From my experience quartz crystal microbalances can only be used as an indication of the rate and thickness during vacuum coating. In principal they can be accurate but in reality this is confounded by factors relating to the deposition conditions. These include.
Depositing more than one material, their different acoustic impedances can introduce an error.
Depositing a highly stressed material (e.g. Cr). The stress will cause the frequency to drift and introduce an error.
The source temperature and the distance from the source. Radiative heat transfer can introduce frequency drift and a thickness error (this should 'cancel out' on cooling, but some monitors don't register this). Crystal cooling is very important.
Charged particle impact during sputter coating. A special crystal holder with an internal magnet should be used.
Stable location. You must fix your crystal holder in a stable position looking directly at the source. If it wobbles around or gets knocked out of alignment you will get an error. Remember some deposition sources are quite directional.
The quartz crystal and substrate are not in exactly the same position so they aren't receiving the same flux from the source.
Position on your substrate. Don't forget if you have a large substrate the thickness deposited across it could vary.
You should never trust the thickness/rate value given by a quartz crystal microbalance without calibration using an independent film thickness measurement such as a stylus profilometer. They can be wildly out. The comparison of the observed thickness and the measured thickness then gives you the tooling factor to calibrate the instrument. The users manual should explain this. You then need to do regular checks and keep a log of the calibration results to check things don't drift. The accuracy of the system will then be how repeatable the results are. Don't forget the stylus profilometer (or other technique) will have its own experimental error. So unfortunately this can be quite a lot of work.
This sort of thing should be covered in any good book on vacuum coating. Hope this helps.
We use QMB's to calibrate deposition rates in the range of sub-monolayers to a few monolayers. That involves frequency shifts of just a few to a few tens of Hz. Our QMBs are installed such that they can easily be positioned (quite precisely) at the substrate position. Evidently therefore, deposition on the QMB is not measured while the actual sample is being prepared. The Evaporator must run stably therefore, for getting consistent results.
Our observation is that we get pretty good repeatability in this way, while absolute thickness values may be off by up to 30% or so. We regularly re-calibrate our QMB readings, usually at the beginning of each new series of experiments. (Not only does this help us to stay accurate but it is also a very good exercise for the students.)
For calibration, we try using systems and methods which allow us determining the coverage to the best possible extent. Mostly these are in situ roughness measurements (useful for layer-by-layer growth) be He scattering (could also do electron scattering, e.g. MEED or RHEED, depending on availability). Alternatively, we use LEED patterns, when we know the structural phase diagram as a function of coverage.
And yes, as David indicates, this may turn out to be pretty laborious, but it pays off. Otherwise, I find David's hints excellent.
With a crystal thickness monitor you are measuring the shift in resonant frequency with the increase of mass on the surface of the crystal. High density materials give a large frequency shift for a small thickness and you will get a pretty good answer. Low density materials give a small frequency shift and give poorer results. In addition, the shift depends on the relative acoustic impedance between the materials deposited and the quartz. Acoustic impedance will depend not only on the material but the microstructure of the film deposited.
1) Thickness monitors give better results if only one material has been deposited on the quartz surface. Different materials in layers require a much more complicated fit to determine thickness accurately and will have a large error bar.
2) Geometry matters. The relative position of the monitor and substrate introduce a geometrical calibration term because the flux of material is not uniform. If you always deposit the same material with exactly the same growth parameters then this can be calibrated for.
3) Temperature. If the crystal temperature changes so does the resonant frequency and that looks exactly like deposition or sublimation. If the crystal cools back to its original temperature after growth you can get a sense of how big this is by letting the monitor run as your system cools.
4) Sticking coefficient. If your substrate is heated and your monitor is cooled your material will in general have a different sticking coefficient. So, while you measure one deposition rate at the monitor the amount of material that sticks to your substrate will be different. For very thin films this also depends on substrate.
5)Crystal age. There are a set of resonances at different frequencies. As the amount of material on a crystal gets large you can reach a point where the circuit that drives the resonance hops between resonances. This will introduce a large error. Even before you get there you can get thickness effects because the fit that is used to determine your material from frequency shift assumes that the material deposited is a small perturbation. This breaks down as the amount of material increases.
Best case scenario, assuming a brand new crystal, a material with a density much greater than quartze, temperatures controlled well, only one material ever deposited in the system, growth conditions always the same and crystal monitor calibrated with a careful post deposition technique, will be about 5% error. Clearly reality is much worse but 10% is clearly doable. I have seen ridiculously compulsive people get to 2%.