From what I've tried to glean from patents and whitepapers, they have memory in the panel so that each sub-pixel can hold its state without being driven by the controller. The controller also can address sub-regions under the control of the application processor.

The net effect is that the app can vary refresh rate and also update a small part of the screen while leaving the rest static. In Garmin smartwatches, this is used to switch between 1 Hz, 1/60 Hz, and a hybrid mode where the overall screen is 1/60 Hz and small part is updated at 1 Hz. E.g. the seconds digits may remain active while the rest of the screen is static.

It also seems like they use more interesting sub-pixel arrangements. The limited color depth may involve different combinations of 1-bit sub-pixels rather than any cells driven into partially twisted states. I've seen diagrams claiming some white (unfiltered) and RGB filtered sub-pixels of different sizes.