Here is some additional information that might be helpful. I separated out random noise from fixed pattern noise by taking duplicate exposures, splitting out the color channels in Iris, and subtracting one image from a duplicate, using an offset to prevent negative numbers. I did this for the green1 channel below clipping and at full clipping.
When all pixels are driven to clipping, the standard deviation decreases since the pixel values are at maximum and shot noise is no longer counted. The remaining noise is is mainly fixed pattern noise, which I would assume is mostly PRNU (pixel response nonuniformity). This can be shown by a simple analysis of variance.