reverse-blocking-absent

A connector feeds a power input with no directional element blocking reverse flow.

Remedy

Add a directional element between the connector and the load: a series FET where the voltage drop matters, a diode where it does not, or a bridge where the input polarity is genuinely unknown.

What it checks

A connector feeds a power-input pin, and nothing on the path between them blocks current flowing the wrong way.

For hardware engineers

Two failures share one mechanism.

In reverse polarity, a connector is wired backwards, or a technician plugs in a supply the wrong way round. Current flows into the board through what should be its return path. On a vehicle this is a qualification requirement. ISO 16750-2 makes reverse voltage a test the module has to survive.

Reverse current (backfeed) happens when two sources share a rail, or a rail is switched off while something downstream still holds charge. Current flows back up a path that was only ever designed to carry it forward, powering a domain that is supposed to be dead. An industrial board carries an ORing FET or an ideal-diode controller to stop it.

Both need a directional element, something that conducts one way and not the other, such as a series diode, an ORing FET or an ideal-diode controller.

A connector-to-power-input path through a series diode is fine; a path with nothing directional on it is flagged; a path crossing an unidentified transistor is inconclusive

The three panels are the rule’s three answers. The middle one is the defect. The right-hand one needs explaining, and the next two sections say why the rule reports it as neither of the other two.

Why a fuse or a TVS does not count

The rule exists separately from input-protection because neither a fuse nor a TVS blocks reverse flow.

A fuse opens on current magnitude. It does not care about sign, so it will not stop reverse current until the reverse current is large enough to blow it, by which point the damage is upstream of the fuse.

A TVS shunts transients to ground. It clamps a voltage spike; it does not block a path.

So “a fuse or a TVS is present” says nothing about whether reverse flow is blocked. On a real design, two review items were bound to input-protection for this reason. They read pass, and nothing tested what they actually asked.

What it will not claim, and why that is deliberate

A path crossing an unidentified transistor is reported INCONCLUSIVE, not as a defect.

A P-FET ideal diode is a transistor plus a bias network. Nothing in a netlist labels that arrangement, and it is structurally indistinguishable from any other FET sitting in a power path. It is also the correct modern answer to reverse protection. A rule that fired on every path with no series diode would false-fail every well-designed ORing-FET board.

A false fail here is worse than the gap it would close. A reviewer who sees this rule fire on a properly protected design learns to ignore it, and then it is worth nothing on the design that really is missing protection.

It used to report nothing instead, and that was wrong for a reason the rule layer cannot see. A review item bound to this rule read silence as a PASS, so the report asserted protection on a path nothing had verified. The rule now says out loud that it could not decide, and a bound item reads inconclusive, which is neither a defect nor a clean bill of health.

A controller the datasheet identifies resolves it. Seed the part with a device_class of ideal_diode_controller, and the rule credits it as a directional element and passes the path. The alias set also accepts the ORing, power-mux and power-path spellings. Seeding the part turns an inconclusive into a real pass, so the message names the exact class to seed.

Orientation matters

A diode only counts when it is fitted the right way round: anode toward the source, cathode toward the load. Fitted the other way it blocks the supply rather than the fault, and the rule reports the path as unblocked.

Only a plain diode counts. A TVS, a Zener and an LED all carry the diode family tag, and none is a series blocking element, since the first two shunt to ground and an LED in a power path is an indicator. A diode whose part type declares no anode pin has unknown orientation, and the rule reports its path as unblocked rather than guessing which way the part faces.

When it passes or does not apply

  • No connector on the net, since the rule is about what enters the board.
  • No power input reachable from it. A connector’s own pins are not loads. A receptacle’s symbol types VBUS power_in and the rule still skips it, so a source port driven by a load switch reaches none (agni issue 935). A clamp array’s own supply pin is not a load either, since the part guards the rail it is biased from (agni issue 1257).
  • A transistor touching the path only through its gate, such as a FET sensing VBUS for presence detection, is not in the path and makes nothing inconclusive.
  • A transistor on the path, when a seeded datasheet identifies it as an ideal-diode / ORing controller (above). An UNIDENTIFIED transistor gives an inconclusive finding instead.
  • A current-limiting power switch as the entry’s only load, when its datasheet names it one and specifies a reverse current. A switch states a reverse current only when it blocks one (agni issue 1248). The pass cites that row and says the switch stops backfeed but not a reversed supply, since such a switch is rated to a few hundred millivolts below ground at its input. On a USB-C port, where VBUS cannot be plugged in backwards, backfeed is the half that matters. Without the datasheet the entry still fails, and the verdict names the IC it could not read.
  • Ground and read-gap (external) nets, excluded up front.
  • A net the board drives itself is not-considered, since its connector is an outlet rather than a way power arrives (agni issue 1016). A converter’s output counts, and the rule reads one as a switch node behind an inductor plus a feedback net within the walk. A power-output pin on a part other than a connector counts too. A fixed-output buck names neither net, so one part’s switch pin behind the inductor and its feedback pin on the walk count as well. The walk crosses a fitted net tie, since a net tie is a short (agni issue 1253). A power-output pin does not count when the walk also finds the input of another part with a power output. That shape is a dual-role USB-C port. A load switch sources VBUS and a PD controller’s power path sinks it, so power can still arrive and the rule still judges the net. The pin counts again when the driver’s own input is wired to a net those parts do not feed, because that shape is a cascade of regulators. A dual-role port whose sink reaches the driver only through a further converter reads as that cascade, and a design with that shape loses this rule on the net. A boost converter’s input reaches its switch node through the inductor too. The rule still judges that input, because its feedback divider sits on the output side.

Fixing a finding

Add the directional element, or record why the path does not need one. A path is a legitimate exception when its source physically cannot be reversed, such as a fixed internal rail rather than a user-facing connector. Write the exception down anyway, because the next reviewer will ask the same question.