I checked that it is terminated at 100 % for 1000fs simulation time. So as you suggested, I increased it to 10ps with auto shutoff min 1e-07 and mesh accuracy 2, steep angle PML of layer number 64. And I did the simulation for the normal incident with broadband BFAST source (487 nm to 787 nm) with pulse width ~ 3fs and offset ~ 9fs, but still, I am getting the reflection of more than 1, even I tried with by adding extra mesh, but still there are no improvements.
Here is the simulation result Reflection and transmission. And for the BFAST source, the problem happens only at low incident angles; for higher angles (more than 35 degrees), the normalization is happening correctly.
I think there might be some other problem.
Can you tell me how to check the diffraction ?