resistor wattage for hotplugging snubber?

[k8943]

Have various boards with a small buck convertor at the input.

Over a few weeks blew 3 of them whilst plugging into a 24v line, despite the fact that the max input voltage of the buck is 28v continuous, 30v peak.

Checked on the scope and indeed was getting some serious ringing at the moment of hotplugging, the first spike going well above 30v.

After some experimentation discovered that 3 * 10uF 50v 1206 ceramic capacitors in series with a 2-3 ohm resistor does a pretty good job of snubbing this in a variety of circumstances. (Those ceramics seem like a space and cost effective solution.)

Thing is, during testing was using chunky through-hole resistors because it was easy to chain them together. The whole charging thing is all over in 50-100uS so am guessing that not enough energy flows to do much damage to any kind of resistor but nonetheless did the following approximate calculation:

watts = total energy stored in capacitor / time
= 0.5 C * V*V / time
= 0.5 * 3 * 10uF * 24 * 24 / 100uS
= 0.5 * 30 * 10e-6  * 24 * 24 / 100 * 10e-6
= 86 watts

That is quite a large number and so feel a bit silly looking at 0.125W 0603 resistors on Digikey. But then why take a 0.5W resistor?

How should I validate a resistor wattage decision? (Only calculations found on the net pertain to steady switching in convertors rather than a one off hotplugging event.)

Taking 86W and setting it equal to I * I * R where  R = 3 ohms, get a current of 5A (for a very short time). Hmmm.

Or do we take the total energy and use the mass and specific heat of the resistor....?

Or just try and if a little 0603 resistor doesn't blow it's OK?

[Siwastaja]

If you want to do it properly, look at the single pulse (non-repetitive) handling characteristic curves on the actual resistor datasheet.

Of course, not all manufacturers provide such full curves, and sometimes they are hidden behind another application note sheet, but by digging around a bit, you'll find curves for a very similar (same package, thin film vs. thick film, same nominal power rating...) product, which you can use as a reference by applying a bit of derating margin, even if it for a different manufacturer, or a different product series.

If you are space restricted, special resistors that can handle much more pulse current than the typical same package counterparts exist. They are more expensive, though.

Note that often the option of using a bog standard electrolytic capacitor does exactly the same, but in a cheaper, smaller and simpler package (one integrated component). The ESR is used as the R. This is important because typically for such snubber, large capacitances (at least 3x the total ceramic no-ESR capacitance) are needed.

[k8943]

Excellent. Exactly as you said found curves (attached - for anyone reading along!) after a few tries.

Will now re-look at electrolytics. Had already noticed that 100uF crushes it, so now it's a search cost, can size and possibly a smaller value. Most of the time they don't bother to give an ESR so guess it's down to trial and error.

Thanks a lot!

[Siwastaja]

Electrolytics tend to be "inherently stable", i.e., their capacitance - resistance ratio never makes them oscillate/overshoot on their own; this is even true for most "low ESR" electrolytics, which still have considerable ESR.

Over full temperature range and product lifetime (i.e., increased ESR at cold temperature, and/or very old capacitor), you may want to use a bigger (than the minimum suggested 3x ceramic C) elcap so that the ESR won't be too high. But elcaps tend to have a lot of excess capacitance available very cheaply, so often just designing in 10x the ceramic C, while also derating voltage, is not a problem.

Do note that if the device you are snubbing is a switch mode converter, e.g. a buck converter which produces quite heavy input capacitor ripple, and if the ceramic capacitance you are snubbing is not quite large enough, the added elcap could be significantly contributing as the switcher bypass cap, heating up internally due to its ESR. (The same issue is possible for your discrete resistor case!) Make sure this is not an issue, since this power dissipation would be repetitive and continuous. Testing is often the only way here, because it's very hard to simulate all the parasitics correctly to see how the ceramics and electrolytics share the AC ripple current.

[k8943]

Drawing 150mA current at 5v from this 0.5A max buck, driven at 24v, measured circa 240mV p-p ripple on its input cap and about 1/3 of that on a 100uF electrolytic. The input cap is a tiny 2.2uF ceramic as recommended by the TI web designer for this configuration.

Not sure how to quantify 80mV p-p on the electrolytic as too much or too little but had already internalised that:

a)  it's better to put 2 or more electrolytics to achieve target capacitance "just in case";
b) case size and price does increase moving from 100 -> 300 uF;
c) perhaps larger electrolytics are better suited to through-hole mounting since it's rather against their nature to build them for a stint in a (hot) reflow oven?

In this application, hotsplugging of boards is an infrequent event (essentially linked to maintenance or configuration), it would seem desirable to minimise investment in input protection. If the strategy is to use the minimum cost/component volume possible, is there a case for the ceramic setup based on:

1) if there is heating it's in an external resistor and not drying out electrolytic => if the energy loss can be tolerated then has no importance and so the capacitance can be designed to handle the hotplug-spike but under-designed from the perspective of the convertor input interaction;
2) no worries about finite lifespan linked to inherent component properties;
3) natural component redundancy (since 3 times 10uF ceramic far cheaper than a larger ceramic or tantalum);

?

[Siwastaja]

Run the device for an hour and see if the elcap heats up. They die prematurely in switching power supplies when they are taken way beyond their ripple current ratings, or when heated by external sources, combined with high ripple current. But in such cases, they are often designed to take the full (or most) of the ripple current, instead of ceramics.

Ripple current is what causes heating, and heat is what causes the damage. So it's OK to measure the heat, it's a perfect indicator of the problem, while ripple current is almost impossible to measure (if you insert a shunt resistor, you are adding inductance and resistance, and thus affecting how the ripple current shares between the caps).

And, quite frankly, if the input bulk/snubber elcap is heating up a lot, that would be a good indicator to increase the amount of ceramic decoupling capacitance; let the ceramics take the largest part of the ripple without heating, and let the elcaps do the snubbing.

If your input elcap is cool to touch (say, cool enough you can keep your finger on it indefinitely without discomfort, i.e., way below 50 degC), the lifetime would be very long, unless it happens to be a bad enough capacitor to dry out anyway, cool or hot. The big players such as Panasonic, Nichicon, etc. have a good reputation again and I believe have learned their lessons regarding electrolyte issues in early 2000's, so I wouldn't hesitate using their products.

Reflow is not a problem for elcaps, they are fine for the reflow heat; but SMD elcaps still tend to be a bit more expensive.

For a snubbing capacitance in parallel to your 2.2uF MLCC, 100uF is of course already an order of magnitude more than required. But then, you won't typically choose the elcap for capacitance, but for ESR and ripple current rating. If the 100uF cap has too much ESR, a larger size (or two paralleled as you say) may still make sense. The best way is to measure.

Tantalum is completely unsuitable for exactly this purpose.

[k8943]

Totally on board now.

May I just double check something? "For a snubbing capacitance in parallel to your 2.2uF MLCC, 100uF is of course already an order of magnitude more than required."

I thought we were taking 30uF as the capacitance required for snubbing on the grounds that experimentally that seemed to be what was necessary - using ceramics - to balance the impedance of the system. (And that value was characterised by the power supply, the cables and so forth - rather than the convertor.) Which would mean that 100uF was about 3 times more than "required"?

[Ice-Tea]

Have you considered an NTC? They are made for this kind of application, so the surge rating shouldn't be an issue. And you could dimension it so that losses in steady state are a lot lower than a fixed 2-3 resistor.

[Siwastaja]

Have you considered an NTC? They are made for this kind of application, so the surge rating shouldn't be an issue. And you could dimension it so that losses in steady state are a lot lower than a fixed 2-3 resistor.

Sorry but you have completely misread the subject. The steady state losses in the OP's question are exactly zero; and AC losses are independent of the actual resistance, defined by the capacitance only.

It's not a precharge current limiter; it's not in series with the supply, but in series with the capacitor. An NTC is a completely wrong component as it has variable resistance; instead you want an optimum snubbing resistance (which, luckily, tends to be a large range in this application). (Of course, the ESR of an elcap is variable too, but that's a cost saving tradeoff; an NTC would be both more expensive, and less suitable, than a standard resistor, so it makes very little sense to suggest one!)

The losses are independent of the resistor value anyway, so the NTC dissipates the same as a fixed resistor.

Also, the NTC doesn't have any time to heat up during the snubbing action at power-on, so it's basically the same as a fixed resistor, just temperature dependent for no reason.


Or, maybe we are discussing about a completely different subject, surge current limiting, but let's make it clear that we change the subject, shall we?

A surge current limiter (such as NTC) does helps with the voltage overshoot, as well, but in a very lossy way (now you have the DC loss all the time, keeping the NTC hot 24/7, just to combat the very low-duty hotplug event!). If the problem is only the surge-induced voltage ringing, and if it's completely solved with the typical snubber circuit, I would not go on to suggest further input surge current limiting, for such low current, low capacitance circuit, unless there is something special there requiring it.

Also note that properly using an NTC as a surge limiter is complex as you have to carefully synchronize bus capacitance discharge with the thermal cooldown of the NTC and even then it tends to be tricky, leading to edge cases where the current isn't limited properly. So for a robust NTC based design, you should analyze for the worst edge case, which is, hotplugging while the NTC is still hot from the earlier run, but the capacitors have already ran empty. And, when you do that, you could as well use a fixed resistor (with the same R as the hot NTC has); using NTC instead would only reduce the average case input surge current, which isn't super meaningful.

In any case, I don't think that the inrush current is a problem. The amount of ceramic capacitance is such low (2.2uF), that to snub it, no massive amount of snubbing capacitance is needed (so we are at around 10 to 100 uF, with series R of at least several ohms), which means, the inrush I^2*t isn't high enough to blow sensibly chosen fuses, or anything like that. The reason why I don't suggest current limiting here, is that it's really a lot more complex subject, so avoid it in simple, low power cases where it isn't needed.

[k8943]

Actually a detail thus far unmentioned is that the TPS62175 doesn't start up for a full 1ms until after the EN pin has been powered. By which time all the voltage spikes caused by hotplugging it to a tangle of wires connected to a voltage source have already occurred and potentially damaged the IC. Since the convertor is not yet switching <1ms there's no inrush current in this period.

[Siwastaja]

Totally on board now.

May I just double check something? "For a snubbing capacitance in parallel to your 2.2uF MLCC, 100uF is of course already an order of magnitude more than required."

I thought we were taking 30uF as the capacitance required for snubbing on the grounds that experimentally that seemed to be what was necessary - using ceramics - to balance the impedance of the system. (And that value was characterised by the power supply, the cables and so forth - rather than the convertor.) Which would mean that 100uF was about 3 times more than "required"?

Well, you can see that if you snub 2.2uF low-ESR cap with another 2.2uF + R (1x), you'll reduce the overshoot from 100% to somewhere around 20-30%. At 14x (30uF) and properly tuned value of R, you are already at practically completely snubbed system with no overshoot (less than a few percent) - it's already overkill, from pure capacitance viewpoint. I have seen "rules of thumb" vary between 1x and 10x! I tend to go with 3xC, but you may need more for other reasons (when choosing an elcap for different parameters than C - namely ESR).

If you are interested, research the math, or just go with something. In any case, as you have the 2.2uF on the BOM already, it would make sense to construct the snubber by placing at least two more (4.4uF or more). They could share a common series resistor, or you could use separate resistors (regarding the wattage; this would spread the heat as well). This kind of solution is often flexible in miniatyrized layouts, if you happen to have some "dead real estate" somewhere - compared to using one big component.

[Siwastaja]

Since the convertor is not yet switching <1ms there's no inrush current in this period.

Note that the converter itself doesn't take a lot of inrush, because it's a current mode controller (well, at least with a current limit!), so that it never goes over the specified limit by active limiting (mere 1.2A max by the datasheet).

The inrush is from the input caps - the 2.2uF plus the snubber caps you add. For example, the 2.2uF cap would have an inrush current of about 2400A (assuming 24V input and 10mOhm ESR), but only for a very short time (time constant of 22 nanoseconds), and from a zero inductance supply - which doesn't exist, hence lower actual inrush current, for a longer actual time - and the overshoot and ringing you see!

Your 30uF + 2R would have inrush current of 12A at 24V. This is for a very short time as well. Approximate with RC time constant of 60 microseconds. Now this is in a meaningful time scale, which is a real possibility with your cable inductance. Hence, it works for snubbing.


For this inrush current reason, don't go overboard with your snubber caps. If you put, for example, a honking 10000uF electrolytic there, it would indeed give you a stable voltage rise with no overshoot, but chances are that you'll blow some small fuse later, or when you try to control your circuit through a MOSFET "rated for 10A", it will die on your 100A inrush because that capacitor would have ESR way below 1ohm, and capacitance large enough so that the spike last for a long time. Just some examples.


Inrush limiting is more often needed in AC power line switchers, because they need to have large input capacitors to store energy because your AC supply is "dead" for such long time during every 50 or 60 Hz cycle. They have no choice but to store the energy if they want to output DC, necessitating the big cap.

[Ice-Tea]

Have you considered an NTC? They are made for this kind of application, so the surge rating shouldn't be an issue. And you could dimension it so that losses in steady state are a lot lower than a fixed 2-3 resistor.

Sorry but you have completely misread the subject. *snip*.

No need to be pedantic about it. You have read the OP and have concluded it's standard snubber. I did not. That's all there is to it.

[Siwastaja]

Sorry but you have completely misread the subject. *snip*.

No need to be pedantic about it. You have read the OP and have concluded it's standard snubber. I did not. That's all there is to it.

Sorry; in any case I hope being overly pedantic did give real information to the OP and the others.

[Ice-Tea]

Sorry but you have completely misread the subject. *snip*.

No need to be pedantic about it. You have read the OP and have concluded it's standard snubber. I did not. That's all there is to it.

Sorry; in any case I hope being overly pedantic did give real information to the OP and the others.

Heheh, that it did 

[k8943]

One way or another, bringing the conversation to the matter or current did help clarify what had to be learnt and have now verified that a 1ohm / 4.7uF combination indeed will work quite well. (Have also in the meantime load/heat tested electrolytics and a bunch of other things!)

Thanks a lot everyone.

[k8943]

With hindsight found this quite interesting. Chipped a way a little and small update.

Tried 1ohm 0603 resistor in the snubber it became 4ohm! However two 2ohm resistors in parallel took the load. (Tested hotswapping with linear bench supply set to 24v, since it seemed to give more violent peaks than switched supply and 28AWG wiring used within project.)

Used 1206 1ohm resistors in testing, then looked more closely at the pulse curves in the post at beginning of thread and realised 2 * 0603 resistors might give better pulse resistance and work out cheaper.

Discovered a 0805 4.7uF ceramic had about the same snubbing effect (with 1ohm) as 3 * 2.2uF 0603 (6.6uF) caps, but 1206 4.7uF cap was way better than both. Size matters?

Preferred recipe therefore 1206 4.7uF ceramic and 2 * 2ohm 0603 resistors.

Capacitor 1276-2789-1-ND is 9.5 cents per thousand.

Then went off in another direction, what about a zener?

For example Digikey: SMBJ24D-M3/IGICT-ND

Voltage breakdown occurs at 27v and the cost per 1000 is 11 cents. Can handle "600W" pulses.

If, say, a pulse ran to 40v and the voltage drop across the diode is (not actually sure from datasheet?) 25v, that leaves 15v that needs to go somewhere. A 10ohm resistor in series => 1.5A current (not really sure how to size this). This is so massively with specs maybe a smaller zener would handle it? However didn’t see an obvious cost saving insodoing.

Do zeners get used for this sort of thing?

[Siwastaja]

Bigger ceramic is better because it has closer to the rated capacitance even at high DC bias voltage. The nonlinearity of a small ceramic amplifies the overshoot effect. In a snubber, it's less effective.

Zeners do work and are used, but the curve isn't very steep, and there is unit variation so that you are limited to reducing the overshoot to about 30-40%, of course a great improvement if you have 100% overshoot in beginning. With an RC circuit, you can get to almost 0% overshoot, and this works over a wider voltage range. But if your aim is to protect the semiconductors, a zener does great job, and you can be allowed to ring more at lower input voltage, since only the worst case matters.

Combining both approaches can be very good.

[k8943]

Thanks. See the advantages of the RC solution better now with context.

Thinking that the "curve" you refer to is the line between the "minimum breakdown voltage" (which is fairly well defined) and the "maximum clamping voltage" (which is indeed terrifyingly high). Sort of see that if the series resistor was chosen to limit the maximum (pulse) current through the zener in such a way that the maximum voltage across it was much lower (than the max), then the voltage across the resistor will be significant enough such that the total voltage drop is nonetheless high. RC it is!

[T3sl4co1l]

...But, the idea of using an NTC here is still an option, no?

May not be practical, simply because the amount of power draw is so low that it won't heat up appreciably.  Well, if that's the case, it won't be wasting much, either, right?

In that case of course, an ordinary resistor would do as well.  Take for example the 4.7 to 10 ohm power resistor commonly seen in fluorescent/LED lamps: this serves as inrush limiting as well as fusing, and has a small impact on efficiency.

- The relative merits of electrolytic and ceramic have been mentioned, so I won't go into further detail.

- TVS is good for this sort of thing, but also as mentioned, it's unlikely to be quite good enough, with so little margin available.  For a 24V supply, I would pick at least a 36V converter, preferably more.  The TVS would be "24V" rated if it's 24V nominal (say from a regulated supply), or 28 or 33V or more if a linear supply or battery is involved, and the peak surge voltage will be in the 40-50V range, hence the higher converter rating.  28V is too low though.

- The last option, that hasn't been mentioned yet: extend the operating voltage, and deal with surge in whatever way is necessary.  This is impractical for heavy loads, but the light load in question here is quite easily handled, say with a DN2625 or similar depletion FET, in a zener-reference-pass-regulator configuration.  (The DMOS with G-D pullup resistor has low voltage drop, until gate voltage is limited by the zener; the high drain voltage rating allows quite loose protection to be used, even a MOV.)

I did this for an automotive application last year; the device drew maybe 20mA at 28V, and withstood 60V continuous and 100V load dump without a flinch.  (The FET limiter was set to kick in at 40V.)

Tim

[k8943]

Thanks for that.

Reply #8 did examine the fixed resistor case but thought I should give it a try.

Discovered that 2.5ohms in series with the 24v Vin did a good  job of limiting the inrush to the 2.2uFh input capacitor of the buck convertor at the moment the RJ12 is hotplugged on the linear bench supply (the most agressive testing environment as it turns out). 2ohms was too little.

Seemed to go from 0->24v in 2.5uS so calculate the current as follows:

I = Cdv/dt = 2.2uF * 24v / 2.5uS = 21amps
Which implies wattage = I^2 R = 21 ^2 * 2.5 = circa 800W for 2.5uS

Looking at the graph in Reply#3 and extrapolating back towards 2.5uS (not on the graph) have the impression a 1206 might not make it so instead searched on 2512

Digikey RMCF2512JT2R70CT-ND comes in at about 4cents per thousand so that would appear indeed to be the most cost effective solution excluding running costs.

If the buck were run at maximum 0.5A and my selected 5v => current at 24v approx 0.5 * 5 / 24 = 0.1A
Losses in resistor = 0.1 ^2 * 2.5 = 25mW

Admittedly that's not a hill of beans.

Intuitively and aesthetically it's not nice to put a series resistance in the power line but not sure what other critiques to bring to this low power situation?

*************
Browsed TI's excellent webbench tool for power convertors with an increased input voltage range (10v-35v instead of 10v-28v - the 10v is in case project is running off backup 12v lead acid) and indeed this increases the BOM cost for the convertor by more than 50% as well as possibly the surface area whilst reducing the efficiency. So no slam dunk.

********
The DN2625 circuit sounds as though it will sound difficult situations very well. In this case of my humble application note that the RDS on is 3.5ohms.

Thanks again for the input. Fascinating how even the most innocent question can big bang into a universe.