It turns out that the electronic transport degrades even further due to light induced degradation of amorphous silicon. One successful workaround to this unavoidable problem is the use of thin cells because a stronger field across the absorber layer ensures collection of charge carriers before they can recombine [Hanak-1982ipvc].
We are left with a problem now: the absorber should be thin because of poor charge collection, but it should also be thick for sufficient absorption. Luckily we have more than one dimension at our disposition, we can decouple the path of light absorption from the path of charge collection. This technique is widely used in crystalline silicon solar cells by simply refracting light rays at inclined surfaces of surface facets. Because silicon wafers are quite thick, the dimensions of surface facets are typically tens of micrometers. Thus, geometric ray optics is completely sufficient for their understanding [Redfield-1974apl].
The obvious application to thin film cells was, of course, to reduce the size of surface features [Deckmann-1983apl].
Once the dimension of surface features is in the same range as the wavelength of light, geometric optics is no longer very useful, we are entering the domain of light scattering. In a very general treatment of the light scattering process, the maximum enhancement of the optical path length was estimated to be 4n2 where n is the refractive index of silicon [Yablonowitch-1982ieee]. The estimate is based on the statistic distribution of energy into the radiation modes that exist in a weakly absorbing film or slab of finite thickness, and the assumption of full randomization. The authors claim that randomization should be most easily achieved by Lambertian light scattering at the interfaces [Tiedje-1984ieee].
Periodic modulations of the refractive index (gratings, photonic crystals, etc.) have been suggested to surpass the 4n2 limit by manipulating the modal structure of the radiation modes in certain spectral regions [Sheng-1983apl, Gee-2002ipvc].