LT3045 crowbar

[matthuszagh]

I'll be using an LT3045 (https://www.analog.com/media/en/technical-documentation/data-sheets/LT3045.pdf) to power an expensive device that I really don't want to damage. The LT3045 has a lot of built-in protections (current limit, thermal overload, reverse polarity input, under-voltage lockout), but I didn't see anything about output overvoltage protection. So, I was planning to add a simple crowbar at the output, consisting of a fuse, a zener, a few resistors and a thyristor. The device requires 15V and a max of 350mA.

Is there any reason why I shouldn't do this?

[David Hess]

The power good feedback function of the LT3045 could be used to trigger the crowbar, or disconnect the output load with a series connected power MOSFET.

[matthuszagh]

The power good feedback function of the LT3045 could be used to trigger the crowbar, or disconnect the output load with a series connected power MOSFET.

This is a reasonable suggestion, though I would only ever expect the output voltage to sufficiently exceed the set point if the LT3045 fails in some way, so I'm a bit reluctant to rely on features provided by that regulator.

[matthuszagh]

More protection is a good thing generally.
They do make integrated overvoltage (and over current, polarity, ...) protection ICs
e.g. the ones linear/analog devices called "surge stopper" and similar.

If you're really worried about making sure the supply is in tolerance e.g. +/- 10% or whatever then
you could use some kind of supply supervisor / watchdog that ensures the correct limits and shuts off the
supply if there is a problem.

A zener based one will be less precise than that generally but you could always use a more precise option like TL431
or something in a similar circuit if you wanted better cut-off-voltage accuracy.

You could also add more TVS protection in this too if you want to.

Thanks for suggesting this; it was very helpful, especially since I didn't know these sorts of protection ICs exist. It seems like the LTC1696 (https://www.analog.com/media/en/technical-documentation/data-sheets/1696fb.pdf) + SCR would be a good fit here. I'll also add a TVS diode. A lot of these protection ICs reenable the voltage when it falls back in range. In my case, I only expect an excess voltage when a serious fault occurs so I'd rather go with the traditional approach of keeping the rail shorted and blowing a fuse.

[iMo]

It depends what is your precious equipment like.. The fuses are rather slow devices. Also when talking for example a transceiver, the circuit around the thyristor's gate should be well decoupled (what slows down the response even more). I tried that once upon a time and rejected the idea after some experiments. I put TVS diodes everywhere then..

[matthuszagh]

It depends what is your precious equipment like.. The fuses are rather slow devices. Also when talking for example a transceiver, the circuit around the thyristor's gate should be well decoupled (what slows down the response even more). I tried that once upon a time and rejected the idea after some experiments. I put TVS diodes everywhere then..

Definitely, and that's why I plan to put a TVS diode there as well. Even fast-blow fuses are slow (as you've pointed out), but the thyristor will shunt the current from the moment it's activated. I'll use one that's capable of handling more than the fuse current so that shouldn't really be a problem. So, TVS will kick in first, followed by thyristor and eventually fuse will blow in failure. And the thyristor will stop the TVS from dissipating too much energy internally.

The overvoltage seen by the load will be very shortly lived thanks to the TVS. I expect the load would be fine without the TVS there, but more security is better.

[Gerhard_dk4xp]

The power good feedback function of the LT3045 could be used to trigger the crowbar, or disconnect the output load with a series connected power MOSFET.

This is a reasonable suggestion, though I would only ever expect the output voltage to sufficiently exceed the set point if the LT3045 fails in some way, so I'm a bit reluctant to rely on features provided by that regulator.

No, it isn't.  During powerup, power is not yet good by definition and that
would trigger the thyristor, preventing to start the system forever.

Gerhard

[Terry Bites]

As noted you cant rely on the LT3045 to trigger a crowbar. You can improve the trip point accuracy and speed by using a TL431 as a precision comparator. A small PNP inverts the 431s output and provides plenty of gate current. C1 creates a miller integrator to filter out fast transients and prevent false triggering. Icrowbar is for illustrative purposes.

R1 R2 set the theshold.
Vt= (1+R2/R1)*Vref
Vref=2.5V for the TL431.

[WatchfulEye]

The power good feedback function of the LT3045 could be used to trigger the crowbar, or disconnect the output load with a series connected power MOSFET.

This is a reasonable suggestion, though I would only ever expect the output voltage to sufficiently exceed the set point if the LT3045 fails in some way, so I'm a bit reluctant to rely on features provided by that regulator.

Definitely don't do this. LT3045 power good only detects undervoltage, not overvoltage. If you try to be clever, and set the trip point above nominal output voltage, then this doesn't work.

LT3045's fast start mechanism means that it does not establish its voltage set point until AFTER it has received the PG signal. If it doesn't get a PG signal, it'll output the full input voltage within a few ms. 

I had exactly this issue one board I built, the issue was a dry joint on one of the resistors on the PGFB divider, resulting in no PG signal being generated. The device remained in fast start mode, which delivers approx 20x the normal reference current to the set resistor.

[ledtester]

There was a similar discussion about two months ago:

https://www.eevblog.com/forum/projects/best-design-option-for-over-voltage-monitoring/

Another approach is to use a dedicated over-voltage protection switch. For instance, I was able to find the KTS1662 which has an integrated MOSFET and sub-microsecond response time:

https://www.mouser.com/datasheet/2/936/KTS1662-04d-1634486.pdf

[tchiwam]

I am playing with quite a few LT3045, care has to be taken on step unload. If you unload the LDO it may overshoot by quite a bit depending on your C-Set and sense.

[wizard69]

It depends what is your precious equipment like.. The fuses are rather slow devices. Also when talking for example a transceiver, the circuit around the thyristor's gate should be well decoupled (what slows down the response even more). I tried that once upon a time and rejected the idea after some experiments. I put TVS diodes everywhere then..

Definitely, and that's why I plan to put a TVS diode there as well. Even fast-blow fuses are slow (as you've pointed out), but the thyristor will shunt the current from the moment it's activated. I'll use one that's capable of handling more than the fuse current so that shouldn't really be a problem. So, TVS will kick in first, followed by thyristor and eventually fuse will blow in failure. And the thyristor will stop the TVS from dissipating too much energy internally.

The overvoltage seen by the load will be very shortly lived thanks to the TVS. I expect the load would be fine without the TVS there, but more security is better.

In the few supplies that triggered a crowbar that I've seen the fuse often remained intact.   A crowbar can destroy a supply in quick order.   Then agian it can be hard to tell what failed first.   In any event you are right in that if the crowbar triggers the load should not see the high voltage, the dead short shold take care of that.

The big problem with crowbars is unintened triggering.   Are you sure your expensive device will never send a transient back to the supply and cuase it to self destruct.   Honestly I'm not a big fan of this approach.

[matthuszagh]

I am playing with quite a few LT3045, care has to be taken on step unload. If you unload the LDO it may overshoot by quite a bit depending on your C-Set and sense.

Interesting. Any configuration you've found to work well? I'm using a large (4.7u) bypass cap for the set pin (same as used for the dev board) so hopefully that will help mitigate somewhat. I also don't expect large load steps - it should be pretty steady but I would prefer not to overshoot and trigger the crowbar.

[matthuszagh]

It depends what is your precious equipment like.. The fuses are rather slow devices. Also when talking for example a transceiver, the circuit around the thyristor's gate should be well decoupled (what slows down the response even more). I tried that once upon a time and rejected the idea after some experiments. I put TVS diodes everywhere then..

Definitely, and that's why I plan to put a TVS diode there as well. Even fast-blow fuses are slow (as you've pointed out), but the thyristor will shunt the current from the moment it's activated. I'll use one that's capable of handling more than the fuse current so that shouldn't really be a problem. So, TVS will kick in first, followed by thyristor and eventually fuse will blow in failure. And the thyristor will stop the TVS from dissipating too much energy internally.

The overvoltage seen by the load will be very shortly lived thanks to the TVS. I expect the load would be fine without the TVS there, but more security is better.

In the few supplies that triggered a crowbar that I've seen the fuse often remained intact.   A crowbar can destroy a supply in quick order.   Then agian it can be hard to tell what failed first.   In any event you are right in that if the crowbar triggers the load should not see the high voltage, the dead short shold take care of that.

The big problem with crowbars is unintened triggering.   Are you sure your expensive device will never send a transient back to the supply and cuase it to self destruct.   Honestly I'm not a big fan of this approach.

This has been worrying me a bit. My design is a bit strange in that the fuse rating (500 mA) is at the very upper limit of what the lt3045 can supply. Max DUT current is 333 mA (15 V). I've also used the current limit functionality of the lt3045 to limit to 385 mA. I figured the crowbar would only trigger if the lt3045 failed (set to trigger at 15.8 V), but maybe it would be possible for this to trigger under normal operation. In that case, the LT3045 would continuously dump 385 mA into a short circuit and never blow the fuse. Also worth mentioning is that the LT3045 isn't spec'd to put out more than 15 V.

The DUT should be pretty quite and I wouldn't expect spikes: it's the combined oven/oscillator supply for a very high quality OCXO.

The manufacturer gives the allowed voltage as 14.25-15.75 V, so that's what I'm trying to meet. I'd be a bit surprised if I can't push the voltage a bit higher without damage, but I really can't say that with any certainty. If I still use the crowbar, I guess I could raise the trigger point and assume the DUT will still be fine if hits 16.5 V or so. Alternatively, I could filter out fast transients with the Miller capacitor as shown in the design a few posts earlier. That one makes me a bit nervous though as it can send a current pulse through to the SCR gate. I doubt it would be enough to trigger it, but still...

Anything you prefer to crowbars? I guess I could go with one of these overvoltage disconnect protection ICs. I tend to put stock in design decisions made in old HP equipment and they used zener+thyristor crowbars all over the place, but maybe that's because there weren't too many alternatives in the 80s.

[matthuszagh]

A few more points:

There's a very useful discussion of crowbars in the Art of Electronics (9.13.1). I'll probably swap out the crowbar I had for the simple one they described using a tl431+triac instead of scr in Fig. 9.102C. That also includes a parallel capacitor at the TL431 reference input to ignore short spikes.

The regulator current limit would prevent the fuse from blowing (unless the lt3045 failed in some applicable way). However, the LT3045 has an internal foldback current limit that kicks in when the in-out voltage differential is greater than 12V (applicable to my case since the crowbar should drop the supply down to about 1V from 15V). That should protect the lt3045. The fuse still won't blow but nothing should be damaged. FYI there's also thermal overload circuitry in the lt3045.

[David Hess]

Below is the 12 nanosecond circuit breaker from Linear Technology application note 13 - High Speed Comparator Techniques.  Of course you probably want supply side current sensing and a MOSFET instead of bipolar switch, but it still gives an idea of what kind of speed is possible.

[tchiwam]

The failures of LT3045 is rather annoying. It will potentially short the input and output together hence I always keep my Vin near the needed Vout.

The drop out voltage is 360mA but the performance is better around 1V above. For my purpose

If possible I keep my Vin under the Absolute VCC of the device I want. In my case a LMK1C1108 I will keep the Vin around 4V.
I blew at least 1 of them by having Vin = 9V and when the LT3045 blew, it blew my LMK1C1108 too... At 4V it is above Abs max of 3.6V but it survives.

ILim resistor I keep at 90% just to be on the safe side. A bit more forgiveness with step loads. With the value in the app note, it blew as mentioned above...
I double the current output I need with the LT3045, if I need 500mA I use 2 of them. Then ILim at 90%
I have added a LT8609 (ultra low EMI from APP note config) 2A in front of the 2x LT3045 driving a max of 450mA Output caps were 10uF and I got them down to 4.7uF, and ordered 2.2uF

This LT8609 + LT3045 is still untested. I may change the LT8609 for a Low noise isolated DCDC at a later stage to avoid current loops.

So yeah, low noise indeed.. Alone I don't like it.

[Geoff-AU]

There's a very useful discussion of crowbars in the Art of Electronics (9.13.1). I'll probably swap out the crowbar I had for the simple one they described using a tl431+triac instead of scr in Fig. 9.102C. That also includes a parallel capacitor at the TL431 reference input to ignore short spikes.

I've made one of these and it works, but if you have a very slight overvoltage (in my case, slowly creeping up on the trigger point) you can kill the triac because the switch-on is a bit soft.  Mine dumps a fairly large capacitor bank so it may be pushing the friendship a bit.  I just wrote a note to myself to replace the fuse AND the BT136, it's easier than improving the trigger circuit since crowbar operation is not a regular event.

The normal higher voltage diac triggers avoid this because the diac turns on quite suddenly.
 

[matthuszagh]

There's a very useful discussion of crowbars in the Art of Electronics (9.13.1). I'll probably swap out the crowbar I had for the simple one they described using a tl431+triac instead of scr in Fig. 9.102C. That also includes a parallel capacitor at the TL431 reference input to ignore short spikes.

I've made one of these and it works, but if you have a very slight overvoltage (in my case, slowly creeping up on the trigger point) you can kill the triac because the switch-on is a bit soft.  Mine dumps a fairly large capacitor bank so it may be pushing the friendship a bit.  I just wrote a note to myself to replace the fuse AND the BT136, it's easier than improving the trigger circuit since crowbar operation is not a regular event.

The normal higher voltage diac triggers avoid this because the diac turns on quite suddenly.

Good to know. I'll probably stick with the trigger circuit provided earlier by Terry then. I expect the PNP should give a nice, strong trigger signal. Like the circuit in the Art of Electronics, I'll use a 1n cap across the TL431 reference to prevent false triggering from fast glitches.

[gman76]

Why not use an OV/UV supervisor, like the LTC2912?  Use it as an independent circuit that monitors Vout.  You can tie the /OV pin to the 3045's EN pin.  You can also tie the /LATCH pin to GND to latch off if the 3045 goes into an OV condition.  The 2912 operates up to 6V but has a built-in shunt regulator if you need to run from a higher voltage.

[Geoff-AU]

Why not use an OV/UV supervisor, like the LTC2912?  Use it as an independent circuit that monitors Vout.  You can tie the /OV pin to the 3045's EN pin.

If the LT3045's pass BJT failed short circuit for some reason, fiddling with the enable pin isn't going to do anything useful.

[nealix]

As noted you cant rely on the LT3045 to trigger a crowbar. You can improve the trip point accuracy and speed by using a TL431 as a precision comparator. A small PNP inverts the 431s output and provides plenty of gate current. C1 creates a miller integrator to filter out fast transients and prevent false triggering. Icrowbar is for illustrative purposes.

R1 R2 set the theshold.
Vt= (1+R2/R1)*Vref
Vref=2.5V for the TL431.

I like what Terry shows above, but due to 10 ohm sense resistor and BJT, it is not going to be great for much current in a load.
But the idea is widely used;   Rather than a crowbar and fuse, which places stress on the power supply parts and is slow, a lot
of people, even in commercial products, are now using a fast comparator and MOSFET switch to OPEN the load rather than short
the power supply.   I use two 40 amp DC load distribution bars that have the MOSFET and comparator built in.   When I dial up
the output voltage on the 50 amp power supply, the moment it hits the trip point everything simply turns off.   No blown fuses.

Neal

[Geoff-AU]

MOSFET to open the load works great and less destructive, as long as you aren't in danger of overvolting the MOSFET.  If you are, adding an extra fuse/sidactor on the input would be an easy protection method that would only activate if things got really bad.  (you can get some properly impressive MOSFETs these days though, so it's unlikely you'd need to).

[matthuszagh]

FYI Based on the interesting discussion here I asked another question concerning a comparison between the crowbar and comparator + MOSFET approaches (https://www.eevblog.com/forum/projects/comparison-of-overvoltage-protection-methods/).