Hi,
I have some experience in MD, however, not using LAMMPS but other MD code. You need to take into account the expected sputter yield for a specific Ar impact energy, impact angle as well as the crystal orientation of the substrate. In many case, the sputter yield is less than 1, meaning that not every single sputtering simulation will end up in a number of Si and/or O being released from the substrate. You need statistics, meaning around 100 simulations with slightly different surface impact coordinates, only in this way you will see sputter yields of around 0.2 e.g. I would also use Ziegler-Biersack-Littmark ZBL potential for the interaction with argon, as it is very distance dependent. The ZBL potential is developed for the representation of high energy processes, where individual atomic cores of the impacting atoms can come very close to each other.
Please take a look to works of Sigmund, Malherba, “nuclear stopping power”, “electronic stopping power” etc.
Regards
Alen