Dear @Rubayet
Thank you for this question. Let me address it using the tutorial PLECS: Thermal Simulation of a Buck-Converter.
Although this tutorial does not exactly match your application, it is well suited for explaining the principle, since the only loss-producing elements are the semiconductors. The resistor in the model represents the load and therefore defines the output power.
Assuming you start from the beginning of the tutorial, where all semiconductors have an On Resistance Ron and Forward voltage Vf set to zero, you can measure the efficiency as
\mathrm{Efficiency} = \frac{P_{out}}{P_{in}} \times 100.
You will observe an efficiency of 100 %. This happens because the semiconductor losses are calculated in the thermal domain and are not fed back into the electrical domain. As a result, the electrical output power does not reflect the actual losses.
The correct approach is therefore to subtract the semiconductor losses from the input power in order to determine the effective output power:
\mathrm{Efficiency} = \frac{P_{in}- P_{loss}}{P_{in}} \times 100 = \left(1 - \frac{P_{loss}}{P_{in}}\right) \times 100
This equation correctly describes the efficiency with respect to the power flowing into the converter. The output power measured at the load resistor, however, is not correct in this case, because the losses are not reflected in the electrical model.
If you are interested in the efficiency at a given output power, you could use your originally proposed equation. However, this is not entirely correct, because the input power must increase to compensate for the losses. This leads to a slightly higher current, which in turn affects the losses again. Using a closed-loop controller to regulate the output power can help to resolve this issue.
Another possibility is to set On Resistance Ron and Forward voltage Vf. This introduces a voltage drop across the semiconductors and therefore reduces the power transferred to the load. While this voltage drop approximates conduction losses, switching losses are still not taken into account. All of these aspects are discussed in the tutorial, and although it does not exactly match your application, it is still worth reviewing.
As you correctly noted, both efficiency equations yield similar results as long as Ploss << Pout. From a practical point of view, it is often more useful to focus directly on the semiconductor losses. This provides a quantity that is easier to interpret and compare.
Finally, I would like to point out that relying on simulation results for efficiency can be challenging. As with any model, only effects that are known and explicitly modeled can be captured. In practice, there are many additional loss mechanisms that are unknown early on or differ when moving to real hardware (e.g., component tolerances, parasitic resistances, manufacturing variations). For this reason, it is often preferable to report losses rather than efficiency when discussing simulation results.