Reverse power flow through step-down DC-DC

[Unixon]

Hi, I guess I need an advice on how to deal with this...

There's a device with two power inputs +12V DC and +5V DC (USB) and a TPS565208 step-down DC-DC converter circuit between them.
It is possible to power this device from either input or both inputs, however when +5V source is connected and +12V source is not connected,
a weird failure may occur if +5V rail is not stable enough -- DC-DC starts oscillating its input node to a voltage it can't handle and then it destroys itself.

Is there an good and/or easy (minimal modification) way to solve this?

[Unixon]

There are significant number of assembled boards which have to be modified in some way so I'm limited in options.
I wonder if a dummy input load will work to drain charge from input caps when +12V source is not connected...

[ajb]

It'd be a lot easier to help you with a schematic.  Also if you're looking for advice on rework strategy, a PCB layout would help with that.

I'm not sure I'm following your description correctly.  Is the output of the converter directly connected to the 5V input to the device?

[Unixon]

It'd be a lot easier to help you with a schematic.

I'm sorry, I think I can't can post whole schematic, only DC-DC part of it, but there's nothing special about it.

Also if you're looking for advice on rework strategy, a PCB layout would help with that.

Not yet. I'm looking for an idea on now to prevent this behavior of DC-DC converter in principle.

I'm not sure I'm following your description correctly.  Is the output of the converter directly connected to the 5V input to the device?

Almost, and that is the part of the problem.

We have two boards, one 3rd party board with USB input and our 2nd board that contains among other things a step-down DC-DC.
When boards are connected, our board with DC-DC supplies +5V rail to the board with USB.
It is up to an end user to decide which combination of power supplies to use.

Recently we found out that if USB power supply is connected and +12V supply is not, sometimes DC-DC blows up for no obvious reason.
It was hard to replicate this problem but eventually I discovered that if 5V from USB is noisy enough,
TPS565208 starts boosting voltage on its input (transferring power back-to-front) to a higher level that it can withstand.

Initially we had a Schottky diode between these 5V rails on 2nd board, but it dropped too much voltage and we had to remove it and replace with a jumper or PTC.
In recent revision of 2nd board there's an "ideal diode" made out of MOSFET and BJT current mirror and we don't have this problem.
But we still have to do something to already manufactured boards of that old revision that should have had a diode.

[eblc1388]

Initially we had a Schottky diode between these 5V rails on 2nd board, but it dropped too much voltage and we had to remove it and replace with a jumper or PTC.

Would it helps if you add the Schottky diode in the following manner(see image)

 

[Unixon]

Would it helps if you add the Schottky diode in the following manner(see image)

Probably yes, but we can't fit that onto existing boards without badly affecting their appearance.
High frequency DC-DC layout is very tight and the board in question is already populated as hell.
Whatever the modification is it must look nice not turn the product into garbage.

[capt bullshot]

The auto destruction mechanism is "as expected" for a synchronous DC/DC converter.
You need to disable the converter if your circuit is fed from the 5V input, or decouple the converter from the 5V rail. If a schottky diode drop is too much of a voltage, one could consider adding another (smaller) schottky diode in series with R7, or simply adjust R7/R6 to provide a somewhat higher voltage to compensate for the diode drop.
You might try to set a higher turn-on voltage for the converter by adding a resistor from R5 / IC3 Pin 5 junction to GND.

[Unixon]

The auto destruction mechanism is "as expected" for a synchronous DC/DC converter.

Why is it even turning on? The output voltage from 5V rail is about the same as its own set voltage, IMO it shoudn't be switching at all.

You need to disable the converter if your circuit is fed from the 5V input, or decouple the converter from the 5V rail.

I would say it must not enable itself when there's no power at input connector, i.e. the correct way of doing this is to teach it to distinguish between input voltage occurring from a connected power supply and a voltage occurring from reverse power flow.

If a schottky diode drop is too much of a voltage, one could consider adding another (smaller) schottky diode in series with R7, or simply adjust R7/R6 to provide a somewhat higher voltage to compensate for the diode drop.

Compensation was one of ideas, eventually it didn't take off because we have onboard circuits coupled with DC-DC without a diode and compensating decoupled rail would drive coupled rail above allowed voltage range.

You might try to set a higher turn-on voltage for the converter by adding a resistor from R5 / IC3 Pin 5 junction to GND.

This may actually help... Should this turn-on voltage be higher than any voltage the converter can boost itself to? Just higher than 5V rail + noise?

[capt bullshot]

As soon as there's high enough a voltage on its input, it'll turn on.
And once it is on, the control loop cannot work, since the output voltage is "fixed" and not under control of the converter. So depending on the actual 5V "input" voltage, the (reverse) "output" voltage will rise in an uncontrolled manner once the converter is on (this applies to synchronously rectified converters, typically not to ones that have a diode rectified output).
The internal high side switch will conduct in reverse through the mosfet body diode. So if you have 5V at the output, you'll have at least 4.3V at the input, even if the converter is disabled. Some special purpose converters add another (back-to-back connected) mosfet for reverse current protection in this case.
IMO, you should set the turn-on threshold as high as possible for your circuit to work as desired when the external 12V supply is connected.
Anyway, it's always a good idea to decouple (active-"ORing") power supplies if they are intended to power the same rail at the same time. You could damage your supplying 5V USB in case the 12V converter is on, as it might supply reverse current into the USB supply. Depends on that supply what will happen. Or what just happened to you, the USB supplied 5V will damage the 12V converter. Some supplies / converters can handle reversely applied power, others not. In general it's safe to assume they can't and take measures to prevent.

[nali]

The TPS565208 enable pin threshold is 1.6V, so you could maybe tack a resistor from there to ground to form a potential divider, so the regulator only works with a higher voltage say 10V.

[Mau5]

Hi,

I don't think that this solution will help you on an already manufactured board..
But I once designed a mosfet switch, that allows you to distinguish between two sources and prevents the described behavior.
The use case was exactly the same. I had a 5V supply from a Buck-Converter and another 5V supply from USB.
The mosfet switch network ensures that only one source is active at a time and has a small voltage drop.
An image of the schematic is attached.

Maybe it would help to just insert a pin-header that either connects the 5V net to the USB or the DC-DC input.
I know this is very cheap but maybe it is the only feasible thing that you can do on a populated board.

[Unixon]

Rising UVLO threshold didn't help, it just insignificantly delays a moment when input caps are charged to a higher voltage.
It looks like 5V rail noise pushes current through the inductor and slowly charges input capacitors until voltage is high enough to overcome UVLO and then DC-DC turns on, works for some time, then it turns off (or doesn't).
This cycle can happen on and off many times, when DC-DC is on I see about 25V on input caps and this is already their rated voltage.

Loading unconnected 12V input with a resistor kind of helped, but required resistor value makes this idea ridiculous -- static heat dissipation by this dummy load ruins everything.

[poorchava]

You can try bodging an N-MOSFET at the output of the DC/DC, source to dcdc, drain to 5V rail, gate to 12V supply, +resistor between gate and source. This way the dcdc will be isolated from the 5v rail unless there is voltage at the 12V port.

I haven't simulated this, but i think it should work (came up with this just now, not sure if I haven't made some logical mistake). Using a normal threshold mosfet (ie not logic level type) you should get a voltage headroom of maybe 3...4 Over the 5V rail before the fet turns on. Kind of the reversed version of pmos reverse polarity protection.

[NiHaoMike]

Loading unconnected 12V input with a resistor kind of helped, but required resistor value makes this idea ridiculous -- static heat dissipation by this dummy load ruins everything.

How about a zener diode, something like 18V?

[langwadt]

maybe change the voltage divider so the nominal output from the switcher is slightly above 5V, so it won't try to pull down the USB ~5V
put a ~15V zener on the 12V input

[Unixon]

How about a zener diode, something like 18V?

Hmm, yes, I've been thinking about this too... I'll try this next.

My concern was that reverse flow might be energetic enough to blow either zener / TVS or converter itself.
Since nothing bad happened with a resistor, I guess a zener diode will also be fine.
p.s. 100R worked for dummy load, 1k didn't, max workable value must be somewhere between them, but hundreds of ohms is too small resistance for this purpose.

[Unixon]

maybe change the voltage divider so the nominal output from the switcher is slightly above 5V, so it won't try to pull down the USB ~5V

It's very difficult to set a fixed voltage with that high precision. Typical allowed range is 5% and DC-DC precision is about 2% even with good feedback divider. We've tried that approach in order to compensate for Schottky diode loss but we couldn't come up with a repeatable result. For both supplies the specs are 5V+/-5% and this make it nearly impossible to be both slightly above another supply and stay withing specs at the same time.

put a ~15V zener on the 12V input

Yes, sure.

[Unixon]

Well, a zener diode at input caps mostly works fine, at least it's the best solution for a nice looking mod that saves the board.
A downside is pretty high energy consumption during self-excitation phase, the inductor and the zener diode (I've used 15V one) are pretty hot, the converter IC -- not so much.
Nothing blows up in short time frame (minutes), don't know if it holds for really long time like days/weeks/months.
Definitely not recommended for battery powered devices...

[capt bullshot]

Try a zener diode and a transistor:
Zener from input node to base of transistor base, transistor emitter to GND and collector to feedback node of the regulator. This should draw the control loop into the correct direction to stop reverse energy flow.
Or connect to collector to the enable pin, to turn off the converter if the input voltage gets too high.

[Unixon]

Try a zener diode and a transistor:

OK, I'll see what works best. Unfortunately, there's no nice way to budge this into an assembled board.
We have 200+ boards with this issue and after installing a mod they must look clean like they've been initially assembled that way.

[fcb]

Wire the EN (enable) pin to the +EXT input (via 10K), you'll only switch on the converter when the external 12V is present.

You'd have to do it before the reverse-polarity protection diode, and you might need a resistor across the EXT input to ensure it doesn't creep up (diode leakage).

[gnuarm]

Wire the EN (enable) pin to the +EXT input (via 10K), you'll only switch on the converter when the external 12V is present.

You'd have to do it before the reverse-polarity protection diode, and you might need a resistor across the EXT input to ensure it doesn't creep up (diode leakage).

+1

Seems like the best idea yet unless you can add the FET like poorchava suggests.  Heck, this is so simple it might be better anyway. 

To do fcb's hack if you can cut the trace from resistor to the diode cathode without disturbing the rest of that trace you can use a wire to connect the resistor pad to the diode anode.  That may show up more than you'd like.  You can also remove R5 and make the connection with a leaded resistor making it look like it belongs there.  I did this on boards that were Rev 1 in order to make them compatible with Rev 2 operation.  Most of the changes were in the FPGA on board, but this was a required hardware mod.  I could have told them I couldn't do it and make $200 per board rather than $50 per board for the upgrade, but I felt I should do this.  Later the customer had rules that precluded even that sort of mod.  No more than one white wire allowed. 

[Unixon]

fcb, gnuarm, thanks, good idea. I know this solution, but it can't be nicely applied to a production board.
I will try to integrate this kind of protection into a new board revision.

To illustrate the problem a little further, I've attached relevant part of PCB layout.
As you can see all power traces are very thick and necessary pads are pretty far away.
Wire budges are prohibited for production boards, all mods like that are only for prototypes.

So, from all solutions suggested until now, there are two definitely applicable:
1) raise UVLO threshold - OK - a 0603 resistor could be nicely soldered between R5 and R6;
2) clamp VIN+ with a zener diode - OK - a SOD80C zener could be nicely soldered on top of C3,C4 or between them, this would allow for better heat transfer away from the zener;
Of course, an experienced person would immediately know this is a budge, but it still looks good enough.

These mods don't solve the problem entirely, but they can buy enough time to rethink one's power supply connections.

[fcb]

+1 for thoroughly testing BEFORE production 

[gnuarm]

Here's something that's bugging me.  Looking at the schematic it looks like the 12V input polarity is backwards.  I know these jacks are not standard, but I've never seen one before that wasn't positive on the inner conductor and negative on the outer sleeve or ring as they call it in telephone work. 

I'm wondering if you've been testing using X2 and not found a bug with X1 that could be very important.

So is X1 wired in ok?