Trinamic TMC5160 stepper controller fails catastrophically!?

[Kremmen]

Anyone else seen this happen? I have designed a board for a simplish application involving a NEMA34 stepper (a Trinamic) controlled by a TMC5160 and that one in turn with a jellybean Atmega320P over SPI.
The driver is supposed to work up to 60V rail voltage and my target is 48V. While testing with a lab supply providing 30V everything works as expected - motor turns and the stepper driver obeys commands from the MCU as planned.
When i changed to the proper 48V supply to my surprise the stepper driver promptly exploded. I replaced the driver chip and same thing (2nd time there was an actual pinhole in the epoxy shell).
What is even more curious is that if i slowly raise the rail voltage using the lab supply, the driver will work with 48V just as expected. It is the turn-on that kills it.

The schematic is practically identical to Trinamic's reference designs in the datasheet (see attached). As far as i can read, no operational or abs max values are exceeded. The highest voltage is 48V and that should be well within the allowed range.

I haven't found anything about power sequencing requirements in the atasheet but this failure surely has something to do with the power ramp-up since slowly rising the supply is OK. And as noted, at 30V it works every time.

Any ideas related to the schema or any other possible cause are appreciated since of course the project is in a desperate hurry to deliver. Aren't they all.

[ajb]

This will depend a lot on the specifics of your supply, power distribution, other loads, etc.  How is everything hooked up?  Is there any protection between the outside world and the driver?

[capt bullshot]

You've probably got a transient overvoltage at turn-on.
One reason for this might be the parasitic inductance of the supply (e.g. just a long enough wire) forming a resonant tank circuit with the bypass capacitors. Ceramic capacitors have low ESR, so that tank might have high Q causing a high voltage spike as a result of the rising edge of the supply voltage.
One simple cure is to add some damping, easily done by adding a bulk electrolytic capacitor to the 48V rail.

[Kremmen]

Thanks for the replies.

I did think of power supply overshoot but kinda discarded the idea for 2 reasons: A proper lab supply should not overshoot and my Rigol does not seem to do that. The second reason is that if you check the schematic, there is 880uF of electrolytes per bridge on board already.

But i will check this in more detail since fixing that particular problem would be a relatively easy cure.

Meanwhile, any other ideas would be very welcome.

Image of the proto asembly as well as the pinhole crater in the stepper driver attached.

[Kremmen]

... and the rise curve of the +48V power supply as measured over one of the bridge electrolytics.

[jeremy]

Email trinamic (and report back here!). They have been very helpful to me in the past.

[Kremmen]

The thought has occured. I will, but was kinda hoping for a quick solution if someone here has encountered this before.

Next thing i'm going to try is supply the driver VSA aux voltage from a separate regulated voltage, say 15-20V and see what that does. The pinhole crater in the driver package is suspiciously close to the VSA and associated ground pins...

[ajb]

Since that's an external-driver part, I assume it has connections to the switched nodes, right? (I haven't looked at the datasheet and the schematic is kind of indecipherable)  It's possible that the problem is actually occurring in the bridge driver section, something is going wrong to cause massive ringing or something that is enough to cause a cascading failure there.  Does the problem occur immediately on power-on?  Does it occur even if you keep the driver disabled?

[Kremmen]

Naah, read the previous. It all works nicely at 30V, no ringing, no symptoms. It even works at 48V if only you raise the voltage slowly enough. I have verified that the rail does not ring or overshoot during "normal" power-up (see scope capture above), the driver just blows.

P.S: just curious: why is the schematic indecipherable?

[capt bullshot]

Didn't notice the electrolytics on the first glance, and your 48V ramp looks harmless, agreed.

Next thing that comes to my mind:
The trinamic chip has more than one GND potential applied: GNDA, GNDD and GND if I see that correctly.
Then you've used inductors to separate these GND potentials. From my experience, this is a real bad idea since this can lead to a high noise level between these grounds and quite high transient voltages occuring across them. You already have differential sensing for the current shunts, so a separate GND for analog / digital / power should not be necessary.  In most cases (at least to my experience), splitting GND planes or having multiple GND potentials leads to more problems than it is intended to solve. Try shorting them (use wire bridges instead of the inductors), even ferrite beads can be evil here. As an alternative, to protect the chip, one can apply beefy anti-parallel Shottky diodes between all the GND potentials.

[Kremmen]

The inductors are actually ferrites but yes. I separated the grounds to avoid ground current loops around the MCU. The only ground common to the MCU and the power stage is the MCU to driver I/O circuit return. Again, while there are reasons to separate grounds and resons to avoid separation, that does not seem to play a part in this.

Actually, i think i found the culprit: Trinamic datasheet appears to be slightly optimistic regarding the max auxiliary voltage of the driver. The logic voltages are generated by internal regulators from an input pin that the datasheet says may be directly connected to the power rail. 48V seems to be a bit much however although 50V is stated in the standard operating range. Once i bodged a small regulator to drop the aux input to about 22V - no problemo any more. At least it will now tolerate turn-on of the lab supply that earlier blew the chip right away. Next to see if i'm lucky with the real PSU as well. But i think so. Fingers crossed.

[Siwastaja]

Even if not related to the issue at hand, get rid of that ferrite bead multi ground plane disaster: use one, continuous ground plane, unless you have very strong reasons and expertise to think otherwise. If you think you could have ground loop issues, you need actual isolation: there's no way around it (optical or magnetic isolator ICs).

[Kremmen]

Well. It definitely is not a disaster since the thing works nicely with no undue noise or oscillations. The issue at hand had nothing whatsoever to do with the grounding arrangement, so maybe discuss this in a more relaxed manner?.

I don't have _strong_ reasons to separate the grounds and will grant that it may not be necessary in this case. But i do have reasons. Commonly power circuits separate the signal and power grounds to avoid noise and ground bounces in the signal circuit grounds. Also Trinamic has so separated the grounds. You might think of the MCU and power stage as 2 separate systems, connected only by the I/F signaling referenced to the signal ground. So where is the benefit of spreading the power ground over that?
BTW i didn't mean ground loops in the cabling sense; more that i don't want the power stage return currents circulating uncontrolled around the MCU. Maybe it won't be a problem anyway.

Edit: I decided to see what effect, if any, a single ground will have, so i removed the "disaster" to get a data point. Probably no big change in any direction but interesting to see.

[mikerj]

Deliberately putting any kind of impedance into your ground paths is a bad idea, and you can't fix ground loops like this anyway.  Careful layout with a solid low impedance ground plane is the way to go.

[enoch@eevblog]

Did you solve this issue? I'm getting the same issue as well.