Understandably, there's a lot of confusion about what exactly such a construction would do. How would it affect the image? Wouldn't it simply darken half of the frame?
The answer--perhaps strangely--is no. To understand what it does, think about a conventional lens shot wide open. Note that in out of focus areas of the image, circles of confusion ("bokeh balls") are created, and are especially visible when a bright point light source not in the plane of focus is projected through the lens onto the image plane.
If you could alter the shape of the aperture in some way, you would see those out-of-focus highlights rendered in the shape of the aperture. This has been a popular trick exploited by numerous photographers to give bokeh in the shape of hearts, stars, or whatever cloyingly cute gimmick is in vogue these days. This is the key insight that should immediately tell you that placing an ND filter over half of the aperture will not darken half of the frame in such a crude way: instead, if you were to obstruct the light entirely on that half where the ND is being applied, you'd see out-of-focus highlights that were shaped like a semicircular disk.
But in the plane of focus, the image would still be sharp, albeit at an exposure exactly one stop slower than if that obstruction were not present (in the case of total density obstructing half the aperture). Diffraction effects at small apertures would be asymmetrical, but you would not notice this unless you were stopped down to, say, f/8 or slower, and you imaged bright point light sources (the familiar "stars" you see radiating from street lights in night photography, for example, would then be asymmetric).
Now, if the density were NOT complete--only partial, the effect would still be proportionately visible in the out-of-focus highlights if shot at fast apertures (or in such a way as to reveal the circles of confusion). But at, say, 3 stops ND, you would still find it very difficult to see any effect on the image in the plane of focus.
Okay, so now we roughly know what such a device would do to the image as it is perceived. What does any of this actually get you? What is this talk about dynamic range?
The idea is that, in the plane of focus, the incident light on the sensor is coming from two halves (as delineated by this hypothetical ND filter). The light "cone" comes to a (reasonably) sharp point on the image plane, but half of this cone is unimpeded, whereas the other half has its luminous intensity diminished by the ND filter, and so is darker. This dual nature (again, in the plane of focus only!) lets the designer of the system theoretically use a sensor for which two neighboring (sub-)pixels are oriented in such a way as to capture these two halves separately. This is essentially a more sophisticated version of the dual ISO Magic Lantern hack, conducted at an optical level; alternatively, it is a variant of the light field imaging idea but with the incoming image split into only two components (light and dark). The result is the recording of a wider exposure range after appropriate signal processing.
The final step of the idea is that if we already have this Dual Pixel AF technology, then the sensor is already essentially adapted to the use of this modified optical construction. But as we have already seen, this idea is not without its drawbacks: it will totally mess up the appearance of out-of-focus highlights.
