TPS92518HV: Shorted to Ground condition

Hi Sumeet,

Can you help me understand why the part dies when the output is shorted? I got a closer look at the part that was damaged and there is a physical crack in the package.

If the controller switches higher frequency to try to raise the output voltage, and the die overheats, then the thermal protection circuit should kick in and protect it, right?

As long as current is regulated, the mosfet should be unharmed, and even if it died, there would be no voltage high enough to go out of the controller's spec.

I can't think of a way a voltage on one of the pins of the controller would go above spec. What exactly kills the 92518 when the output it controls is shorted?

Thanks,
Nick

Nick,

The part should not die simply due to a short. The part does not "switch at a higher frequency to raise the voltage". In fact since the off time is an inverse function of the output voltage, the off time will extend very long which is why there is a need to set up a max off timer to not have it be indefinitely long. So in fact the max off timer ensures that the switching frequency does not go too low.

This is a constant current regulator - it does not try to raise the voltage - voltage is not a controlled parameter. It tries to regulate current, no matter what the output voltage wants to be (within the range of 0 to VIN - headroom).

If you are seeing damage to the device the cause of that would have to be something else happening when the short occurs - one example could be if there is a long cable between the output of the TPS92518 and the LEDs being shorted, there can be an inductive ringing that goes above or below (especially if it is several volts negative) abs max. This is just one example but we would need to look at everything else in the system and what happens at the time of the short. 

The short in and of itself would not damage the part. As the shunt FET topic describes, a fully shorted condition (with a FET of perhaps a few hundred mohm), is not considered a fault condition, and we routinely use that type of dimming.

Regards

Sumeet

Hi Sumeet,

I'm the customer who prompted Nick to start this thread and who is currently bringing up a prototype with the TPS92518HV.

I have to figure out how the damage occurred and add something to prevent it from happening again.

I have previously damaged the part by unplugging the output cable while the circuit was active, I added a diode to Vin from the output to ensure that doesn't happen again.

Based on the information in this thread it sounds like I need to add a diode from the output to ground to ensure the output never goes too low.

Let me describe what I was doing when I caused this failure.

I had just gotten the part to operate with its default parameters. It was running roughly 500 mA peak with a 50% duty cycle. It was only running at 156 KHz based on the voltage ripple of the output. And the voltage ripple for the 42 volt output was almost 2 volts so I wanted to measure current ripple, just to get a feel for the performance of the part.

I added a 1 ohm shunt resistor in series with the load and hooked up the scope across the resistor. I thought I had put the resistor on the ground side of the output, but instead I had actually put it on the output itself. So I shorted the output to ground through the scope. The prototype was not powered at this time.

I powered the system, loaded my FPGA and then the output enabled by default, there was a few hundred millisecond delay and then a snap sound. Then I noticed that Vin was 5VDC (down from 57.5 VDC) and current from the bench supply was limiting at 1.7 amps. I turned off the system, saw what I had done and then facepalmed.

That part is cracked and dead and as you probably know, they are not easy to replace. I'll probably use the other three channels to debug and the write this board off.

To get the board to work again I had to remove the fuses that feed both sides of the part. It is interesting to note that the Vin pin has roughly 7 ohms short to ground and the EN input is roughly 26 ohms.

Do you think the issue was the output bouncing below ground because of circuit inductance? Will adding a diode to ground on the output prevent that from killing parts in the future?

Thanks much for your help.

Brian

Hi Brian,

thanks for the clarification. Since you had a shorting resistor locally near the LEDs (even though it was accidentally shorting them as you described) - I dont believe this is due to inductive oscillation. The inductance would not be that high. That is usually when you have long cables between the LED driver output and the LEDs being shorted remotely.

In this case the likely explanation is the current ratcheting up due to the Max Off Timer not being appropriately set to handle the short.

As I described above, the max off time is required when the output is very close to zero. Here is the ratcheting sequence:

1. The high side FET turns on and due to the very low output, the inductor current rises very fast. It may exceed the set peak current due to the leading edge blanking time (which sets a minimum on time)

2. When leading edge blanking time is over, the high side turns off, and the off time begins. Now you need the off time to be set long enough for the inductor current to ramp down sufficiently such that the next rise in inductor current due to min on-time does not cause the current peak to further ratchet up. Remember the ramp down will be very slow since there is almost zero voltage across the inductor to induce it to de-energize - you essentially have only a negative voltage of one diode and whatever the "shorted" output voltage is to cause the inductor current to ramp down. 

The max off time now has a default value of 80h (reset value if you dont program it otherwise at startup) which may not be sufficient to prevent ratcheting as described above.

If the max off time is correctly programmed then shorting the output should be possible - however in your scenario there was no opportunity to do so.

Hope this helps.

Regards

Sumeet

Hi Sumeet,

First, let's make sure I understand.

If the toff max is too small, then the current may not fall below the target output current during the off phase, and when the transistor turns back on, the current will ratchet up even farther. This cycle will continue until the transistor, (in my case) one of the sense resistors, or some other series component (like the fuses in my case) fails due to the excessive current.

At this point, current flows from the output in multiple amps, because -- effectively -- the transistor is stuck on. Components begin to fail. When a component fails open circuit, the inductor suddenly has no sink for all this current, so the voltage somewhere in the chain (either into the MOSFET, one of the sense resistors or some other series component) rises rapidly. Or if the input fuse opens up, then the voltage into the part drops rapidly, going far below zero, killing the TPS92518HV. If the voltage rises on the output, I have a protection diode that will likely pop due to the amount of power it's attempting to shunt, if it's the sense resistor or MOSFET that dies, then the nets attached to those components rise to hundreds of volts which then kill the TPS92518HV (after it's own protection diodes pop).

Is that the mechanism of destruction?

Too bad the TPS92518HV doesn't ensure that the current measurement has fallen below the target current before turning the MOSFET back on. That would prevent this issue and it's only widening one gate per controller, right?

I also think the default for Toff(max) should be 255 and there should be a warning in the documentation about shorting the output with Toff(max) set to a low number.

For our project, the simple solution is to set Toff(max) = 255 to ensure the current can come down to a respectable level no matter what the output voltage looks like.

What are the disadvantages of setting Toff(max) to 255? What cases exist where that is not desirable?

Thanks again for your help and thanks for describing the intricacies of this controller.

Brian

Hi Brian,

Thanks for your reply.

Correct - if the inductor valley following minimum Ton is not deeper than the steady state valley, then the inductor current will ratchet indefinitely.

The most common mechanism for damaging the TPS92518 is uncontrolled switching of the gate driver which cause excessive power dissipation because at that point the switching frequency is set by only Ton (min) and Toff (max) which can be high.

To address your two questions:

The reason the part does not wait for the inductor current to return to a specific valley before starting a new cycle is simply that this is not a closed loop controller - in a closed loop part you can have an error amplifier set a valley threshold and have the off time set that way. In this part the off time is open loop - set by the timer based on output voltage (Toff timer) or as in this case overriden by the max off timer when the output is close to 0. This is a trade-off to be able to achieve fast response to fast dimming inputs either through the PWM dimming pins or through shunt FET dimming.

You can set the Toff max to 255 but this can result in very low frequency of inductor current when you are in fully shorted condition. Also, your ripple in shorted condition may not match the ripple in other conditions. In normal conditions the ripple is set by the output voltage, while in short it will be set by Toff max of 255. This can result in the average LED current being different (avg = peak - ripple /2).If output short is not an operating condition for you, and all you need is a way to prevent ratcheting, this could be acceptable. 

However do keep in mind that if you power up into a short at the output, the Toff max will still be the default value.

Hope this helps. Let me know if this answered your questions, and if it did, please hit the resolved button.

Thanks

Sumeet

Sumeet Kulkarni said:

The most common mechanism for damaging the TPS92518 is uncontrolled switching of the gate driver which cause excessive power dissipation because at that point the switching frequency is set by only Ton (min) and Toff (max) which can be high.

Why doesn't thermal overprotection kick in and turn off the output?

Is my example of the open circuit condition due to damaged component causing very high or very low voltages on 92518 pins less likely?

I'm trying to think of ways to ensure the state of the output doesn't cause damage to the system, as the lighting fixture installers often make wiring mistakes. It will happen that the outputs will be shorted, while we are looking for loads. It will happen that they disconnect a fixture while we are driving it at 1 Amp. In other words, besides the diode from output to Vin, what else can I do to protect the part?

Sumeet Kulkarni said:

If output short is not an operating condition for you, and all you need is a way to prevent ratcheting, this could be acceptable. 

Oh, you mean in the case where there is external dimming, right? We don't do external dimming, so I think this will work for us.

Sumeet Kulkarni said:

However do keep in mind that if you power up into a short at the output, the Toff max will still be the default value.

In our application we don't power up on the on condition so this won't be an issue. The first thing I'm going to do is put 0xFF into that register for both channels.

Brian,

Very fast switching of the gate drive could also result in excess shoot through current through it - resulting in damage due to that reason rather than excess heat from the power dissipation. I apologize - I should probably not have said excess power dissipation is the likeliest reason.

Also, I am not excluding your reasoning of an external component failing and causing a voltage spike at one of the IC pins - it is very hard to determine what exactly might have happened without waveforms to look at. 

If you worry about the output being disconnected during 1A operation - yes, I would certainly use that diode back to VIN to clamp the output. That would be sufficient.

Regards,

Sumeet