Skip to content

Campaign: atom_fishbone_coupler/rb780_atom_topology_seed_v1

Assembled 2026-09-05 14:54 UTC.

Engine: atom_fishbone_coupler · path: engines/atom_fishbone_coupler/campaigns/rb780_atom_topology_seed_v1/

README

Rb-780 protected-aperture topology seed v1

This campaign prepares the from-scratch Tidy3D topology stage for the actual atom-collection target. It is not yet an optimized device result.

Fixed physical requirements

  • Rb target line: 780.24 nm.
  • A vertical, exactly 1.0 µm diameter vacuum column at the atom/tweezer site.
  • 500 nm SiO2 slab (n = 1.45) surrounding the protected aperture.
  • One +x, 800 nm-wide output guide and a forward TE0 collection objective.
  • Three spectral samples (779.24, 780.24, 781.24 nm) so discovery is not rewarded for an unresolved, single-frequency spike.
  • 50 nm topology pixels, 160 nm filter radius, 80 nm feature target, and an erosion/dilation penalty.

The protected vacuum structure has priority 100, above both the static output guide (10) and topology region (0). It therefore remains vacuum even if the raw or filtered optimizer density is nonzero there.

Retained preflight

preflight/manifest.json records Tidy3D 2.11.2, Python 3.12.13, the exact simulation, raw and transformed density, aperture mask, signed-distance field, contours, geometry digest, and SHA-256 hashes. The density-to-contour round-trip mismatch is zero on the 160 x 100 design grid. This preflight made zero cloud API calls.

dry_run/preflight.json is the paid runner's no-action path. A paid run needs the explicit --execute switch, uploads one simulation for a maximum-cost estimate, and starts only if twice the per-task estimate for all 30 Adam steps plus the initial step fits a 10 FlexCredit sub-cap and its persistent ledger. The full conservative bound is reserved before Adam starts; a crashed run leaves the reservation in place, and a successful run charges the bound because the wrapper does not expose a stable list of internal task IDs.

Status boundary

The current handoff is a boundary candidate, not a physical device. It has no measured Q, mode volume, Purcell factor, or desired-channel beta. A forward TE0 topology objective is useful for discovery but is not the physical quantity

beta_desired,+ = Gamma_TE0,+ / Gamma_total.

Promotion requires binarization followed by independent, remeshed Q/V/channel measurements, branch tracking, power closure, a two-mesh comparison, and all fabrication gates. No gradient from the topology stage is trusted by the boundary optimizer.

Commands

Offline, repeatable preflight:

PYTHONPATH=src:engines/atom_fishbone_coupler/package \
  python engines/atom_fishbone_coupler/tools/prepare_atom_topology.py \
  --output engines/atom_fishbone_coupler/campaigns/\
rb780_atom_topology_seed_v1/preflight

Paid runner dry run (still no API call):

PYTHONPATH=src:engines/atom_fishbone_coupler/package \
  python engines/atom_fishbone_coupler/tools/run_atom_topology.py \
  --output engines/atom_fishbone_coupler/runs/<run_id>

Review the preflight and available repository credit ledger before adding --execute.