From the paper [1]:
Each round of attack consists of three steps.
In the first step, the cache is primed, i.e.,
the attacker completely fills some of the
cache sets with its own data. The attacker
then waits some time to allow the
victim to execute. Finally, the attacker
probes the cache by measuring the time
it takes to access the previously-cached
data ...
From a brief glance, it looks like they then use the timing to feed into a deep neural network classifier.[1] https://arxiv.org/abs/1811.07153 (thanks leni536)