I used COMSOL model to calculated Brownian noise in substrate. This was done for cross checking my model simulation. The result from model is within 2% compared to half infinite model calculation.
I followed Levin's Direct approach to calculate Brownian noise in substrate, basically, to calculate the elastic energy inside the substrate under the applied test force. This can be done using COMSOL and analytical calculation. The comparison between the two is shown below.
U is the stored energy in substrate.
Note: I used the same COMSOL model for TE noise calculation. I just asked it to produce the strain energy in the substrate (no spacer).
The simulation is very close to the analytical result. So I think my spacer-cavity model and all the factors in the calculation are correct. The TE calculation is a little more complicated, since I have to calculate the gradient of expansion in COMSOL and it might be wrong somewhere. I'll check that.