Sometimes one does not have to be able to "see" in order to get superior results. I once interfaced an optical wafer stepper to produce superior optical wave guides that performed favorably when compared to e-beam processes.
I made sure my CAD programs produced symmetrical stepping on the nearest 10th micron centers, and all symmetrical features where always encoded into the same 100x reticles. Thus if there were any flaws in the features due to wave length considerations or convolution effects, they would be similar in regard to both sides of any bilateral symmetry through two levels of 10x reduction.
The flaws in question were smaller than we could measure, but so what? The important thing is that they were the same on the adjacent optical paths that were brought together in order to cancel. (These were Mach Zehner devices)
We achieved an unbelievable 40 db of cancellation between the symmetrical optical paths.
The result is that we actually made wave guides with better phase cancellation than should have been possible with the theoretical limit of resolution thus the feature detail of the devices we made.
Sometimes the opposite happens. It is possible to "see details" that under normal circumstances should not show up, except for the nonlinear effects when getting close to 1/2 wave length limitations.
In one case we stepped a 10X reticle of an acoustic surface wave device with 100s of taps that in regard to all measurements looked perfect in the 10X reticle. However, when the actual device was made, artifacts from the stepping a 100X reticle on 10th micron centers that clearly indicated the pattern of stepping of the 100X reticle while making the perfect looking 10X reticle were clearly visible in the 1X device.
