3-channel LED driver MOSFET overheating (intermittently)

[paw]

Hi all. As always I plead ignorance and beg forgiveness for anyone who I may offend by posting this.

I replaced a cheap and dodgy IR LED RGB driver with my own at least as flawed and dodgy version (although infinitely better in some respects). I found that on two occasions the lights started flickering and the FETs would show signs of overheating (>160 deg C). Interestingly they would work for hours at a time with no problem on other occasions (at various PWM duty cycles).

My limited understanding is that FETs are great while ON and OFF even at very high currents (as per data sheet) but while switching the impedance is unfavourable and due to junction(?) capacitance duration of the switching (while charging) might be longer than desired. I started wondering if for whatever reason my FETs were not completely switching ON or OFF.

I've used NTD5804NT4G 40 V, 69 A, Single N?Channel, DPAK

I am switching ~ 8A at 12V which is not nothing I realise that. I just hoped that at 1kHz PWM and relatively low R the FETs would cope with board dissipation and air cooling... Which it does every time but twice so far (70C max normally). I provisioned for pull-down (10k) but left it off as it doesn't make them any happier.

The Micro PWMing the FETs is running at 5V. The VGS (th) is 2.0V-3.5V. There is a 11k resistor for driving the gate from the micro. Isn't 11k a bit high?

I would have liked the switching characteristic to be cleaner but I don't really know what to expect. I'm attaching CH1 (Yellow) Gate, CH3 (purple) Drain without load.

Any suggestions appreciated.

P.S. Allow me to add that the layout attached is my very first. I would have done few things differently today (decoupling, trace thickness, ground plane etc). I welcome any comments still but would love to understand my FETs better.

EDIT: I din't hope hope. Back of the envelop calculation of Vdrop (at 5mOhm or even 25mOhm) and 8A load at 12V gave me under 1W each dissipated on each FET.

[Zero999]

How can you expect to be able look at the waveform on the drain without a load up pull it up to 12V?

Yes, 11k does sound on the high side. At 8A the conduction losses shouldn't exceed 768mW but if it's taking a long time to switch, the switching losses will be higher.

[paw]

Ignorance and lack of fundamental understanding?

I'll pull the drain up to load it a bit and check. Will see what the switch on looks like with 11k and 1k.

Attached 11k and 1k. I like the 1k better. The FET should be fine with 1k right? I'm not going to burn the gate?

[Zero999]

Why would it burn up the gate?

A MOSFET's gate is a capacitor. The faster it can be charged and discharged, the better. You could even remove the resistor and it'll probably work but this isn't recommended. Try some low value such as 100R.

[paw]

Ah so we are protecting the MCU pin peak current rather than protecting the MOSFTET right?

[Zero999]

Yes, connecting a MOSFET's gate can directly to an MCU output can cause problems due to the current surges.

There will also be some inductance which can cause ringing.

The voltage on the magenta trace looks wrong to me. Are you sure you didn't have the switch on the probe set to x1 and the oscilloscope set to x10? One tenth of those voltage readings seems more sensible.

Notice how the voltage on the yellow trace briefly stops rising/falling, just as it turns on/off and that's it's more noticeable when the gate resistor is higher?

It's because  the capacitance of the MOSFET's gate increases just as the threshold voltage passed and it starts to turn on/off. This is due to the Miller effect: the parasitic capacitance between the drain and gate form a negative feedback path. When driving MOSFETs it's often more helpful to use the gate charge figures listed in the datasheet, rather than the gate capacitance.

See links below:

http://www.vishay.com/docs/73217/73217.pdf
http://www.onsemi.com/pub_link/Collateral/AND9083-D.PDF
http://www.irf.com/technical-info/appnotes/mosfet.pdf