AIPE PCB Agent

AIPE PCB Agent is the PCB design agent in the AIPE framework. It turns a converter’s requirements into editable KiCad schematics and board layouts, runs KiCad’s own checks on them and packages the results for engineering review. This page explains the framework it belongs to, its role there, how the agent is organised, and a 5 kW SiC converter it has designed.

Explore the repository

The AIPE Framework

AIPE (AI for Power Engineering) is the open engineering ecosystem I am building through AIPE Labs. It treats power-converter design as a coordinated, iterative optimisation process rather than a chain of separate tools. Specialist agents each own one engineering step: electrical sizing, schematic drawing, circuit simulation, device and magnetics selection, thermal and field analysis, mechanical CAD and PCB layout. They exchange explicit inputs, assumptions, results and evidence through a shared Engineering State.

A workflow engine runs these steps as a checkpointed task graph. Results from later stages, such as magnetics or finite-element parameters, feed back into the circuit model, and every result is recorded against a verification gate. Software checks do not replace engineering review, measurement or hardware qualification, which remain separate gates. The first full benchmark is a 10 kW dual-active-bridge converter (800 V → 400 V, 100 kHz).

AIPE design workflow: specification, electrical design and schematic sketching, circuit simulation, device, magnetics and thermal design, field simulation, CAD and PCB agents and a review package, coordinated by AIPE-Workflows, AIPE-Core and AIPE-Registry Specification ratings, operating range and design constraints Electrical design analytical sizing Sketcher checked SVG schematics Circuit simulation AIPE-Simulation-Skills · switching waveforms, control Device, magnetics and thermal design Semiconductor and Magnetics Databases · Thermal Design Field simulation AIPE-Field-Simulation · thermal and magnetic FEM CAD Agent FreeCAD geometry PCB Agent KiCad schematics and boards Review package report and evidence for human engineering review feedback: iterate and reconcile Coordination layer AIPE-Workflows task graph and checkpoints verification gates feedback iterations review reports AIPE-Core shared Engineering State SI units and provenance evidence conventions interface contracts AIPE-Registry capability manifests tool and dependency index integration maturity Each stage reads the state and returns results with evidence
AIPE design workflow. Specialist agents (left) run in sequence with feedback, while the coordination layer (right) carries the shared state, evidence and verification gates. The highlighted block is the agent described on this page.

Role of the PCB Agent

PCB Agent works at the physical-design stage, downstream of AIPE Electrical Design and alongside the CAD Agent, which produces mechanical geometry such as transformers and heat sinks. It takes the converter’s electrical decisions and turns them into boards that an engineer can open, check and revise.

InterfaceContent
InputsRequirement files: converter specification, insulation-coordination table and stack-ups, component selections with sources, interface pin maps, floorplan and routing inputs
OutputsNative KiCad projects, ERC/DRC and layout-check records, manufacturing exports and review documents
Hand-offAIPE-Workflows re-runs the KiCad ERC and DRC on the generated boards for its CAD/PCB gate; the results then go to human engineering review

A clean ERC/DRC result is a statement about the generated CAD files under the recorded rule set. It does not establish electrical safety, insulation compliance or manufacturability. The agent interface defines these inputs, outputs and limits.

How the Agent Is Organised

The agent is built around one shared circuit model. The model records every part, pin and net, while schematic drawing and board geometry are generated from it separately. Checks compare the CAD files with the model and feed their findings back to the stage that needs another pass.

Flow chart of AIPE PCB Agent: requirement files and a knowledge base feed a shared circuit model and design rules, which drive schematic generation and board construction, followed by verification, exports and hand-off, with findings fed back to the model, placement or routing
Internal organisation of AIPE PCB Agent, from requirement files to checked and packaged KiCad projects.
PartResponsibility
Requirement filesYAML/JSON files hold every number the design uses: ratings, parts, interfaces, insulation distances, stack-ups and floorplan regions.
Knowledge baseDesign skills, circuit blocks, component records and layout rules, each traced to manufacturer or reference-design sources.
Circuit model and rulesConverter-specific Python modules build the part and connection model; net classes and KiCad design rules are derived from the insulation table.
Schematic generationGroups parts by function, draws and routes hierarchical sheets, and refuses to write a sheet whose connectivity differs from the model.
Board constructionSets the stack-up, places parts with insulation-aware spacing, forms copper zones and routes connections with a deterministic grid router.
VerificationRuns native kicad-cli ERC and DRC with schematic parity, then checks netlist equality, routing quality, layout metrics and copper islands.
Exports and recordsWrites editable projects, PDFs, Gerber, drill, BOM, placement and STEP files, with SHA-256 manifests linking each output to its sources.

The converter-specific modules supply only design data and policies, while the shared tools do the CAD generation and checking. This lets buck, DAB and future converter projects reuse the same operations. The automation architecture and task workflow describe the modules and their acceptance gates.

Example: 5 kW SiC Buck Converter

This example shows a complete task, from the specification to two routed boards. The agent was asked to design a SiC synchronous buck converter; its design requirements specify:

ParameterDesign value
Nominal input voltage400 V
Output voltage200 V
Output power5 kW
Nominal output current25 A
Switching frequency50 kHz
Power devicesWolfspeed SiC MOSFETs
ControllerTI TMS320F28335 DSP

From Requirements to Boards

The task followed the agent’s workflow:

  1. Requirements and engineering. Values the task left open, such as the input range, cooling and ripple targets, were recorded as labelled assumptions. Calculations and the knowledge base then selected the parts: Wolfspeed C4MS025120K SiC MOSFETs, UCC21710-Q1 isolated gate drivers with UCC14141-Q1 bias modules, a custom 272 µH High Flux toroid and isolated current and voltage sensing.
  2. Architecture. The agent compared board arrangements and chose two boards: a power board for conversion, gate drive, sensing and protection, and a DSP control board connected through a 2 × 25 mezzanine.
  3. Schematics and boards. It generated hierarchical schematics, placed the parts with insulation-aware spacing along each board’s isolation barrier, and routed both four-layer boards.
  4. Checks and outputs. Both boards reached zero native ERC/DRC findings with schematic parity, and the exported netlists matched the circuit model exactly. The task then produced the review package.

Power Board

The 240 × 180 mm power board groups the SiC switching stage, isolated gate drivers and bias supplies, DC-link capacitors, output filter, sensing and protection circuitry. The layout separates the high-voltage power path from the low-voltage interface used by the control board.

KiCad 3D view of the SiC buck power board, showing the switching devices, large magnetic component, capacitor banks and control-board connector
SiC buck power-board CAD view, showing the component arrangement and control-board interface.

Control Board

The 120 × 80 mm control board brings together the TMS320F28335 DSP, local supplies, signal conditioning, PWM and protection interfaces, and debug connections. The mezzanine connector stacks it on the power board’s low-voltage strip, keeping the control circuitry on a separate PCB.

KiCad 3D view of the control board, showing the central TMS320F28335 DSP, support circuitry, debug headers and mezzanine connector
DSP control-board CAD view with local supplies, signal interfaces and mezzanine connection.

Task Outputs

The task produced two editable KiCad board projects and their schematics, together with BOMs, placement files, Gerbers, drill files, STEP models, check reports and hashed manifests. The design is a CAD-complete prototype for review: no board has been fabricated, energised or measured.

A later clean-sheet revision (V3) redesigned the same converter after an independent review, moving the control board beside the power board on a keyed cable and mounting the heavy magnetics on the chassis. The V3 package and the example history link each revision to its design records and review packages.

AIPE Overview · Source Code · Back to Research