Simulation
Using SPICE to validate AI-generated circuits
PCBBench editorial · Published October 14, 2026 · Editorial policy
Use SPICE to test a defined electrical requirement under stated conditions. A successful simulator exit or a plausible waveform does not prove that the complete board is correct; the result depends on the circuit model and the measurements chosen.
Choose the observation before running the circuit
Specify the expected operating point, transfer behavior, transient response or limit. Record numerical acceptance ranges so a generated waveform cannot be accepted by visual impression alone. Preserve the stimulus and measurement definitions with the candidate design.
Validate model and pin assignments
Confirm that each modeled device maps correctly to the schematic pins and that the model covers the behavior being tested. Ideal components can be useful for a first check, but they may omit startup, saturation, protection or package effects that matter to the requirement.
Exercise the boundaries
Run relevant input, load and tolerance conditions rather than a single nominal operating point. Treat a convergence problem separately from an electrical failure until its cause is understood. Preserve failed runs and simulator warnings instead of discarding them from the evidence.
Practical checklist
- Define measurable pass conditions.
- Record simulator and model versions.
- Check model pin mapping.
- Test relevant corners and startup states.
- Keep logs and unmodified testbenches.
A concrete example
Test a generated sensor amplifier across its required input range.
- Evidence to accept
- The measured output stays within the allowed error and operating range for the stated conditions.
- Evidence that is insufficient
- A nominal small-signal waveform is used to claim correct startup and rail behavior.
Common questions
Does a simulated circuit need a physical test?
Simulation reduces uncertainty within the model. Manufacturing, assembly and unmodeled behavior still require other evidence, including bring-up where appropriate.
Should an agent edit its own acceptance test?
For a scored evaluation, keep the acceptance test outside the candidate's editable files or verify it has not changed.
Primary references and further reading
Use the matching device documentation and project configuration when applying this checklist. These references explain the underlying tools and evaluation approaches.
How this fits PCBBench
The AI schematic generation benchmark describes the proposed inputs, outputs and checks for this capability. The pilot snapshot contains partial measured attempts; it is not a finished ranking.
Related reading
- How to test AI schematic generationTest an AI-generated schematic for electrical intent, real components, valid connectivity and reproducible verification.
- KiCad ERC checklist for AI-generated schematicsReview electrical-rule results without hiding connection mistakes, changing severity settings or overlooking wrong component values.
- Netlist equivalence for AI schematic validationValidate pin-level circuit intent while allowing equivalent drawing layouts and legitimate alternative implementations.
- SPICE vs ERC: simulation and schematic checksCompare circuit simulation with electrical-rule checking and choose evidence for each requirement.