Building the FEMM-XSPICE Simulator: Technical Architecture Meets Agentic Workflow
Jason Owens and Phillip Mobley of Resonance Group presented FEMM-XSPICE, a project connecting magnetic-field simulation with circuit analysis to support motor-controller development.
The architecture brings together KiCad, the NGSPICE/XSPICE circuit engine, and the FEMM magnetic-field solver through a Python bridge. The bridge coordinates the simulation, updates geometry and mechanical state, and passes calculated values back to the circuit engine for subsequent time steps. The presentation included voltage and current plots alongside animated motor simulations, progressing from a simple two-phase example to a more complex motor geometry.
The speakers described practical challenges throughout that work: running FEMM through Wine on Linux, keeping the simulations synchronized, handling convergence and changing time steps, and balancing accuracy against runtime. They also explained why a plausible-looking result still needs scrutiny against the underlying physics.
The second half examined the AI-assisted workflow behind the project. Topics included choosing models for the task, keeping documentation available, organizing persistent memory, and breaking implementation into phases with independently tested components. Shared terminal sessions and checkpoint reports provided ways to inspect an agent’s work. The discussion emphasized stating assumptions, keeping useful logs, reviewing results, and recording lessons so later development could build on earlier troubleshooting.
