Second Source · Circuit Replacement Lab

CHALLENGE THE DESIGN

Replacement circuit

Behavioral LDO model · 3.3 V fixed output

READY TO TEST

Original component awaiting baseline test

NO ACTIVE RUN

The evidence comes first.

Run an investigation to compare the original and proposed replacements.

RAIL MINIMUM— V

JUNCTION ESTIMATE— °C

TESTED PROPOSALS0

SPICE TESTS0

LOAD-STEP RESPONSE 40 mA → peak → 40 mA

Original Selected

Investigation log

AWAITING INPUT

⌁

A hypothesis, a test, a decision.

Each proposal is recorded before the simulator runs. Failures stay in the evidence.

ⓘ

Know what is being verified. Genuine ngspice execution with simplified behavioral models, three input corners and one load-step profile. Thermal values use your board assumption. Package compatibility, stability, hardware reliability and supply availability are not verified.

A record you can inspect.

Immutable input snapshots, proposals, simulator logs and independent checks.

No experiments yet. Start an investigation in the workbench.

Small catalog. Explicit assumptions.

Three candidates with source-linked specifications. No claim of drop-in compatibility.

What this model captures

Dropout-limited output target, a reference-derived LM1117 current clamp and rated-current approximation for the other parts, a nominal 40 µs control response and 0.08 Ω output resistance (20–80 µs and 0.04–0.12 Ω in the assumption sweep), capacitor ESR, and a stepped current load. ngspice solves the circuit afresh for every proposal.

What these results do not establish

These are lab-authored behavioral models, not manufacturer transistor-level models. LM1117 dropout follows a nominal TI reference curve; the other parts use linearly scaled datasheet endpoints; control dynamics and current clamp are assumptions. ESR stability, startup, noise, production variation, PCB layout, pinout and thermal protection require further analysis and hardware tests.

Junction temperature = ambient + estimated dissipation × user-selected board θJA. It is a steady-state estimate, not electrothermal simulation. Accuracy is checked as a separate specification budget; its full corner variation is not simulated.

Offline versus live AI

Offline mode uses a visible deterministic candidate ranking. Live mode calls NVIDIA Nemotron through Nebius Token Factory to propose the next experiment after receiving previous failures. Neither mode can assign the evaluator’s verdict.

A replacement is a hypothesis.
Test it.

Replacement circuit

The evidence comes first.

Investigation log

A record you can inspect.

Small catalog. Explicit assumptions.

What this model captures

What these results do not establish

Offline versus live AI

Bring the agent online.

A replacement is a hypothesis.Test it.

Replacement circuit

The evidence comes first.

Investigation log

A record you can inspect.

Small catalog. Explicit assumptions.

What this model captures

What these results do not establish

Offline versus live AI

A replacement is a hypothesis.
Test it.