LT8645S current at no load with sync pin float

Do you have the BIAS pin connected to the output and what is the output voltage? The internal regulator draws from the BIAS pin when BIAS>3.1V otherwise it draws from VIN which results in higher power dissipation. If you connect BIAS to ground it forces the internal regulator to draw from VIN.

This is my circuit schematic. I tested six chips, among which only one is drawing 50mA, while the rest are all above 100mA.

With RT=17.8k the switching frequency is around 2MHz. With SYNC/MODE tied to VCC (you said it was floating in your first post) it's operating in pulse-skipping mode with spread spectrum modulation. The 24Vin also seems to be the worst case for the BIAS pin current. Although there's no plot which shows the current at 24Vin, 15Vout, and 2MHz it's probably consuming at least 25mA BIAS pin current. That doesn't account for anywhere near 2.5W of dissipation though.

Have you checked the switching frequency with a scope? If it's really unstable you could possibly have some unexpected behavior and excessive current draw.

It might also be worth simulating the design in LTpowerCAD with components close to what you're using. The 47pF Cff cap might be a bit high. You probably want it closer to 4.7pF.

I attempted to switch to a 4.7pF capacitor, but the current did not decrease. Ultimately, I replaced the chip with a brand new one and measured the current to be approximately 50mA. Thus, I can only speculate that the damage to the chip might have been due to reflow, transportation, or storage issues.

The switching waveform doesn't look very good considering it should be pulse skipping at no load. Are you sure the component values on your circuit are correct? Are you sure it's soldered properly? These parts are relatively difficult to desolder and solder when using a 4 layer 1oz/layer or higher board.

Is there something wrong with the layout that you may have missed? I've never had an ADI/LT part not work when implemented properly.

I can temporarily accept this result of 50mA. It's very difficult to manually reflow this chip; I only have one PCB, and the sample quantity is too small to observe any additional phenomena. Currently, the switch node waveform is stable and clear, and it still operates normally when tested with a 3A electronic load. Furthermore, referring to it as 'no load' is not precise, given that the bias pin is connected to Vout, which constitutes a load.

Even if BIAS is drawing 25mA from 15Vout that's only 0.375Wout. At say 40% efficiency that's only 0.938Win, which isn't close to the 2.5W you're seeing. Something isn't right.