Rubidium cell atomic clocks, on the other hand have a vapor cell which is a closed tube, and doesn't get expended. Thus Rubidium clocks have far longer service lives. You can find them surplus on Ebay for less than $200. The stability is slightly less, but still far more accurate than most electronics technicians need to do adjustments and calibrations.

[Update - Far more than you ever thought you wanted to know about atomic clocks] The main difference between an atomic clock and a quartz crystal oscillator is that in a crystal oscillator, you have a continuous signal that is generated from the circuitry and passed through the crystal, and amplified in a feedback loop. A crystal oscillator can be made with a single transistor and a few passive components.

In an atomic clock, there is a stabilized quartz oscillator, which actually keeps time, but it is then steered up or down in frequency slightly by using a harmonic of that frequency, along with a very slight amount of frequency modulation, to inject into a cavity containing the atom of interest.

In Cesium (Caesium outside the US) beam clocks, the physics package is all packed inside a large vacuum tube. Inside the tube is an oven with microscopic holes in the top containing cesium metal, the oven is heated to about 200 degrees. A small stream of individual atoms exit. Those atoms pass through a magnetic field which diverts them according to spin. The atoms then drift through a microwave chamber where there is a 9.129 Ghz signal imparted. If the frequency of the signal exactly matches the natural resonance frequency of the cesium atom, its spin will flip.

On the other side of the drift chamber, a second magnet selects only those atoms which have flipped state, all other atoms are diverted and trapped. Then the atoms encounter a hot wire with high voltage where they are ionized, them passed through another magnetic field which is a mass spectrometer, this serves to filter out impurities and contaminant that are present in the tube. There is a fine slit that allows entrance only to the cesium beams (now regardless of their spin), and they imping upon a plate in a photomultiplier.

The net result of all this physics is that if you have exactly the correct frequency, there is a DC current of a microampere or so output, too high or too low, and the output drops quickly. The line width is on the order of a few hertz.

To sample against this, the quartz oscillator is multiplied in frequency the appropriate amount using multipliers and phase locked loops, and combined with a very small amount of frequency modulation at 137 hertz.

The DC output of the tube varies with the modulation... if it increases with increasing frequency, the reference clock is too slow... if it is out of phase, the reference clock is too fast, and if you get a 274 hz second harmonic, you are right on frequency.

They take about 10-20 minutes to warm up and lock.

There are some adjustments, and it is possible to have them locked on the wrong frequency if you aren't careful.

Cesium clocks, used, are on the order of $5000

--

Rubidium clocks are easier to use... they are also quartz oscillators probing with FM signals, but instead of a stream of atoms, there is a lamp with one isotope of rubidium which is then passed through a chamber with microwaves and the other isotope of rubidium... if the frequency is just right, the light transmission dips by 1%, and this is used to lock the oscillator.

Rubidium standards are about 10 times less stable, but have far longer service lives, and can be had used for about $200.