Validation
What we've checked, and what we haven't.
Numerical sanity checks against textbook estimates, consistency between the deck engine and the built-in templates, and automated tests. None of this is calibration against measured devices yet.
Sanity checks · default parameters
Simulated against analytic estimates.
| Device | Check | Simulated | Reference |
|---|---|---|---|
| Si NMOS · Lg 0.5 µm, tox 5 nm, Na 3e17 | Linear Id at Vg 1.5 V, Vd 0.05 V | 31.8 mA/mm | 30 mA/mm, long-channel formula |
| Vth | 0.33–0.45 V | 0.36 V analytic | |
| SS / DIBL | 76 mV/dec / 11 mV/V | Typical for this geometry | |
| AlGaN/GaN HEMT · 20 nm Al0.25GaN | 2DEG under the gate at Vg 0 | 1.1 × 10¹³ cm⁻² | 1.1 × 10¹³, charge-control estimate |
| Vth | ≈ −4.7 V | −4 to −5 V typical |
Consistency
The deck engine reproduces the templates.
The same devices, written as decks and as built-in templates, should agree. They do, within the margins below.
| Device | Metrics compared | Agreement |
|---|---|---|
| Si NMOS | Vth, SS, gm, Ion | Within 1.2 % |
| AlGaN/GaN HEMT | Vth, gm, SS, 2DEG density | Within 1 % (Ion 8 %, because the meshes differ) |
The tests check software behaviour — validation, saving, exports, search bookkeeping and solver runs — not scientific accuracy.
The benchmark ladder
How we plan to earn trust, step by step.
Analytic cases
Textbook estimates for threshold, linear current and 2DEG density. Started — see above.
Established TCAD
The same structures in an established simulator, with tolerances recorded.
Published data
Reproduce device results reported in the literature.
Measured devices
Calibrate against a partner lab's I–V and C–V measurements, with mesh-independence checks.
Open issues
- Convergence. The full transfer sweep of the NMOS dielectric example stops near −0.4 V. Completed points are kept and the run is marked partial.
- HEMT peak current. Around 2.2 A/mm, above typical measured values, because self-heating and contact resistance aren't modelled.
Full model limits are on the physics page.
Have measured data?
We're looking for labs willing to share I–V and C–V measurements for calibration. We'll publish the comparison with your permission.