230VAC dimmer MOSFET failure at surge test

Our dimmer prototypes work fine, but fails surge test, as per 61000-4-5, 1/50us.

Circuit is isolated dimmer for 230VAC different types of bulbs. It uses 2x N MOSFET as switch. V503 detects zero crossings. N504/N505 are isolated drivers, in parallel to provide more power=faster switching.

Mains side passes test for 0,5kV, 1kV and 1,5kV L-N, tested in OFF and ON states. Bulb side fails for 0,5kV L-N @ ON state (because of peak current), passes 1,5kV @ OFF state (voltage suppressor works). 50% dimming is not tested yet.

What I tested, but not helps: protect each transistor with separate varistor, without LC filter (L503/C503), without CMChoke L502 on mains side, additional gate-source capacitors 2n2, additional gate-source TVS 15V bi-dir, moving output N before L502. Most often only one of FETs gets broken (but not always the same) with almost shorted 3 terminals, no smoke or package holes. Load is 60W tungsten bulb.

I have limited space on PCB and components prices, can't add second CMC.

L503 is 470uH/2A

• The location of your CMC is somewhat odd. Is there anything else connected to that branch besides the optocoupler? – Edgar Brown Feb 1 at 15:18
• What’s the value of L503? What if you make a symmetrical Pi filter instead of LC? Can you cheat and stop switching during any overvoltage transient? RC snubber across each MOSFET? Can you have a varistor to ground? What does the voltage waveform across drain-source look like during testing? – winny Feb 1 at 15:29
• Model the surge and the circuit and simulate it. Examine what’s doing the damage. – Andy aka Feb 1 at 17:30
• @EdgarBrown I agree. There's nothing more there. Moving output N before L502 not helps. – Bogumil Feb 4 at 10:47
• @winny L503 is 470uH/2A. I tested ON/OFF states, not 50% or any other. I'll try symetric Pi or snubber. Varistor to ground - You mean to isolated side? – Bogumil Feb 4 at 10:49

The main issue here is ( no pun intended) the lower main line injection of 0.5kV with a 1us rise time and Miller capacitance coupling of impulse voltage back to the gate drive is probably exceeding the Vgs levels. This rise time equates to a the driving spectral energy band of $$\f_{-3dB}=0.35/t_R= 350kHz\$$