Dear Friends and Prof,
I have confused with MSE and mean deviation in Ellipsometer data fitting or modeling. is it MSE and mean deviation is same? what is the maximum values of those, for an acceptable model?
I use Nano-view ellipsometry instrument. I appreciate your idea and Thanks in advance.
http://research.ihost.com/qprc_2009/papers/QPRC_Invited_Session_5/Pintar_QPRC_09.pdf
Depends upon what system you are trying to model. If it is a transparent thin film, you can have MSE as low as 2 or 1 but if you are having a complex nanostructure then you might have to settle down with more, may be more than 10 or 15. In principle, there is no target value. The developed optical model should simulate real world structure and be physically acceptable that is follow Kramer Kronig consistency. It all depends upon how you collect raw data.
For thin films on silicon, measurement at 70 degree angle of incidence will suffice but if there is surface roughness, inhomogeneity in film thickness, then depolarization needs to be taken into account.
For transparent substrate, you have to consider backside reflection. Then for anisotropic samples, you will need generalized ellipsometry. You can also do multiple angle measurements to collect more information from the sample if its a complex one so lot of options are there and collection of proper raw data is very crucial.
Visit the link for more info: https://www.jawoollam.com/resources/ellipsometry-faq
Impossible to answer without knowing details of your system.
basically MSE = minimum distance between experiment and model.
MSE can also be scaled with experimental error (biased estimator)
Without knowing which unit you use to express these data we cannot give absolute number to what you call an acceptable model.
It is true the value will depend on the complexity of your sample, factors of 10 can be found between acceptable models depending if you talk of a bare piece of glass or silicon or a multilayered mirror coating.
Try to reach MSE in the same range as what you would get on a 100-1000nm oxide on silicon.
I'm not familiar with customary names used in ellipsometry, maybe they are used differently than in other areas. The definitions known to me are:
MSE = \sqrt{ < (xi - yi)2 > }
and mean deviation = < |xi - yi| >, where stands for average value (just an arithmetic mean), while xi are measured values, and yi are corresponding model values. Try the case xi = yi + C (C is any fixed number not equal to zero as C=0 is trivial) and you will see that the two measures may be equal to each other. But usually MSE will have smaller value than mean distance. Perhaps this is one of the reasons why it is often preferred in laboratory practice.
Thank you Firoz Ahmad, Mukul Bhatnagar, Michel Stchakovsky and Marek Wojciech Gutowski for your effort
@ Michel Stchakovsky, you were mentioned "Try to reach MSE in the same range as what you would get on a 100-1000nm oxide on silicon". it is worth to consider sio2 layer as a error reference --> awesome idea.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling