High-Speed Workbench (SI, PI, EMI)
Impedance, pairs, return paths, crosstalk, interfaces, power and EMI — computed on your routed copper, with the evidence.
Who does what
The top lane is you; the bottom lane is WireFrame.
Say what it is made of
Your board house's stack-up and tolerances. Until you confirm it, results stop at WARNING — nothing can PASS on a guess.
Accept the rules
Accept what is right. Targets live on net classes: impedance, skew in length or in time, vias, stub length.
Approve & apply
Only a person can approve. Applied as one undo step, and rolled back from the same page if you change your mind.
Find the fast signals
From the net names — USB, Ethernet, DDR, QSPI, SD — it proposes impedance targets, pairs, length groups and whole interfaces.
Check the routed copper
Every rule computed on the copper you routed, each result with its value, limit, source, method and assumptions.
Propose fixes
A failing result becomes a proposed change, previewed on a copy of the design and checked again there.
Write the evidence
The revision, the stack-up, every result with its method, every fix and who approved it — and what was not checked.
A fast signal has to be more than connected. The High-Speed Workbench checks the copper you actually routed: impedance with a 2D field solver, pair skew and matching in time, return paths and plane edges, crosstalk, decoupling and PDN impedance, and an EMI estimate against CISPR 32 or FCC. Whole interfaces — USB, PCIe, Ethernet, HDMI, MIPI, DDR, RGMII, QSPI, SD — are checked against the layout rules their specifications publish, each row with its source. Nothing reads PASS that cannot be proven, and nothing changes the board until a person approves it.
What it does
Evidence on every result
The value, the limit, where the limit comes from (spec, vendor, physics or rule of thumb), the method and its assumptions — e.g. USB 2.0 at 107.8 Ω against 90 Ω ±15 % (Renesas R01AN5119).
Stack-up you can trust
Laminate library (FR-4, Isola, Megtron 6, Rogers), fab-tolerance corners, and the controlled-impedance table for the board house. An unconfirmed stack-up caps every result at WARNING.
Simulate what you routed
Reflections and terminations, a coupled pair's common mode, fly-by and tee nets, and Touchstone S-parameter channels — cross-checked with ngspice.
Interfaces as a whole
Twelve profiles with cited rows, from USB 2.0 to DDR4 byte lanes in picoseconds.
Fixes you approve
Proposed → previewed on a copy → approved by a person → applied as one undo step. The AI Copilot can propose; it cannot approve.
A report for the review
HTML with figures, Markdown or JSON: every result, every fix, and what was not checked.
No PASS that cannot be proven
A result is PASS only when everything it depends on is known. Until you tick that the stack-up was confirmed with your board house, every result is capped at WARNING; a missing input — a rise time, a load step, a DC current — makes it UNKNOWN, never PASS. Results go STALE the moment the board changes, and the Workbench notices within half a second. The status line counts them, for example "Stack-up ASSUMED · 10 FAIL · 33 WARNING · 26 UNKNOWN · 10 PASS".
One geometry, one impedance, everywhere
The same engine sits behind the routing readout, DRC, Design Rules and every Workbench page: closed forms where they are accurate, and WireFrame's 2D quasi-static field solver wherever a result is close to its limit. The stack-up behind it holds each layer's copper and roughness and each dielectric's laminate, with Dk and Df interpolated from datasheet points and kept causal outside them. Fab tolerance turns every impedance into a range at the board house's corners.
Thirteen pages, in the order you ask the questions
Start here walks six steps — stack-up, fast signals, rules, check, fix, report — with the board's state at each. Then Stack-up, a Calculator with ten tabs, Simulate, Rules, Checks, Interfaces, Power, EMI, an Inspector for one net layer by layer, Fixes, Report, and Learn, which explains the idea behind every number with a button that shows it. On h730duino, a four-layer STM32H730 board, it recognises 30 fast signals by name before you set a single rule.
What it measured, on a real board
The USB pair on h730duino computes to 107.8 Ω against Renesas's 90 Ω ±15 % — 97.3 to 119.0 Ω across the board house's corners, outside the window even at its best. The proposed fix changes the pair's class to 0.220 mm wide with a 0.128 mm gap on F.Cu, which brings it to 90 Ω. The Inspector shows the same pair's halves differing by 101.81 mil, 14.6 ps.
The details
| Checks | Impedance · pair skew · length and time matching · via count and stub · return path · plane edge · return via · crosstalk · decoupling · PDN impedance · current capacity · emission · insertion loss |
|---|---|
| Statuses | PASS · WARNING · FAIL · UNKNOWN · NOT RUN · STALE |
| Interface profiles | USB 2.0 HS, USB 3.x, PCIe, Ethernet MDI, HDMI, MIPI D-PHY, LVDS, SATA, DDR3/DDR4 byte lane, RGMII, Quad/Octal SPI, SD/eMMC |
| Laminates | Generic FR-4, Isola 370HR and FR408HR, Panasonic Megtron 6, Rogers RO4350B and RO4003C |
| Exports | Controlled-impedance table (CSV, HTML) · report as HTML, Markdown or JSON |
| Fixes | Proposed → Previewed → Approved → Applied, one undo step; roll back from the page |
How to start
- Open a routed board and choose View › High-Speed Workbench.
- On Start here: confirm the stack-up, accept the fast signals WireFrame recognised, run the checks.
- Pick a failing result to read its evidence; Suggest a fix, preview it, approve and apply.
Questions
Does meeting an interface's rows mean the link passes compliance?
No, and the Workbench says so wherever it says compliance: meeting every row means the copper follows the published layout rules. Whether the link passes its own electrical test is decided by that test, on hardware.
Why is everything a WARNING?
The stack-up is still assumed. Enter your board house's stack-up and tick "I confirmed this with my board house".
Can the AI Copilot change my board from here?
It has the same tools as the pages and can propose and preview fixes, but applying refuses any patch a person did not approve in the Workbench, on the same revision it was previewed on. Every number it quotes comes from an engine run.