Result

splitter

All design points

splitter

Answered

12.6 ms

282× vs JoSIM

Both margin edges served with a conformal interval from a held-out calibration split.

Fluxon splitter. Same heads, a visibly tighter band -- the band is fitted per quantity, not assumed.

Functional verdict

Predicted functionalyes

P(functional)1.0000

Headgbt_knobsAUROC 1.000

Conformal set size1singleton · answers

Margins, relative to nominal

Bias margin0.800–1.200×±0.006 / ±0.014

Ic margin0.797–1.100×±0.011 / ±0.010

Bias width0.400±0.020 · derived

Sourcegbt_knobs

What this margin is, exactly

Measured under reference feeder F0, the reference drive. width varies up to 2.17x across nine defensible reference feeders, and 2.7x (jtl) / 6.0x (splitter) across designs within the operating band. This value is for THIS design under feeder F0; it is not a property of the cell type.

interval at alpha 0.1 · calibration n=1200 · width is derived as hi − lo and carries the sum of both edge half-widths, so it can never contradict the edges beside it

fluxus-s · captured from the same service the API serves · reproduces the JoSIM oracle, not silicon; nothing here is fab-calibrated. The cell library targets MIT-LL SFQ5ee+ and is literature-reconstructed, not foundry GDS. Margins are served for 3 of 8 library cells.