setting), low equivalent noise charge ($\text{ENC} = 564\,e^-$ at $C_d = 150\text{ pF}$ and $1\,\mu\text{s}$ peaking time), and noise within 10% of the DUNE LArASIC. A multi-channel version is slated for fabrication in early 2027.
  • Extending Dynamic Range to 10 pC (V2A and V2B) & Discussion:

    • To handle up to 10 pC without saturating the first charge amplifier ($CA_1$), BNL is developing two architectures for early 2027 fabrication:

      • CHARMS250V2A: Uses dual parallel gain paths (high-gain for low noise, low-gain for high charge) with a comparator-driven switch indicator ($V_{sw}$), consuming ~8.2 mW/channel. Responding to Nicolas on mid-pulse saturation, Prashansa explained that thresholding after the first shaper stage is fast enough to switch in the low-gain path before the delayed signal exits the third shaper stage, preserving the full pulse.
      • CHARMS250V2B: Uses dynamic feedback capacitance switching ($C'_{f1}$) combined with a dynamic reset current ($I_{RQI}$) increase to handle kHz event rates, consuming ~6 mW/channel and achieving lower high-gain noise ($660\,e^-$ vs. $1110\,e^-$ at LNT, $C_{in}=200\text{ pF}$) because it avoids the input capacitive charge splitting used in V2A.
    • Additional Q&A Points: Prashansa and Chris confirmed the readout FPGA will sit in the cold (reusing the DUNE prototype FPGA). Nicolas also flagged TSMC's eventual phase-out of 65 nm CMOS; BNL expects interposer demand to extend 65 nm availability somewhat, is considering a future move to 28 nm, and agreed to join DRD7 electronics discussions.

3. Impact of a Thick Solenoid on HitPF Performance

  • Motivation & Setup: Placing the solenoid between the ECAL and HCAL (à la ALEPH) avoids dead material in front of the ECAL, but introduces a ~400–500 mm spatial gap and ~$1.3\text{–}1.5\,X_0$ of dead material before the HCAL. The study evaluates the impact of a 50 cm thick solenoid (compared to a 7 cm baseline—and noting ALEPH achieved 45 cm 30–40 years ago) on HitPF (Mask3D clustering + GATr energy regression/PID) using $Z \to uds$ events.
  • Key Findings & Discussion:

    • Photons & Charged Hadrons: Photon clustering efficiency, energy response, and resolution remain virtually unchanged between thin and thick solenoids. However, in Q&A, Nicolas and Brieuc noted that even the baseline 10 GeV photon energy resolution is ~2× worse than standalone BDT calibration, likely due to Mask3D over-segmenting showers or missing peripheral cells at high energy.
    • Neutral Hadrons & Fakes: Neutral hadrons suffer lower energy response and degraded energy resolution at higher energies due to dead-material energy loss, alongside up to a ~2.5× increase in fake neutral hadron clusters. Nicolas and Giovanni pointed out that baseline neutral hadron efficiency and resolution are surprisingly poor—likely driven by PID misidentification (e.g., classifying neutral hadrons as photons at low energy) and low high-energy training statistics—and recommended evaluating HitPF on single-particle neutron guns, training on higher-energy ($t\bar{t}$) samples, and comparing against CLD benchmarks.
    • Event-Level Impact: Reconstructed $Z$ invariant mass resolution degrades slightly from $\sigma/\mu = 3.3\%$ ($\text{RMS}_{90} = 3.0\%$) to $\sigma/\mu = 3.7\%$ ($\text{RMS}_{90} = 3.2\%$) with the thick solenoid—still outperforming current PandoraPFA benchmarks on ALLEGRO ($\sigma/\mu = 5.5\%$). Minimizing solenoid thickness remains a priority, and Jacopo's next steps include adding realistic carbon-fiber cryostat walls and support structures.

4. ALLEGRO ECAL Barrel Angular Reconstruction

  • Barycentre Reconstruction ($\theta$ and $\phi$):

    • Bug Fix & Clustering Improvement: Fixed a half-fine-cell offset in cell center calculations that shifted the polar angle ($\theta$) S-curve away from $(0,0)$, and confirmed that linear energy weights avoid the sawtooth S-curve artifacts caused by log-energy weights in early layers. To fix low-energy shower truncation and arbitrary sorting in the $5\times 11$ sliding window, an iterative $\theta$-only re-centering algorithm was implemented (re-centering in $\phi$ was disabled because it splits converted $e^+e^-$ photon showers separated by the magnetic field).
    • Neural Network Regression: Adding 30 per-layer inputs (barycentres and energy fractions across 10 layers) to the MLP and switching to more conservative inter-quantile ranges ($q_{68}$) improved $\theta$ resolution by ~10–20% over previous paper plots, yielding $\sigma(\Delta\theta) = 1.301/\sqrt{E} \oplus 0.047\text{ mrad}$ for LAr/Pb and $1.179/\sqrt{E} \oplus 0.049\text{ mrad}$ for LKr/W with bias $< 0.01\text{ mrad}$. Applying a similar NN to $\phi$ corrects the energy-dependent bias caused by the $50^\circ$ electrode inclination, improving azimuthal resolution by 25–30% and reducing bias by an order of magnitude.
  • Non-Projective Direction Reconstruction & Discussion:

    • Moving beyond standard line fits of layer barycentres (which suffer from large $\phi$ biases due to electrode inclination), an 87-input MLP and a cell-level Transformer were tested to reconstruct photon trajectories without an interaction-point constraint. Both ML approaches substantially improve angular and impact-parameter ($z_0, d_0$) resolutions and suppress biases.
    • In the discussion, Mattia noted that correcting per-layer S-curves (now available via Giovanni's open PR) helps break projectivity bias in classical line fits, and mentioned he is already testing shifted or additional strip layers to present at the next meeting. Nikiforos added that because ALLEGRO reconstructs true 3D photon direction (including $\phi$, unlike ATLAS), future tests on strongly non-pointing photons should highlight ALLEGRO's distinct advantage. Giovanni is moving his code to the central ALLEGRO repository for Mattia, Nikiforos, and Filomena to continue studies with electronic noise and topoclusters.
There are minutes attached to this event. Show them.