Optimize a PCB design with JITX and HFSS#
Choose the starting point that matches your existing project plan and data.
Start from a solved Ansys Electronics Desktop (AEDT) project. Compare the reconstruction with the source solve before accepting it as your baseline.
Start from a written specification. Use the first whole-board HFSS solve you accept as your baseline.
For either path, prepare the solver host first. The agent uses jitxlib-ansys to export the native JITX design to AEDT and run HFSS. Authorize the campaign budget before running HFSS, including the first correlation or launch calibration solve.
Prepare the solver host#
Complete JumpStart 0. Provide a compatible AEDT installation and HFSS license on the solver host. Obtain jitxlib-ansys and the HFSS optimization plugin from JITX. Install the library in the project’s Python environment and enable the plugin alongside the JITX design skills. Start a fresh agent session after installing skills.
Set ANSYSEM_ROOT to the intended AEDT installation. Ask the agent to report the package, plugin, runtime, and AEDT versions it will use. Run these checks from the project environment on the solver host.
python -c "from importlib.metadata import version; print(version('jitx'), version('jitxlib-ansys'))"
python -c "from jitxlib.ansys.pipeline import AnsysPipeline; AnsysPipeline.checkenv()"
jitx design export hfss --help
Check HFSS license availability with your Ansys license administrator or license manager. Once the JITX project exists, start its runtime with jitx runtime start --background and check jitx runtime status from the project directory.
Start from a solved AEDT project#
Supply a copy of the source project (.aedt or archived .aedtz), its companion .aedb directory when applicable, and the solved traces or exported reference data. Include the fabrication rules, port definitions, and instructions for any changes requested after correlation. Agree on correlation tolerances before the agent rebuilds the design. Replace the bracketed fields in this prompt with your inputs.
Read the solved project, reference data, fabrication rules, and port notes
in [input directory]. Reconstruct the source as a native JITX design.
Use JITX routes, vias, padstacks, and pours. Preserve the geometry and solve
settings, including any source rule violations. List those violations.
Build the layout and pause so I can inspect it in the JITX board viewer.
Then export a separate AEDT project and show me the side-by-side comparison.
Use [correlation tolerances]. Keep proposed changes separate until I accept
the baseline. Bring any unexplained difference to me before tuning.
Use the board viewer to trace each signal from entry to exit. In AEDT, compare the original with the JITX export before spending a solve.
Inspect |
Compare in the source and the export |
|---|---|
Stackup and copper |
Map layer names to physical order. Check copper and dielectric thicknesses, permittivity, loss tangent, fill material, and soldermask. Match the presence or absence of each plane, its net, and its openings. |
Routes and vias |
Check route shape, padstack by layer, drill and pad sizes, antipads, and via polygon count. In cross-section, locate the end of the plated barrel and the start and stop of each backdrill void. Compare corresponding objects by volume and face count to catch differences hidden by equal bounding boxes. |
Measurement setup |
Match port type, polarity, numbering, impedance, and measurement planes. Compare boundaries, airbox clearance, adaptive solution frequency, mesh settings, sweep, and extraction settings. |
Follow the campaign procedure to solve at the source settings. Overlay the original and reconstructed traces. If a trace exceeds tolerance, have the agent mark the largest difference and its frequency or time. Review any out-of-tolerance result before tuning. Accept the discrepancy in writing or request a correction and another solve within budget. Keep both projects and log each approved change, its reason, and the first affected variant.
Differential-via example#
For a ball grid array (BGA) escape, the following are illustrative correlation tolerances for the reconstruction. They measure agreement between two representations of the same design.
Quantity |
Acceptance check |
|---|---|
Differential input reflection (Sdd11) |
The two magnitudes in dB differ by at most 0.25 dB across the source sweep. |
Time-domain reflectometry (TDR) |
The two impedance traces differ by at most 3 Ω in the agreed time window, using the source rise time and reference plane. |
Launch capacitance |
The estimates agree within 0.5 fF using the same extraction on both models, either the TDR dip or the admittance (Y) parameters. |
For this escape, first meet the BGA-end Sdd11 limit across the specified band, then minimize the largest absolute TDR excursion from 90 Ω in the launch window. Use a TDR pass/fail limit only if your specification sets one. Report Sdd22 and TDR at the escape end separately. Hold routes, via locations, and ports fixed. Vary the layers carrying nonfunctional annular rings and the size of the antipads within fabrication limits.
Check the exported cross-section of each variant. For an escape that requires a reference pour immediately below the exit, verify that the pour meets clearance rules, that the plated stub reaches the specified stop layer, and that the backdrill void reaches it from the drill side. Verify the specified prepreg fill and the soldermask beside the outer copper. A change to port type or sweep requires a recorded comparison before attributing a response change to the vias.
AC-coupling capacitor example#
Preserve capacitor placement and breakout routing to establish correlation. For optimization, allow the capacitor and its breakouts to move while holding the package, layer assignment, stackup, and endpoint locations fixed. Inspect the exported cutout for floating copper and vias bisected by the cut boundary. For a stripline wave port, check that its top and bottom edges contact the intended reference planes across its full width.
Compare padstacks, voids, return loss, insertion loss, and TDR for each routing layer. Then solve the combined arrangement. To report power-sum crosstalk for each victim pair, sum the coupled powers from the specified aggressors at each frequency before converting to dB. Measure placement area with the same bounding rectangle around capacitor lands, signal-via pads, and return-via pads in both layouts.
Start from a written specification#
In JumpStart 5’s diplexer runbook, a common port feeds low- and high-band branches. The specification defines which port should pass or reject each band. Measuring at the connectors includes the launches in the loss budget. Give the agent the electrical and mechanical requirements, fabrication rules and stackup, available parts, and vendor models with this prompt.
Build a native JITX design from [specification, stackup, and vendor files].
List the ports, reference impedances and measurement planes. For each band,
turn insertion loss, return loss and isolation into a target with units,
direction, and the nominal or worst tolerance corner where it must hold.
Show me unresolved choices. Use the available component models for a
circuit-level estimate and identify the margin left for layout and launches.
Wait for my approval of the architecture, then show me the build plan.
Build the approved design and pause so I can inspect the JITX layout.
For the diplexer, review the branch topology, component values and tolerances, and the predicted losses at the common junction. Record the approved topology, part families, substrate, and reference planes in rough-path.md. After approving the build plan, inspect the native layout for port placement, ground returns, and copper at the component lands.
Make a two-connector through fixture from the board’s launch geometry. Agree on launch loss limits and measure the fixture within the calibration budget. Use the accepted launch at the board ports. The first accepted whole-board HFSS solve establishes the diplexer baseline. Compare later board variants with that solve and score them against the approved specification.
Run the optimization campaign#
Complete this brief with the agent using the jitx-hfss-optimization skill. Record the targets and geometry constraints in the section’s PRINCIPLES.md. You supply the fixed-geometry decisions in pin-answers.json and authorize the work in campaign-budget.json.
Campaign brief |
What to specify |
|---|---|
Primary objective |
Name the metric, port or pair, reference impedance and plane, frequency band or time window, units, limit, and direction of improvement. |
Other measurements |
List each additional pass/fail limit, its applicable tolerance corner, and quantities to report for diagnosis. For TDR, specify the rise time and its definition. |
Design changes |
Identify geometry that must stay fixed, what may change, and fabrication limits. Record approval for any change to those boundaries. |
Solve budget and stopping conditions |
Set allowed solve tiers, rounds, calibration solves, solve-hours, and expiry. Define success and any plateau criterion. Reserve enough time and authorization for a final confirmation solve. |
Inspect the export. Define an
Experimenton the native design for each section with its ports, solve setup, and results to extract. Use the HFSS exporter’s--no-simulateoption to generate the AEDT project. Open it and check the cutout, connectivity, ports, and setup. Resolve discrepancies and review export warnings before launching a solve.Measure the baseline. Run the accepted setup under the campaign budget. Check HFSS convergence and verify that the response covers the required ports and frequency bands. Save the design revision, solve settings, and reference traces. For a reconstructed design, complete the source correlation at this point.
Tune one section at a time. Record the proposed change and its predicted effect. Rebuild, check fabrication rules and fixed geometry in the export, then solve. Compare the result with the baseline and preceding variant under equivalent conditions. Investigate failed geometry or convergence checks before drawing an electrical conclusion.
Confirm and hand off. When tuning stops, run the selected design at the confirmation tier while the budget is active. Check convergence and frequency coverage, then score the required metrics. For the diplexer, use the same solved board response with component models at nominal values and at the specified tolerance corners. Report each nominal and worst-corner result, including the limiting frequency and component values. If the budget expires before confirmation, identify the best measured variant and mark confirmation pending.
Deliver the native JITX project, its exported AEDT geometry, the baseline and final traces, and a side-by-side comparison. Include the approved modeling changes, source rule violations, and unresolved discrepancies. Fill in one row per metric, recording the frequency, time, or corner of the worst case. The entries below are a reporting template.
Metric and condition |
Target |
Baseline |
Final |
Result or shortfall |
|---|---|---|---|---|
BGA-end Sdd11, worst in band |
≤ [limit] dB |
[dB at frequency] |
[dB at frequency] |
[pass or dB over limit] |
BGA-end TDR, launch window |
Minimize excursion from 90 Ω |
[peak deviation, Ω at time] |
[peak deviation, Ω at time] |
[final minus baseline, Ω] |
Diplexer insertion loss, low band, worst corner |
≤ [limit] dB |
[dB at frequency, corner] |
[dB at frequency, corner] |
[pass or dB over limit] |
Escape-end TDR, launch window |
Report only |
[min/max Ω] |
[min/max Ω] |
[change in excursion] |