Simple High side switch keeps breaking

[pimswinkels]

hello everyone,

For a project I need a low cost, small footprint high side switch that can switch 48V DC up to 1.5A. I came up with the follwing circuit:


The picked mosfet is a 60V 5.5A P type MOSFET, the WST06P06.
Link to datasheet: https://www.lcsc.com/datasheet/lcsc_datasheet_2409302300_Winsok-Semicon-WST06P06_C5242059.pdf

The voltage divider including the 12V clamping Zener is to make sure the gate-source votlage stays within the max gate-source voltage of 20V. This design has however proven to be very delicate and breaks easily. Mostly on the first time the circuit is switched on with a load (100mA usually enough). One time this circuit has switched a 1.5A load for over 20000 times though without any issues. (2 seconds on, 2seconds off)

When playing around with the circuit, I came up with an even simpler design, using a 36V zener to shift the P-mosfet's gate up by 36V:


This circuit works very well with no load, but again fails when a small load of just 100mA is connected. After faling the zener diode becomes extremely hot (>150C) when the input signal for the high side switch is high, suggesting a low resistance path from the P-type mosfets source to its gate.

In both situation the broken mosfet will always conduct from its source to its drain, making it an "always open"switch.

The circuit is realy simple, so the solution must be really simple as well. For some reason I can not find it though.

Some things that might be the cause:
-Slow switching times caused by the 47k pullup resistor from source to gateresulting in mosfet staying in it linear region for too long.
-Slow response time of the zener diode (should only be an issue in the circuit where the zener is used to clamp the Vgs of the p mosfet).
-Mosfet picked too small. - unlikely, but could be the case. I already tried the slightly beefier VB8658 as well.

Hope anyone can help me with this. I have been stuck on this small circuit for way too long now...

thanks in advance!

[voltsandjolts]

What is the load? Resistive, or perhaps inductive, like a motor?

Edit: You're outside the DC SOA, see attached, 50V 0.5A. So you need to switch the fet fully on (or off) in less than 1ms, at 1.5A Id.

[Ian.M]

Even with a supposedly non-inductive load, if there is much wiring, the wiring inductance may be problematic.  Add a free-wheeling diode, anode to ground, cathode to MOSFET drain to clamp the back-EMF from the load or wiring inductance and see if your problem goes away.  For a non-PWMed load the diode continuous current rating can be less than the peak load current as long as its  surge current rating is greater.

Also consider the supply side wiring inductance - you may need decoupling at the MOSFET  source.

[pimswinkels]

Thanks for the quick response!

The load I tested with was a constant current load set to 100mA. This means that it will not act resistive, but indeed non linear. in the final usecase it will drive a load that has internal buck converters, so the initial switch behavior will be as if there was a capacitive load i guess?

As for the inductive wiring: I did not think  of this when designing these prototype boards. I wanted to keep the design as clean as possible, but could indeed have done with some input decoupling and a freewheel diode. I will test this tomorrow to see if that helps.

As for the SOA: I think 1ms is still really quite long, and I have designed the switching time to be shorter than that (just a few microseconds max).

I will try adding capacitande and a freewheel diode tomorrow. I will also try to get some clear scope pictures of the behavior

[temperance]

The load I tested with was a constant current load set to 100mA.

The constant load probably behaves like a dead short with no input voltage applied because the control loop is saturated.

[Ice-Tea]

As for the SOA: I think 1ms is still really quite long, and I have designed the switching time to be shorter than that (just a few microseconds max).

Are you quite sure? Seems longer at first glance...

[voltsandjolts]

^yeh, plus some Miller effect too.

[Peabody]

Is it the P-channel mosfet that fails?  If so, what is your source for that part?

Does the first circuit fail if the input voltage is something lower, like 20V?

Have you tried inserting a series resistor (220R or 470R) between the Enable source and the gate of the 2N7000?  That would be placed after the R22/R2 connection point, so as not to create a divider.

Are all these connections soldered, or are you using a breadboard with jumpers?  I'm just thinking that if the zener connections aren't good,  the P-channel would be blown. That might explain why it worked once for 2000 cycles, but often fails.
 

[temperance]

As for the SOA: I think 1ms is still really quite long, and I have designed the switching time to be shorter than that (just a few microseconds max).

Are you quite sure? Seems longer at first glance...

From the gate-Charge characteristics with 48 V input voltage and two 47 K resistors, the turn on time is about 13 µs.

[Ian.M]

It is a characteristic of many budget electronic loads, that in the absence of input voltage, the  OPAMP driving the pass transistor gate rails to try to maintain its drain current at the setpoint.  This means that it will act as an almost dead short till the OPAMP can discharge the gate back into the linear region, which takes a significant time vs SOA timescales due to the OPAMP's limited slew rate and output current.

Retest with a real resistive load  or with a high-end electronic load that has a limiting circuit to make it act resistive with less than its minimum voltage across it and there is a good chance your high side P-MOSFET will survive!   A workaround for cheap electronic loads is to provide an alternative source of current before the supply you are testing is switched on, e.g. an alkaline cell or a bench PSU set to 1.5V, feeding the load positive terminal via a diode, which will maintain approx 1V across the load and stop its OPAMP railing.

[Alex Eisenhut]

Why is the body diode shown as a Zener diode?

[PCB.Wiz]

For a project I need a low cost, small footprint high side switch that can switch 48V DC up to 1.5A. I came up with the follwing circuit:
...
The picked mosfet is a 60V 5.5A P type MOSFET, the WST06P06.

A 60V part in a 48V system is not much margin, especially with no shown supply capacitors or clamping.

[PGPG]

Link to datasheet: https://www.lcsc.com/datasheet/lcsc_datasheet_2409302300_Winsok-Semicon-WST06P06_C5242059.pdf

Reading the thread I expected at least DPAK package and going to datasheet I see SOT23.
It says Total Power dissipation = 1.5W.
If you have 100mA constant current load than at first moment of switching on you have 50V*0.1A = 5W and you plan to switch on higher currents.

There is something wrong with this datasheet. From Fig.8 it looks that 60V/0.5A is ok for DC. SOT23 part dissipating 30W continuously is suitable for the Guinness Book of Records.

[Zero999]

Perhaps adding more transistors to speed up the gate drive will help? Here's an example of a circuit I designed awhile ago. It will work with 48V, but the components might need to be changed.

[pimswinkels]

Hi all,

Wow, thanks for all the responses!

 First i want to answer some questions:
- everything is soldered on a custom pcb. Implementation errors like half connections in a breadboard are luckily not one of the prameters I have to take into account
- The 30W continuous power dissipation that can be dducted from the datasheets SOA seems indeed very unrealistic and does raise some questions about the rest of the datasheet...
- The p mosfet body diode being drawn as a zener diode has to do with me being lazy and just copying a symbol from my library without paying too much attention. The pinout is verified by me and a colleague though.

I have changed some things and retested the board

I have added:
- a 1uF input capacitor between the input voltage and gnd as close to the P mosfet as possible
- a 1n4148 diode as freewheel diode as close to the mosfet as possible

I have changed the load:
- the load wil only be used in resistive mode.

I then started with a no-load test. this gave the following results:

switch on behavior

and then the switch off behavior. Both zoomed in quite close and zoomed out completely:

switch off zoomed in


switch off zoomed out

As you can see the gate of the mosfet is completely settled at about 300us. Significantly longer than expected, but still well within the 1ms max switching time.

When turning on the load at 480R though, the circuit broke again after just a few switching cycles. I did manage to get some scope images though:


480R load total view

480R load switch off

The weird thing is that now the bottom mosfet (the 2n7002) broke. The gate-source resistance measured just 16R, where it should be in the mega-ohm range.
I am really surprised by how I could have broken this mosfet though. It only broke when I enabled the load. I just dont undertand how switching on the load can affect this mosfet enough to break it....
the enable signal is a simple 2 second on, 2 second in 5V signal generated by an arduino. nothing special...


does anyone have a suggestion?

[PCB.Wiz]

I am really surprised by how I could have broken this mosfet though. It only broke when I enabled the load. I just dont undertand how switching on the load can affect this mosfet enough to break it...

Which circuit did you use ?
It shows how sensitive MOSFETS are to overvoltage damage, hence my comment about 60V parts not really  having enough margin for 48V switching.
 If you live-plug this, you need to also consider the power cable inductance + input capacitor ringing effect, which can be significant.
 I would change to 100V or more parts, and add voltage clamps, then retest.
Lcsc have more capable mosfets in TO252 for similar price to your link.

[Ian.M]

Even in resistive mode, I wouldn't trust most electronic loads that are based on an OPAMP + MOSFET control loop.   If you have a calibrated current probe with >1 MHz bandwidth, you can check the max. inrush current when voltage is suddenly applied to the load, but if not, your best bet would be a real power resistor!

[Smokey]

This is an off the shelf part.  Search digikey for " power distribution switches, load drivers".  Filter by" high side". 

[pimswinkels]

Hello and happy new year everyone!

I know you can buy this as an off the shelf part, but since I need to implement this circuit 10 times per module for 1000+ modules, it is worth trying to minimize the cost and build it out of low cost components. Sub 1 euro high side switches being able to switch 48V 2A are very scarce, and are still over budget.

I am going to try this exact circuit with beefier mosfets and a true resistive load to see if that improves the situation. I will also purchase a current probe to get some better insights into what is happening exaclty.

I will get you posted!

[mtwieg]

There is something wrong with this datasheet. From Fig.8 it looks that 60V/0.5A is ok for DC. SOT23 part dissipating 30W continuously is suitable for the Guinness Book of Records.

Yeah the SOA figure is just nonsense. These curves are for a case temperature held at 25C, but even so completely implausible and should be ignored. Also noticed that figs 1 and 2 suggest a typical Rdson of around 63mohm at Id=-3A and Vgs=-10V, contradicting the table value of 80mohm. I wonder if they simply mixed up data from completely different parts in the datasheet...

That being said, if the load is just 100mA then I think SOA shouldn't be a problem even for a SOT23 package (unless there is something particularly awful about this part inside). Looking at other SOT23 FETs with similar specs like the FDC5614P and SSM3J351R show decent margin at pulse widths <1ms.

I didn't see anyone else ask, but is there any capacitance on the output? Load capacitance along can kill these circuits even without any DC loading.

Echoing previous comments above about the electronic load possibly drawing much larger currents than expected as output voltage rises from 0V. This can be an issue even if the load is set in resistance mode. I would test with simple power resistors as a load, or measure the actual load current waveform in some way.

Ringing/osillation due to parasitic inductance is also something to think about (adding some RC snubbers on the input and output should help), but based on your waveforms it doesn't look like the culprit at the moment. However your first plot seems to show the gate voltage dropping by around 20V, which is dangerous. Perhaps this waveform was from the second "simpler" circuit, while the other waveforms came from the first circuit.

[BennoG]

Just tested with my ET5410 electronic load.
When set to 500mA the inrush is 120A, so don't use a electronic load for this.
The inrush is about 180 uS above 80A

Benno

[pimswinkels]

Hi all,
I did bite the bullet and bought a current probe. A first test did indeed show that the constant current mode of the electronic load draws a 100A peak when I switch on the power supply. The peak is just a few microseconds, but I am still surprised that some mosfets survived multiple switch cycles like this....

The resistive mode does have some current overshoot as well, but not that significant (the peak is about 2x the expected current for the configured). A significant overshoot, but not something that should break the circuit when the load is set to emulate a 480R load (100mA at 48V).

I just received some power resistors to guarantee a resistive load (realisticly slightly inductive due to the cables, but I will use a freewheel diode to mitigate that).

Will test today and post the results by the end of today. Hopefully I will get somewhat closer to the desired result!

[Thomas]

I saw high transient currents into my Itech IT8512+ eload too. Then I measured the input capacitance to be ~4.5uF.
Maybe yours have something similar?
I made a label and put on the front of my eload.

[mk_]

Hi all,
I did bite the bullet and bought a current probe. A first test did indeed show that the constant current mode of the electronic load draws a 100A peak when I switch on the power supply. The peak is just a few microseconds, but I am still surprised that some mosfets survived multiple switch cycles like this....

Swithingtime is way to fast, so try to slow down the Highside-FET with a small cap between G and S, just to keep the current within the abs max ratings and to avoid fast transients when there is no path the current can go..
Yes, this creates heat within the die but if there only some uF to charge for your stepdown this could solve problems because of ringing as your parts are working near their abs max. voltagelimits.
Your final load (stepdown) should not power up until the highside-caps are more or less fully charged, so add an RC- delayed enable.