Yes—at least in principle. On many CMOS designs, readout/gain can be controlled in ways that allow different effective sensitivities across the sensor, but it requires sensor hardware and processing designed for it.

There is prior art: Magic Lantern’s “dual ISO” alternates ISO by sensor lines on some Canon cameras, and some Fujifilm and Sony sensor designs have used mixed-sensitivity or mixed-exposure approaches to extend dynamic range. These methods show that varying sensitivity within one frame is technically possible.

However, doing it as “top half ISO 100, bottom half ISO 400” has practical drawbacks:

• the transition between regions could be awkward or visibly abrupt
• the split would be fixed unless the camera offered a way to move/rotate/feather it
• added complexity may not be worth it when grad ND filters, bracketing, and HDR processing already solve the problem well

So the answer is yes, but it’s mostly a design and usability question rather than a fundamental technical barrier. In practice, camera makers have generally favored whole-sensor HDR methods, dual-gain/dual-native-ISO approaches, or multi-shot bracketing over a user-positioned split-ISO sensor.