ML dual_iso is as a result of reverse engineering. It was noticed that there were more then one amplifier (attached to the sensor) that could be adjusted separately. This allowed dual_iso, but does have some drawbacks.
Canon's solution is bespoke. It is likely that there are still more then one amplifier (column gains iirc?), however, instead of finding more DR by adjusting these amplifiers separately, with all of the drawbacks associated with, Canon doesn't flush the photosites when they are first read, and instead retains the photosite "memory" for more than one amplification pass.
ML....
2 lines of image signal at ISO xx
next 2 lines of image signal at ISO yy
next 2 lines of image signal at ISO xx
next 2 lines of image signal at ISO yy
next 2 lines of image signal at ISO xx
next 2 lines of image signal at ISO yy
etc...
etc...
Then a post application is used to match ISO yy brightness to ISO xx brightness.
Canon scans the entire image at ISO xx, then scans it again at ISO yy (and presumably could continue to scan it however often they want, to a point). Basically this is ML dual_iso with none of the drawbacks.
Oh and I assume all of the post processing is done in camera. Would be nice if they allow a raw of each scan to be saved.