Bending loss decrease with larger bending?



I have a problem with using the bend analysis model. My customized file is attached to this discussion. My primary concern is that I am getting a decrease of bending loss with more bending, i.e., smaller the bending radius, which is obviously false.


In fact, these values are both smaller than that of without bending,

When I tried with the original example waveguide_bend.lms, I find the correct correlation, i.e. bending loss increases with more bending. Also, the power coupling does seem to decrease in the right direction but I am really puzzled by the loss here.

I read posts regarding boundary conditions and mesh cell size, thickness etc. but this does not seem to avert the false trend. Could you give me an insight on this? (The second file is the original example from Lumerical)

wavguide_bend_SiN_d.lms (248.6 KB)
wavguide_bend.lms (242.3 KB)

Thank you!


Dear @ymkim

The bending loss you are calculating is in the order of 10^-9. This value is very small because you have a very large bend radius (~100um). For these bend radius (and generally for bend radius above 5-10um), you can assume that your bending loss is zero. Thus, what you are calculating, within numerical errors, is zero for both cases (bend radius of 80 and 100um) and I will not worry about the exact value.



Thank you for your feedback!


One more question, regarding this topic.

When I simulate the bend radius less than 70 um, I no longer get TM mode.
With further reduction the TE mode also goes away, i.e. no physical mode.
I compared with symmetric structure where the width and height are both
230nm, then I get both modes until much smaller bend radius as expected.
Hence, I am assuming that not seeing the TM mode for asymmetric structure
with bend radius less than 70 um is a plausible result. The question I have
is that I don’t see any significant increase of loss towards 70 um radius.
It remains a very small loss but the mode disappears.I just want to double
check with you whether this can still be an arbitrary result of simulation.

Thank you for your help.


Hi @ymkim

When you are decreasing the bend radius, it becomes a bit challenging for the software to find the proper mode specially if the index value is set off. A good technique to help software to calculate the modes is to start with a very big bend radius, and slowly decrease it while manually updating the value for n as is shown in the screenshot below:

Please note that you should be able to see the modes for any bend radius even if they are very lossy.
This link might be a good one to read:

Waveguide Bend loss - Choosing the simulation size, mesh, BC's

I hope this was helpful



Dear Behzad

Thank you once again for your pin-pointed advice. This solves my inquiry.



Hi Behzad

Actually, I came across another parameter which seems to influence the
When you look under FDE parameters > material tab, there is a starting
I am simulating at 493nm which I put into the eigensolver currently.
However, depending on what range I put into FDE parameters, for an example
instead of starting at 1.5 um put, 0.3um, this drastically changes the loss
without bending.
Could you comment on this please? Should I set the range of my desired
wavelength into FDE as well as in the eigensolver window?
Thank you for your feedback


Here is my file attached once again.

wavguide_bend_SiN_final.lms (249 KB)


Dear @ymkim

Software uses a multi-coefficient model to fit the experimental material data i.e. real and imaginary part of the indexes. The materials that exist in the software come with a built in fit parameters that are optimized for broadband simulations, however, you can choose a new fit parameters for the existing material or the new sampled data. In this case, software will create a copy of the material (to make sure that it is not confused with the builtin material).

The Material tab under FDE does the same as Material Explorer however, I recommend using Material Explorer so that you can see the actual fit. Now lets look what happens to the material fit when you change the frequency range:

You can see that while the imaginary part is zero, your fit data, which will be used in simulations, introduce some material loss. This fit will change if you change the frequency range, and I expect that this is the reason why you are getting different value based on the material fit settings.

If you want to avoid this, it is a good idea to check the material fit all the time and make sure that you have a good fit in the frequency range of interest. If you are not getting a good fit in the range of interest, sometime its a good idea to break a broadband simulations into a few narrowband simulations. For more information regarding material fit please visit these pages:



hi bkhanaliloo,
I have small query regarding bend loss similar to above .
my query is if we fitted material and calculated sweep effective index from some wavelength band.
and next we remove the fitting and making narrowband simulation to calculate mode loss (several times individually).
why I am targeting like this for exact effective index match and dispersion calculation fit is necessary?
with the same fitting we expect more loss. if we remove fitting we get less loss reasonable.
is this kind of procedure is ok.


Hi @Nanda

Sorry for my late reply. Can you please clarify your question more?



Dear @bkhanaliloo if I have straight wave guide of width 400 nm and height 200 nm and its length is 2cm. Can you tell me about its loss? Is it also negligible,although its length is so large?


Dear @LS1519201

It will depend on the material loss and loss associated with the mode of interest. Once you simulate the modes in FDE solver, it will give you the mode loss (in units of dB/um or dB/cm) which includes these two loss factors.


Dear @bkhanaliloo On the page of bent wave guide

It is written that "The propagation loss of this mode in this case is about 0.0012 dB/um ."
but in formula
propagation loss used is 0.009 dB/um intead of 0.0012 dB/um. What is the reason?


Dear @LS1519201

That seems to be a typo and should be 0.009 dB/um as is shown in the screenshot. Also the loss calculation should be modified to be:

Loss= -2*10*log10(overlap)+LossPerBend(i);


LossPerBend = Loss[dB_m/um] * 2*pi*R[um]/4;

and we have overlap = 0.988



Dear @bkhanaliloo, Thanks for your reply. Your answer is loss is 0.009 dB/um. Reference to page"Waveguide Bend loss - Choosing the simulation size, mesh, BC’s"
aya_zaki calculate the loss 8.57 dB/cm using the same lms file of example. when we convert this loss to dB/um it should be 8.57/10,000 dB/um which is 0.000857 instead of 0.009.both these value are very different with each other.what is reason?


Dear @LS1519201

It looks like I made a mistake too when I converted loss from dB/cm to dB/um. My apologies.

You and aya_zaki are rigth, it should be 0.000857 ~ 0.0009. This will affect the final results as well, and you can use the equation that I provided in my previous post to calculate total loss.



Dear @bkhanaliloo, Thanks for your reply.
Sir I am still confused about the calculation of loss,When I find the loss using FDTD solutions with same parameters as in mode solution bend wave guide example,then loss value come 0.0089 dB/um instead of 0.000857 dB/um as calculated by aya_zaki in mode solution.Why FDTD solution value is different from mode solution value.I want to tell you that how I calculate the loss in FDTD soln. I used this formula Loss=[10Log (output (T))]/piR. I attached my fsp file,please tell me where is issue to find the loss ,because I want to find the loss using FDTD solutions. [bend_waveguide.fsp (398.2 KB)
I used this method which is similar to aya_zaki post
waiting for your reply,
Best regards


Dear @LS1519201

You will need to do convergence testing in both FDTD and Mode solutions to make sure that your results converge, and if there is disagreement, I will trust FDTD results more.

Regarding your question, the loss that you are seeing is the total loss that includes loss coming from overlaps and radiation loss. This means that if you want to obtain radiation loss, you will need to subtract mode mismatch between the straight and bend waveguide (overlap) from your calculations.



Dear @bkhanaliloo
sir can you please guide me that how can I do convergence testing in FDTD solutions.