← Insights

Design review

What a review finds that DRC misses

ERC and DRC prove a board is legal, not that it's right. Four functional errors that pass every rule check — and why a review is the step that catches them.

2026-11-10 · 6 min · Article

A board can pass every electrical-rule check and every design-rule check and still fail on the first article. ERC and DRC are necessary, and neither of them is the thing that catches a wrong-but-legal design. The gap between a board that is legal and a board that is right is what a review is for.

What the rule checks actually verify

ERC checks that the netlist is internally consistent: electrical conflicts, connectivity faults, pin-type violations. DRC checks that the layout obeys the fab’s geometry and your own rules: clearances, widths, manufacturing breaches. Both are essential, and both answer a narrow question — is this board legal? Neither knows what the circuit was meant to do, which rail a pin should sit on, or whether a value is right for the load.

ERC DRC
Works on netlist / schematic layout geometry
Guarantees electrical & connectivity consistency manufacturability
Can’t know design intent, function whether the right rail, value or domain was chosen
Passes anyway a legal-but-wrong netlist a buildable-but-wrong board

A clean pass is not evidence of a correct design. It is evidence of a legal one.

Four errors that pass both checks

A supply pin that isn’t supplied. On a board with two I²C repeaters, one device’s supply pin was tied to 3.3 V through a 200 kΩ resistor instead of to the rail. Below the part’s UVLO threshold the bus pins sit high-impedance, so two channels were simply dead. Two other repeaters of the same part, on the same board, were wired correctly — so the design knew the right pattern; one instance didn’t follow it. ERC sees a net on the pin and is satisfied; it can’t know that a bias resistor is not a supply.

The wrong voltage domain. A level-sensitive input on a 1.8 V device was driven from a 3.3 V GPIO. The part is over its absolute-maximum rating from the first power-on. Nothing in the netlist is inconsistent and nothing in the layout is illegal, so it ships — and comes back later as intermittent field returns.

An I²C segment with no pull-up. One I²C domain was split into about 60 net segments across two masters, four repeaters and an eight-channel mux. Some segments need an external pull-up and some don’t, because a repeater supplies its own internal current source. Whether a given segment is correct depends on what the devices do, not on whether the copper connects — which is exactly what a connectivity check doesn’t model.

A capacitor with no stated rating. A 16 V ceramic on a 14 V rail with transients is a judgement call; a part with no voltage rating on file is a hole that reads as a pass. On one board, 88 risk-bearing capacitors had no rating recorded. A rule check has nothing to flag, because there is no value to check against.

Why the review is the weak step

Reading a design against what it was supposed to be is different work from running a checker, and in practice it is often the thinnest-resourced step — done from memory, under time pressure, without the datasheets open. That is not a failure of skill; it is a failure of method. A review needs access to the actual design (schematic and BOM before layout, ODB++ after), findings that trace to evidence — a datasheet line, a pin, a net — rather than opinions, and a way to keep what was learned so the next board starts ahead of this one.

Where Probe fits

This is the gap Probe is built for. It reads the netlist, the datasheets and the BOM, traces copper across connectors, and reviews the design against spec — citing each finding to its source and abstaining when the data doesn’t support an answer. It runs self-hosted, so the design never leaves your network. It doesn’t simulate: simulation answers how does this behave, which is the right question only after the first one — is this the board I meant to draw? An independent review of a 14-layer FPGA carrier board surfaced 67 findings of exactly this kind — see the CERN carrier-board review.

ERC and DRC are the foundation, not the finish line. See how Probe reviews a board, or book a 30-minute review with your schematic, BOM and ODB++.

Book a demo of Probe

Show us your board — we review it and go through the findings together.

Prefer to email? info@esharp.se