Source · soonGet started
PUBLIC BETA
QT-1Studies S-1 to S-8

SStudies

A library of studies for programmable photonic circuits, not for one chip. Every study takes --design <pack>, reads its pass thresholds from that pack’s gates.yaml, and runs a Monte Carlo over virtual chips at each of 3corners. Studies that do not apply to a design, such as a two-photon gate on a classical mesh, are simply not run. Need a study your design does not fit? Tell us.

Table S-1. Study summary

IDStudyQuestionPacks
S-1Drive envelope and lossCan every heater reach its phase range, and what is the loss on the longest path?All packs
S-2Calibration yieldWhat fraction of chips calibrate under the error gate?All packs
S-3HOM and timingHow much two-photon interference survives the mesh, and how long does a scan take?All packs
S-4Two-photon gateWhat fidelity and rate does a post-selected CZ gate reach?Packs with six or more modes
S-5Drift and recalibrationHow long does a calibration hold?All packs
S-6AI effective bitsHow many effective bits does the mesh deliver as an AI layer?Packs with an AI encoder
S-7SensitivityWhich input dominates each result?All packs
S-8Cross-design comparisonWhich design is better, each judged by its own gates?Lists of packs

S-1Drive envelope and loss

Method. Sweep drive codes at each resistance corner, apply the driver limits, and sum the loss budget along the longest optical path.

Output. Reachable phase per heater, refused codes, and fibre-to-fibre loss with its evidence label.

S-2Calibration yield

Method. For many seeds, generate a virtual chip, calibrate it on simulated power readings, then test it on random target unitaries.

Output. Yield curve: the share of chips under the pack’s error gate against split spread, crosstalk and loss.

S-3HOM and timing

Method. Hong–Ou–Mandel visibility against path mismatch, split error and source purity; counts per point and scan time per detector type.

Output. Visibility curve, scan time and path-length tolerance.

S-4Two-photon gate

Method. Post-selected CZ gate; fidelity and success rate against indistinguishability, residual calibration error and double-pair fraction.

Output. Fidelity and events per second against pair probability.

S-5Drift and recalibration

Method. Run the drift model over 30–60 minutes and find when the error gate or the AI bit depth falls below its line.

Output. Recalibration interval for the gate and for 4-bit and 6-bit operation.

S-6AI effective bits

Method. Embed matrices, read out with every noise term on, train with hardware-aware noise, and report bits and task accuracy.

Output. Effective bits against phase error, per heater type.

S-7Sensitivity

Method. Vary one input at a time over its range and record how each study output moves.

Output. Ranked table of inputs per output; it orders the measurements and foundry questions.

S-8Cross-design comparison

Method. Run S-1, S-2 and S-3 (and S-4 where the pack allows) on a list of packs with the same seeds and the same question.

Output. One table across designs and architecture variants, each scored against its own gates.

Every result carries the worst evidence label of its inputs. On the public host, calibration is ideal and the report says so.