← Journal Process · · 7 min

Running a hardware design review before you order boards

How we run a hardware design review before ordering boards: who should attend, what to check in schematic and layout, and how to track fixes.

Milo Perović
Founder & lead engineer

  • DFM
  • Manufacturing
  • Power
  • QA
Running a hardware design review before you order boards

Ordering a board spin is cheap compared with what a missed fault costs once it is on the bench: a week waiting for the respin, a rework session with fine wire, and a schedule that quietly slips. A proper hardware design review is the cheapest insurance we know against that, yet many small teams skip it or reduce it to a quick scroll through the schematic the night before the files go out.

This post is for founders, product managers and engineers who want a review that catches real problems rather than producing a feeling of diligence. It covers when to hold one, who should be in the room, what to check, and how to turn findings into changes that actually make it onto the board.

When a hardware design review earns its time

We split reviews into two gates rather than one big meeting. The first happens when the schematic is complete but before layout starts in earnest. The second happens when layout is routed and the fabrication outputs are generated, but before anything is ordered.

The schematic gate catches architectural and electrical mistakes while they are still cheap to fix: a regulator that cannot supply the radio’s peak current, a missing pull-up on a boot strap pin, an I2C address clash. The layout gate catches physical mistakes: footprints, return paths, antenna keep-outs and assembly problems. Merging the two into one review at the end means electrical changes arrive after routing, which is where they hurt most.

For a small connected device, each gate typically takes two to four hours of preparation by the designer and one to two hours of meeting time. That figure grows with board complexity, the number of power domains and whether there is RF on the board.

Who should be in the room

The designer presents but should not be the only reviewer, because people reliably read their own schematics as they intended them rather than as they are drawn. We aim for three or four people with different angles:

  • A second hardware engineer who reads every net, not just the blocks they find interesting.
  • A firmware engineer, who checks pin assignments, peripheral conflicts, boot modes, debug access and whether the wake sources match the power plan.
  • Someone from enclosure design at the layout gate, to confirm connector positions, button and LED locations, mounting holes and height restrictions against the CAD.
  • Whoever owns testing or production, to ask how each board will be programmed and tested at end of line.

The firmware seat is the one most often left empty, and it is the one that finds the most expensive problems. A pin that cannot be used as a wake source, or a UART shared with the bootloader, is invisible to a pure hardware reader.

What we check at the schematic stage

A checklist keeps the review from drifting towards whatever the most vocal person cares about. Ours is organised by subsystem, and we tick items against the actual sheets rather than from memory.

Power

  • Every rail’s worst-case current, including radio transmit peaks, against the regulator rating with margin. Wi-Fi modules such as the ESP32 family can draw several hundred milliamps in short transmit bursts, which sizes both the regulator and the bulk capacitance.
  • Input protection: reverse polarity, overvoltage and inrush, and what happens when the user plugs in a poor-quality supply.
  • Power sequencing and brown-out behaviour, especially for parts with more than one supply pin.
  • Quiescent current of every always-on part, summed and compared with the battery budget.

Microcontroller and interfaces

  • Boot and strap pins pulled to the correct state, and nothing external that could fight them during reset.
  • Debug and programming access (SWD, JTAG or UART) brought to a header or pads that will still be reachable once the board is in its enclosure.
  • Bus pull-ups present once, not on every sensor breakout, and sized for the bus speed and capacitance.
  • Logic level compatibility between every pair of connected parts.

Protection and passives

We check ESD protection on every connector and user-touchable surface, voltage ratings on capacitors (ceramic capacitors lose a large share of their capacitance under DC bias, so a 10 µF part on a 5 V rail may deliver far less), and derating on resistors that dissipate real power. Our post on ESD protection design for small connected devices covers that part of the checklist in more depth.

What we check at the layout stage

At the layout gate we review the board in the layout tool and in 3D, not from PDFs alone. The questions shift from “will it work” to “can it be built, and will it behave”.

  1. Footprints. Every new or custom footprint checked against the manufacturer’s recommended land pattern, with pin 1 orientation confirmed. One wrong footprint is the most common reason for a dead first spin.
  2. Return paths. Fast signals and switching regulator loops routed over unbroken ground, with no splits under high-speed traces.
  3. Switching regulators. Input capacitor close to the IC, the hot loop small, the feedback trace kept away from the switch node.
  4. RF. Antenna keep-out respected on all layers, matching network footprint present even if initially populated with zero-ohm links, and the antenna not shadowed by batteries or metal in the enclosure.
  5. Thermal. Copper and vias under parts that dissipate heat, and hot parts kept away from sensors whose readings they would skew.
  6. Manufacturing. Fiducials, panelisation tabs, test points for end-of-line testing, silkscreen that is readable and not over pads, and component clearances that suit the assembler’s capabilities.

We also run the design rule check with the fabricator’s actual rules loaded, not the tool’s defaults. A clean DRC against the wrong rules proves very little.

Recording findings so they get fixed

A review that ends with a vague list in someone’s notebook is mostly wasted. Every finding goes into a single tracked list with four fields: a reference (sheet and net, or designator), a short description, a severity, and an owner. We use three severities:

SeverityMeaningRule
Must fixThe board will not work, or will fail testing or certificationBlocks the order
Should fixRisk, margin or manufacturability concernFix unless there is a written reason not to
NoteImprovement for a later revisionLogged for the next spin

The designer closes each item with what changed, and a reviewer confirms the change before the files are released. That closing step is where most teams cut corners, and it is where fixes quietly go missing. It also gives you a record of why the board looks the way it does, which matters when someone else picks the project up months later.

Fitting reviews into the build stages

Reviews scale with the maturity of the design. At an early engineering build, the review leans towards function and debug access: extra test points, jumpers and population options are welcome. By the time you are close to production, the focus moves to cost, assembly yield and removing the options you no longer need. We describe how those stages relate in EVT, DVT and PVT: build stages for small hardware teams.

Whatever the stage, keep the checklist as a living document. Every fault that escapes to the bench should add a line to it, so the next review is a little sharper than the last.

What we would do before the next board order

  • Hold two gates: schematic complete, then layout complete, never just one at the end.
  • Put a firmware engineer in the schematic review and an enclosure designer in the layout review.
  • Work from a written checklist organised by subsystem, and tick items against the real files.
  • Verify every new footprint against the datasheet land pattern and pin 1 marking.
  • Run DRC with the fabricator’s rules, and review the board in 3D against the enclosure model.
  • Log every finding with severity and owner, and have a second person confirm each fix before release.
  • Add every escaped fault to the checklist afterwards.

None of this needs special tools. It needs a few hours, the right people and the discipline to close every item before the order button gets pressed.

Working on something similar?Free scoping call →