I worked at Integrated Plasmonics in 2012–14, the major problem you will face is fluidics and sample handling not so much the sensor design. You should consider

I worked at Integrated Plasmonics in 2012–14, the major problem you will face is fluidics and sample handling not so much the sensor design. You should consider stochastic plasmonic features grown using dna self-assembly on the surface of a cmos sensor (lots more recent work on DNA-based plasmonic nanostructures). I worked on varieties of ways to characterize a chemical signal even from stochastically distributed plasmonic features by running calibration during manufacturing. We were able to use support vector regression to characterize known chemical reaction signals on a per sensor basis.