I have found that the correlation of parameters is a drawback of ellipsometry models.  If possible, try to characterize Nb2O3 layer first.  This can be accomplished by either depositing a simple film or covering part of the substrate during the deposition of SiO2.  Then you can model the second layer( SiO2.)  You may also have an interface between the two layers, which of course can be modeled, but introduces more parameters as well. 

In this case I would measure the film thicknesses with X-ray reflectometry (XRR) and use those thicknesses to calibrate the ellipsometry model. 

Have you try a profilometer to measure thickness?

Dear Harry, Unfortunately we don't have a XRR in our lab and problem is not calibration.

Dear Beatriz, Profilometer can only measure 1 layer unless manufacturing process changes so we can have step profile where each step is indicative of one layer which is impossible or very time consuming in our case

I agree with Gisia, that for easier processing try to evaluate first the NbO layer and then multilayer. The choice of model is very important step and it have to take into account many parameters. My colleagues developed the PJDOS model for SiO2 and many other oxides. Look to the publications below.

P.S: Glass is not good substrate for ellipsometry, especially is it has high non-uniformity. So model for such thin films on glass would be tough! Analyze glass at first.

Article Ellipsometric characterisation of thin films non-uniform in thickness

Article Models of dielectric response in disordered solids

Article Application of sum rule to the dispersion model of hydrogena...

Also, you are probably doing ellipsometry in reflection (almost always).  In that case, is the back of the glass polished or rough? Reflection from back surfaces should be included in  the model.  Or, the back surface can be roughened (one could try other methods as well) so that the reflection from back surface is not collected in the detector.

Too bad that you don't have XRR available because I agree with Harry that this might be a better technique for you. Is there another lab with this capability that you could use so that you can compare results. Have you performed cross section TEM? Eitherway, if you are going to use ellipsometry it is just fitting a model to your data so it is never going to be perfect even with a high GOF. While high GOF is important I would just focus on using a model that makes physical sense.

Dear Sir,

In general we have three golden rules to validate the ellipsometry results: first is to use the widest optical wavelength range possible in the regression process. Second is to avoid parameter correlations in the optical modeling (in most cases using multi-angle evaluation helps to find unique results, as well as minimizing the number of fit parameters). Third rule is to maximize the goodness of fit in terms of R2 which helps to differentiate in between competing optical models.

In case you compare results obtained with optical dispersion fixed (literature refractive index used in the evaluation) with those models where the index was floating (optical functions considered as regression parameters), it seems reasonable that you receive better fit quality if the material composition was also included in the fit process. The reason is that you might not have exactly the same material in reality than the material used anno for the literature reference measurement, but your ellipsometer should be able to follow the material preparation variations in terms of refractive index / extinction. Finally it is important to check the physical feasibility of the resulted dispersion, whether it is reasonable for the given material (ie. comparing the index to literature reference data and evaluating the difference).

I have contacted you in private to check the actual measurements and results together.

Dear Gisia Beydaghyan, Yes with glass substrates I always scratch the back side. I also attached a black barrier which increased the GOF a little.

Dear Richard Padbury, I'm trying to make a setup that we can constantly do measurements for quality control on our multi-layered samples, and since we already have a Semilab ellipsometer I'm trying to utilize that now.

Dear Peter Basa, Thanks for your reply and your email. I got back to you in email.

I would contact J.A. Woolam, they are very helpful in answering questions about ellipsometry.

For transparent substrate it is also useful to use some kind of knife edge illumination but this is only available for imaging ellipsometry

https://www.youtube.com/watch?v=OxF901UhCNw

Best regards

Christian

Another technique we have used successfully is to deposit the film to be characterized on 2 different substrate types simultaneously, then look to have both models converge to a comparable solution.  For example, we use 1 Si wafer and another with a thick SiO2 layer (pre-characterized).  The "correct" model should be consistent for both substrate measurements.

Other options: 1. SEM cross-section is a nice technique to confirm thickness as well, if the coating is sufficiently thick.  2. Step-height measurement in a white light interferometer (be careful to match the phase on reflection for both sides of the step) or stylus profilometer (probably too thin on your sample).

Good luck.

Confirm your film thickness through RBS Technique.

Hi Amirhessam,

Your glass substrate is transparent. Therefore, in order to characterize thin films on transparent substrate using Spectroscopic Ellipsometry, then you have to suppress the substrate backside reflections as these reflections are "incoherent" with the reflections coming from the top surface thin film and the front side of the substrate. To do this, you can eliminate the substrate backside reflections during analysis. However, it is recommended that you roughen the substrate backside before SE measurement, or use a scotch tap to cover the substrate backside in order to physical eliminate the substrate backside reflections.

Here is a good reference 

http://www.jawjapan.com/toolbox/Newletters/General/transparent_substrates.pdf

In addition, for a "Transparent" Film on a "Transparent" Substrate, see the following reference paper:

https://www.researchgate.net/profile/Andriy_Kostruba/publication/280242532_Method_for_determination_of_the_parameters_of_transparent_ultrathin_films_deposited_on_transparent_substrates_under_conditions_of_low_optical_contrast/links/5600051d08aec948c4fa017b/Method-for-determination-of-the-parameters-of-transparent-ultrathin-films-deposited-on-transparent-substrates-under-conditions-of-low-optical-contrast.pdf?origin=publication_detail&ev=pub_int_prw_xdl&msrp=4OsZFOLWwuD8chbawx2NY4dAm5oRasaSjI4aGmOAMmP5bwzopWOkkKobzOX03lcHS0gOpLgnkBDWSBGGYDOKQNORTFA3QFeOpg05oSn8rzX4aonHsqahXoBg.sfvSe1WkaBA6VCdqI7UFe1KPxcoTsJRU8iw0vmA-4YXVJoC1GuaKT_dS5vn27m98R66cT0DmR3reVdcnmTiuH8qkDmF7tTCR5myL1w.jNMDAiLJTB4XBJ8VZDA3eK4kP6UPugNwStV9XRfO3T87fdmAlVuBiwI8jUIJHEdg1zhEyJWE6NA1l5ks7akfUQUB0ycWEPJNjWPGRg.ivxMdYWzPXTA5e4slgaO9Gsdoo7IUVcweuQ6Dxvs0_8aKioK45LvYU-27v9IV036qvHTqXARJltcRKcBvm03Q594AjeKTmkTeiDYMA

Mostly likely, the transparent substrate backside reflections are the source of error in your data.

Good luck

Article Method for determination of the parameters of transparent ul...

If you suppress the glass substrate backside reflections as outlined above and remeasure the spectra  again after treating the substrate backside, then by using the database data for (n and k) for the two sub-layeres, you will be able to get the two sub-layer thicknesses perfectly. At this case you have two inputs of Psi and Delta for the multilayered structure and thus they will be enough to obtain the two output thicknesses d(SiO2) and d(Nb2O3).