Strange shoot through in simple class AB Mosfet amplifier

[Kokoriantz]

After inventing the class abD, I decided to put an end designing analog amplifiers. So this is my ultimate one. It is not brilliant in THD numbers, 0.01-0.03%, but it features a unique open loop response the shift only -5° at 20khz instead of habitual -90° which requires 60db NFB at 20khz for perfect sound. This amp has only 36db NFB flat to over 100khz.
The problem I have is on transient square wave, instead of having the active side to be commanded inactive, on the contrary it gets commanded to over act for a short period. The scope shows for positive and negative slopes with 2us scale.
Any idea what is the problem?     

[magic]

Probably it's Miller feedback.

The upper driver can sink significant current. It turns on the 9240 which pulls the output up, sending significant current through the upper capacitor, which is easily sunk by the driver.

An equal current flows through the lower compensation cap, but the low side driver can't sink it being an NPN.

Probe the bases and emitters of the drivers and see what happens.

If EF drivers are used in front of a common source stage, compensation should be routed back to the bases of the drivers, not to the gates. Generally, global Miller compensation is normally routed all the way back to the output of the input stage.

[Kokoriantz]

Thank you.
I halved the millers, nothing happened, I compensate from the bases, the shoot through greatly reduced but the double LTPs are still commanding wrong, you can see the red curve when increased, the green should start falling. It doesn't, it goes also for a while increasing.
  By putting the miller on the outputs, this transconductance amplifier doesn't require Zobel, now it does.

[Kokoriantz]

Yes, yes, I now understand. It is the miller that is injecting opposite command from the output, nothing wrong with the LTPs.
Thank you magic, thank you very much.

[Kokoriantz]

I inverted the polarity of the drivers, as I lost 1.4v on the base resistors hence bias current of the ZLTPs, I lost 3db on the NFB. Good point is it doesn't need millers to be stable. The shoot through has totally disappeared. I have a phenomenal slew rate to get 1Mhz power bandwidth.