Skip to content

v1.5.0 - Validation Complete

Choose a tag to compare

View all tags
@MarkWSavage MarkWSavage released this 19 Jul 12:18
v1.5.0
13cc733

Closes out the validation pass across all three of PANDA's core responsibilities -- geometry, particle generation/transport, and energy/charge scoring -- against real experimental data, other simulation tools, and PANDA's own robustness across device scale. See Documentation/PANDA_VALIDATION_SUMMARY.md for the full account.

Validation

  • Hitachi HM68512 proton-SEU cross-validation against real 1994 NASA GSFC accelerator data: agreement within ~2x at Qcrit ~50fC, 20-27 um³ sensitive volume, across three independent aspect ratios.
  • McNulty et al. 1989 / CUPID comparison improved by adding a surrounding silicon volume and secondary-neutron biasing (log-RMSE ~1.1 → ~0.91 decades); remaining high-energy residual gap recorded as open/understood-cause.
  • Fixed a comparison-methodology bug in compare_creme_panda.py: it was comparing CREME-MC's ideal-100%-collection reference against PANDA's collected-charge (not deposited-charge) cross section. Corrected, agreement is good (log-RMSE ~0.29 decades, full-range coverage) -- the previously-reported "intermediate-charge hump" does not survive the corrected comparison and is no longer attributed to differing nuclear reaction model families.

Geometry/physics robustness

  • Region-scoped production cuts auto-scaled to device geometry, fixing a real deep-submicron charge-sharing fidelity gap (up to ~32% shift at the high-charge tail for a 50 um device).
  • Fixed a Geant4 navigator boundary artifact producing non-physical LET spikes at short step lengths.
  • Confirmed silicon recoil LET ceiling (~14.4-14.8 MeV·cm²/mg) holds up under 10x more statistics; confirmed deep-submicron (80nm-node) scale operation with a consistent, ~10x lower LET ceiling.
  • Added heavy-metal (Au, W, Pb, Ta) dead-layer materials; confirmed real neutron- and proton-induced fission-fragment recoils up to LET ~35-41, well above silicon's own ceiling, with severity clustering in pairs (Ta/W mild, Au/Pb severe) rather than scaling with atomic number.
  • Tested and rejected two plausible-sounding hypotheses: a gold package-lid LET-ceiling increase, and an ~80 um surrounding-volume "shell ceiling" -- both recorded as closed negative results.
  • Added approximate angle-of-incidence support (thickness/cos(theta)).

Correctness fixes

  • Fixed extreme near-zero-charge events.csv artifacts via a pair-creation-energy per-step floor.
  • Fixed a GUI visualization stall (G4 vis-queue backpressure) via discard+accumulate.
  • Fixed a misleading biasing ratio message in SEEBiasingOperator.
  • Default the GUI's Plot Metric dropdown to Deposited charge, matching the RPP-model 100%-collection convention used by CREME-MC/CUPID.

See README.md and Documentation/PANDA_MASTER_DESIGN for details.

Assets 2
Loading