Live 8 × 8 µm y-dipole SiN/SiO2-slab atom-slot optimizer

Worker: not active · stage: projected_q_over_v_continuation · status: running_fixed_beta_pure_q_continuation · checkpoint age: 1503.2 min

This independent clean restart starts from a uniform density-0.5 SiN topology inside a planar SiO2 stack. The oxide occupies the complete x–y simulation plane, including the x and y PML cells. An explicit immutable air body cuts the full strip -0.5 µm < x < 0.5 µm through the oxide, SiN, and complete transverse/PML extent. A centered y-oriented atom and symmetric TE-like source select the Ey cavity family. Full x/y/z symmetry retains one octant of the 8 × 8 µm physical design.

Stage A uses raw Adam on the fixed absolute Q=10,000 ideal atom time trace. Once temporal improvement clearly slows with a trustworthy Q≥100, Stage B switches to adaptive-trust L-BFGS-B on pure same-pole harmonic Q using only the raw 50 nm SiN pixels. The next diminishing-return handoff unlocks SiN and per-side SiO2 thickness while preserving t_SiN + 2t_SiO2 ≤ 1.000 µm. Gradual fixed-beta continuation and then topology-free smooth cubic-spline boundary optimization follow under the same pure-Q scalar. Q/V, atom-local volume, Purcell, fields, pole continuity, and trust health remain live telemetry.

The separate 12 × 4 µm first-step air-strip-width sweeps compare independent density-0.5 temporal seeds from 1.0 to 2.0 µm for 100 nm SiN and 300 nm SiN without altering this archived campaign.

quantity	current
temporal Adam updates	197 · plateau not detected (16/20 trend samples)
pure-Q substage / accepted since temporal handoff	joint_stack / 55
pole accepted updates / trials	825 / 6702
Q	4011.8
best recorded Q (trust gate shown above)	4153.47
field / energy Q	4305.52 / 4011.8
Q-estimator ratio	1.07321
pole wavelength	834.441 nm
source-center wavelength	827.606 nm
field-monitor wavelength	834.383 nm
design-region slab thickness	0.427735 µm
Stage-B thickness coordinate / range	True / 0.050–0.500 µm
SiO2 thickness per side	0.215482 µm
Stage-B SiO2 coordinate / maximum	True / 0.475 µm
stack-thickness constraint	t_SiN + 2t_SiO2 ≤ 1.000 µm
SiO2 x–y extent	complete simulation plane, including x/y PML
air-strip cutout extent	complete y/z simulation span, including PML
physical mode volume	35.2435 µm³
normalized mode volume	60.6586 (λ_pole/n_air)³
Q / normalized atom volume	68.473
Purcell from Q/V	5.20332
fitted atom-pole amplitude at source clear	0.000565168
absolute atom-trace peak / handoff	0.234464
source spectral power at pole / actual carrier	0.961621
reciprocal guided-Purcell proxy	7.04699e-08
guided-Purcell proxy / handoff	0.5224
Stage-B volume treatment	none — pure log harmonic Q is optimized
three-window V span	0.0578487
active Q/V run time / window	2.65 / 0.8 ps
gradient backend / storage	checkpointed / 212 lean checkpoints
structural backend decision	exact nondispersive PML tape exceeds the automatic memory budget
exact reversible PML tape size	10.9784 GB
frozen measured value+gradient speedup	—x
Q/V fit start / nominal lifetime per window	0.25 ps / 0.125 τ_amplitude
precision-triggered tape growths	4
final ringdown field-power / energy fraction	0.430423 / 0.40645
finite-step adjoint direction verified	True
ideal temporal fidelity	0.493914
active objective / value	pure log harmonic Q / 8.3317
stage plateau / reason	False / collecting_two_optimizer_memories
plateau samples / evidence window	16 / 20
objective slope / detectable slope	— / —
trust-floor restarts since progress	0 / 2
practical fixed-beta Q/V gain over one memory	0.0592407% / 1% limit
practical plateau samples / required	11 / 11
practical trust radius still expanding	True
healthy / cumulatively pruned curvature pairs	1 / 544
L-BFGS / current-gradient predicted-ascent ratio	0.135332
replay-triggered curvature recoveries	72
maximum last density motion	0.0262905
trust radius	0.00376506
binarity score (1 is binary)	0.984335
trainable gray fraction (0.05<rho<0.95)	2.12515%
archived tangent-stage Q/V floor	43.2415
projection beta / eta	15 / 0.5
fixed-beta accepted Q/V updates / transition	16 / objective plateau
material parameterization	unfiltered_centered_tanh_projected_50nm_pixels
hard atom access	1.000 µm full-y air strip, immutable
pixel / grid	50 nm / [101, 95, 35]
non-PML z span / material-stack-to-PML air gap	2.50 / 0.82 µm
z PML	0.50 µm per side
z-padding convergence audit	not_run
1.50→3.40 µm replay: residual / field-Q / energy-Q change	nan% / nan% / nan%
binarization / material projection / filter	True / True / off
physical waveguide escape eta	not part of the active objective

Q, V, objective, and optimizer history

The Q axis is linear. A pole is not trusted until the centered Ey and total energy decay estimates agree. Mode volume uses the geometric mean of their three same-pole windows and the same fitted pole wavelength; the local permittivity is exactly air by the hard geometry constraint. This run imports no geometry, pole state, or optimizer memory from earlier campaigns.

Ideal and measured atom response

The target has fixed amplitude and phase for Q=10,000, beta one, V=0.1 µm³, and perfect Ey-polarized overlap. It is not rescaled to fit the simulation.

Geometry and fields

Downloadable layout

Layout status: provisional density-0.5-contour GDS preview. The ZIP includes the GDS, JSON stack/metric manifest, and README. Before the spline handoff the GDS is intentionally marked as a thresholded preview; after handoff it is the fabrication-oriented smooth boundary representation. GDS is planar, so use the manifest for SiN and SiO2 thicknesses.

GDS bundle

Current GDS

Layout manifest

Exact campaign methodology