A current rating is the lower of two computed numbers

Both limits are calculated under conditions you will never hit, and the published rating is whichever comes out lower. Neither is a measured value.

The thermal limit asks how much current raises the junction to its maximum temperature, assuming an ideal heatsink holds the case at 25°C. In the standard form it is the square root of the temperature rise divided by the junction-to-case thermal resistance and the hot on-resistance.

Because it depends on resistance and thermal impedance, it behaves like a figure of merit. A die with lower on-resistance shows a higher thermal limit, and the number falls as the case temperature rises.

The package limit comes from somewhere else entirely: the bond wires, the lead frame, or the copper clip that carries current from the die to the pins. It is largely independent of temperature.

Exceeding it does not destroy the device immediately. The documented long-term mechanisms are wire fusion, thermal degradation of the molding compound, and electromigration — a reliability boundary rather than an instantaneous one.

The published rating is the lower of the two, which is why a part can be package limited while its die is capable of far more. In that Toshiba example the silicon could carry 340A, but the package capped the printed number at 150A.

The same pattern appears across the industry: a 60V TO-220 part published at 200A is package limited while its silicon limit is closer to 350A.

Three tells that say which limit you are reading

You can usually identify the limit in under a minute, without any calculation at all. Three features on the datasheet give it away.

Tell one: does the number change at 100°C? A thermally limited rating falls with case temperature, because the whole computation is about how much heat the die can pass. A package limited rating holds steady, because bond wires do not care how hot the case is.

In one 40V family, a 290A label reads 205A at a 100°C case, while a 160A label in the same footprint reads 160A at both temperatures.

Tell two: does the sheet say so? Some manufacturers print a footnote such as current limited by package directly under the ratings table. When they do, take them at their word.

Tell three: is there a board thermal figure? A sheet that publishes a junction-to-PCB or junction-to-ambient number is telling you the part is meant to be evaluated in an application.

A sheet that gives only a junction-to-case figure — 0.48°C/W, in one case here — is leaving the board calculation to you, while the competitor next to it prints 35°C/W on 1in² of 2oz copper.

Every part in that table is a legitimate 40V MOSFET in a surface-mount power package. The two D2PAK parts hold their number because the package is the binding constraint; the others drop because the die is.

Bars are scaled to each part's own 25°C rating, so the bar length shows retention rather than absolute current.

Where the two limits come apart

When a family spans a wide current range in one package class, the thermal limit eventually stops climbing and the labels keep going. Two examples from the same 40V ladder show what that looks like.

The first is a pair of rungs that print identical thermal numbers: 310W dissipation, 2.07 W/°C derating, 0.48°C/W junction-to-case, 840A pulsed current and 2500mJ avalanche energy. One is labelled 210A and the other 290A.

Both dissipation figures reproduce from the same relation — 2.07 W/°C across 150°C is 310W, and the temperature rise over the thermal resistance is 312W. The package budget was fully used at the lower label.

The second is the body diode. Four rungs of that ladder rate the diode at 100% of the channel current. The 290A rung rates it at 210A — 72%.

The diode is the same silicon and the same path as the channel, so equal ratings are the norm. The reason for the exception is dissipation: diode conduction loses current times forward voltage, and that voltage has a 1.2V maximum on this sheet.

At 210A that is 252W, inside the 310W budget. At the full 290A it would be 348W, outside it. The printed diode figure is a power limit wearing a current label.

Read the diode row as a power statement and the exception stops being a mystery. It also tells you where it matters: synchronous rectification and freewheel duty, where the diode conducts for a real fraction of the cycle rather than a transient.

The condition columns move under your feet

Even after you know which limit applies, the numbers underneath it are measured under conditions that change from sheet to sheet. Four of them move often enough to check every time.

On-resistance test current. Every sheet quotes a maximum on-resistance at a stated drain current, and that current is not the headline rating. Across one 40V family it appears at 20A, 40A, 80A or 10A depending on the sheet.

Where two rows on the same sheet use different currents, the low-gate row is almost always tested at the lighter one, which flatters it.

Avalanche test inductance and rail. Avalanche energy is measured into an inductor, and the stored energy depends on the inductance value. One family mixes 1mH and 0.5mH, and one sheet tests at the full rated rail rather than half of it.

The conversion back to a current needs the right circuit form, and the more conservative form gives roughly half the value of the simple one when the supply is close to the breakdown voltage.

Switching test conditions. Turn-off delay and fall time depend on the test rail, the load and the gate resistance. Two sheets tested at a 20V rail with 3Ω and two at 30V with 2.5Ω are not printing comparable nanoseconds.

The reference temperature in the ratings header. Some sheets head their absolute maximum ratings at a 25°C case, others at a 25°C ambient. Those are not the same condition, and the difference is routinely a factor of thirty.

None of this is hidden or dishonest — it is all printed, just not where you look first. The condition column is the part of a datasheet row that carries the actual information; the numeric column only means something once you have read the condition beside it.

When to choose by current rating, and when not to

Choose by current rating only when the two candidates share a package, a thermal path and a stated basis. That is a narrow case, and it usually means two parts from the same family.

Choose by on-resistance when the application is unknown. That is the standard cross-reference guidance, and the parts in this article show why: their labels run 80A to 290A while their maximum resistances run 2.0 to 5.0mΩ in almost the opposite order.

Do not choose by current rating when the packages differ. A package-limited part and a thermally limited part can print the same number and mean entirely different things by it.

Do not choose by current rating when the duty is a body diode. Rate that path by the power it will dissipate, not by the current printed next to it.

And do not choose by current rating when the case will be hot. Re-read the 100°C row, where the ordering changes shape: an 80A label against a 290A label is a 3.6× gap at 25°C and a 2.6× gap at 100°C.

The 290A part also carries 239nC of gate charge against the 80A part's 117nC, so the larger number costs drive current as well as board area.

Frequently Asked Questions

Q1: Is a MOSFET's current rating a measured value?

A: No — it is calculated, and the published figure is the lower of two computed limits. One limit comes from the die's thermal path assuming an ideal heatsink; the other comes from the package's internal connections. Neither is tested on every part.

Q2: Why does one part's rating drop at 100°C and another's stay flat?

A: Because they are limited by different things. A thermally limited rating falls with case temperature since it is computed from the die's heat path. A package limited rating holds steady because bond wires and lead frames do not change with case temperature.

Q3: Can I compare current ratings between two manufacturers?

A: Only if the package, the thermal path and the stated basis all match. Different manufacturers pick different limits, use different thermal resistance tolerances, and define steady state differently — variations large enough to move the printed current by a fifth between essentially identical products.

Q4: What should I compare instead?

A: On-resistance at a stated gate drive, plus the thermal path. On-resistance is the usual cross-reference metric when the application is unknown, and it is more meaningful than the current label. For switching converters, folding it together with gate charge gives a better loss picture.

Q5: Does exceeding the package current limit damage the part?

A: Not immediately. The documented long-term failure mechanisms are wire fusion, thermal degradation of the molding compound and electromigration. It is a reliability boundary rather than an instantaneous failure threshold, which is why it does not announce itself on the bench.

Q6: Why is the body diode sometimes rated below the channel?

A: Because diode conduction dissipates current times forward voltage, not current squared times resistance. With a forward voltage around a volt, the same current costs far more power through the diode than through the channel, so the printed diode figure is a dissipation limit expressed as a current.