Limiting inrush current into an electrolytic capacitor

[philpem]

I've got a project "in the lab", which is about to go into production... and there's a bit of a snag. Part of the power supply is done with DC-to-DC converters (the little black plastic potted things) to isolate critical parts of the circuitry. To reduce the noise these push back onto the incoming 24V power supply, there's a 100uF low-ESR capacitor across the 24V input, behind a reverse-polarity protection diode.

Total power consumption of this thing is 50mA, plus about 200mA for some motors and stuff which are usually running but can be switched off by the electronics block under command from a PC.

Problem is, there's been a power supply spec change. We've gone from a 24V rail sourced from a "dumb" power supply (a switcher with an output fuse and a crowbar OVP) to an all-singing-all-dancing "smart protected power supply" which does open and short circuit detection. For bonus points, the "smart" PSU has fast-switching outputs - essentially a relay driven from the output filters, via the open/short protection circuit. The dV/dt is essentially as fast as having a lab PSU on and ready, then switching the power to the unit with a mechanical switch, only without the contact bounce. There is no soft-start.

If the device being powered draws more than 400mA for 200us, the PSU will declare the output as shorted and trip. The 100uF capacitor and power supply components (DC-to-DCs) are taking it way over this limit - by the highly scientific method of "stick a 1R resistor in series and measure across it", I'm getting 7A (falling exponentially, but not fast enough) for a couple of hundred microseconds, more than enough to trip the PSU.

I can't make any modifications to the shiny new power supply or its wiring, but I can modify my circuit's power supply all I like within reason. I'm not keen on adding a series resistor to the low-ESR cap as it'd impede the filtering ability of the capacitor, and the voltage drop over a series resistor on the 24V input might cause problems for the DC-DCs (this device could potentially be on a long run of cable and needs to run down to about 20V). An inductor (I've tried up to 100uH)

Can anyone think of a way I could reduce the inrush current enough to get by the limit, without impeding my capacitor's ability to filter the DC rail?

Cheers,
Phil.

[Ice-Tea]

NTC

[acbern]

the standard solution for an inrush current limiter is a fet. you need to drive the on-resistance during power-on, e.g.  in the simplest implemantation by an RC combo
there are circuits on the web also. ntc and relais are not recommendable, mtbf is bad and limited number of cycles (overall/per given time)

[jpb]

the standard solution for an inrush current limiter is a fet. you need to drive the on-resistance during power-on, e.g.  in the simplest implemantation by an RC combo
there are circuits on the web also. ntc and relais are not recommendable, mtbf is bad and limited number of cycles (overall/per given time)

Is the mtbf of relays that bad? I'm curious as instruments such as my bench DVM are full of clicking relays and have been going ok for decades.

I'm in the process of building a linear power supply myself and was thinking of simply using a relay with perhaps an SCR to control the coil to switch out a series R.
It has the advantage of being fail safe - if the relay does fail then the resistance isn't switched out so the voltage drops.

[suicidaleggroll]

A fet with an R/C combo on the gate is really dead simple to implement and works quite well.  Something like this:

[jpb]

A fet with an R/C combo on the gate is really dead simple to implement and works quite well.  Something like this:

The problem I had when looking at FET circuits was the knee voltage. A relay presumably can pass several amps whilst dropping less than 0.1V, most FETs have knee voltages much higher than this.

[Ice-Tea]

the standard solution for an inrush current limiter is a fet. you need to drive the on-resistance during power-on, e.g.  in the simplest implemantation by an RC combo
there are circuits on the web also. ntc and relais are not recommendable, mtbf is bad and limited number of cycles (overall/per given time)

I'm a bit confused. How is a NTC not recommended? Why is mtbf bad? I don't recall they have a limited #cycles (I presume this is meant for the relay?).

Not doubting you, rather looking for more insight. I've used NTCs in designs quite a bit and would like to know if it's something that's gonna bite me in the ass one day

[ovnr]

Is the mtbf of relays that bad? I'm curious as instruments such as my bench DVM are full of clicking relays and have been going ok for decades.

I'm in the process of building a linear power supply myself and was thinking of simply using a relay with perhaps an SCR to control the coil to switch out a series R.
It has the advantage of being fail safe - if the relay does fail then the resistance isn't switched out so the voltage drops.

Relays are also noisy. And is your series resistor capable of being switched in continously if the relay does fail? Most in-rush limiters designed to be switched out will dissipate too much power to be safely enabled all the time.

I'm a bit confused. How is a NTC not recommended? Why is mtbf bad? I don't recall they have a limited #cycles (I presume this is meant for the relay?).

Not doubting you, rather looking for more insight. I've used NTCs in designs quite a bit and would like to know if it's something that's gonna bite me in the ass one day

Relays die. Often at very inconvenient times...

NTCs dissipate a fair bit of power and there is a fairly significant voltage drop. You may not be willing to put up with it; I rarely do. They also run hot (kind of the point), and having components running at 100+ deg C can introduce other issues (heat transfer to other components, etc).

[Ice-Tea]

Yeah, well, I wasn't proposing a relay. And you have a valid point about power consumption of an RTC. Can't be beat in terms of cost and simplicity though. And I'd argue that it's more reliable as well. Would prefer a surge on a NTC rather than a FET...

[mikerj]

I'm a bit confused. How is a NTC not recommended?

NTCs that provide a reasonable value of resistance when cold tend not to have very low "on" resistance
Their behavior change with ambient temperature
They have lengthy reset times.
Purely from my own experience they don't seem to be the most reliable of components.

They provide a perfectly adequate and cheap solution in many cases, but are not always appropriate.

Why is mtbf bad? I don't recall they have a limited #cycles (I presume this is meant for the relay?).

Mechanical relays have a finite number of operations they can make during their life, primarily limited by degradation of the contacts.  MTBF is not an appropriate measurement for a relay, since they may last almost indefinitely if held in one state, instead MCBF (Mean Cycles Between Failure)

[nuno]

Remove the electrolytic, check if the new better converter filters the noise...

[SeanB]

Mix both circuits. The NTC will limit the cold current, then the power FET will switch on and bypass the NTC. That way you have both low dropout voltage and if the FET dies short circuit ( the most common failure) you have the power supply overcurrent safely switching all of the unit off for repair. As well the NTC will never warm up unless the FET does not turn on. You can simply have a non resettable thermal fuse in series with the input, thermally bonded to the NTC and the FET, to disconnect the power if the unit gets too hot from this or other reasons. 110C fuses are cheap, reliable and not likely to fail at normal operating temperatures.

[suicidaleggroll]

The problem I had when looking at FET circuits was the knee voltage. A relay presumably can pass several amps whilst dropping less than 0.1V, most FETs have knee voltages much higher than this.

Could you describe a bit more what you mean by this "knee"?  I'm not following.  24V is so far beyond the Vgs threshold of any old P-FET that it's a non-issue, and 0.4 ohms of on-resistance is painfully easy to beat as well.

[rx8pilot]

I have recently had to deal with this and I went through many possibilities except a relay. My goal was to limit the inrush but not limit the outrush. This was to charge 8200uf capacitor bank which is 10 820uf caps in parallel, so the ESR is fairly low. The purpose of the capacitance is to ride through droops and deal with current spikes with minimal voltage drops. In the end I am using a FET with an ideal diode controller which blocks incoming voltage, but lets the caps discharge with VERY low resistance of about 3mOhms. To charge the caps I have a 50 Ohm resistor from the main bus. When the voltage of the caps approaches the bus voltage, the FET turns on to a Vgs of +15v. The ideal diode controller has a charge pump that will drive N-FETs which (in general) are lower loss. It also allows me to control the cap bank from a uController.

Conceptually, you have a similar setup but at much lower currents it seems.

I have never actually used an NTC, but from the little research I have done they seemed much too lossy and restrict current in both directions. I needed to be able to discharge the capacitors very fast to make my circuit work.

[jpb]

The problem I had when looking at FET circuits was the knee voltage. A relay presumably can pass several amps whilst dropping less than 0.1V, most FETs have knee voltages much higher than this.

Could you describe a bit more what you mean by this "knee"?  I'm not following.  24V is so far beyond the Vgs threshold of any old P-FET that it's a non-issue, and 0.4 ohms of on-resistance is painfully easy to beat as well.

The very simple circuit I was looking at had the FET in series with the smoothing cap before the voltage regulator which was a LDO one. As it was only for my own use in the UK I was trying to maximize efficiency (whilst staying linear) by having only a very low voltage drop from the rectifiers to the regulator. My total voltage allowance was therefore rather small and I didn't have much to drop across a FET.

My mistake I realise now was I was assuming I needed to drive the FET beyond the knee but of course it is fine to operate it in the linear region - I am used to small signal FET design so I intrinsically assume an above-the-knee bias point 

[rx8pilot]

what is the "knee voltage"?

[ajb]

Probably overkill for your current requirements, but there are hotswap-oriented high-side switch controllers that provide programmable soft-start and current limiting.

[jwm_]

How about a lm317 in current limiting configuration. Or a LDO version of the same if you can't handle that much of a voltage drop.

like so:
http://www.ae5d.com/images/LM317.png

[suicidaleggroll]

The very simple circuit I was looking at had the FET in series with the smoothing cap before the voltage regulator which was a LDO one. As it was only for my own use in the UK I was trying to maximize efficiency (whilst staying linear) by having only a very low voltage drop from the rectifiers to the regulator. My total voltage allowance was therefore rather small and I didn't have much to drop across a FET.

My mistake I realise now was I was assuming I needed to drive the FET beyond the knee but of course it is fine to operate it in the linear region - I am used to small signal FET design so I intrinsically assume an above-the-knee bias point 

I still don't know what you're talking about.

When you use a P-ch FET in a soft-start circuit, the goal is to gradually increase Vgs (well, decrease, from zero to negative), which gradually reduces Rds.

When voltage is first applied, Vgs is 0, Rds is infinite, and no current flows.  As the RC filter allows Vg to drop below Vs, Rds starts to decrease, and current starts to flow with almost the full 24V drop across Rds.  After some time (configurable with your R and C values, probably on the order of milliseconds), Vg will drop to its final voltage at or near ground, Vgs will be the full 24V, and Rds will drop to the nominal Rds(on) of just a few milliohms, and your voltage drop across the FET in steady-state will be on the order of a few mV at your operating current.

It essentially acts like a variable resistor that runs from infinite resistance to the FET's nominal Rds(on) over an amount of time you control with the R and C values.

[jwm_]

A FET configured in this way is often refered to as a 'constant current diode'.

I tend to use an lm334 or lm317 (depending on limit) for this though as i already have a bunch of those but may not have a suitable fet on hand and are equally easy to build. a caveat of the lm317 is that it doesn't like a reverse voltage so isn't quite a constant current diode so you may need to take that into account.

[SArepairman]

I don't think a relay timer is a bad solution so long you pick an appropriate relay. It's just more expensive.

I have seen quite a few failed NTC's before. They defiantly do fail into a brittle mess.

 I never had a product that used a relay timer though (well, I don't get to take apart gear that was not designed with dollar slashing in mind very often). I built my own though. I build an entire module that used a small transformer and a LPS / 555 timer and transistor to trigger a beefy relay which shorted out a beefy inrush current limiting resistor.

Are you guys sure that a power electronics solution at the same price range as a decent relay will be more reliable? In college my electromechancis professor (who had advanced in industry ranks as well) spoke about relay inrush current limiting circuits a few times in a manner that suggested they were used in the power industry and they were a respectable solution to this problem.

Keep in mind that a NTC is slow to cool down as well.

You guys got SERIOUS relay hate haha , sounds like you are specifying crap!

[jpb]

what is the "knee voltage"?

Where the IDS(VDS) curve leaves the linear region and levels off (though there is generally still a slope) - this is where the electron velocity in the device saturates. For  linear amplifiers the load line (class A) should stay in this saturated region.

Normally the knee voltage is a lot less than the overall voltage swing so it doesn't matter but for low voltage devices (battery operated) the knee voltage significantly reduces the potential linear voltage swing (class A amps). I was once involved in a project to design a FET to reduce the knee voltage to allow operation down to 1V (this was with GaAs devices at high frequencies).

[rx8pilot]

This is why I hang out in this forum. Nice.

[jpb]

I still don't know what you're talking about.

That makes two of us!

To try and explain my original thinking, I come from a background of working with GaAs FETs (schottky barrier gates) and on the whole, unless you're designing a mixer, you aim to operate the device where VDS is large enough for the electrons to be at saturated velocity (ideally the current curves are flat with VDS but in fact they keep rising but this is due to 2D E field effects rather than changes in electron velocity). The point at which the curve levels off (as far as it does) is the knee voltage and so I'm programmed to assume that you want VDS greater than this (around 0.5 to 1V for small signal devices).

Of course, in the case of using it as a variable resistor it is fine to operate it below the knee. I just wasn't thinking straight - I blame the fact that I've been out of the industry for more than 10 years and senility is setting in.

[AG6QR]

For a commercial product that does this, see

http://www.mfjenterprises.com/support.php?productid=MFJ-4403

This is a power conditioner, intended for the ham radio market, where radios nominally run off of 13.8V supplies, maybe in vehicles.  Vehicles can have lots of transients on their supplies, and radios can have transients in their power consumption as they switch between transmit and receive, or as SSB transmitters change output power with modulation. 

That power conditioner is basically a 4 Farad (notice no prefix on the Farad unit) capacitor bank, with some associated circuitry around it to limit inrush and provide a bit of protection.  There's a downloadable manual there with a block diagram and detailed schematic.  It uses a resistor to limit inrush.  Once the capacitor bank is charged sufficiently, a relay bypasses the resistor.

Not that that's the only or best way to do it, but here's hoping someone finds it interesting to look at how someone else approached a similar issue.

[Ice-Tea]

I have seen quite a few failed NTC's before. They defiantly do fail into a brittle mess.

Just wondering, was that on secundary or primary voltages? I've mostly had them on secondary, DC stuff and never seen any returns or failures. And that's on products that have been produced in the millions.

[philpem]

Yikes, I had no idea this issue would be so contentious...    Thanks for the responses, guys!

NTC

That was my first thought, but in the actual circuit it doesn't seem to work. The NTCs I've found which are specified as inrush limiters require a considerable current through them to maintain a low(ish) voltage drop. There's also the issue that they run hot, which is going to be a problem in this application - for safety reasons, the finished product is potted into a solid block. Getting heat out of that block will be tricky.

I have recently had to deal with this and I went through many possibilities except a relay. My goal was to limit the inrush but not limit the outrush. This was to charge 8200uf capacitor bank which is 10 820uf caps in parallel, so the ESR is fairly low. The purpose of the capacitance is to ride through droops and deal with current spikes with minimal voltage drops. In the end I am using a FET with an ideal diode controller which blocks incoming voltage, but lets the caps discharge with VERY low resistance of about 3mOhms. To charge the caps I have a 50 Ohm resistor from the main bus. When the voltage of the caps approaches the bus voltage, the FET turns on to a Vgs of +15v. The ideal diode controller has a charge pump that will drive N-FETs which (in general) are lower loss. It also allows me to control the cap bank from a uController.

Conceptually, you have a similar setup but at much lower currents it seems.

I have never actually used an NTC, but from the little research I have done they seemed much too lossy and restrict current in both directions. I needed to be able to discharge the capacitors very fast to make my circuit work.

I'd never heard of an "ideal diode controller", but having looked them up, they sound interesting. Sadly they're also a bit above my target BOM cost, but still something worth putting on the old "this exists, maybe it'll be useful in future" list

A fet with an R/C combo on the gate is really dead simple to implement and works quite well.  Something like this:

That's nifty. After a bit of thinking I came up with something similar, but I'm curious - why put the capacitor on the drain like that as opposed to, say, between source and gate with a resistor to ground (and a resistor shunting the FET's drain and source to set the current limit)?

Thanks,
Phil.

[Ice-Tea]

the finished product is potted into a solid block. Getting heat out of that block will be tricky.

Yep, that pretty much disqualifies the NTC. It's a device that relies om some thermal equilibrium to work, assuming a device-air thermal interface.

Tinker with that and all betts are off. My guess would be that cooling it wouldn't be too much off an issue, the "block" will do just fine but rather the opposite: it will be cooled too much which will lead too a high impedance of the device.

[dazz1]

Hi
I have a problem at the moment with the inrush current to a 12VDC to 5VDC SMPS tripping the BMS protection on a LiFePO4 battery.  I need to get the inrush current below 70A for less than 1ms.
Has anyone considered using a EMI ferrite and diode in parallel to act as a LD snubber??

Dazz

[Mad ID]

Hi
I have a problem at the moment with the inrush current to a 12VDC to 5VDC SMPS tripping the BMS protection on a LiFePO4 battery.  I need to get the inrush current below 70A for less than 1ms.
Has anyone considered using a EMI ferrite and diode in parallel to act as a LD snubber??

Dazz

How much current does the device consume in normal operation? 12/70 = 0.17 -> it seems to me that adding 0.2R series resistance solves your problem.

The other option is to place a dv/dt limiter with one PFET and drain-gate capacitance, see this:
https://electronics.stackexchange.com/questions/123710/pmos-gate-driver-using-bjt

[TERRA Operative]

Welders often have an NTC with a relay that shorts across the NTC to take it out of circuit once the inrush has passed.

Seems to work well and we almost never had to replace the relays.

[Ian.M]

I have a problem at the moment with the inrush current to a 12VDC to 5VDC SMPS tripping the BMS protection on a LiFePO4 battery.  I need to get the inrush current below 70A for less than 1ms.
Has anyone considered using a EMI ferrite and diode in parallel to act as a LD snubber??

Although theoretically a LD snubber approach could work, its probably not going to help in real life.   The EMI ferrite would have to handle most or all of the peak inrush current without its core saturating as if it saturates it might as well be a piece of wire^* . Also its got to have enough inductance to 'hold off' the current enough to get it under 70A, without its resistance being too high and causing you problems with voltage drop on load.   Its therefore going  to need enough core area not to saturate too soon + enough turns of thick enough wire to get the inductance without much resistance, which pushes up the core size even further just to get the winding area required! 

TLDR: The inductor needs to be impractically *<expletive>* *<F_word>* *massive*.^#

*Holding off *part* of the spike before saturation may be sufficient to reduce the remaining inrush current to an acceptable degree by allowing the input caps to charge enough so that the remaining voltage differential when it saturates isn't high enough to drive an excessive current through the circuit resistance.

# Highly technical terminology here - ask for clarification if you need it. 

Edit: Changed to reflect that limited saturation of the inductor *may* be acceptable

[salbayeng]

Has anyone considered using a EMI ferrite and diode in parallel to act as a LD snubber??

Dazz

Yep.
  this was for Solar project where we had up to 48 controllers hanging off the DC bus each with a 1000uF capacitor, the solution was to use a big inductor (5mH? at 20A) about 20mm dia x 25mm long, with a schottky diode across it,  also put two 1milliohm MOSfets across the inductor. we also staged 3 of these with about a 100ms delay. The resulting inrush current was less than 40A (i.e comparable with maximum load current) .  The MOSFETs were driven with a simple comparator, when the stage before reached 20v, the next stage would come on. The inductors make an audible clunk when the power is applied ,  so the whole system would go brrrrup when cycled up.
We originally tried low ohmic resistors, but they have low thermal mass, so we thought copper has a low thermal mass and filter inductors are cheap with milliohm resistances, so we tried them,  the whole "soften the edge with inductance" thing was a bonus. We also put a 10A PTC in series with the inductor, so if there was a dead short the PTC would trip, and there wouldn't be enough volts to pull in the main MOSFETs.
Overall we were quite happy with this approach, it seemed quite bulletproof, we were powering the system from a 24 battery, so failure modes were potentially spectacular.
We switched the negative rail BTW, this makes it easier with NMOSFETs, and allows earthing of the positive rail (less corrosion this way)

[edit] and also note what Ian says above ,  the optimal inductor is always physically too big! as an aside, I once worked on a pulsed laser flashlamp supply, this had an air-cored toroidal inductor about 10" across, this is part of the "pulse forming network" that discharges a big capacitor into the big flash tube, I think it's about 100us to discharge it,  more of an "outrush limiter" than "inrush", but functionally similar.

[salbayeng]

I should also add that it won't really work with a small ferrite bead. A piece of ferrite can only store so many watt-seconds of energy before it saturates, so you really need an air-cored or open bobbin inductor.
The energy at which a core saturates is proportional to its mass, so for big powers you need big cores! can't escape basic physics.

The other thing to get your head around is the I²T value of a component, this value is typically used to define circuit breakers and fuses and the like, but all components (diodes, transistors, resistors, coils, wires, bits of PCB trace) will act like fuses at some I²T . in any circuit containing an assortment of components in series, the one with lowest  I²T will blow first with a short circuit fault, ideally this component would be the fuse/PTC/breaker.

[Ian.M]

So *THREE* biggish inductors, three Schottkys, three MOSFETs and a sequencer for the same, each stage nibbling a bit off the peak surge!   It all sounds a bit Rube Goldberg, but as it did the job reliably, and was within the size and cost constraints, it must have been sound engineering.

[dazz1]

Hi
My application draws about 90W but the battery bms is tripping with no load on the smps.   If I just put an inductor in series with the smps capacitors, it would ring like a bell and may blow the smps on over voltage.  The diode would damp the ringing.

I just need to knock the top of the inrush current so I am going to try an LD snubber without anything else.  It should also smooth out all current spikes improving emc.

[salbayeng]

@ ian , well the power distribution PCB was pretty big (400m x 100mm), it needed to host 3 groups of 16 MTA156 connectors.  It was made with 3oz copper, and had a pair of 10mmx3mm busbars bolted to the back of the PCB, so wasn't particular pretty. There was a whole bunch of stuff on the PCB like lightning protection and comms drivers, and current sensing, and diagnostic LED's, (bit difficult to push 19200bd down 16 parallel cables totalling 30nF). and there were 4 of these distribution PCB's in each enclosure, with a 40AH 24v battery, with 200A fuselink, and 4 63A breakers.

[David Hess]

Given the requirements I would probably use a current limiting MOSFET in series as described but there is another way which could work.  The problem is the surge current into the input capacitors and not into the switching voltage regulators so use the current limiting MOSFET in series with the input capacitors instead.  The resistance added to the ESR is not great and with the transistor inserted on the ground side, a less expensive n-channel device can be used.

The other way I would consider is using LCR decoupling instead of just a large capacitor.  The added inductance means that a much smaller capacitor can be used limiting surge current.

[T3sl4co1l]

If you're going with big inductors, you might as well toggle them on and off, add a catch diode to recycle the stored energy, and some caps to deal with input and output (stray) inductance.  Now you get an active current limit with no losses, so it can handle enormous overloads (like whole seconds of shorted output).

In other words, a current mode buck converter.  Which suggests some methods.  Find a buck regulator/controller, current mode, that supports 100% duty cycle (with internal charge pump, or add one yourself).  Set it to regulate an output voltage higher than nominal, and it'll sit there saturated all the time.  But because it's current mode, it regulates current when an overload occurs.

You can use fairly crappy parts for this, since they only need to operate for a short period of time.  Well, the transistor wouldn't be crappy, it'll be low Rds(on) for low quiescent losses -- but it can be absurdly so, which will be slow and lossy in switching, which is fine.

The downside is it's kind of a lot of stuff, just for a seemingly simple problem, and in this case, probably not much actual energy.

I made a few of these for my lab,



which operate in this way, also supporting bidirectional operation, and not needing a ground return if you don't have one (in which case operation is thermally limited, since it has to dissipate the stored energy rather than recycling it).  Control is an original low power circuit: battery life is over a month from a 9V, or it can be powered externally.

For lower energy purposes, perhaps a "hot swap controller" would do?

Tim

[Evan.Cornell]

Tim, if one wanted a purpose made IC to do what you're suggesting, take a look at http://www.analog.com/en/ltc7860. Not cheap, though.

[T3sl4co1l]

With voltage regulation too, oh hey, that's that thing Simon was looking for--

Tim

[Evan.Cornell]

They have some other options that also regulate, but linearly, instead of a buck.

[rx8pilot]

Tim, if one wanted a purpose made IC to do what you're suggesting, take a look at http://www.analog.com/en/ltc7860. Not cheap, though.

I have been using a similar device - the LTC4356 that can limit current by way of using the MOSFET to create a linear regulator. It quickly violates the SOA of the MOSFET so it can only deal with short-term events before it simply shuts down. It does a good job of dealing inductive spikes and inrush current from hot patched capacitive loads.

Perhaps it is not the most simple approach, but the end results are good. At first glance, it looks like the LTC7860 is capable of dealing with longer-term events.

[T3sl4co1l]

Similarly, I've used this before:
https://www.digikey.com/product-detail/en/texas-instruments/TPS2491DGSR/296-26925-1-ND/2255225
(there are also different voltage and cost versions I think)

And yeah, ride-through or short-circuit capacity is limited by SOA, which is in turn limited by physical size of the component.

Get the cheapest, lowest-Rds(on) D2PAK you can, and you should be able to handle a couple milliseconds of fault condition without trouble.

Consider that we're often talking hundreds of watts under a fault condition, while a transistor might handle a few mJ (small SMTs), tens of mJ (DPAKs), or hundreds of mJ (D2PAKs and THT).  With power and energy figures like this, it's hard to have more than a few milliseconds of active operation before it has to stop, or it blows up.

There's not much you can do at the PCB/assembly level to improve this -- there's just not enough time and material to carry the heat away fast enough.  You can improve the retry rate, with heatsinking.

You could also use simplified logic to go between a saturated transistor, and another (smaller) saturated transistor in series with a big fat resistor, to try and soft-start a load.  This doesn't have the same dynamic range as the active version (pass transistor): it may deliver "too much" short-circuit current, and not enough startup current when the voltage drop is small (which means the switch still has to pull in the remainder, so should still be a linear current-limiting type!).

This isn't suggesting anything new -- startup resistors have already been mentioned in this thread.  It's just another way of thinking of it: an array of switched resistors (including zero or one of them!) makes a few-number-of-bits power DAC, while a linear element is of course the analog version.

The advantage with resistors is, basically anything wirewound has metric shitloads* of energy capability.  Apparently the TE Connectivity aluminum-body wirewound resistors can handle quite a lot of power, like 100J, at good prices.  It's hard to find anything else that can dissipate as much energy per dollar of BOM cost.

*That's a technical term.

Tim

[salbayeng]

You can improve the SOA to infinite time (to some extent) by adding a smd PTC with one terminal butted up against the drain tab,  in an overload event the temperature of the pad gets to 120C and the PTC opens.
Its not perfect , as the PTC adds 10milliohm,  and you still need to have some limiting impedance in the rest of the circuit to limit short circuit current to around 100A. as the PTC is a bit sluggish. The SOA is good up to 100ms by virtue of the MOSFET thermal impedance, and good from 2sec up by virtue of the PTC, but there is a bit of vulnerability in the 100ms-1s range.

In all my motor controls I use this approach, and I also sense the voltage drop across the  PTC + MOSFET , and throttle back the PWM when the voltage exceeds 100mV or so. This provides inrush protection with high inertia loads, thermal protection of the MOSFET, and short circuit protection. (the power supply is a  12v gelcell, so the theoretical short circuit current is several hundred amps)
So far I have had zero failures of about 4000 transistors protected this way.

[splin]

When you use a P-ch FET in a soft-start circuit, the goal is to gradually increase Vgs (well, decrease, from zero to negative), which gradually reduces Rds.

When voltage is first applied, Vgs is 0, Rds is infinite, and no current flows.  As the RC filter allows Vg to drop below Vs, Rds starts to decrease, and current starts to flow with almost the full 24V drop across Rds.  After some time (configurable with your R and C values, probably on the order of milliseconds), Vg will drop to its final voltage at or near ground, Vgs will be the full 24V, and...

... you have a dead FET since most have absolute max Vgs of +/-20V

Sorry. I know it's an old post but since the thread has been revived...

[Mad ID]

When you use a P-ch FET in a soft-start circuit, the goal is to gradually increase Vgs (well, decrease, from zero to negative), which gradually reduces Rds.

When voltage is first applied, Vgs is 0, Rds is infinite, and no current flows.  As the RC filter allows Vg to drop below Vs, Rds starts to decrease, and current starts to flow with almost the full 24V drop across Rds.  After some time (configurable with your R and C values, probably on the order of milliseconds), Vg will drop to its final voltage at or near ground, Vgs will be the full 24V, and...

... you have a dead FET since most have absolute max Vgs of +/-20V

Sorry. I know it's an old post but since the thread has been revived...

You accoung for that by using a R divider, not just R. Limits Vgs to say 10V and still has dv/dt operation.

[Circlotron]

Some say NTC thermistor.
Some say relay and resistor.
You can get over all your inrush problems by using an NTC thermistor *and* a relay together. This has a number of advantages.
1/ You can use a larger value NTC than you could a resistor and still have it pull up to near final voltage.
2/ After the relay switches on the NTC begins to cool, so if there is a power interruption the NTC is all ready to soft start.
3/ Power dissipation (except for relay coil)is practically zero, unlike NTC by itself.

As far as relays being unreliable, in this particular case it would be switching off under zero current conditions, so no arcing. At switch on, provided the NTC has been given enough time to heat up sufficiently, the voltage across the relay contacts will be relatively low. The big one though is contacts sticking, and one point many people overlook is that relays work best if the coil voltage is interrupted suddenly and allowed to flick up to a decent voltage. The worst thing you can do is put a reverse diode across the coil winding. This causes the flux to decay slowly and the armature comes away from the pole piece gently, so there may be insufficient kinetic energy in the moving armature to separate the contacts if they have microwelded together. For a 12 volt relay I often use a 75 volt zener across the switching transistor to allow the flux to drop fast. The relay makes a distinctly different sound compared to when it has a reverse diode across the winding.

[T3sl4co1l]

You can improve the SOA to infinite time (to some extent) by adding a smd PTC with one terminal butted up against the drain tab,  in an overload event the temperature of the pad gets to 120C and the PTC opens.
Its not perfect , as the PTC adds 10milliohm,  and you still need to have some limiting impedance in the rest of the circuit to limit short circuit current to around 100A. as the PTC is a bit sluggish. The SOA is good up to 100ms by virtue of the MOSFET thermal impedance, and good from 2sec up by virtue of the PTC, but there is a bit of vulnerability in the 100ms-1s range.

Rather, this is applicable when you have an SOA giving you ~100ms of hold time.  I'm guessing your application is low voltage, 12V, possibly 24?

I've used the transistor-PTC pair before; it works for power levels about 2-10 times the transistor rating.  Higher than that, and heat flow is probably too slow to save the transistor and either the transistor has to shut off, obviating the PTC, or the transistor fails shorted and the PTC does its job normally with only a one-time assistance from the transistor.   Lower than that, and obviously the transistor isn't going to overheat very much, and it's pretty much fine without worrying about the PTC (but if you have a high ambient temperature rating, that lowers the threshold and the PTC becomes worthwhile again).

Also on the order of 100ms, you get some help from thermal connections, mostly for direct connections like a TO-220 greased to a heatsink (not a thermal pad or insulator), or a slight improvement to a D2PAK on heavier copper and solder filled vias in pad.  Downside, the amount this helps by is difficult to model or test.

Tim

[salbayeng]

Yes Tim, really only suitable for 12-24v,  at 48v the circuit impedances and the quadrupling of I²T conspire to make it difficult to ensure survival of the transistor,

with only a one-time assistance from the transistor.

yeh but at least you can unsolder the dead transistor without having a charred crater there
I've also used the PTC between the MOSFET (or bipolar) and an inductor, for switching applications, particularly where the the MOSFET (or halfbridge) is driven from a microcontroller, and you can't trust the micro to always output the correct duty cycle. (or frequency).  We did have sporadic failures of one transistor in a boosting half bridge  (5 in 500) , but could never replicate the failures on the testbench, using a different inductor, and doubling the frequency seems to stop the failures. 
Having a liberal sprinkling of PTC's on a PCB helps with field servicing, just tell the tech's to put their fingers on the all the yellow and green components.
We have also had booby trap issues with the ATMEGA328, where one of the PWM pins is used for JTAG, and factory fresh chips have the pullup turned on for this pin.

[dazz1]

Hi
I looked at the simplest option of a series inductor shunted with a reverse biased diode.
Basic simulation showed I needed a 1mH inductor to keep the in-rush current below 70Amps on  a 12V battery fed system.  The shunt diode does very little, with only about 5 amps flowing when it conducts.
In addition, even with the diode shunt, the L/C combination showed a relatively slow harmonic decay.

This was fixed in the model by adding a series resistor to damp the oscillation.  By itself, adding a resistor was sufficient to reduce the in-rush current below 70A (the battery trip current).  Adding a resistor would be simple but inefficient.

Installing an 1mH inductor that doesn't saturate at 70A is not an option.  Too big, too heavy and too expensive.  An active in-rush limiter is the most obvious solution.

I found this paper  http://www.mosaic-industries.com/embedded-systems/_media/pdfs/application-notes/motorola-an1542-active-inrush-current-limiter.pdf that describes simple circuits that do what I want.   I don't need overcurrent protection because the LiFePO4 battery includes a crowbar circuit as part of the internal BMS.  I already have a standard fuse as well.    I like the idea of the PTC thermally linked to the transistor but if the transistor is rated to take more power than the battery protection, I wouldn't need the PTC.

What I don't currently have is a battery fuel gauge so combining the in-rush current limiter with a fuel gauge on the same PCB would make sense.

Dazz