Simulation surrogates · Applied AI
Replace expensive simulation
with learned physics.
We build custom surrogate models for the simulation problems other tools can't touch — superconducting circuits, biotech, robotics — trained on your simulator, calibrated to your process, with confidence bounds on every output. Fluxus, our superconducting design studio, spans qubit chips and RSFQ logic — and its verification API runs 200–500× faster than SPICE.
Fluxus — RSFQ circuit verification · MIT-LL SFQ5ee+
import cogitan
# load your cell library and run verification
results = cogitan.verify(
path = "./layouts/sfq5ee+",
pdk = "mit-ll-sfq5ee+",
batch = True,
)
for cell, r in results.items():
print(
f"{cell:<14} {r.delay:>5.1f}
f" ± {r.sigma:.1f} ps {r.status}"
)$ cogitan verify ./layouts/sfq5ee+
SPLIT_v1_a 17.2 ± 1.4 ps ✓
SPLIT_v2_b 18.9 ± 1.1 ps ✓
DFF_x3_c 22.1 ± 5.8 ps → spice
AND2_v1_d 14.4 ± 1.6 ps ✓
...
─────────────────────────────────
1,000 cells 4.8s (SPICE: ~85 min)
997 pass · 3 → spice fallback
Wall-clock time · chip-scale verification
Verification is one desk. Fluxus is a full superconducting design studio — qubit Hamiltonians, chip-yield Monte Carlo, inverse design, and RSFQ margins. Explore Fluxus →
200–500×
faster than SPICE at chip scale
99.1%
functional accuracy on held-out chains
2.1 ps
chain delay RMSE vs JoSIM, held-out
100%
DRV detection from layout, held-out cells
The problem we solve
Simulation doesn't scale.
In every physics-constrained engineering domain, full-fidelity simulation is accurate but slow. Slow simulation means slow iteration — no design-space exploration, no confidence scoring, no gradient-based optimization.
In RSFQ, a 1,000-cell block takes 5–10 minutes in SPICE. A full chip is an engineering day — per iteration. The constraint isn't compute. It's the simulation paradigm itself.
Workflow
01
Share your simulation data
Cell library and reference simulation outputs for your process node. MIT-LL SFQ5ee+, IPHT, SeeQC, AIST, SkyWater, or proprietary.
02
We train the surrogate
A learned surrogate calibrated to your junction parameters, cell geometry, and design rules. Training runs on your data.
03
Explore at 200–500×
Run design-space exploration with per-cell confidence scoring. Low-sigma predictions pass. High-sigma cells are flagged.
04
Validate the shortlist
Your reference simulator validates final candidates. 5–10× wall-clock savings per design loop. Every shipped design is ground-truth verified.
Live today · demo portal
Don't take our word for it.
Two suites are live on the demo portal — built with the same pipeline we'd point at your simulator. Every prediction carries a risk percentile, every model abstains where its reference physics can't label, and the validity reports publish the classical baselines our own models had to beat.
demo.cogitan.ai
RSFQ screening — Fluxus
SuperconductingMargin, timing, yield, and design-rule screening for RSFQ logic cells — layout-native, validated on MIT-LL SFQ5ee+.
100% DRV detection · 99.1% chain match, held-out
~10³× vs. JoSIM margin runs
Battery-life screening
Lithium-ion500-cycle degradation from 11 cell-design and duty knobs — an emulator of PyBaMM aging simulations.
0.49% retention-curve MAE · 91.0% conformal coverage
~5×10⁵× vs. the aging simulation
Your simulator
NextThe same pipeline pointed at your solver: fixed-scope feasibility study first, validated endpoint after — evidence before commitment.
Validity report incl. the classical baseline
Abstention + risk percentile on every prediction
Cogitan
Superconducting is the first stack.
The same approach — learned physics surrogate, calibrated uncertainty, differentiable for optimization, trained on your data — applies to any domain where simulation is the bottleneck and ground truth is expensive to generate.
Fluxus, our superconducting design studio, now spans both qubit (cQED) chips and RSFQ logic. Robotics and thermal surrogates are live in production on our API. Biotech and molecular systems are next. If you have a hard simulation problem, we'd like to understand it.
Get in touchProduct roadmap
Have a simulation bottleneck?
We'd like to understand it.
RSFQ was the proof of concept — three production surrogates are live on the API today. If your domain has physics-constrained design and slow verification, the same approach applies. Bring us the problem.
Get in touch