Hi,

I have a MOSFET that is failing frequently (once every 30-40hrs or so). The circuit is shown below.



Q4 eventually fails ON after continued use. I have several identical circuits running on the board for high side switching of other components with no reported failures. I'm assuming (maybe incorrectly) that this one is failing because its load runs in very close proximity to a strong RF noise source.

The MOVs in the circuit are an attempt to stop overvoltage on the MOSFET nodes due to noise spikes (particularly when the lamp is ignited).

I haven't yet installed shielding on the wire (shown in green on my diagram), and don't know if it could help, but would appreciate any and all advice/suggestions. Should the shielding be connected to chassis ground, or the signal ground?

This is driving me insane....

 

What is the MOV rating?
The shield should be connected to signal ground.

 

sorry i forgot to include the MOV specs. they are both VC080512A250DP.

Package 0805, Working voltage 12V, Energy rating 0.1 Joule, Clamping Voltage 27V,

 

3 out of 4 identical circuits on the same board have now failed today. an excellent friday afternoon! is there an obvious problem with the way i designed the circuit (ignoring RF problems)?

It ran fine for most of the week, then this morning the first circuit failed. Followed by me trying to recreate events that cause the problem.

It seems the chance of the problem happening is much higher when these two conditions are met: 1. wires running near the lamp/ballast, 2.the microcontroller signal is high when the lamp is sparked (this corresponds to the highest peak RF/EMI emissions)

 

You need a TVS not a MOV. MOVS 'wear' out after a certain # of spike events and are quite slow to react possibly allowing the leading edge of spike events to 'batter' the FET.
If you have a brushed DC motor as well you will get more spike events.
For simplicity get yourself a bidirectional TVS of appropriate max Vclamp rating to protect your FET.
Perhaps this one:
http://www.newark.com/multicomp/1-5ke15ca/tvs-diode-71a-21-2v-bi-directional/dp/91R0614
Also place a 4.7 Ω to 22Ω resistor in the gate drive of the P FEt and clamp the Vgs with a zener rated above the normal gate drive voltage but below the max spec Vgs on the Pfet. This handles any gate ringing and transients.
Place a 0.1uF (50V) ceramic cap across existing clamping diode in the circuit. That diode needs to be a fast switching type, not a 1n4007.

 

You need a TVS not a MOV. MOVS 'wear' out after a certain # of spike events and are quite slow to react possibly allowing the leading edge of spike events to 'batter' the FET.
If you have a brushed DC motor as well you will get more spike events.
For simplicity get yourself a bidirectional TVS of appropriate max Vclamp rating to protect your FET.
Perhaps this one:
http://www.newark.com/multicomp/1-5ke15ca/tvs-diode-71a-21-2v-bi-directional/dp/91R0614
Also place a 4.7 Ω to 22Ω resistor in the gate drive of the P FEt and clamp the Vgs with a zener rated above the normal gate drive voltage but below the max spec Vgs on the Pfet. This handles any gate ringing and transients.
Place a 0.1uF (50V) ceramic cap across existing clamping diode in the circuit. That diode needs to be a fast switching type, not a 1n4007.

Thanks for your suggestions!

The diode I am using is a 1N4448 (with a recovery time of 4ns) (the emi spikes are generally 40-100ns in length). Could you explain the advantage of adding capacitance to the diode?

I have a few backup boards populated, so will be adding zeners to the gates of the p-mosfets. What is the most likely cause of a drain source short on a pmos? overvoltage on gate? overvoltage on source?

Thanks

 

I also wonder if using a 1N4448 diode is appropriate here. They are rated at 100ma max forward current. Are they still operational?

 

Usually a positive or negative going spike on the gate will do it in. Most FEt failures I have had (NFETS) appear to be a Vds short.
Gate drive transients or high current switching can cause parasitic Vgs or Vgd capacitance transient voltages on the gate causing noisy partial Vth switching, linear mode ops and FET burnout. I had a particularly noisy case (>100A switched currents with 5Mhz ringing) that needed a 0.1uf low esr cap ON the gate/source (in parallel with an 18V zener) to control that to prevent parasitic false Vth switching.

A cap to ground across the flywheel diode will reduce the conducted noise from your 'transmission' line causing circuit probs.

If you are switching high currents.....even transient high currents your copper traces to the Drain & Source should be augmented with thick solder or solder braid to avoid voltage bounce causing the gate voltage to momentarily go out of spec. The zener will help with that. Do you have a low inductance ground plane? If your supply current is coming from a little distance (higher inductance power rails) you will need a 10uF electrolytic or tantalum (preferably) & a ceramic low esr 1uF cap pair (on the power rail) placed close to the FETs to provide the currents during switching otherwise voltage bounce ensues.

 

Vgs for this part is nominally -2.5v, and absolute max is -8v. Looks like -12 Volts gate to source is applied when q3 is saturated. Could this be a problem? -2.5v gs should be quite enough for this circuit. Also, why not put a suppression device at the end of the line where the load is?

 

@Domophone

If r3 fails or you have bad solder joints (or accidently placed 510 ohm), it could fail. Once it fails, there is no path to discharge the gate and the (Q4) p-channel mosfet will stay on. Make sure you have good connections to the board and that gate of Q4 is secure as well.

 

Vgs for this part is nominally -2.5v, and absolute max is -8v. Looks like -12 Volts gate to source is applied when q3 is saturated. Could this be a problem? -2.5v gs should be quite enough for this circuit. Also, why not put a suppression device at the end of the line where the load is?

That would do it alright.

 

Try the circuit that I modified.

 

You don't show how the xenon lamp is powered and grounded. This lamp wiring can have a huge effect on how noise gets into the fan circuit.

Also, does the FET fail when the lamp strikes? The lamps I worked with needed tens of kilo volts to start the lamp conducting.

 

I also wonder if using a 1N4448 diode is appropriate here. They are rated at 100ma max forward current. Are they still operational?

Hi, thanks for the response. From the datasheet i'm looking at, the 1N4448W is rated for 250mA (i'm using a sod-123 footprint version). They all seem to be intact and functioning within spec. I'll up their size next time i populate a board. Would Panasonic's DA22F2100L be appropriate?

 

Vgs for this part is nominally -2.5v, and absolute max is -8v. Looks like -12 Volts gate to source is applied when q3 is saturated. Could this be a problem? -2.5v gs should be quite enough for this circuit. Also, why not put a suppression device at the end of the line where the load is?

I think this may be my problem, or at least related! When i started the design, i was using 5V fans. I forgot to check the pmos specs when i switched to 12V! Instead of using a voltage divider on the gate, i was just going to switch the pmos to an NTF2955. it has an max Vgs of -20 (and a threshold max of -4). max Vds of -60 instead of -20 which i figure can't hurt with any transients.

I've added caps in parallel with the loads (at the load end of the wires).

Thanks for the tips.

 

You don't show how the xenon lamp is powered and grounded. This lamp wiring can have a huge effect on how noise gets into the fan circuit.

Also, does the FET fail when the lamp strikes? The lamps I worked with needed tens of kilo volts to start the lamp conducting.

I've attached a diagram of how the lamp is powered/grounded


The blue lines from the ballast to the lamp are ~3ft long. Lamp strike is 35-40kV, once running and stable it is generally ~25Amps @ 14 volts with moderate switching noise.

Is my grounding scheme OK on the Lamp/ballast? I get confused when grounding shields despite having read up on it quite a bit over the years.

 

Is Q4 getting hot? Can you touch it with your finger? This breakdown mode suggests to me that the junction temperature got too high, and the entire part diffused itself into a resistor. Is the microprocessor running on 5 volts, or even 3.3 volts? It could be that when the processor thinks it's putting out a "1", the voltage never rises above the processor supply. Or even if the processor output is open drain (open collector), the circuit counts on a resistor to bring the gate voltage up to 12 volts, and turn Q4 off. If this is happening thousands of times per second, the transistor is spending a lot of time in the linear region, and that would overheat it. If that's the case, the best cure is to use a MOSFET driver chip that takes the logic input, and translates it into a clean 0 - 12 volt drive, with fast rise and fall times. The TC4401 is an example of such a driver chip, but there are many to choose from.

If you have an oscilloscope, or can borrow one, it would be helpful to look at the waveform at the mosfet gate.

 

If you have an oscilloscope, or can borrow one, it would be helpful to look at the waveform at the mosfet gate.

Here are Turn-On and Turn-off waveforms, respectively, for Q4:


Q4 is not getting hot or even warm during operation

 

Here are Turn-On and Turn-off waveforms, respectively, for Q4:
...

Q4 is not getting hot or even warm during operation

If the transistor is not even getting warm, then I retract my theory. Must be something else. I notice there is some ringing that could be snubbed out, but then again, if the transistor is not getting warm, then that can't be it, either.

 

Here are Turn-On and Turn-off waveforms, respectively, for Q4:
View attachment 92892 View attachment 92891

Q4 is not getting hot or even warm during operation

It looks like you are still using the FDT434 as the threshold is very low. If you are going to use that FET use a divider - say 470 ohms and 470 ohms. Things will switch faster and the diode ringing won't be so large.