I simply do not believe one could find a "back door" looking at a chip in a SEM. It sounds to me like you are describing destructive physical analysis whose purpose is to make sure requisite manufacturing practices are being followed.
Have a look at this video from ChipWorks http://www.youtube.com/watch?v=Il5sTZKBLO0
See the schematics? They've created those from scratch by deconstructing the chip. (I can say with certainty that this is the case because I'm familiar with the original schematics for this part. The ChipWorks ones are much neater!)
Doing this for a larger, all-digital chip is substantially the same. In that case you can probably step up from identifying individual transistors and identify the standard cells directly, since they tend to have distinctive-looking gate structures.
In the cross section case you are going to see a few dozen transistors out of the millions in a design of any complexity.
It would be remotely feasible to discover some sort of shenanigans if you knew the exact layout of the design, which would basically mean you are a foundry yourself. In that case you might get lucky and spot some difference between the mask you made, and the mask that was used to produce the part under inspection.
But the scales involved make this not believable to me. It would be roughly like scanning the whole of, say, America, and checking every street and intersection of every town, and comparing it against some known quantity to see if something changed in Springfield Missouri.
Maybe somebody could automate this, but the chemical processes for removing layers is less than perfect. Those strands of metal stretching across the ASIC have some built in tension, and if you remove the layer of glass above them, then tend to spring up and jumble. Good luck trying to do something with that.