Skip to content

Pulsim — User Guide

Pulsim is a header-only C++23 circuit simulator with a Python frontend, designed for power electronics and SMPS workloads. The whole solver lives in a few .hpp files; the engine pre-factors every reachable switch configuration into a piecewise-linear state-space cache, then steps through time at near-machine-speed.

If you've used SPICE before, the mental model is similar but the implementation is very different — see the mental model page for the one-page summary.

When to use Pulsim

  • You're building power-electronics converters (buck, boost, flyback, full-bridge, 3-φ VSI, LDO, …).
  • You want a scriptable, Python-first workflow with full access to the solver internals.
  • You need to run thousands of duty-cycle / load-line sweeps quickly — the kernel's PWL cache makes each switch combo essentially free after the first solve.
  • You're OK driving things from a CircuitBuilder (Python or C++) or YAML — there is no schematic GUI yet.

When NOT to use Pulsim

  • You need a full SPICE language frontend with .MODEL cards — Pulsim has its own device catalogue, not SPICE-compatible.
  • You're doing RF / high-frequency PCB simulations with distributed elements — Pulsim is lumped-only.
  • You need built-in convergence retries that mask physically dubious circuits — the kernel's Newton is principled but unforgiving; expect to think about your circuit.

Reading order

  1. Getting started — install, run the first transient, see a plot.
  2. Mental model — what the Graph, DevicePool, PwlStateSpaceCache, and Newton refresh do.
  3. Tutorials — six walk-throughs from simplest to most involved:
  4. 01 — RC charging from a pulse source
  5. 02 — Buck converter (YAML + switch_fn)
  6. 03 — Isolated flyback (transformer + commutation)
  7. 04 — Three-phase voltage-source inverter (SPWM)
  8. 05 — LDO with op-amp feedback (VCVS + MOSFET)
  9. 06 — IGBT boost with realistic gate drive
  10. API reference — Python surface, one page.
  11. Gotchas — Newton convergence corner cases and the workarounds that ship in 1.0.

Post-hoc analysis (v1.5)

  • Losses & junction temperature — device_loss_summary + device_thermal_summary. Reads resistors, inductors (with optional Steinmetz / iGSE core loss), ideal switches (PSIM E_on / E_off), and switched diodes (Q_rr / E_rr_ref); pipes the per-device loss through a Foster network for T_j(t).
  • @fast_block — Pulsim's PSIM-Custom C Block / PLECS-C-Script workalike via Numba LLVM JIT. Write the control law in Python, get native-speed code without a C compiler in the user's PATH.
  • YAML composite devices + chain wiring — induction_motor / hysteretic_inductor in circuit: plus wire_chain_from_yaml(loaded, chain_spec) for the control layer (IM mechanical, JA observer, SMO/MRAS sensorless).

Where the source lives

Tree What's inside
core/include/pulsim/ The entire C++23 solver (header-only)
core/tests/ Catch2 unit + integration tests
python/pulsim/__init__.py Python re-export module + simulate() helper
python/bindings.cpp pybind11 bindings
examples/*.yaml YAML showcases referenced from these tutorials
openspec/specs/pulsim-* Authoritative capability specs
openspec/changes/pulsim-* Historical change proposals