load-switch-trip-above-fet-rating

A controller-based load switch trips above the continuous drain rating of its external MOSFET.

Remedy

Lower the switch’s current-limit setting below the FET’s continuous drain rating, or fit a FET rated above the trip point. As drawn, the limit protects nothing.

What it means

A load switch is built from a switch controller, an external MOSFET and an external sense resistor. The controller trips at a current higher than the MOSFET’s continuous rating.

Both numbers are vendor values and the finding cites the page each came from. The third input, the sense resistance, comes from the design, because no datasheet knows it.

For hardware engineers

An integrated load switch is one part. Its current limit and its pass element are specified together by one vendor, in one document, and they already agree.

A controller-based switch is three parts chosen by three separate decisions:

  • The controller states a sense threshold, a voltage such as 50mV. That is all it states. It has no current limit of its own, because it has no idea what shunt you will put in front of it.
  • The sense resistor turns that threshold into a current. ITRIP = V(OCP) / Rsns. A 50mV threshold across 10mΩ trips at 5A; across 50mΩ it trips at 1A. Same controller, five times the current.
  • The MOSFET carries that current, and has its own continuous drain rating.

Nothing in the design flow checks that the third number is bigger than the first two put together. When it is not, the current limit does nothing useful. The FET reaches its own rating while the controller is still comfortably below its trip point. The protection existed for the FET, and the FET dies. A high-side MOSFET that dies usually dies short, and then the whole rail reaches the load it was switching.

The effective on-resistance of such a switch is likewise the external FET’s RDS(on), not any number on the controller’s sheet. The finding reports it, because the reviewer’s next question is “what does the FET dissipate at the trip current”.

What derating this does and does not cover

The rule compares against the FET’s printed rating, which is a 25C figure. Real designs size well under it. A trip point that sits below the printed rating therefore passes here and may still be badly derated.

The rule checks the unambiguous half on purpose. “The current limit is set above the vendor’s own number” is a defect nobody argues with. “The current limit is set at 80% of the vendor’s number and the part is in still air behind a connector” is a thermal argument, and this rule does not attempt it.

How the three parts are found

  • The FET is a transistor whose gate terminal is on exactly one net. The role comes from the naming lexicon’s role vocabulary, so a house that calls its gate DRV declares that in --conventions rather than patching the engine.
  • The controller is the single non-transistor part on that gate net whose datasheet states an overcurrent threshold. The rule identifies it by what its sheet says rather than by a device-class keyword, so a part that declares a sense threshold is a current-limiting controller, whatever its description string reads. This also settles integrated-versus-controller for free, since an integrated switch has no external gate net to be found from.
  • The sense resistor is the single resistor where the controller also touches every one of its nets. This is Kelvin sensing, expressed structurally. Two dedicated pins measure a shunt, one on each of its two terminals. A series part that merely sits nearby has a far terminal, and the controller does not touch it.

A feedback or programming divider between a controller’s output and its sense pin shares that structural signature. Their values differ by orders of magnitude. A shunt drops tens of millivolts at amperes, so it is milliohm-class. A divider must not waste current, so it is kilohms. One ohm sits in the empty middle, so the resolver accepts only a resistance of 1Ω or less.

When it reports no finding

No finding always means “I could not tell”, never “this is fine”.

  • With no seeded datasheet set, check.Available gates the rule to not-applicable without --params.
  • The gate net carries two candidate controllers, or the controller has two candidate shunts. Unresolvable, so no verdict rather than a verdict computed from the wrong part.
  • The FET’s gate is on more than one net, or it declares no gate pin at all.
  • The controller states no overcurrent threshold, or states it in a unit the parameter layer does not recognize. Real controller sheets print this row in millivolts, so a millivolt row is reduced to volts and compared. A unit with no entry in the conversion table is skipped rather than scaled by a guess, so the rule still fails toward silence instead of toward a current a thousand times too large.
  • The FET is unseeded or states no continuous drain rating. The FET then reports not-considered with that reason. Pulsed drain current is deliberately not accepted in its place.
  • The shunt’s value is not stated in ohms in the design. The reader may never have normalized a component’s value, or normalized it without a unit. Either way the value is no evidence of a milliohm shunt. This is the live limit on formats. The loader’s value pass (classify.StampValues) recognizes only a few property spellings (value, val, partvalue, component_value). An EDIF exporter may spell the property otherwise, and then the design carries no readable number and the rule reports nothing there.
  • A zero-ohm shunt is skipped, because dividing by it gives infinity, which every comparison downstream would read as an enormous current and report as a defect.

For software readers

The rule joins two things that had no prior arithmetic between them. check.ExternalFetLoadSwitches resolves the topology and the trip current; the rule adds the rating comparison.

The trip current is the first calculation in the engine to cross two units (WS3-085). It goes through check.OhmsLawCurrent(volts, ohms) (amps, ok). That function is a named physical operation rather than a general dimension algebra. Every other consumer compares within one unit, and the accessors already gate that, so a dimension system would be a lot of machinery with no second caller. The signature states the unit contract, and ok is false for a non-positive or non-finite resistance.

DatasheetProv carries exactly two citations, the FET’s rating first (the endangered part) and the controller’s threshold second. The on-resistance is quoted in the message with an inline citation but is not in DatasheetProv. The verdict does not depend on it, and the review’s data-trust gate rates a finding by its weakest citation.

Fixing a finding

Three levers, and they are not equivalent:

  1. Raise the shunt’s resistance to bring the trip current down. Cheapest, and it costs a little more dissipation in the shunt.
  2. Choose a FET with a higher continuous rating. Right answer when the load genuinely needs the current.
  3. Choose a controller with a lower threshold. Rarely the lever, since the threshold is usually fixed by the part.

Check which of the three numbers was the intended one before changing any of them. The rule reports the FET because the trip current exceeds the FET’s rating, not because the FET is necessarily the wrong part.