I noticed that when a battery is connected and I turn the PWM to a low duty cycle, a negative current flows from the battery into the cicuit. The current can get very high, here 1.35 Amps, even more than 3 Amps. A few days ago, I turned the potentiometer to a low PWM duty cycle setting and then I heard a pop. Here is the destruction tally:

1 IR2104 chip

1 INA3221 board

1 Arduino UNO

1 diode

Obviously, the design has to be changed.

I suspect that when the output voltage of the buck converter is lower than the battery voltage, the gate of the MOSFET Q3 is still open, but the battery now pushes current through the MOSFET to ground. There is almost no resistance, which can lead to high currents and the destruction of components.

One possible solution would be to keep the PWM high enough through software, but that setting depends upon varying conditions. It is much easier and safer to just add a diode that prevents current from going into the circuit from the battery. A 1N5822 Schottky diode costs me a voltage drop of about 0.3 Volts, which is not bad, but still reduces the efficiency of the circuit. Eventually, I want to look into the technique of using a MOSFET as a diode which can reduce the voltage drop.