The best general method is ion beam analysis (Rutherford backscattering, RBS,  and related techniques) since there is no sample preparation and analysis times are typically a few minutes;  the results are absolute - that is,  they do not require sample-matched standards or other assumptions about the sample; and problems like roughness and other sample problems (contamination etc) are also determined by the method.

Vincent Dolique mentioned that RBS accuracy is near 1%:  this has been long claimed but only very recently been critically demonstrated by us:  see Colaux & Jeynes and literature cited therein (High accuracy traceable Rutherford backscattering spectrometry of ion implanted samples, Analytical Methods 6 (2014) 120-129).

In this case you would use "Total-IBA" techniques (see Jeynes et al Nuclear Instruments and Methods in Physics Research B 271 (2012) 107–118);  that is,  use PIXE (particle-induced X-ray emission:  comparable to XRF) as well to get extra sensitivity to the Al signal (very poor signal-noise with a Ge substrate in RBS) which would be determined by RBS mainly through the energy loss.  PIXE/RBS takes the same time as RBS - you just need to add an X-ray detector!

Alternatively you could use HI-ERD (heavy ion elastic recoil detection:  this is like RBS but looks at the recoils rather than the scattered particle) with a heavy beam that will recoil Al.  Many labs now offer HI-ERD (using time-of-flight spectrometers).

Comments on the other techniques mentioned:

TEM is claimed to be accurate,  but for thin films there is a problem interpreting the contrast very precisely and the technique is much less accurate than expected.  See the critical NPL review on the determination of native oxide thicknesses (Seah et al Surface & Interface Analysis 36, 2004, 1269-1303).  Of course,  TEM is very time consuming and only looks at a very small region.

Ellipsometry on the other hand is fantastically precise (see Seah et al,  2004, op cit) but not necessarily accurate since interpretation depends on the model.  Spectroscopic ellipsometry is a wonderfully rich technique,  but the interpretation can be difficult (I like Barradas & Keddie on this:  Phys Rev E, 59, 1999, 1269-1303 and Solid State 43, 1999, 1095-1099).

Profilometry is very direct and sensitive - I used to measure films

A simple method is use a mask (can be as simple as a shadow metal clip or a piece of tape) and then run a profilometer over the step height. This is good for > 150 nm depending on the precision of the profilometer. This can also be verified with an in-situ crystal monitor within the deposition chamber if present.

For choosing the best method it all really depends on basic characteristics of the film. For instance if the film is very thin, if it has a significant amount of roughness, and the atomic contrast between film and substrate….

So the “best ”method can vary from a profilometer for thick films, X-ray reflectometry for very thin films, elipsometry, or Rutherford back scattering (RBS) TEM etc…. The difficulty to prepare good samples for cross section TEM may also be an issue for some films.

So, to make e long story short, the selection of the best method depends on the specific details of the sample you wish to measure. It also depends on the knowledge you have on the measurement technique (calibrations may be also an issue) and the physical parameters of the film to be measured.

There is no ultimate best method that is good for everything! 

Spectroscopic ellipsometry is one of the best techniques for this, it can give you film thickness upto angstrom order precision. However, if you don't need that much accuracy, you may go for cross sectional SEM

I would suggest using spectroscopic ellipsometry. You gain a lot of information from this technique, such as thickness, refractive index and extinction coefficient of your thin film.

Ellipsometry is the best choice,but if you have step better to afm or profilometry ,

cross-section tem or sem can tell thickness but with not too much accuracy

The choice of technique would actually depend on how thick your Al film is.

Ellipsometer works great only for thicknesses that allows reflection of incident polarized light from both interfaces of Al film (Al/air and Al/Ge interfaces). Profilometry/AFM are easy to use if there exists a step kinda structure.

Another way would be to etch a small feature within Al film and compare optical interference pattern on unetched section with the etched section.

Electron microscopy techniques can always be used if nothing else works.

It is important to know the error which is accetable for you on the thick (1,2 or 10%?) and the time to do it.

TEM is an accurate method but with a very long time.

RBS is interesting , the error is close to 1%.

A mask and a profilometer is easy to do it but it is not a very accurate method.

You can use also a mask and SEM analysis.

The best general method is ion beam analysis (Rutherford backscattering, RBS,  and related techniques) since there is no sample preparation and analysis times are typically a few minutes;  the results are absolute - that is,  they do not require sample-matched standards or other assumptions about the sample; and problems like roughness and other sample problems (contamination etc) are also determined by the method.

Vincent Dolique mentioned that RBS accuracy is near 1%:  this has been long claimed but only very recently been critically demonstrated by us:  see Colaux & Jeynes and literature cited therein (High accuracy traceable Rutherford backscattering spectrometry of ion implanted samples, Analytical Methods 6 (2014) 120-129).

In this case you would use "Total-IBA" techniques (see Jeynes et al Nuclear Instruments and Methods in Physics Research B 271 (2012) 107–118);  that is,  use PIXE (particle-induced X-ray emission:  comparable to XRF) as well to get extra sensitivity to the Al signal (very poor signal-noise with a Ge substrate in RBS) which would be determined by RBS mainly through the energy loss.  PIXE/RBS takes the same time as RBS - you just need to add an X-ray detector!

Alternatively you could use HI-ERD (heavy ion elastic recoil detection:  this is like RBS but looks at the recoils rather than the scattered particle) with a heavy beam that will recoil Al.  Many labs now offer HI-ERD (using time-of-flight spectrometers).

Comments on the other techniques mentioned:

TEM is claimed to be accurate,  but for thin films there is a problem interpreting the contrast very precisely and the technique is much less accurate than expected.  See the critical NPL review on the determination of native oxide thicknesses (Seah et al Surface & Interface Analysis 36, 2004, 1269-1303).  Of course,  TEM is very time consuming and only looks at a very small region.

Ellipsometry on the other hand is fantastically precise (see Seah et al,  2004, op cit) but not necessarily accurate since interpretation depends on the model.  Spectroscopic ellipsometry is a wonderfully rich technique,  but the interpretation can be difficult (I like Barradas & Keddie on this:  Phys Rev E, 59, 1999, 1269-1303 and Solid State 43, 1999, 1095-1099).

Profilometry is very direct and sensitive - I used to measure films

The ellipsometric techinque on an aluminum film is not really suitable because light can easily be reflected by the substrate. Moreover, depending on the layer roughness (in the case of aluminum strongly depends on the deposition temperature), the ellipsometric analysis can results in a very unprecise measurements. 

Very thin metal film can be better seen with direct technique such as RBS or XPS. TEM analysis can be an interesting solution if the sample preparation is carefully done (lamella done using low energy focuse ion beam for example).

Cross sectional SEM works if the film is thick

Do afm by doing a scratch on the film (or deposit Al layer on half of Ge substrate covering the other half do AFM then)

For large area thin films, XRF and electrical four point probe for metal films, and ellipsometry  for dielectric films are good and non destructive popular techniques. 

You can Measure it by: SEM, TEM, UV visible, Ellipsometry

I think the best method is SEM

For thickness

I think the best method to measure the thickness of thin film is by SEM image Provided that the measurement is by cross-section

I think the best method will be to cut the sample with diamond cutter or wire edm and do the cross section in FESEM.

http://www.ecti-thailand.org/assets/papers/411_pub_24.pdf

The best method is SEM, FESEM but FESEM is good for thin films or membranes and calculated by Microstructure measurment.