Checked last night. Need around 1 hour to have the condition repeat. The most hot is 1.2V LDO then 2.5V and the least is 3.3V. The output voltage of the 2.5 is triangular shape of the noise and the highest amplitude. nearly no pulsation on 3.3 one and some on the 1.2. At the disconnect happening the pulsation on the 2.5 and 1.2 rise by 40mV. Not very good in absolute value as the scope was using is not as accurate. Speeding cooling of LDOs (1.2 & 2.5) with cooling liquid would bring the pcb to acting state. So have two suspects now. Will try changing one at a time. Can you specify which exactly ones are used in the latest PCBs?