Ну и зачем вы пишите мне эти глупости?

Их метод находит уникальными 2МБ области памяти. Никакой однородности ячеек в одной партии и в помине нет.

Затем им нужно хорошенько постараться чтобы задолбить ОС таким образом чтобы получить достаточно уникальных 2МБ областей со всего объема ОЗУ.

Из статьи:
Takeaway 3: Different bits have different probabilities of
flipping. The distribution of these probabilities is unique across
contiguous 2 MB chunks of memory and persistent within each
2 MB chunk

However, we cannot guarantee access to the same 2 MB
regions on a user’s device, since memory allocation is handled
by the OS. Thus, if we obtain two different 2 MB chunks of
memory on a user’s device during two different sessions and
compared their distributions, we would incorrectly conclude
that different devices were used during these sessions. One
way to overcome this challenge would be to hammer multiple
2 MB chunks during each session. For example, suppose the
user’s device has 1 GB of memory which corresponds to 512
different 2 MB chunks of memory. Taking inspiration from
the birthday paradox [1], if we were to to hammer 64 chunks
each in two different sessions with the user’s device, then the
probability that at least one chunk would overlap between them
is over 99.9%. We show how we derived this number from first
principles in Appendix A.