regulator-output-exceeds-abs-max

Remedy ¶

Reprogram the regulator’s output to a voltage the downstream part is rated for, or move that part to a rail that already is. Both numbers come from vendor documents, so a design change settles this without needing a measurement.

What it checks ¶

A regulator’s datasheet output voltage against the absolute-maximum supply rating of a part it feeds. Both numbers come from vendor documents, and the finding cites both.

It is the first rule that compares a parameter on one part against a parameter on another part across the net joining them. Every datasheet rule before it read one spec against one rail.

For hardware engineers ¶

A regulator makes a rail. Everything on that rail is exposed to whatever the regulator puts out. If a 5V regulator drives a rail and a 3.3V part sits on it, the part is outside its absolute maximum from the moment power comes up. That is outside the envelope the vendor will stand behind, not merely marginal.

Absolute maximum is a stress limit, not an operating range. Exceeding it can destroy the part immediately, or damage it enough to fail months later in the field, and the second costs far more.

The check follows the supply one series element out, so a ferrite bead or a series resistor between the regulator and its load does not hide the connection. It deliberately goes no further. Unlike a surge, voltage does not fall off along a supply path, so a wider walk would make every part on the board look connected to every regulator. The rule would then start pairing parts that share no supply at all.

Why it is not the same as supply-exceeds-abs-max ¶

supply-exceeds-abs-max asks a similar question, but it takes the rail voltage from the net name, where +5V means 5V. That is a naming convention, and a convention can be silent or wrong. A rail named VOUT_A carries no voltage in its name at all, and a rail named +5V that became 3.3V in a respin still reads as 5V.

This rule reads the number off the regulator’s own datasheet instead, asking the same question on evidence rather than convention.

Neither replaces the other. A connector or an unseeded module can feed a rail, and neither has a seeded datasheet. That rail still has only its name to go on, so the older rule keeps its job.

For software engineers ¶

The rule joins two projections, and before it each was read on its own. One is the part’s supply-limit rows (SupplyAbsMaxLimits). The other is the driving part’s output rows (OutputVoltageLimits). Net membership within check.SupplyPathReachHops connects them.

Where several rows are comparable the rule picks the extreme in the endangering direction, meaning the highest output the source can present, against the lowest absolute maximum the load declares. An adjustable regulator endangers its load at the top of its range, and a part is endangered at its weakest rating, so any other pairing would under-report.

The finding carries both citations, so Finding.DatasheetProv is a slice. The review’s data-trust gate rates a finding by its weakest citation. With a single slot, a conclusion drawn half from a low-confidence extraction would still have counted as a hard Fail.

When it does not compare ¶

• The rule needs a seeded datasheet set. It reads the params tier, so check.Available gates it to not-applicable without --params. An unseeded design reads unevaluable, never clean.
• Both parts need a seeded spec. A load with no spec, or with no absolute-maximum supply row, is reported not-considered with the reason. A regulator with no spec is not treated as a source.
• Under-specified rows are not compared. A parameter with text-only conditions or no max bound is skipped rather than coerced (see architecture/datasheet-layer.md). Since agni issue 148, a row printed in a prefixed unit (mV, kV) is not in that category. The parameter layer reduces it to volts through its one conversion table, and only a unit that table does not recognize is skipped.

Fixing a finding ¶

Either the wrong regulator is feeding the rail, or the wrong part is on it. Check which of the two the design intended before changing a value. The rule reports the load as the subject because the load is at risk, but the defect is as often the supply.