Thanks for the suggestion,
I tried changing the maximum mesh size as you said and it seems to have helped refine the mesh, at least visually. However, I set up a parametric analysis, changing this size by a small amount, and I again got different solutions for different mesh sizes. Now, my instinct says to just decrease the mesh size until the solutions converge, but this seems extremely inefficient and will probably lead to me running out of memory. For example, if I decrease the mesh size in all directions to something suitable for the smallest dimension (nm), I would also apply a nm-scale mesh to the other dimensions (mm).
Is there a way to assign a mesh size for each individual dimension?