Externally speaking, for just a switch, you can disregard most of the datasheet.
What you need to find are:
Vceo and/or Vcbo: must be greater than supply voltage (page 2, 40V >> 3.3V is fine),
Ic max: must be greater than the required current load (page 2, 200mA?),
Pd max: must be greater than the required power dissipation (page 2, 250mW; more on this later),
hFE min (at condition ___ ): must be good enough for the application (page 3),
Optionally, switching speed t_(d, r, f, stg), fast enough for application (page 3).
Note that PNP datasheets are sometimes written in consistent units, i.e., Vce is literally voltage from positive collector to negative emitter, current is positive if flowing in, negative if flowing out, etc. So you get a storm of negative values. Sometimes you get inconsistently signed values.
Personally, I don't give a shit, magnitude is fine by me. I already know what direction it's supposed to be. That's what the little arrow says. If you've got 20V across an NPN or a PNP, you don't have to guess which direction you're doing it in -- the presence or absence of magic smoke will tell you at a glance!
Ok, switch. Using the "current controlled" model is fine here (though don't lose sight that B-E voltage really is the thing that matters, we're just using an excess of controlled current to get there safely!). So, we need Ic / hFE base current. Or, we can get up to Ic = Ib * hFE current out of it. Ah, but that's linear. hFE drops in saturation. How much is safe? Depends. You get lower saturation for lower hFE (= higher Ib for the same Ic -- when it's saturated, we're in charge of what "hFE" can be; as a device parameter, hFE is only valid in the linear range, as mentioned).
Typically, a design goes like this:
- Pick a transistor with min hFE 100 or so (it'll actually be up to 300 at higher temperature, maybe 500+ with manufacturing tolerance included -- many types are binned by rough hFE range, giving you some control over this if necessary/desirable)
- Pick some safety margin for saturation (usually 3 to 10 x)
- Apply that base current (if from a logic voltage or whatever, use a current limiting resistor)
Easy.
The safety margin works thus: a transistor with typical hFE = 200 might saturate quite nicely (Vce(sat) < 0.5V)) at hFE = 100 (i.e., if we're switching 10mA, use 0.1mA Ib). But it won't be very good, because at low temperature, hFE drops and it might come out of saturation. (On the upside, the increased power dissipation might heat it up enough to resume a near-saturated condition?) A factor of 3 is pretty good insurance there, so you might use 0.3mA Ib instead. Now it's pretty well saturated (Vce(sat) < 0.2V, say?), and pretty well guaranteed to remain so, even in the friggin' arctic, with floor-sweepings quality transistors. But maybe that's still not good enough, maybe the load is 10mA normally, but spikes to 30-100mA! We can keep it nice and solidly saturated under most of those conditions, by raising Ib still more. And the highest peaks might be dubious, but we can give them a good try. So such a switch might be driven at Ib = 1mA instead. It's not usually a big cost to draw more base current, so this is OK, and it makes it stupidly robust*.
*Except that, if the load is prone to shorting as well, it'll pull the collector voltage up from Vce(sat), moving into the linear range. Where it will draw 100-200mA. At 3.3V supply, a short can be expected to cause 3.3V * 0.2A = 0.66W dissipation, which exceeds the continuous 0.25W rating. It won't burn out immediately, but it won't be happy after a second or three. Because of cases like this, it can be desirable to keep base current more modest, thus limiting the short-circuit current (at least crudely).
Traditionally, transistors are rated for saturation at this condition, hFE = 10. Not sure why it's so low, really, because 30 would be comfortable for most, but, I guess it comes from the olden days when transistors were truly all over the place. Always take note of the condition, though: high voltage power transistors will be rated for hFE as low as 5, or even 2; high gain (often "low Vce(sat)" or "superbeta" types) might be specified for hFE = 80 or more.
Anyway, page 3 shows Vce(sat) at hFE = 10 (Ic = 10 or 50, Ib = 1 or 5), and it's under 400mV.
Now... if you were planning on drawing more than 50mA Ic, you should probably consider a different transistor. They give the highest rating for 50mA, and the maximum is only four times this. Double, 100mA, would still be okay -- they don't guarantee parameters at this condition, but we can at least guess that it won't be too bad. Referring to the hFE plot on page 4, it's clear that hFE is dropping sharply in the 50-100mA range. But the 25C curve crosses 100mA at around 70 hFE, which still isn't too bad. So it will probably be quite well saturated at the hFE = 10 condition, even at 100mA. (Note that all the curves are pretty well tanked by 200mA, and don't even show data beyond Ic max.)
Now, if you need to handle more than 100mA, like switching USB power (as your drawing seems to suggest..?), that's completely unreasonable, and you'll need a better one.
BTW, Ic(max) for BJTs is usually a physics limitation. Current crowding and all that. It simply can't carry any more current (hFE drops)! As a switch, or in pulsed operation, it's not at all unreasonable to operate at 25-50% of Ic max. You'll probably cook it if you try doing that in the linear range (i.e., Vce > Vce(sat)), but that's a thermal problem.
On a related subject:
MOSFET ratings are usually thermally driven, so beware: a D/D2PAK device will typically be rated for Id(max), DC, at *package* thermal dissipation (50-200W!), which is one or two orders of magnitude off the typical PCB-mounted capability (~1-5W?). In normal use, an "on" MOSFET can be assumed to act more-or-less like a resistor, so dissipation can be calculated accordingly. The MOSFET *physics* limit is much higher than the continuous rating; this is more-or-less captured by the Id(pulsed) rating instead. MOSFETs have a constant-current characteristic, like BJTs, but "below the knee" (where it looks like Rds(on) resistance) goes up to much higher voltages, usually ~10V or so. The point where the highest Vgs curve levels off, is the point if Id(pulsed). So the rating is essentially, "if you want to blow up transistors, this is how much shit you're gonna see flying". At lower currents (and voltages), it can simply be treated as a resistor.
(On a related related subject: BJTs in saturation kind of act like resistors, too. They're not nearly as linear, which means they stink for "audio mute" type switches. You can still do calculations, assuming a BJT has an "on resistance" -- but it's not usually as useful, because you still need to know Ib anyway.)
Tim