Gold nanoparticles can frustrate locally a total internal reflection, acting as point-like sources over a dark background, see the sketch.This is far better than shadowing uniform light.
I feel it less good than a notch in a metal screen, because in the transparent material, the displacement current d(D)/dt concentrates before entering the nanoparticle, stretching the radiating zone, and dilutes at the sides, reducing the radiated field and sharpening its pattern.
As the conduction current is nearly compensated by nearby displacement current dilution, the particle's protruding position must be the main contributor to radiation.
Surface smoothness elsewhere must be far better than 20nm. Semiconductor processes have worse requirements.
A small notch in the reflecting surface has nearly the same effect as a gold nanoparticle. I'd try the already suggested methods: mechanical action of the tungsten tip; local electric current from the tip, which may bombard the surface with H and F ions alternating...
Because total reflection puts limits on the angle of incoming light, a lens can't concentrate light as strongly as on a punched screen.
A monomode fibre carries concentrated light, so a small notch there radiates more. Bigger notches are made for fibre gratings. The fibre should be cladding-less (ĦOle!) or maybe have a soft polymer cladding over a hard ceramic core. Alternately, an optical chip made by semiconductor processes can have a light guide with a notch (20nm being a present-day size), and allows a protective cover. Nice compact source, especially if the laser diode is integrated. But I prefer a small hole in a metal screen covering a laser diode.
Marc Schaefer, aka Enthalpy