If this would be interesting to someone else in the future here is the answer for this question from Joe Nabity (author of NPGS writing software http://www.jcnabity.com):

"In general, when the deposited material hits the surface there will be nucleation and growth of small islands which seems to match what is seen on the edges of the lines in your images.

If you consider how the resist blocks the incoming metal during the evaporation, a ‘perfect’ edge between metal and no metal will only happen when the incoming metal is completely collimated. If there is any spread to the incoming evaporated metal, the edge will have a transition region which changes from no metal to the full thickness.

Have you considered the geometry of the entire setup where you estimate the expected spread on the sample from the incoming angle of the deposited material based on the crucible distance, size of the deposition source, and thickness of the resist?

Another possibility is that if the opening in the resist changes size during the evaporation (either by shifting of the resist or by build-up of the deposited metal), it could cause an apparent spread to the deposition along the edges of the structure."

I think his reply answers  my question. 

In reply to Pavlo's great answer, would perhaps a thinner resist stack help to resolve the issue? My second guess at a remedy would be a short gold etch to remove the satellites. This one is a great non-hazardous gold etchant: https://gmwgroup.harvard.edu/pubs/pdf/471.pdf

Hi Pavlo,

I'm not sure this is related to the evaporation process, since you've mentioned it happens only in the double layer regime. What is your alternative - only 950 PMMA? If your metal stripe is thin enough, it might do. 

Also, think whether the particles are distributed within the width you designed, or outside of it.

Other things to consider

1. EBL dose

2. Slightly change thickness of layers (thinner 495, thicker 950)

3. Aggressive developer?

4. Contacting supplier :)

Thank you for the inputs and sorry for the late response. To answer your questions.

I have tried to etch using dilute I2/KI buffer to remove the gold using the negative lithography approach, but it goes way too fast and leaves a rough edge. The RIE Cl2/Ar or Cl2 dry etch works much better, but leaves Au/AuCl deposits (need to optimize this further). Another simple solution is to dip in the HF buffer to remove few nm of underlying SiO2. The particles are not well attached to the main structure and the majority of them is removed this way. I have not thought about the SAM masking. Thank you for the reference.

The particles are distributed outside the designed pattern. The lower layer resist thickness used to make the structures in the image was at ~180nm. I think thinner lower resist layer would help and I should give it a try. Thank you for the suggestion! 

Try pumping down the vacuum chamber on your evaporator longer. The better vacuum, the more collimated the incoming material by minimizing scattering. If you have poor vacuum, you may actually want to decrease the source-sample distance even though the angular effect may be worse.