For me the 2N4401/2N4403 replaced the 2N2222/2N2907 long ago, and I have yet to find an application where they cannot replace the 2N3904/2N3906 as well.
Well, I've made plenty, but in the context of jellybeans, that'll be mostly true. (Though it's not clear how "jelly" you mean your "beans" to be, here.
) 4401/3 are, or at least can be, quite low leakage, and good hFE even at low currents, so you aren't gaining much from the small junction of a 3904 -- just the capacitance, where speed matters.
2N4401/3 themselves go quite fast, I've had sub-20ns edges out of them before, but you have to cook quite hot to get there, 20mA say. 3904 do the same at lower current, of course, and MPSH/MMBTH10/81 even moreso.
My jellybean transistor list would also include a fast low capacitance pair like the KSA1381/KSC3503 or KSA1142/KSC2682. When needed they really have no alternative.
My jellybean high power pair has become the D44VH10/D45VH10 which are ring emitter transistors suitable for fast regulators, audio amplifiers, and such in place of the 2N3055/2N2955.
I would definitely not put CRT drivers in a jellybean category; but if so, then the 3904/4401 difference above is definitely relevant. But in any case, their "beaniness" is just a matter of classification.
Having small-junction drivers for high-performance audio amps, and the occasional special-purpose driver (Idunno, piezo? legacy/vintage projects like CRTs?), is definitely nice, and that's one of the best types.
D44xxx are, surprisingly -- shockingly even? -- old designs, but pretty much as good as any modern Sanken, onsemi, etc. device today. They date to what, the 70s or so? Very underrated parts, it seems like.
Hmm, not sure how general that perception is; the naming scheme I think is from GE? So that dates it pretty far back at least. It may be just that, a naming scheme, with all matter of types there, and only the D44VH10, and close relatives, survived the test of time. *shrug*
As with 2N2222, my gripe with 2N3055 is the trash spec; you don't know what you're going to get. Most parts today will be high performance epitaxials, comparable to MJ15020-something or other for example, but almost anything can fit the spec so you don't really know for sure. You can at least rest somewhat soundly knowing that nobody's making those trash (relatively speaking) homotaxials anymore... but you also have no guarantee you won't be shipped vintage parts perfectly in spec. Safer to just buy the thing you want in the first place.
If it's not an outright counterfeit, of course. People have seen 2N3904-size dies inside the poor things before... But, that's more of a supply chain issue, than a matter of specs.
That leaves an RF pair, a fast saturated switch pair, and a low current pair, but so many parts have been discontinued that they are difficult to select now.
BFT/BFR92 used to be an amazing pair, but the PNP's now long gone (probably still good stocks of it to be found, if you really need to?), and variants of the NPN even I think are obsolete or soon to be. Even MMBTH10 is on the chopping block...
For small signal stuff, the 2N3904 is a fairly decent RF transistor. It's generally better than the other jellybean types in this respect. This assumes it is a genuine 2N3904 made with process 23.
This isn't always a good thing, as the 2N3904 is going to be able to go unstable at higher frequencies than other jellybean parts like the BC547B or 2N2222A. The 2N3904 can easily go unstable up at 800-900MHz and should be able to oscillate above 1GHz. This sets it apart from the others. Some extra care is therefore needed with the PCB layout for some RF amplifier circuits.
Speaking of, did you realize -- actually, you probably do since mentioning it by name, but I don't recall seeing this until relatively recently when I was flipping through databooks. Anyway, 2N3904 is (or, was) gold doped!
http://www.bitsavers.org/components/national/_dataBooks/1971_National_Transistors.pdfWhich should indeed run pretty fast, but should also have poorer hFE at low Ic due to stronger recombination, and I think most modern parts don't do that so much. I wonder if they indeed use gold doping still, or skip it entirely. Or perhaps the dosage is different.
That would explain an observation I saw recentlyish -- not so recently I have the link handy unfortunately, or recall precisely what was in question, but it was something like comparing the storage time of various components, and I was shocked that someone mentioned BC847 has quite long storage; or maybe it was (E-B or C-B) recovery. My measurements of 2N3904 disagreed, but if others are not gold-doped where 3904 is [still], that might explain that in part.
As for RF stuff, much like MOSFETs, there's no jellybean as such -- the market is constantly changing, and no part remains relevant for very long.
There's still the most prominent of HEXFETs, IRF540s and such, IRFPs for higher power, stuff like that. Performance stinks, but not enough to matter -- in relative terms, it's bad, it's about 5x worse than modern equivalents, but in absolute terms, you just want to switch a solenoid or something, hell yeah, it works just fine.
The same technology (HEXFETs, or more generally anything planar, non-trench, non-superjunction) lives on, reborn in boutique form, as "linear" MOSFETs -- at least, as far as I know, this is what Littelfuse (nee IXYS) is doing with such product lines, and, I don't think anyone else is offering "linear" parts as such so it's pretty much their little corner.
Shockingly, a lot of SJ parts boast full SOAs, despite the power density being higher than ever. Not sure how or why, but, hey, I'm not complaining. You might (rightfully) wish to spot-test parts and make sure the SOA isn't lying to you, but the checks I've done, have been promising.
Have even seen some IGBTs with DC SOAs, which... okay, I guess!?
Really curious how they pull that off, or, why?, but, there it is.
But yeah MOSFETs, shop for the Vmax, Rds(on) and Qg you need, and that's about it. See what Pd is available, packaging, go back and forth with a few design iterations, and settle on a combination that works.
Honorable mention for product families, if not parts specifically, or manufacturers: (They all make these types, so take your pick.)
1. At low voltages, pretty much everything is trench something or other.
1a. TI's NexFET, being the most exceptional; they, uh I guess still use a trenching process but not in the active region..! They use a sort of wrap-around structure that acts like a single-piece SJ structure, as well as a cascode. The Crss is *extremely* low, to the point the Miller step is barely perceptible.
2. At high voltages, pretty much everything is SuperJunction, with the "tell" that the Coss(Vds) curve drops precipitously at ~20V. (It's not a planar junction at all, C(V) is allowed to drop much much much faster than (1 - V/VJ)^-m, and indeed can take absurd values like infinite or negative incremental capacitance because of how it's shaped.) This is true even for cheap Taiwanese and Chinese parts, which are offering very good specs these days.
2a. These are available in "fast" and "slow" grades, as far as body diode recovery. Most things, it doesn't matter, but for resonant converters, you might prefer fast recovery. The avalanche rating may vary between these types as well. Compare Vishay E vs. EF series for example.
3. Mainly when you need the speed, but also for efficiency, and higher voltages or more compact designs: consider wide bandgap types. SiC have very attractive ratings, improved speed, and only modestly increased drive voltage requirements; voltage ratings go to 1700V or more (apparently there are 10kV parts, both MOS and IGBT, but you probably have to sign over all sorts of special information to even peek at datasheets, let alone real parts?). GaN are mostly low voltage, with newer families pushing over 600V now; they're
scary fast, so must be used carefully. GaN have no avalanche capability by the way (or, I've yet to see one so rated), so I mean that doubly so.
SJ bears the distinction of having, effectively, dielectric losses; basically as the pillars deplete, they get cut off (they act as opposing JFETs pinching each other off), and charge has to flow through their lengthwise resistance to escape. This is avoidable with very slow edge rates (>100s ns?), but most of the time you just have to accept it. This limits the efficiency of resonant converters (a bigger device has lower conduction losses but proportionally more dielectric losses), so choose different technology if absolute maximum efficiency is required. (As I understand it, SiC are basically ye olde ~200V Si MOSFET designs, ported to a new material; and I mean that literally in terms of feature size and stackup, if obviously not in terms of doping and whatever. So, they aren't SJ, and don't suffer from its loss mechanism. The higher voltage types might.)
Don't be afraid to use SMTs, either; DFN packages have extremely low inductance, and can be heatsunk quite effectively, in combination with wide pours, thermal vias and inner planes. You shouldn't be dissipating much power, anyway; use switching to your advantage, and optimize Rds(on) vs. Coss to suit the application. (Lowest possible Rds(on) is rarely the way to go! That said, as figure-of-merit keeps improving, you can pick quite low-resistance types these days, and still reap improvements.)
Basically, buy some types as needed, and say in a production context, write up some detailed selection rules for how to choose substitutes. Expect EMI changes if Qg, FoM, etc. differs notably, and watch for avalanche, dV/dt etc. where ruggedness matters (not that anyone offers repetitive avalanche ratings anymore, and you shouldn't be making use of it in a real application anyway; get a TVS in there if you must!).
Tim