TPS548D22: set the frequency to 650khz, but the real frequency is 545khz~590khz

Frank,

Based on the waveforms and the switching frequency, I am guessing that this was taken at very light load well below 50% of the inductor peak to peak ripple.  At these light loads, even in forced continuous conduction mode (FCCM) actual switching frequency will decrease slightly as the negative inductor current at the end of the low-side FET on-time will cause the switching node (SW) to rise up to VIN + Vfd while both FETs are off.  That increases the effective On-time beyond the forced ON-time generated by the on-time generator.  The additional on-time reduces the switching frequency.

If you increase the load current above the critical conduction point so that the switch node is forced negative during the dead-time the switching frequency should increase to the programmed switching frequency.