Hot swap needs care and attention to work well, yeah. And not just in the protection. I've never really had to do a serious hot swap design so I'll leave that to others.
(The rest of this might be longer than you are looking for, but it took younger me longer to work out than I care to admit. So I'm writing down something so that it will hopefully take others less time than it took me.)
Back to the protection devices... and first, a bit about TVS parameters. You specify a TVS that starts conducting just above the maximum voltage you want to have on the line. Note that TVS diodes are not precision devices, so you can simply never use them as precision clamps. Sometimes you can't even use them as suitable protection devices, period, because their trigger voltage and their clamp level are too far apart. These cases are uncommon, in my experience, but very real, and usually involve DC or repetitive crud rather than simple ESD.
The manufacturer gives a "max reverse working voltage" V_rwm and if you read that as "use on lines of this nominal voltage" you won't go too far wrong. (V_br is actually defined as "highest applied voltage that gives a current through the diode of some given value or less".) But remember that this is
not a fundamental characteristic of the part, it's a characterization, and sometimes exceeding V_rwm might be a perfectly good thing to do. For example, I once put a 5V TVS across a 5.5V power rail because I didn't care about a potential 1mA of leakage there.
The breakdown voltage V_br is usually the key specification of the part, and is the measured voltage to achieve a given current through the diode. You'll see V_br as minimum, maximum, typical, or some combination of those; min and max are most useful, in that order. This specifies the "knee" of the part, or when it actually starts to turn on. The reason I knew the 5V TVS I mentioned would be OK at 5.5V is that its minimum V_br was 1mA at 5.5V, and that was acceptable to me. If it was 100mA at 5.5V, I could not have used that part. V_rwm didn't really matter to me, V_br did. In other circuits this will be reversed. But that's what they mean.
Another thing to remember is that TVS diodes and zener diodes are (usually*) the exact same thing. It's just that they are optimized for different things (peak power vs continuous power/stability/binnability). You can swap one for another in a pinch, and there are several series that are rated for use as both. (Like my favorite SOD323 zeners,
Vishay's PLZ series!) You'll do
better with the correctly-optimized part, but don't let dogma stop you from using the good-enough choice.
Except... I said "usually*", with that ugly asterisk thing. Zener diodes are crap below,
very roughly, five volts. And people like to protect data lines on 3.3V and 1.8V and lower parts, but you can't always get the performance you want out of zeners. So manufacturers started offering thyristor designs instead (aka SCRs aka "snapback diodes" aka I think there's others I'm forgetting). These are
just fine if you (1) recognize you're buying a thyristor (2) recognize you're not buying a zener and you shouldn't treat it like a zener and (3) confirm that the snapback release point is going to be OK in your application and you're not going to get stuck in SCR latchup. (#3 is the important one. It's uncommon, but possible, especially if you're not aware to look for it.) Snapback diodes are rare above 5V, the definite minority for 5V TVS parts, common at 3.3V, and the majority below 3.3V. If not otherwise clearly indicated, the best way to recognize a snapback part is from the V-I curve having that characteristic SCR shape to it. If they bother to give you the curve. Now that I've mentioned that snapbacks are a thing, I'm going to forget about them, because that makes life easier.
So. Example time. Let's say I have a TTL (5V) UART leaving the board on a connector and I decide a TVS (per line) is an appropriate filter. (I might want more if the cable is long or the environment harsh, but usually, I consider a single TVS to be enough.) You like SOD-323, I like SOD-323 (though I'm going with smaller more and more often these days...) so let's search the catalog and come up with
STS321050U182! (Also, it was in my library because I've used it before.) In my opinion this is a good part for this job. The 180pF is not great for a data line, but this is a UART and UARTs are slow so who cares. (Substitute actual analysis as required. Especially if you're doing more than writing a forum post.) We see it's rated to take a 30kV hit (that's a great rating, they don't really come any higher), and has a 350W 8/20us peak power rating (you can do better here but at the cost of worse C or being a snapback, so this is also very good). It conducts max 1uA at 5V and hits 1mA no earlier than 6V. It's unidirectional rather than bidirectional, which is appropriate because there are no negative voltages in our design. (Bidirectional devices are almost never appropriate unless you have a negative rail. Almost. So prefer to exclude them by default in your searches.) So far, so good. This is really about what it looks like to protect a data line with a TVS diode.
To finally start addressing your first question.... You see that I have V_rwm at 5V and my signal voltage is 5V. So, no, those aren't usually very different. Because TVS clamping is so coarse, I usually need the diode to kick in as soon as it can, and so the only reason I'd go for a V_rwm far removed from the signal max voltage is either availability (what I can get; or what I am willing to pay for, since some signals just aren't important enough to spend resources on), or if I really need to optimize leakage (this is not common).
Now, clamping voltages. Hit the diode with an ESD pulse and it'll see a lot of volts, break over, and start to conduct. A lot. Diode voltage drops scale with current. Put 1mA through the diode, it drops less voltage than if you put 1A through it. It's also exponential-ish, so 1A to 2A is a lot less trouble than 2A to 3A. These guys handle enough amps that we're definitely in the "more trouble" end of the curve. Our STS321050U182 is actually pretty clear about how it's been characterized when this happens: they used 8/20us tests (very common; still good to see stated; also 10/1000 is preferred with beefier parts), and they've been nice enough to provide us with two data points. (Not everyone gives you two.) Data point number one is that when you shove Ipp=1A into the poor thing (pulsed at 8/20us!) it drops Vcl=9.0V max. Data point number two is that when you shove 22A into it, it drops 12V typical or 15V max. The 350W P_pp number, incidentally, is from 15V × 22A = 330W, which is close enough to 350W in this business. (It tends to round better for some parts in the family than others, and they stick the same rating on 'em all.) So Digi-Key and Mouser both call this a "15V" clamping TVS diode. That's direct from the datasheet and completely correct. It's also bullshit because it lost the "22A" part of the conditions. You must, when you are comparing TVS diode performance,
compare at the same conditions. Is a "10V" clamping diode better? Maybe! But probably not if "10V" means "V_cl typical 10V at 10A". And, for that diode I just made up, guess what Digi-Key will list its clamping voltage as? Yeah. "10V". Even though I'd take this "15V" diode any day (for clamping ability alone anyway), since if you hit them with the same pulse, this one's clearly going to do better. Because of this wrinkle, it is hard to comparison shop TVS diodes. I end up with a couple series I know well and just use those repeatedly if I can. Also, the footprints are common, so it's easy to change later if I must.
This stuff all tends to show up in peak power ratings, so when searching for a new TVS diode, I will typically start by picking a package, then a max capacitance I will tolerate, then start evaluating parts from highest peak power to lowest. (Along with the usual price and availability concerns, etc, etc.) Peak power rating is
not everything, but it's a great way to direct the search.
OK... so... you're still exposing that poor UART pin to 15V on a hit? Right? That's bad? It's not as bad as you'd think. You put the TVS diode near the connector or ingress point (you did do that, right? seriously, this stuff is important), so it has a low-inductance shunt loop to get rid of the crap. You have the pin farther away (maybe), with at least a little inductance in between them. That helps a lot. The pin isn't defenseless; it has its own ESD network, and it is a lot easier for that small network to do its job now that there's a big mean signal TVS diode (don't tell it about power TVS diodes, lest it realize it's not big
or mean) over there protecting
it and taking the worst of the hit. And it's unlikely you actually managed to get
that much energy into the system (for really fast events like ESD, you have to talk energy rather than just power -- power doesn't vaporize metal layers all by itself) with an ESD hit. See, most of those ratings I just discussed were IEC 61000-4-5 8/20μs surge test ratings. There's a
lot of energy in them. ESD is fast. Lots of volts, lots of amps, yes, but it's over so fast it doesn't have time to do too much damage... if there's any amount of protection.
In general, if you stand a snowball's chance in hell of surviving surge, you are going to do just fine with ESD. No, this is not a "perfect" strategy. But, hey, a single diode is pretty easy, pretty cheap, and does pretty darn well. You really don't need a third layer of zeners in between, and they're not going to do any better even if you do have them unless you are super careful in your design. Remember, a TVS is just a differently-optimized zener, so you won't be able to clamp any tighter if you're handling any real power. And if it isn't handling real power, what's the point of your added part? The series impedance between your existing two protection networks is the first place to spend your time if you are worried. Making sure the nasties get dealt with by the big guy instead of the little guy can bring some real gains.