Claims such as "1 stop more sensitivity" with every generation are a little hasty, IMHO.
Hang on - "with every generation" - who claimed that? No-one in this thread, anyway.
Progress in MFD sensitivity has been painfully slow. You can study the sensor datasheets down the years, or generations if you prefer. Microlenses give an instant boost of about 1 stop. That's as good as it gets really. Readout noise has only halved, gradually, in the past decade (and that's only for the best cases: several backs being made now are still close to the noise of the backs a decade ago). CMOS readout noise has come down far more rapidly.
But I also don't buy claims of noise-free 20+MP-CMOS-cameras at this point.
Depends on your threshold for "noise-free", but if you are happy with readout noise of around 2-4 electrons, Canon are already there at mid ISOs and Nikon's D3x is there even at somewhat lower ISOs.
The great thing is that we don't need to rely on claims - this has been empirically measured, and anyone with such a camera can repeat and verify the measurements for themselves.
I suspect that at this stage, with readout noise almost beaten, FPN and PRNU mark the last remaining frontier where high end CMOS DSLRs can make further genuine noise reductions in everyday photography. Calibrating out PRNU wouldn't push the ISO range any further - the impact would be more visible in finessing the already excellent smoothness of high-signal regions of the image. It might be argued that there is no perceptible need for this, but having reached the point of greatly diminishing returns on other sensor issues, the DSLR manufacturers will always want a new technical improvement to tout over their rivals.
Ray