That's pretty cool (it is), but the production cost of a wafer is
about the same whatever you put on it.
Only if the wafer goes through the same number of process steps, in the same foundry machines. The whole point about the OPs message is that to date an FF sensor could only be made using a 3-step reticle stitch, increasing the number of process steps for the wafer, reducing yields and significantly impacting costs. Reticle stitching is not required for APS sensors using stepper kit available in most fabs. If Canon have, and there is limited evidence of this, eliminated the stitch process then they will reduce their costs to levels commensurate with conventional chip yield estimates.
Your earlier point is quite correct, the larger the device the fewer you can get on each wafer and the higher the probability of a catastrophic defect being included in any device. This determines yield and hence, given fixed wafer process costs, the device cost. However, reticle stitching changes that model significantly - wafer processing costs increase and yields reduce, both due to the additional number of processes being implemented for reticle stitching.
I have been involved in several designs of sensors requiring reticle stitching and I can assure you it has a major impact on yield, and hence cost, over and above the area effects that you refer to. On a single stitched device, depending on the level of circuit complexity in the stitch region, you can expect yield to fall by as much as 50% (I have seen worse, but that was atypical), while process costs increase by around 20%. With a 3 stitch process I would not be surprised to see yields fall and process costs increase further. That accounts for a significant cost hike over and above the area effects that you are considering.
The OP suggests that Canon may have eliminated the stitching process for FF sensors using their new stepper and hence brought the costs of FF sensors into line with the predictions that your simplified view of life suggests.
The linked article states that recent devices on 1Ds cameras show no visible stitching artefacts (which are usually pretty easy to spot visually) whilst devices from the D3 still clearly show 3 stitch interface artefacts (and shows them in photos).
Irrespective of what you would like everyone to believe, stitching does impact costs - and significantly. Eliminating the requirement for reticle stitching through the use of a larger field coverage stepper, would have a major impact on the cost of FF sensors.
