Order 0: predictions

Drafts. These files are regenerated with the boards until the pre-registration release (preregistration.md); from then on they are frozen and never edited.

Coupons

coupons/ holds the expected S-parameters of every modelled coupon stick of each upload, at nominal parameters on FR408HR (fr408hr/) and on the EM528 alternate (em528/), with the FR408HR geometry in both:

  • two-port sticks: reference plane to reference plane (what the multiline-TRL-corrected measurement shows), 30 MHz-6 GHz in 30 MHz steps, renormalized to 50 Ω from the TRL line’s Zc;

  • the C-pads of A16: S11 at each pad’s reference plane, 6 MHz-6 GHz in 6 MHz steps (the measurement grid; 0.1-1 GHz is the band that is fitted; the step from the line to the pad is left out, about 0.5 % of the pads’ capacitance difference);

  • scalars.json: the numbers the held-out criteria test, from the model’s own |S21| minimum on a 10 kHz grid, not the 30 MHz files: the ring A11’s notches n = 1 and 3 (FR408HR 1.924 and 5.793 GHz) and the stub A12’s notch (5.500 GHz). The ring’s arcs and the stub carry Hammerstad’s T-junction reference planes;

  • lines.csv: every line family’s εeff, α (dB/cm) and Zc on the 6 MHz measurement grid.

The model is the coupon forward model: the 2D quasi-static RLGC of each line family (Djordjevic- Sarkar dielectrics, Huray roughness, Wheeler’s conductor loss, the shipped 2D tables) cascaded with closed-form discontinuities. The launches are not modelled: the calibration removes them. The thru and the reflect are calibration standards and have no file.

python -m yapnr.rf.coupons expected --stackup OSHPARK-4L-FR408HR --upload M \
    --out docs/rf/order0/predictions/coupons/O0-M

Every stick and its prediction

Raw solver files are byte-for-byte copies of the cloud tasks’ outputs (yapnr commit 917222a, image ghcr.io/studio-fug/yapnr@sha256:f41a4760…, native float64); the loss-corrected files derived from them are in corrected/ and are the predictions the criteria use (pre-registration). “FR408HR” is the thickness-equivalent substrate of nominal FR408HR (M-eq, W-eq) for the solver and nominal FR408HR for the coupon model; “EM528” likewise (M-eq-em528, W-eq-em528); the solver’s raw nominal view (zero-thickness copper on M-nom, W-nom) is published besides.

O0-M

Stick

FR408HR

EM528

Method

A01 thru

none between its reference planes (a calibration standard: the identity); the launches as tier-1 data see them: openems/a01.s2p

–

openEMS 3D (launch)

A02-A05, A04R

coupons/O0-M/fr408hr/ A0x-M-line.s2p

coupons/O0-M/em528/

coupon model

A06 reflect

none (a standard; its reflection is not assumed)

–

–

A07 verification

A07-M-verify.s2p

the same

coupon model

A08, A09, A10

A08-M07-variant.s2p, A09-M14-variant.s2p, A10-MMK-variant.s2p

the same

coupon model

A11 ring

A11-M-ring.s2p, notches in scalars.json; fdtd/a11-ring-m-eq/

the same; fdtd/a11-ring-m-eq-em528/

coupon model; our FDTD (forward, R1’s settings)

A12 stub

A12-M-stub.s2p, notch in scalars.json; fdtd/a12-stub-m-eq/

the same; fdtd/a12-stub-m-eq-em528/

coupon model; our FDTD

A14 switch-term

A14-M-switch.s2p

the same

coupon model (0402 as 100 Ω + 0.6 nH)

A15 tag, DC

no RF port

–

–

A16 C-pads

A16-C6-pad1.s1p, A16-C12-pad2.s1p

the same

coupon model

R1

run0b/r1-m-eq.s3p, corrected r1-m-eq-loss-corrected.s3p; raw nominal view references/r1-m-nom/, corrected r1-m-nom-…; openems/

references/r1-m-eq-em528/, corrected r1-m-eq-em528-…

our FDTD, refine 2; openEMS 3D (FR408HR)

O0-W light

Stick

FR408HR

EM528

Method

B01 thru

none (a standard)

–

–

B02-B04

coupons/O0-W/fr408hr/ B0x-W-line.s2p

coupons/O0-W/em528/

coupon model

B05 reflect

none (a standard)

–

–

R1t

references/r1t-w-eq/, corrected r1t-w-eq-…; raw nominal view references/r1t-w-nom/

references/r1t-w-eq-em528/

our FDTD

D2 window

if the owner ships the nearest miss d2-star-sched: D2/runs/d2-star-sched/ (W-eq, three grids), D2/variants/ (W-eq wide, W-nom); corrected d2-star-sched-…

D2/variants/d2-star-sched-w-eq-em528/

our FDTD

O0-D

Stick

FR408HR

EM528

Method

D1

D1/runs/d1-star/ (M-eq: coarse, fine, finer), D1/variants/d1-star-m-eq/ (3-7 GHz), d1-star-m-nom/; corrected d1-…; openems/

D1/variants/d1-star-m-eq-em528/

our FDTD; openEMS 3D (FR408HR)

R1 copy

as O0-M’s R1

as O0-M’s

–

A01

as O0-M’s A01

–

–

A20, A04

coupons/O0-D/fr408hr/ A20-M-line.s2p, A04-M-line.s2p; A04 with its launches: openems/a04.s2p

coupons/O0-D/em528/

coupon model; openEMS 3D

The loss correction

python -m yapnr.rf.order0 correct --out docs/rf/order0/predictions/corrected writes the corrected file of every demo and reference prediction and corrected/summary.json (raw and corrected worst cases over 4.25-5.75 GHz, Δα and the ratio at 5 GHz per substrate). Each substrate uses its own loss lines: M-eq run 0b’s (run0b/), W-eq the compute stage’s (references/line-w-*), the others the predictions stage’s (lines/).

Our FDTD on the held-out resonators

fdtd/ holds the forward runs of the ring A11 and the stub A12 (the tuned O0-M geometry from reference plane to reference plane, on the references’ 0.05 mm grid, the stub’s open end snapped to the grid: 8.279 mm drawn, 8.30 mm simulated), and fdtd/resonators.json their notches next to the coupon model’s. The criterion stays the coupon model’s post-fit notch; these are the optimizer’s own solver predicting a held-out resonance before fabrication.

openEMS (D-O0-13a)

openems/: D1, R1 and the 0.40 mm line with 43 µm copper on the nominal FR408HR stack at two meshes, and A01/A04 with the Cinch launches; the models are ../openems/ (order0_em.py).