Edit: At second glance, it seems liks they are actually using Cesium now (some other demonstrators use Rubidium). The principle is the same, but the hyperfine splitting is 9.192631770 GHz (by definition) and the laser wavelength is 894 nm.
I thought Rubidium fountains were easier to minaturise. I can't see where I can find out how the device is made - there's not much detail.
Ah - you're right, it's Caesium. https://ww1.microchip.com/downloads/en/DeviceDoc/A3405FE2-C1...
I used to run an NTP server, in the pool. But it was stratum 2; it would stroke my ego to run a stratum 1 server. But 3.5 kilodollars: I'll wait.
Fountains need to have a minimum size in order to achieve a sufficient flight time when throwing the atoms up on a ballistic trajectory. Obviously, they also have to be very carefully aligned with respect to gravity.