I'm looking at this MOSFET for example, the IRLS3034, which according to its datasheet, can handle a continuous current of 343A.
The wire lead looks like it's about 1mm wide. It looks like it will melt at anything over 50A and instantly vaporize at anything close to 200A -- and this is not prevented by a heat sink.
What is going on here?
This is more of a "fusing" question, so I'll give a "fusing" answer:
Fusing current is quite high for short durations. A 1mm copper wire (about 18 AWG) claims 83A for 10s, 250A for 1s, etc. (See: https://en.wikipedia.org/wiki/American_wire_gauge )
Wire ratings assume long runs, so that heat is only flowing laterally out from the wire, not along its length. This suggests the existence of a critical length, such that if we increase the length further, the rating (for temperature rise or fusing) does not change significantly. That is, d(rating) / d(length) ≈ 0.
If we have a very short run, considerable heat flows along the length of the conductor, into neighboring heatsinking materials: the transistor package (and indeed the die itself, and heatsink; these transistors use quite generously sized bondwires, or even clips), wide pours on the PCB, and neighboring components. Thus, the conductor can dissipate far higher power density, compared to a long run where its only dissipation is through convection to air (or including conduction along a PCB, when considering PCB trace ratings). Put another way: we are quite far from the d(rating) / d(length) ≈ 0 case, and in particular, this derivative is strongly negative at small values (short lengths ↔ high currents).
As others have mentioned, there is also an element of notoriety, as IR (now Infineon), among others, used (and, I suppose still use) rather optimistic rating schemes to drive marketing of their components.[1]
The other one to watch out for is power dissipation, often measured by immersion in nucleated-boiling refrigerant at the specified temperature (when they say TC (case temperature) = 25°C, they mean every point on the case..!).
(I think the engineering "gimme" in these datasheets, is the RΘCS parameter, suspiciously large in most cases -- actual resistance of this interface as-such is usually more like 0.1-0.15 K/W for a TO-220 for example, but they typically publish values around 0.5. But it's hard to know for sure, and best practice is to build a setup to measure junction temperature.)
Personally, I wouldn't use this component much above 50A or so. There are better packages available, anyway: surface-mounted DFN types have no lead length at all, spreading out current and heat immediately into PCB pours and neighboring components; thermal pads can sink several watts even from the smaller types. Not that this answers the question, of course, but as you can surmise, sometimes good engineering means evading certain question entirely.
[1] It may be worth adding some discussion on marketing, here.
Consider the use-case of an engineer looking for a part that can handle, say, 20 or 30 A comfortably, with little heatsinking. Well, the best / most responsible way to do that, is to sort by RDS(on) -- you know you need, whatever, 1 or 2 mΩ, and add in a safety factor for max plus tempco, and there's your selection. But one might also sort by nominal current rating. Or perhaps one is even searching in a catalog/database that doesn't have RDS(on) listed at all(!), or their data are inaccurate(!!).
For example, see that ye olde IRFZ46N is rated for 37/53A (depending on temperature), but the power dissipation would be unacceptably high (~50W), so let's choose something, oh let's say, \$\sqrt{50}\$ times higher rating, to get that down to a comfortable 1W or so. (Underlying assumption: RDS(on) scales inversely with sqrt(Imax) for given PD. It'll actually be better than this, because with derating, we [probably] aren't running at TJ(max). Or, we still can [run up to TJ(max)], without hitting any new limits.)
So we enter something 250-300A in the search query, and see what drops out.
Now, consider this from a marketing perspective: if you specified all your low-RDS(on) transistors as package-limited, your catalog has a hard cap at, whatever, 100, 200, 300A. Those particular values are highly repetitive, and the value loses meaning with respect to any other capabilities of the device. There is no differentiation among parts anymore, and no meaningful selection is possible by this metric.
Put another way: you've introduced a nonlinearity into an otherwise linear (or, at least one-to-one) process, and consternation ensues.
So, as said above -- it's probably better/safer to be sorting by RDS(on) in the first place; but it seems a needless complication to force this on others accustomed to searching by current rating.
It would be irresponsible, of course, to use just that rating, alone; so they add extra notes and parameters, explaining what it means, and specifying what you can really do. And, it's not like you're going to push anywhere near that in practice anyway (who even direct-heatsink-mounts DPAKs?*), so it's mostly not an issue. It's just more wrinkles in something you might otherwise hope is more straightforward; but such is life.
*It can be done with heavy copper PCB, including metal core and machined-insert types. Veeery specialized, but... they do exist(!). On a more consumer level, aluminum core single-layer PCBs are quite affordable, and have quite excellent thermal performance; the downside is the single layer, which makes SMPS layout tricky, for example. Or you can order them with multiple layers, even plated thru-holes (isolated from the core(!)), but at added cost, or lesser availability.
Marketing is full of weird edge cases, human habits, and psychology -- for better, and for worse. This particular rating can seem skeevy at face value, but at the same time, it isn't completely useless, and retains a certain honesty to it (or, bragging rights, under another perspective); just maybe not in the way one might hope [continuous, actual, in-circuit rating]. The world is full of trade-offs and "package deals", and we must be alert to watch for them, and curious and critical enough to dig in and see what's really going on. There's no need to assume malice; there may well be an underlying explanation -- even if not a good one, or if not obvious to us as end users or what have you.
On that note, I wish manufacturers were more open about what ratings and standards they follow, in terms of internal standards, test procedures/conditions, or industry standards. The latter are entirely closed (or at least, I haven't seen anything leaked), so we can at best only infer their existence from what manufacturers do publish; or perhaps related standards (MIL-STDs for example, are open). The relevant document is... ugh why is it so hard to search for these things anymore... here's a newer Infineon explainer:
Datasheet Explanation Update for Infineon’s Automotive MOSFETs | Infineon
but it's their own branded parts, not legacy IR (or current products under the same development house, or branding).
...Aha, finally found it:
Continuous dc Current Ratings of International
Rectifier’s Large Semiconductor Packages | Application Note AN-1140 | International Rectifier
This is the contemporary reference that applies to the part/family/ratings in question.
Mind, I'm no marketer, nor have I worked for a semi company; these are just my observations and assertions, and shouldn't be taken too seriously/literally (apply critical thought, recursively, as always!).
(Anyone who has [worked], comments are welcome!)
Clearly some impractical conditions are being imposed, nicely described by Tim Williams. Here is an approach to estimating a more practical limit. They provide the Junction-to-Ambient thermal resistance \$R_{\theta JA}\$ as 40 \$^{\circ}C/W\$. First, note that this is under very specific mounting conditions described in an application note they refer to (AN-994), which is using a one square inch PCB pad, 2 Oz copper. I would take this parameter to be optimistic, and consider doubling it for conservative design purposes. But let's suppose you did the thermal design very carefully and actually achieved their 40 C/W number, just for the sake of calculation.
If ambient temperature is 25 \$^{\circ}C\$, and you allow the maximum junction temperature of 175 \$^{\circ}C\$, then you can have about 150\$^{\circ}C\$ temperature rise at the junction, and therefore dissipate about 150/40 = 3.75 W. A graph is presented on the datasheet of the normalized \$R_{DS(On)}\$ vs junction temperature, and we can see that at the maximum junction temperature of 175 \$^{\circ}C\$ this resistance rises by a factor of approximately 1.8 relative to 25 \$^{\circ}C\$. They quote \$R_{DS(On)}\$ of 1.7 m\$\Omega\$ for a 10 V gate-source voltage, for a pulse run at 2% duty cycle, so let's be conservative and suppose the junction was at 25 \$^{\circ}C\$ for the datasheet number, and scale it up by 1.8, to get an \$R_{DS(On)}\$ of about 3 m\$\Omega\$.
Then we can compute the maximum current from \$I_{max}^2R_{DS(On)}=3.75 W\$, which gives us about \$I_{max}=35 A\$.
Again, the thermal resistance they quote may be optimistic, so I would consider derating this further.
They're being kind of dishonest. A bit, anyway.
That 343A figure in Absolute Maximum Ratings is stated as 'Silicon Limited'. Look at the fine print in Note 1:
From the data sheet.
Got to watch out for those marketing folks...
Package limit is 195A, and that is with the case held to 25°C, a practical impossibility in most situations.
The lead length is short and we don't know what they are assuming for the termination of the leads, perhaps another infinite heat sink.
Typically "package limit" for TO220 tends to be given as more like 75A, not sure if that is a better package or just more optimistic specs.
In any case, ~200A continuous is not going to happen in most any real life situation.
